Development of an optimal spatial decision-making system...

i

Development of an optimal spatial decision-making system using approximate reasoning

Submitted by

David Thomas Bailey

BEng(Hons) QUT

A thesis submitted in partial fulfilment

of the requirements of the degree of

DOCTOR OF PHILOSOPHY

Research Centre for Built Environment and Engineering Research

Energy and Resource Management Research Program

Faculty of Built Environment and Engineering

Queensland University of Technology

Novemer 2005

ii

TTAABBLLEE OOFF CCOONNTTEENNTTSS

TABLE OF CONTENTS II

LIST OF FIGURES VI

LIST OF TABLES VIII

LIST OF ABBREVIATIONS IX

ABSTRACT X

STATEMENT OF AUTHORSHIP XI

ACKNOWLEDGEMENTS XII

1 INTRODUCTION AND METHODOLOGY 1 1.1 INTRODUCTION 1 1.2 THE PROBLEM 2

1.2.1 Context and fundamental elements of infrastructure site selection 2 1.2.2 Background 3 1.2.3 Use of Approximate Reasoning 5 1.2.4 Case Study 6

1.3 RESEARCH OUTLINE 7 1.3.1 Research Question 7 1.3.2 Aims 8 1.3.3 Objectives 8 1.3.4 Scope 8 1.3.5 Justification 9 1.3.6 Methodology 10

1.4 THESIS STRUCTURE 13 1.5 MULTI-DISCIPLINARY NATURE OF THE RESEARCH 14

2 PROBLEM DIAGNOSIS AND PRELIMINARY LITERATURE REVIEW 17

2.1 INTRODUCTION 17 2.2 PROBLEM DIAGNOSIS 18 2.3 INTRODUCTION TO LOCATION PROBLEMS 19

2.3.1 Problem classifications 20 2.4 DECISION SCIENCE TECHNIQUES 23

2.4.1 Map algebra 24 2.4.2 Multicriteria evaluation 25

2.4.2.1 Specifying and standardising evaluation criteria 28 2.4.2.2 Criterion weighting 33 2.4.2.3 Alternatives and the decision matrix 34 2.4.2.4 Aggregation via decision rules 36 2.4.2.5 Sensitivity analysis 39 2.4.2.6 Limitations of MCE 39

iii

2.4.3 Artificial intelligence and soft computing 40 2.4.3.1 Fuzzy logic 41 2.4.3.2 Neural networks 41 2.4.3.3 Genetic algorithms (GA’s) 42

2.5 TECHNOLOGY PLATFORMS 44 2.6 DISCUSSION 44 2.7 CONCLUSIONS 46

3 APPROXIMATE REASONING 47 3.1 INTRODUCTION 47 3.2 FUZZY LOGIC 47

3.2.1 Fuzzy sets 48 3.2.2 Fuzzy numbers 50

3.3 APPROXIMATE REASONING IN MULTICRITERIA DECISION-MAKING 51 3.3.1 Fuzzy MCE 52 3.3.2 Fuzzy inference systems 55 3.3.3 Pairwise comparison methods 57

3.4 USE OF APPROXIMATE REASONING IN LOCATION PROBLEMS 57 3.5 CONCLUSIONS 59

4 SPATIAL DECISION SUPPORT SYSTEMS 61 4.1 INTRODUCTION 61 4.2 OVERVIEW 62

4.2.1 Decision support and expert systems 62 4.2.2 Basic concepts of spatial decision support systems 64

4.3 COMPONENTS OF A SDSS 66 4.3.1 Geographical Information Systems 66 4.3.2 Dialog 68 4.3.3 Data 69

4.3.3.1 Spatial data representation 70 4.3.3.2 Raster data and cell size 71 4.3.3.3 Non-spatial data (Attribute Data) 72

4.3.4 Models 73 4.4 DEVELOPMENT AND IMPLEMENTATION 75 4.5 CONCLUSIONS 76

5 PROBLEM ANALYSIS AND CONCEPTUAL SYSTEM DESIGN 79 5.1 INTRODUCTION 79 5.2 CAUSES OF CURRENT LIMITATIONS ON SDSSS FOR SITE SELECTION 79

5.2.1 Multiple decision-makers 81 5.2.2 Uncertainty 82 5.2.3 Simplicity 84 5.2.4 Control 85

5.3 A CONCEPTUAL FRAMEWORK 85 5.3.1 Why use Approximate Reasoning? 85 5.3.2 Catering for multiple decision-makers 87 5.3.3 Handling Uncertainty 89 5.3.4 Creating simplicity 91 5.3.5 Giving decision-makers control 92

5.4 CONCLUSIONS 93

iv

6 ALGORITHM DESIGN 95 6.1 INTRODUCTION 95 6.2 ARAISS 95

6.2.1 Framework 96 6.2.2 Notation 98 6.2.3 Linguistic term sets 98

6.2.3.1 Semantic definition 101 6.2.3.2 Term generation 102 6.2.3.3 Uncertainty scaling 103

6.2.4 Dynamic weighting 105 6.2.5 Generating suitability values 106 6.2.6 Aggregation and output parameters 108

6.2.6.1 Utility 109 6.2.6.2 Safety 111 6.2.6.3 Consensus 112 6.2.6.4 Certainty 113

6.2.7 Adjusted aggregation and alternative exploration 114 6.3 ARAISS SIMULATION EXERCISES 114

6.3.1 Validating ARAISS 115 6.3.2 Example simulation 116

6.3.2.1 Data inputs 117 6.3.2.2 Results 118 6.3.2.3 Interpretation 119

6.4 CONCLUSIONS 119

7 INFRAPLANNER 121 7.1 INTRODUCTION 121 7.2 OVERVIEW OF THE PROTOTYPE SYSTEM 122

7.2.1 Target application 122 7.2.2 Target audience 123 7.2.3 Dialog design 123 7.2.4 Database 125 7.2.5 Model 126

7.3 DEVELOPMENT PROCESS 127 7.3.1 Planning 128 7.3.2 Research 129 7.3.3 Analysis and design 129 7.3.4 Construction 130 7.3.5 Implementation 131

7.4 THE INFRAPLANNER PROTOTYPE 131 7.4.1 Project tools 134 7.4.2 Creating maps 135

7.4.2.1 Suitability Maps 136 7.4.2.2 Decision Maps 140

7.4.3 Exploring maps 142 7.5 VALIDATING INFRAPLANNER 143 7.6 DISCUSSION 143

8 A CASE STUDY USING INFRAPLANNER 145 8.1 INTRODUCTION 145

v

8.2 THE PROBLEM 146 8.3 PROCEDURE 148 8.4 RESULTS 163 8.5 DISCUSSION 171

9 CONCLUSIONS 175 9.1 INTRODUCTION 175 9.2 SUMMARY OF RESULTS 175

9.2.1 Answer to the Research Question 175 9.2.2 Achievement of the Research Aims 176 9.2.3 Achievement of the Research Objectives 176

9.3 RESEARCH OVERVIEW 177 9.3.1 Planning and research 178 9.3.2 Analysis and design 179 9.3.3 Construction 180 9.3.4 Implementation and feedback 181

9.4 VALIDATION 181 9.5 KEY FINDINGS 183 9.6 DIRECTIONS FOR FUTURE RESEARCH 184 9.7 CONCLUDING REMARKS 185

REFERENCES 187

A: PUBLICATIONS 199

B: THE BRISBANE AIRPORT ENVIRONMENT 203

C: MATLAB CODE 211

D: ARCOBJECTS VBA CODE 233

E: ANZIIS QUESTIONAIRE 345

vi

LLIISSTT OOFF FFIIGGUURREESS

FIGURE 1.1: BRISBANE AIRPORT................................................................7

FIGURE 1.2: DEVELOPMENT PROCESS LOGIC MODEL.....................12

FIGURE 1.3: DISCIPLINES CONTAINED IN EACH CHAPTER ............15

FIGURE 2.1: CLASSIFICATION OF MULTICRITERIA DECISION PROBLEMS .......................................................................................................22

FIGURE 2.2: FRAMEWORK FOR MCE ......................................................27

FIGURE 2.3: DECISION MATRIX ................................................................36

FIGURE 3.1: FUZZY MEMBERSHIP FUNCTION FOR THE TERM ‘APPROXIMATELY THREE’ ........................................................................49

FIGURE 3.2: TRAPEZOIDAL FUZZY NUMBER TPZ(A,B,α,β) ..............50

FIGURE 3.3: FUZZY INFERENCE SYSTEM ..............................................56

FIGURE 3.4: THE VARIABLE SLOPE AS A FUZZY MEASURE............58

FIGURE 4.1: ESRI ARCGIS............................................................................68

FIGURE 5.1: SOURCES OF CURRENT LIMITATIONS ON SDSSS .......81

FIGURE 5.2: RELEVANCE MATRIX ...........................................................87

FIGURE 5.3: THE SUITABILITY TERM ‘OK’ AS TFN(0.3,0.5,0.7) ........89

FIGURE 5.4: FOOTPRINT OF UNCERTAINTY.........................................90

FIGURE 5.5: HOW A QUANTITATIVE UNCERTAINTY ASSESSMENT AFFECTS THE PRIMARY MF ......................................................................91

FIGURE 6.1: ARAISS FRAMEWORK ..........................................................97

FIGURE 6.2: TERM GENERATION ...........................................................103

FIGURE 6.3: FUZZY OUTPUTS FROM EQUATION 6.15 ......................118

FIGURE 7.1: THE INFRAPLANNER TOOLBAR .....................................131

FIGURE 7.2: HOW INFRAPLANNER TOOLS FIT INTO THE DECISION-MAKING FRAMEWORK ........................................................133

FIGURE 7.3: SETTING PROJECT INFORMATION................................134

vii

FIGURE 7.4: CREATING A NEW TERM SET ..........................................135

FIGURE 7.5: THE DISCRETE CRITERION MAP USER FORM...........136

FIGURE 7.6: CREATING A DISCRETE CRITERION MAP...................137

FIGURE 7.7: THE CONTINUOUS CRITERION MAP USER FORM ....138

FIGURE 7.8: CREATING A CONTINUOUS CRITERION MAP ............139

FIGURE 7.9: THE DECISION MAPS USER FORM .................................140

FIGURE 7.10: CREATING DECISION MAPS ...........................................141

FIGURE 7.11: MAP EXPLORATION..........................................................142

FIGURE 8.1: BRISBANE AIRPORT LAYOUT..........................................147

FIGURE 8.2: UNCONSTRAINED ALTERNATIVES................................150

FIGURE 8.3: CREATING A CONTINUOUS SUITABILITY MAP FOR COMMUNITY IMPACT ................................................................................152

FIGURE 8.4: CREATING A DISCRETE SUITABILITY MAP FOR ZONING ...........................................................................................................152

FIGURE 8.5: BAC TRAFFIC IMPACT SUITABILITY MAP ..................156

FIGURE 8.6: BAC COMMUNITY IMPACT SUITABILITY MAP .........157

FIGURE 8.7: TRAFFIC IMPACT SUITABILITY MAP FOR THE RESIDENTIAL COMMUNITY.....................................................................158

FIGURE 8.8: COMMUNITY IMPACT SUITABILITY MAP FOR THE RESIDENTIAL COMMUNITY.....................................................................159

FIGURE 8.9: PERFORMING AN AGGREGATION .................................162

FIGURE 8.10: UTILITY.................................................................................164

FIGURE 8.11: UNCERTAINTY ....................................................................165

FIGURE 8.12: RISK ........................................................................................166

FIGURE 8.13: CONFLICT.............................................................................167

FIGURE 8.14: ADJUSTED AGGREGATION.............................................169

FIGURE 8.15: SITES OF INTEREST...........................................................170

FIGURE 8.16: ALTERNATIVE EXPLORATION......................................171

viii

LLIISSTT OOFF TTAABBLLEESS

TABLE 1.1: DISCIPLINES INVOLVED IN THE RESEARCH..................15

TABLE 2.1: MAP ALGEBRA OPERATORS ................................................25

TABLE 2.2: TYPES OF ATTRIBUTES, A BRIEF DESCRIPTION, AND DESCRIBING AUTHORS................................................................................30

TABLE 4.1: DIMENSIONS OF A DSS ...........................................................64

TABLE 4.2: GIS SPATIAL ENTITIES IN VECTOR AND RASTER ........71

TABLE 4.3: COUPLING METHODS.............................................................73

TABLE 6.1: SEMANTIC DEFINITION OF PRIMARY TERMS.............102

TABLE 6.2: DECISION-MAKER RELEVANCE .......................................117

TABLE 6.3: DECISION-MAKER 1 INPUTS...............................................117



TABLE 6.6: FINAL OUTPUTS IN LINGUISTIC FORM..........................119

TABLE 8.1: LINGUISTIC TERMS...............................................................148

TABLE 8.2: CRITERIA DEFINITION.........................................................151

TABLE 8.3: CRITERION WEIGHTING .....................................................160

TABLE 8.4: DECISION-MAKER RELEVANCE VALUES ......................161

ix

LLIISSTT OOFF AABBBBRREEVVIIAATTIIOONNSS

AR Approximate Reasoning

ARAISS Approximate Reasoning Algorithm for Infrastructure Site

Selection

BAC Brisbane Airport Corporation

DSS Decision Support System

GA Genetic Algorithm

GIS Geographical Information System

GMCLP Group Multicriteria Location Problem

MCE Multicriteria Evaluation

MADM Multiattribute Decision Making

MODM Multiobjective Decision Making

OR Operations Research

OWA Ordered Weighted Averaging

QUT Queensland University of Technology

SDSS Spatial Decision Support System

VBA Visual Basic for Applications

x

AABBSSTTRRAACCTT

There is a recognised need for the continued improvement of both the techniques

and technology for spatial decision support in infrastructure site selection. Many

authors have noted that current methodologies are inadequate for real-world site

selection decisions carried out by heterogeneous groups of decision-makers

under uncertainty. Nevertheless despite numerous limitations inherent in current

spatial problem solving methods, spatial decision support systems have been

proven to increase decision-maker effectiveness when used. However, due to the

real or perceived difficulty of using these systems few applications are actually in

use to support decision-makers in siting decisions. The most common difficulties

encountered involve standardising criterion ratings, and communicating results.

This research has focused on the use of Approximate Reasoning to improve the

techniques and technology of spatial decision support, and make them easier to

use and understand. The algorithm developed in this research (ARAISS) is based

on the use of natural language to describe problem variables such as suitability,

certainty, risk and consensus. The algorithm uses a method based on type II

fuzzy sets to represent problem variables. ARAISS was subsequently

incorporated into a new Spatial Decision Support System (InfraPlanner) and

validated by use in a real-world site selection problem at Australia’s Brisbane

Airport. Results indicate that Approximate Reasoning is a promising method for

spatial infrastructure planning decisions. Natural language inputs and outputs,

combined with an easily understandable multiple decision-maker framework

created an environment conducive to information sharing and consensus building

among parties. Future research should focus on the use of Genetic Algorithms

and other Artificial Intelligence techniques to broaden the scope of existing

work.

xi

SSTTAATTEEMMEENNTT OOFF AAUUTTHHOORRSSHHIIPP

Except where duly acknowledged in the text, this thesis contains neither:

a) Material which has been previously published under my name, or which has

been submitted as part of another degree or diploma.

b) Any other persons work, either published or unpublished.

This thesis is presented as an original contribution based on my doctoral research

at QUT, and has not been submitted elsewhere, under my name or that of any

other individual.

David Thomas Bailey

20 November 2005

xii

AACCKKNNOOWWLLEEDDGGEEMMEENNTTSS

My time at QUT has offered many opportunities to share and interact with

friends and colleagues on all levels. It should be no surprise that the work

contained in this thesis has grown from the foundation of help and support of

those around me.

Associate Professor Ashantha Goonetilleke, a most capable and tireless

individual with a passion for his role in academia and lively sense of humour, led

my supervisory team. What follows in these pages would not have materialised

without his guidance and faith in my ability.

Associate Professor Duncan Campbell joined me as an associate supervisor late

in my research but has made a contribution heavily out of proportion with the

length of time we have collaborated. His immediate enthusiasm, continuous

encouragement, and unquestionable academic prowess have been indispensable.

Dr John Hayes and Dr Mohamed Deriche made very early, central, and specific

contributions in guiding me with GIS and fuzzy logic respectively. Their input

seemed to both give direction and open up new possibilities at the same time, and

was highly appreciated.

Several other researchers and administrators at QUT have been fantastic sources

of support, advice, and also laughter. I certainly owe a sincere measure of

gratitude to Jenny, Shelley, Wael, Jack, Ramid, Lars, Steve and many others.

Finally I would like to confirm the axiom that writing a thesis places a large

stress on ones home life. To my partner Helga, whose love and support I could

not live without, my greatest thanks.

1Chapter 1 Introduction and Methodology

Chapter 1

IINNTTRROODDUUCCTTIIOONN AANNDD MMEETTHHOODDOOLLOOGGYY

1.1 Introduction

There is a recognised need for the continued improvement of both the techniques

and technology for spatial decision support in infrastructure site selection. In fact

while spatial decision support systems have been proven to increase decision-

maker effectiveness (Crossland, Wynne et al. 1995), few applications are

actually in use to support decision-makers in siting decisions (Maniezzo, Mendes

et al. 1998). This research focuses on the use of Approximate Reasoning to

improve the techniques and technology of spatial decision support. The processes

developed in this research were incorporated into a new Spatial Decision Support

System and validated by their use in a real world site selection problem at

Australia’s Brisbane Airport.

This thesis describes the development of a new Approximate Reasoning

Algorithm for Infrastructure Site Selection (ARAISS) and it’s implementation in

the ‘InfraPlanner’ prototype Spatial Decision Support System (SDSS). The

InfraPlanner SDSS allows a group of decision-makers to give natural language

assessments of multiple evaluation criteria, and receive natural language

feedback on site suitability. It is the product of an attempt to create a system to

cater for the challenging dynamics of infrastructure site selection whilst being

simple to use and easy to understand.


1.2 The Problem

Selecting strategically suitable locations for future infrastructure developments is

a fundamental activity performed by planners and managers of the environment

and infrastructure. These site selection decisions usually have a major impact on

the success of the development, however this is often the least scientific part of

the planning process. The site selection process is, by necessity, ill-structured,

requiring submissions from multiple stakeholders comprising government,

business and community groups, with disagreement amongst them being

commonplace. The source of disagreement is often the measurement or

weighting of a qualitative variable such as environmental or social impact, and

there may be no completely accurate way of determining which party is correct.

In such cases the stage is set for an acrimonious battle where emotion overrides

commonsense and the decision is made on political rather than practical grounds.

It is little wonder that solving these problems is described as a ‘surprisingly

difficult task’ (Carlsson and Fuller 1996).

In such a demanding environment it is essential for decision-makers to have

access to easily interpretable information, and tools for its analysis and

dissemination. The core problem of this research is the lack of software tools

available to decision-makers involved in infrastructure site selection decisions,

and the poor uptake of those tools where they do exist.

1.2.1 Context and fundamental elements of infrastructure site selection

The goal of an infrastructure site selection problem is to select an optimal site for

the successful deployment of an item of infrastructure. Examples include the

location of airports, industrial facilities and educational resources. Such decisions

are commonly made at a strategic level, as they are dominated by high-level

strategic concerns such as the triple bottom line of environmental, social and

economic impacts. It is the strategic aspects of infrastructure site selection that

have been the focus of this research, as this is where most conflict and


uncertainty arises. This research originates from recognition that existing

decision-making methods have limitations that make them either totally

unsuitable or inherently problematic in their implementation when faced with the

strategic issues involved in real world infrastructure site selection problems.

From a decision science perspective infrastructure site selection belongs to a

special class of decision problems referred to as ‘multi-criteria decision

problems’ in the decision science literature. Infrastructure site selection problems

posses four key characteristics that make them particularly challenging:

1. A large number of spatial alternatives

2. A heterogeneous group of decision-makers

3. Multiple evaluation criteria with an explicit spatial component

4. Uncertainty

These characteristics are not specific to infrastructure location problems alone, so

throughout this thesis the more generic term ‘Group Multicriteria Location

Problem’ (GMCLP) is used.

1.2.2 Background

Identification of the need for software-based spatial decision-making tools dates

back to the 1960s, when a new computer based spatial information processing

technology known as a Geographical Information System (GIS) was first

developed. GIS became widely commercially available in the early 1980s and

quickly became the platform of choice for spatial decision-making. The new term

Spatial Decision Support System (SDSS) was coined to signify the use of GIS in

a decision-making context, and countless specific techniques were developed and

implemented, however most were specific to a particular problem.

One promising, almost universally applicable technique, borrowed from the

domain of Operations Research, was Multicriteria Evaluation (MCE). MCE

methods serve to investigate a number of choice possibilities in the light of

multiple criteria and conflicting priorities, and form a natural framework for


further refinement to spatial decision-making techniques. The integration of

MCE and GIS seemed a natural fit, and Piotr Jankowski developed a practical

framework for MCE and GIS integration, in his often-cited 1995 paper

(Jankowski 1995).

MCE has often been used in GIS for Infrastructure site selection, however both

MCE and GIS continue to mature. Current theory and applications exhibit

several limitations that are particularly problematic when dealing with

GMCLP’s. The problems tend to stem from four main causes.

1. Decision-makers find the methods, and systems in which they are

implemented, difficult to use.

2. There is no robust method to accept inputs from a heterogeneous group.

3. There is generally no capacity for decision-makers to express a level of

uncertainty in their judgements.

4. Difficulty in understanding complex analytical methods leaves decision-

makers without a sense of control.

It may therefore be postulated that developing a SDSS for infrastructure site

selection should be driven by the need to produce a system that is easy to use,

has a robust method to accept inputs from multiple parties, is capable of dealing

with uncertainty and delivers a sense of control to users.

MCE has been shown to provide a reliable framework for such systems but to

further improve on their capabilities requires the integration of new techniques. It

is a fundamental hypothesis of this thesis that Approximate Reasoning (AR)

provides a practical means to augment existing methods, and more adequately

address the needs of decision-makers involved in infrastructure site selection. AR

methods allow the use of approximate ‘linguistic’ terms in an analysis, thereby

creating a more natural way for decision-makers to input preferences and receive

feedback.

Finally, one may enquire as to why multiple criteria or linguistic techniques are

needed when the predominant attitude is one of economic rationalism. According


to the economic rationalist perspective, private sector location problems reduce

to the single criteria problem of maximising profits (Beckman 1968). However

due to social pressure and legislation most major developments must now

consider at least three types of criteria - environmental, social and economic - to

satisfy the more enlightened paradigm of the “triple bottom line” (Pullar and

Pettit 2000). Public sector decision-making has always typically involved

consideration of multiple criteria. Even the simplest models consist of two, being

equity and efficiency (Morrill and Symons 1977). The use of MCE and AR

offers the ability to consider these, and other, disparate human concerns in a

constructive, quantitative way, and as such may play a part in producing better

outcomes to the challenging location problems of the twenty first century.

1.2.3 Use of Approximate Reasoning

Many decision-making methods use integers as a means to quantify decision-

maker assessments so they may be processed in an analytical model, however

real world decision-making is subject to uncertainties that make the use of

integers unrealistic. Uncertainty is a fundamental quality of infrastructure site

selection decisions, and much uncertainty is based upon the limited ability of

decision-makers to quantify qualitative evaluation criteria that are more easily

described by statements of natural language. For example stating that ‘a good

site for development will have a low environmental value’ may prove extremely

difficult to express in a quantitative way. In the previous sentence there are two

linguistic variables, one may be termed ‘site suitability’, and the other is

‘environmental value’ which must be ‘low’ for site suitability to be ‘good’. To

make constructive use of such input requires a method for dealing with these

linguistic variables in a numerical model.

Approximate Reasoning (AR) is a fuzzy logic based technique that can be used

to augment the MCE process by utilising fuzzy set methods to characterise and

operate upon imprecise inputs. In this approach linguistic inputs are quantified as

fuzzy numbers and manipulated with specialised fuzzy computation techniques.

Utilising AR and linguistic variables enables users to overcome some difficulties


encountered with regular MCE analysis. Fuzzy numbers provide a convenient

way to represent uncertainty, and procedures for criteria standardisation can

benefit from a universal linguistic suitability scale. Linguistic variables also

provide an easy means for decision-makers to input their opinions and value

judgements without needing to understand complex mathematical formulas,

thereby increasing the usability of spatial decision support tools in the real world.

Fuzzy methods have been implemented for spatial decisions in a limited way, via

inference systems or symbolic approaches. However there has been an absence

of fuzzy MCE techniques. This may be due to the fact that these methods tend to

require unfeasible amounts of processing to evaluate large numbers of

alternatives with a fuzzy algorithm. This thesis postulates that the rapid increase

in processing power over the last decade combined with the simplification of AR

methods allows the successful integration of fuzzy MCE into a SDSS.

1.2.4 Case Study

This research would have been impractical without access to a real world

example of the complexity encountered in infrastructure planning, and the future

development of the Brisbane Airport site provided an ideal case study. During

the next twenty years, improvements to deal with a forecasted trebling in

passenger movements will be phased in on the 2700 ha Brisbane Airport site,

which is situated on the Eastern Australian coastline just outside Brisbane and

adjoining Moreton Bay. There are many environmental, social, economic and

operational issues involved when locating future airport facilities, and it is this

rich decision-making environment that makes the project a valid example. A

comprehensive description of the Brisbane Airport site shown in Figure 1.1 can

be found in Appendix B.


Figure 1.1: Brisbane Airport

1.3 Research Outline

This research could be broadly classified as a combination of theory-focused

research consisting of the theoretical development of a new Approximate

Reasoning Algorithm for Infrastructure Site Selection (ARAISS), and design-

focused research consisting of the practical application of the theory in a new

SDSS (InfraPlanner). The research question, aims, objectives, scope, justification

and methodology are defined in the following subsections.

1.3.1 Research Question

If it were necessary to define the one question that has guided this research it

would be the following:

“Can Approximate Reasoning be integrated into a GIS based SDSS to mitigate

current difficulties with SDSSs utilised for Infrastructure Site Selection?”


It was hypothesised that the answer to this question was yes.

1.3.2 Aims

The aim of this research has been to create new knowledge at the intersection of

several disciplines surrounding the problem of infrastructure site selection, and

ultimately to produce a viable method of aiding decision-makers through the

complexities of site selection decisions. The confluence of Physical Planning,

Decision Science, Fuzzy Logic, Soft Computing, Decision Support and Expert

Systems, Geographical Information Systems, and Software Design, has opened

up many possibilities for new developments in site selection. This research has

aimed to capitalise on new developments in these areas by creating a practical

means to integrate them in a useful way.

1.3.3 Objectives

In keeping with the division of theory focus and design focus, two specific

research objectives were defined:

1. Develop a practical infrastructure site selection algorithm based on an

Approximate Reasoning ‘linguistic’ approach.

2. Develop a new spatial decision support system based on the algorithm

developed in objective 1.

The first objective resulted in the creation of ARAISS and the second in the

creation of the InfraPlanner system.

1.3.4 Scope

There were several factors that limited available resources and imposed

boundaries on the scope of this research. One of the main considerations was

time, which imposes a need for a clear focus and limits work considered

secondary to the research objectives. Moreover, as the research objectives


required a multidisciplinary approach, there was a consequent limitation on the

scope of work conducted in each area.

When considering the disciplinary scope of this thesis, the primary guiding

factors were the research objectives to design and implement a strategic decision-

making model. It therefore follows that the main contribution of this thesis lies in

the confluence of decision science and decision support systems. All other

disciplines served to provide specific tools and context to this main area of

research into what are the most promising ways to aid human beings make better

decisions. This thesis provides one context specific example, focused on

infrastructure site selection using an Approximate Reasoning model in a system

built inside a Geographical Information System.

Data security also posed a practical limitation on the scope of work undertaken,

particularly in terms of the validation problem used. It proved difficult to obtain

spatial data on areas outside the Brisbane Airport grounds whilst satisfying the

data access and security concerns of both the Brisbane Airport Corporation

(BAC) and Brisbane City Council. Consequently, the practical example outlined

in Chapter 8 utilised only data from within the Airport grounds, as was kindly

provided by BAC.

1.3.5 Justification

The problem at the centre of this research is of great significance. Infrastructure

planning decisions have a major effect on society, with most major projects

having significant environmental, social and economic impacts. The case study

used for this research offers a prime example of the impact site selection

decisions have on society. Brisbane International Airport is a major driving

engine for the economy of South East Queensland, and the planned development

of Airport site will involve a capital expenditure of around 1 Billion AUD. The

airport also provides a notable direct source of employment in the Brisbane area.

Important environmental considerations both inside the Airport boundary and

within its sphere of influence are many, as the Airport contains areas of saltwater

mangroves, and is located next to environmentally sensitive Moreton Bay. The


site also possesses three creeks, a river and a floodway, creating a clear need for

environmentally conscious development decisions. Social responsibilities,

including protection of cultural heritage sites and minimising impacts on

surrounding suburbs are also a major factor. Yet despite the enormous

importance of the site selection decisions involved in Airport planning, there is

no Spatial Decision Support System Currently in use. This is situation is typical

of many large-scale projects (Maniezzo, Mendes et al. 1998).

The basis for investigating the use of Approximate reasoning in site selection

also has a solid foundation in fact. Approximate reasoning works by quantifying

the uncertain, human elements of a problem. This is of immense practical value

as it is human intuition which is frequently the basis for decision-making (Turban

1995). Combining Approximate reasoning with GIS is now a realistic

proposition as GIS has grown to become a mature technology (Sui and

Goodchild 2001), with newfound flexibility and data processing power.

1.3.6 Methodology

As the theoretical outputs of the research were to be implemented into a practical

design process, it was decided to structure activities from a design perspective

and follow a Decision Support System development methodology. At a basic

level the method used in this research consisted of four stages:

1. Planning and Research:

• Needs assessment, problem diagnosis & definition of system

objectives.

• Review of relevant literature and gather other information.

2. Analysis & Design:

• Conceptual design of the InfraPlanner system.

• Development of the decision-making algorithm.

3. Construction:

• Coding and debugging of the InfraPlanner prototype.

4. Implementation and Feedback:


• Peer review of the model via publication and focus group.

• Testing and evaluation of InfraPlanner in a real world validation

problem.

• Critical assessment of the prototype and suggestions for future

improvements and research directions

Each step in the process required specific resources to conduct the activities

necessary to produce the desired outputs, and a Logic Model was employed as

the structure to organise the elements of the process. Logic modelling is a

resource management tool used to document the underlying reasons behind a

program of activities. In a logic model the program is divided into six elements.

1. Resources are the raw materials available

2. Activities make use of the available resources

3. Outputs are the tangible results of an activity

4. Customers are those who receive the outputs

5. Outcomes (Short, medium, or long term) are the reason for undertaking the

activities

6. External influences are those influences that are beyond the scope and

control of the program

The six elements of the process are grouped in the columns of a Logic Model and

arrows are used to signify connections between them. The logic model detailing

the sequence of events involved in the InfraPlanner development process is

shown in Figure 1.2, and further described in Chapter 7.

12 Chapter 1 Introduction and Methodology

Users / planners /

experts Statement of user needs

Developer User needs documented

Literature Research:

Critical review

State of the art report

Developmer/ other

researchers

Technology consultant

Analysis & Design:

Conceptual design

Decision-making model

& Design specification

Ideas for improvement documented

Planning: Interviews / focus groups

Software developer

Technology selection

Software developer

Construction: Coding

‘Infraplanner’ prototype

BAC users

Implementation: Prototype used in

validation problem

Ongoing system Adaptation and

better site selection decisions

GIS technology

External Influences: BAC staff turnover, technological development, advancements in operations research, political environment.

Resources Activities Outputs Customers Short term outcomes Long term outcomes

Figure 1.2: Development Process Logic Model

13Chapter 1 Introduction and methodology

1.4 Thesis structure

This thesis is structured to follow the development process outlined in Section

1.3 from start to finish, and as such can be broken into four major parts. Thesis

chapters are structured in the following way.

Chapter 1 Introduction and methodology

Part 1 Planning and Research

Chapter 2 Problem diagnosis and preliminary literature review

Chapter 3 Approximate reasoning

Chapter 4 Spatial decision support systems

Part 2 Analysis and Design

Chapter 5 Problem analysis and conceptual system design

Chapter 6 Algorithm design

Part 3 Construction

Chapter 7 Creating InfraPlanner

Part 4 Implementation and Feedback

Chapter 8 A case study at Brisbane Airport using InfraPlanner

Chapter 9 Conclusions

The Planning and Research section comprises three chapters. Chapter 2 details

the process of identifying the primary objectives of the Spatial Decision Support

System (SDSS), and the preliminary phase of the literature review. This first

review covered existing techniques and technology used in spatial decision-

making. Chapter 3 extends the review to provide a detailed introduction to the

Approximate reasoning techniques used in the new site selection algorithm

developed in this research. Finally, Chapter 4 focuses on the technical aspects of

implementing decision-making techniques in a SDSS.


The Analysis and Design section comprises two chapters. Chapter 5 analyses the

main limitations on current Spatial Decision Support Systems, and proposes a

conceptual framework for mitigating these limitations. Statements about desired

capabilities are made, and conceptual ideas to achieve these capabilities are

discussed. Chapter 6 details the design of a new Approximate Reasoning

Algorithm for Infrastructure Site Selection (ARAISS). The mathematical

processes in the algorithm are derived and discussed. The process of testing

ARAISS for validity outside GIS using MATLAB software is also illustrated.

The Construction section comprises one chapter. Chapter 7 describes the

practical implementation of the ARAISS algorithm in a Spatial Decision Support

System created by customising ArcView GIS software using Visual Basic for

Applications (VBA).

The Implementation and Feedback section comprises one chapter. Chapter 8

describes how InfraPlanner was implemented in an experiment using a three

decision-maker six criteria site selection problem at Brisbane Airport.

1.5 Multi-disciplinary nature of the research

It was clear from the outset that achieving the research objectives required a

multi-disciplinary approach, which tends to compound complexity. Rather than

focusing on one aspect of the problem, the objectives required work to be carried

out in three broad categories, where each category contained at least two separate

disciplines, as shown in Table 1.1.


4

2

4

3

3

5,7,8,9 6

Table 1.1: Disciplines involved in the research

CCaatteeggoorryy DDiisscciipplliinneess IInnvvoollvveedd

Infrastructure location problems Physical planning

Decision science

Approximate Reasoning algorithm Fuzzy logic

Soft computing

Spatial Decision Support Systems Decision Support and Expert Systems

Geographical Information Systems

Software design

The category of disciplines inherent in each chapter varies throughout the thesis,

and is described in Figure 1.3. It may be seen from Figure 1.3 that the major

research effort was spent in the intersection of the three primary areas. Each of

the disciplines invloved contained significant sub-problems, which were outside

the scope of the research program. An example is the problem of how to best

represent a linguistic term as a fuzzy number, a significant issue in the area of

approximate reasoning. This problem is discussed in Chapters 3 and 5, but was

beyond the scope of this research.

Figure 1.3: Disciplines contained in each chapter

Infrastructure Location Problems

Approximate Reasoning

Model

Spatial Decision Support Systems


This page intentionally blank

17Chapter 2 Problem Diagnosis and Preliminary Literature Review

Chapter 2

PPRROOBBLLEEMM DDIIAAGGNNOOSSIISS AANNDD

PPRREELLIIMMIINNAARRYY LLIITTEERRAATTUURREE RREEVVIIEEWW

2.1 Introduction

The first step in the development process was to generate a clear statement of the

problem and survey existing approaches to its solution. To maintain the focus on

a practical application, a focus group was assembled, consisting of planning and

infrastructure managers from industry, and academics from QUT. The meeting

led to a consensus on key statements about the aims, objectives, and scope of the

proposed Spatial Decision Support System (SDSS). Once the groundwork had

been laid, an extensive literature review, encompassing the techniques and

technology of spatial decision-making, was conducted. Planning and research

initially overlapped as feedback from the literature review provided the impetus

for a more detailed description of desired functionality.

This chapter provides key statements from problem diagnosis, and a preliminary

review of the decision science techniques and technology most commonly used

for location problems. A detailed discussion of Approximate Reasoning

techniques, and the technical aspects of SDSSs follow in chapters 3 and 4.


2.2 Problem diagnosis

The initial meeting for problem diagnosis and needs assessment was based

around the discussion paper in Appendix C, and took contributions from

professionals involved in planning, infrastructure development and management,

and environmental management. Also present were QUT academics from the

fields of Geographical Information Systems, Software Design, Mathematics, and

Environmental Management. Key statements emerging from the meeting were

as follows:

• The system will support decision-makers in planning, infrastructure

development and management, and environmental management with site

selection decisions.

• The system should be able to accommodate qualitative variables such as

socio-economic and environmental impacts.

• The system should accommodate multiple criteria and multiple points of

view of the measurement and weighting of those criteria.

• Outputs from the system should be graphical where possible, preferably

in a mapping format.

• The modelling capabilities of the system should be transparent and easily

understandable.

• The system should aim to aid decision-makers, not replace them.

These basic statements of desired functionality were then used as the focus for a

state of the art literature review. The first stage of the review focused on the

existing techniques and technology involved in spatial decision-making.

Specifically, the review was conducted to answer the following questions:

1. What are the analytical techniques used in the solution of the type of

Infrastructure site selection problems encountered by decision-makers at

BAC, and planners in general?

2. What are the major limitations of these techniques?


3. What technology platforms are used in the analysis of Infrastructure site

selection problems?

4. What are the most promising methods for advancing current techniques and

technologies?

The remainder of this chapter is drawn from the first stage of the review.

2.3 Introduction to location problems

Solving location problems is an everyday activity performed by individuals and

groups who use spatial information to make decisions about such things as where

to live, where to shop, and how to manage the environment and infrastructure

(Jankowski, Andrienko et al. 2001). The primary objective of these problems is

to identify the most desirable location for a facility or service (Maniezzo,

Mendes et al. 1998), such as locating a new airport, allocating law enforcement

resources, or buying a new home.

The choice between competing locations is made according to how well each

location satisfies a set of conditions. These conditions, commonly referred to as

evaluation criteria or simply criteria, will vary across space and are unique to

each location problem. They may encompass issues such as maximisation of

utility, minimisation of detrimental environmental and social impact, and ease of

accessibility (Nijkamp, Rietveld et al. 1990). The term ‘criteria’ is generic and is

used to convey the concepts of both objectives and attributes. The primary

objective may also be referred to as the goal, and is usually the top level of a

hierarchy of sub-objectives (Saaty and Kearns 1985). These sub-objectives are

operationalised by assigning measures to achieve them, called attributes

(Malczewski 1999). For example if the objective is to minimise environmental

damage when locating an industrial facility, an attribute chosen to represent this

objective may be the number of acres of bushland lost.


When criteria are conflicting, it is inevitable that trade-offs will need to be made.

In order to optimise the trade-off process, it is essential to specify how relatively

important each criterion is. This is usually a subjective process whereby the

decision-maker assigns weights to each criterion according to his or her

preferences (Bogetoft and Pruzan 1997). However spatial decisions are often

made by groups of decision-makers, to satisfy the needs of multiple stakeholders.

Such situations are described as group decision-making, and in a group

environment where decision-makers are autonomous and heterogeneous it is

inevitable that conflicts will occur (Chu-Carrol and Carberry 2000). These

conflicts generally arise because of the diverse values of the groups or

individuals involved, which lead to different weighting of criteria, but conflicts

may also arise from the definition of criteria, or the decision-making process

(Bogetoft and Pruzan 1997).

A site selection decision is essentially a choice between alternative sites. Each

alternative will have a set of outcomes (consequences) in relation to the various

evaluation criteria, however the set of outcomes is seldom completely

deterministic, and some level of uncertainty usually enters the decision-making

process (Spradlin 1997). Sources of uncertainty are generally two fold. Firstly

there may be some uncertainty about the validity of the information upon which

the decision is to be based, such as the reliability of an expert opinion (Keeney

and Raiffa 1976). Secondly there may exist some unpredictability about future

events and the state of the future environment in which the decision outcome

dwells, such as the weather or economic outlook. Types of uncertainty are also

twofold, the first being stochastic, as described by a probability distribution of

the alternate states of attributes and outcomes, and the second is fuzziness

(imprecision in data), as described by fuzzy set theory (Bellman and Zadeh

1970).

2.3.1 Problem classifications

This research was driven by infrastructure site selection problems, which are

referred to here by the more generic term Group Multi-criteria Location

Problems (GMCLP’s). GMCLP’s are complex real world decision problems with


the objective of finding an optimal site for a facility or service from multiple

alternatives, using multiple evaluation criteria and the opinions of multiple

stakeholders.

GMCLP’s belong to a general class of decision-making problems referred to as

multicriteria decision problems. Classification of these problems is summarised

in Figure 2.1. It is widely accepted that multicriteria decision problems can be

broken into two categories. Multiattribute decision-making (MADM) problems

involve a finite or relatively small number of discrete alternatives, whereas

multiobjective decision-making (MODM) problems have a relatively large or

infinite number of feasible alternatives (Jankowski 1995). MADM and MODM

have also been referred to as discrete and continuous decision problems (Hwang

and Yoon 1981), as MADM implies a discrete number of pre-specified

alternatives, whereas in MODM the alternatives are generated during the solution

process. It is important to note that if there exists a direct correspondence

between objectives and attributes, a MODM problem becomes a MADM

problem, as the objectives may be completely defined by a limited number of

attributes in this scenario.

Group decision-making, where more than one set of goals or preferences is

considered, is then distinguished from individual decision-making, where the

objectives are agreed. This distinction is made on the grounds of conflicting

objectives rather than number of decision-makers. The level of uncertainty

provides a third division between deterministic problems, where all relevant

information is known, and probabilistic or fuzzy problems, where there is some

uncertainty. In real world decision problems uncertainty is commonplace, and

deterministic problems are rare.


Figure 2.1: Classification of Multicriteria Decision Problems

Modified from: (Malczewski 1999)

A Rigorous definition of GMCLP’s is suggested here and defines them as:

‘the selection of an optimal location from a large number of spatial alternatives

by a heterogeneous group of decision-makers using multiple evaluation criteria

under uncertainty.’

MMUULLTTIICCRRIITTEERRIIAA DDEECCIISSIIOONN

PPRROOBBLLEEMMSS

Multiattribute decision problems

Individual

Certain Uncertain

Probabilistic Fuzzy

GGMMCCLLPP’’ss

Group

Certain Uncertain

Probabilistic Fuzzy

Multiobjective decision problems

Individual Group

Certain Uncertain

Probabilistic Fuzzy

Certain Uncertain

Probabilistic Fuzzy


They contain the following four key attributes:

1. A large number of spatial alternatives:

The alternatives under consideration are numerous enough to make manual

analysis impractical i.e. the problem is non-trivial

2. A heterogeneous group of decision-makers:

Multiple parties are involved in the decision process and there is no guaranteed

consensus among them

3. Multiple evaluation criteria with an explicit spatial component

The decision is based on multiple, conflicting criteria that vary across space

4. Uncertainty

The relationship between the available raw data and site suitability is subject to

some kind of uncertainty

The GMCLP discussed in Chapter 8 offers a practical example of the type of

location problem defined above. It involves locating a new industrial facility

somewhere on the 2700 ha Brisbane Airport site. Stakeholders include the

Brisbane Airport Corporation, The Commonwealth Government and Community

representatives. Decision-makers wish to satisfy six evaluation criteria, which

include issues such as environmental value and community impact, that are hard

to quantify and subject to disagreements among parties, as well as uncertainty in

measurement.

2.4 Decision science techniques

Decision science, also referred to as decision analysis, operations research,

systems engineering and management science, has a long history. Put simply it is

the application of scientific method to everyday decision-making. Decision

science seeks to apply logical reasoning to decision problems in a structured

way, thereby making the decision process explicit and repeatable. It also offers a

means to look inside a particular decision and make explicit how and why it was


made. Decision science has found many applications in engineering, the military

and business management. Although there is evidence that formal decision-

making methods in military strategy date back thousands of years, the field of

decision science is commonly assumed to have originated during World War II,

when scientific methods were applied to strategy in antisubmarine warfare by

T.C. Koopmans.

There are a multitude of formal decision-making methods. However those that

have been applied to location problems are relatively few and fall into three

broad categories.

1. Map algebra methods

Map algebra includes standard spatial functions and simple overlay methods

that screen out sites based on Boolean operators or simple arithmetic.

2. Multicriteria evaluation methods

Multicriteria evaluation methods offer the ability to rate criterion outcomes on a graduated scale and choose the relative importance or weight of each criterion.

3. Artificial intelligence methods (soft computing or geocomputation)

These methods include neural networks, fuzzy systems and evolutionary algorithms. They are usually complex in nature and offer great potential for complex spatial problems.

The following Sections provide an overview of these three groups of methods,

particularly the widely applied family of multicriteria evaluation techniques. A

more detailed review of the Approximate Reasoning methods is provided in

Chapter 3.

2.4.1 Map algebra

Map algebra, or overlay analysis, is the most basic level of spatial analysis. It

involves the use of simple arithmetic, Boolean and relational operators to

combine input maps. Table 2.1 provides a sample of map algebra operations.


Table 2.1: Map algebra operators

TTyyppee OOppeerraattiioonn

Arithmetic Subtraction, Addition, Multiplication, Division

Boolean And, Or, Not

Relational Less than, Greater than, Equal to

The basic concept of overlay analysis using map algebra in planning problems

with environmental criteria was first documented in the profoundly influential

work ‘Design with Nature’ (McHarg 1969). Map algebra is the basis of all more

complex methods, and provides a powerful tool for site selection (Tomlin 1990),

whilst being easy to use and understand. However several authors have pointed

out that the use of Map algebra alone tends to oversimplify analysis (Hopkins

1977; Hobbs 1980; Pereira and Duckstein 1993), and more powerful tools are

needed for complex decision environments. The next level of complexity in

analysis is provided by multicriteria evaluation methods.

2.4.2 Multicriteria evaluation

The term Multicriteria Evaluation (MCE), which emerged in the 1970s, is used to

represent a variety of methods for solving multicriteria decision problems.

Multicriteria evaluation has widely and consistently been recognised as

appropriate to deal with spatial decisions eg. (Pereira and Duckstein 1993;

Eastman, Jin et al. 1995; Jankowski 1995; Laaribi, Chevallier et al. 1996;

Malczewski 1999). It is suitable for discrete multiattribute decision making

(MADM) situations such as spatial allocation, and can handle quantitative and/or

qualitative data (Pettit and Pullar 1999). MCE methods serve to investigate a

number of choice possibilities in the light of multiple criteria and conflicting

priorities (Voogd 1983), and have been described as weighing independent

criteria in terms of judged relative importance or value (Smith 1980).

Applications of MCE in site selection are many and have included such diverse

problems as urban waste management (Haastrup, Maniezzo et al. 1998) power


plant siting (Hobbs 1980) ecosystem management (Ji 1996; Prato 1999)

allocation of educational resources (Kwak and Changwon 1998) and sustainable

development (Nijkamp and Giaoutzi, 1993; Sharifi, et al., 2002). MCE is

considered by many to be an immature, non-comprehensive sub-discipline of

Operations Research (Bogetoft and Pruzan 1997). Should current patterns

continue, MCE will continue to evolve and be refined over time.

While individual MCE processes may vary, they generally fit within the same

overall framework, as shown in Figure 2.2. There are two major approaches to

organizing the sequence, and they find their division at the stage of identifying

alternatives and criteria. A value-focused approach seeks to identify criteria

(values) first, and an alternative focused approach will first identify alternatives

(Keeney 1992). Keeney (1992) goes on to argue that the value focused method is

superior as values are more fundamental than alternatives.

There are five major steps involved before reaching the final recommendation

stage, which can broadly be defined as:

1. Specifying and standardising the evaluation criteria

2. Weighting the evaluation criteria

3. Identifying feasible alternatives, and bringing them together with criteria in a

decision matrix

4. Performing an aggregation

5. Performing a sensitivity analysis


Figure 2.2: Framework for MCE

modified from (Malczewski 1999)

CHOICE

DESIGN

INTELLIGENCE

Problem Definition

Criterion Weights

Constraints

Decision Matrix

Alternatives

Decision-makers Preferences

Aggregation via Decision

Rules

Sensitivity Analysis

Recommendation

Standardised Evaluation

Criteria


In a more general sense decision-making processes consist of three phases,

intelligence, design and choice (Simon 1960). The MCE framework in Figure 2.2

is shown in context with these three phases of the decision-making process. It

can be seen that the intelligence phase comprises the systematic collection of

data and information about the problem, the design phase involves processing the

information, and choice entails the selection of a solution based on the processed

outputs. The five major steps within the three phases are explained in detail

within the following sections.

2.4.2.1 Specifying and standardising evaluation criteria

There is no single technique for specifying evaluation criteria in all cases, and

several approaches may be used in parallel. The advice of experts such as

ecologists, social scientists, and economists, is usually used to define, measure

and rate complex criteria (Pullar and Pettit 2000), whilst the popular Delphi

technique may be used to formalise the input of opinions from these experts and

formulate relevant factors (Pettit and Pullar 1999). The Delphi method involves

anonymous inputs made by a group of heterogeneous experts who are given

feedback between rounds (Author 1999). However this does not eliminate the

danger of people’s personal feelings taking priority over facts, and an

examination of relevant literature and/or a rigorous analytical simulation should

also be used (Keeney and Raiffa 1976).

The criteria for a given location problem are operationalised by attributes which

are seldom homogeneous. Attributes may fall into several, sometimes

overlapping, categories, and are often described by the varying scales upon

which they are measured.

At a basic level the scales are generally considered to be one of four types:

1. A Nominal scale is a list of names (labels) used to identify the state of an

attribute. Using this scale it is possible to see if two attributes are equal or

not.


2. An Ordinal scale provides a relative ordering. Using this scale it is possible

to compare attributes using the less than and greater than operators.

3. An Interval scale is a continuous scale using equal intervals from an

arbitrary zero point. Using this scale it is possible to perform additions,

subtractions and scaling by a constant.

4. A Ratio scale is a continuous scale using equal intervals from an absolute

zero point. This scale permits all the previous operations plus multiplication

and division.

Table 2.2 provides further classifications for attributes based on the opinions of

various researchers. A given attribute may be a member of several of these

groups.


Table 2.2: Types of attributes, a brief description, and describing authors.

TTyyppee DDeessccrriippttiioonn RReesseeaarrcchheerr

Factor Measured on a continuous scale (Eastman, Jin et al. 1995)

Constraint Boolean (0 or 1) serving to

limit the alternatives under

consideration

(Eastman, Jin et al. 1995)

Proxy Indirectly related to the

objective

(Keeney and Raiffa 1976)

Direct Directly related to the objective (Keeney and Raiffa 1976)

Benefit Maximisation is desirable (Hwang and Yoon 1981)

Cost Minimisation is desirable (Hwang and Yoon 1981)

Qualitative Measured on a qualitative

nominal or ordinal scale

(Voogd 1983)

Quantitative Measured on a quantitative

interval or ratio scale

(Voogd 1983)

Natural Measured on an established &

commonly used scale requiring

no subjective input


Constructed Measured on a scale

constructed by subjective input


Deterministic Certain (Malczewski 1999)

Probabilistic Described by probability theory (Malczewski 1999)

Linguistic A natural language term

semantically defined by a fuzzy

membership function

(Zadeh 1976)


A consequence of having to use incommensurate attributes (i.e. attributes having

different scales of measurement) is the necessity to perform transformations to

derive a standard scale upon which they may be commensurately compared. The

type of transformations used will depend upon the level and type of uncertainty

in the raw data as well as the qualitative or quantitative nature of the data (Voogd

1983).

Linear scale transformation methods are suitable for quantitative, deterministic

data. The most common linear transformations are the maximum score and score

range procedures (Voogd 1983) given below.

Maximum Score Transformation:

(benefit criterion) (2.1)

(cost criterion) (2.2)

Where: ijx' is the transformed score of the ith alternative wrt the jth attribute

ijx is the raw score of the ith alternative wrt the jth attribute

maxjx is the maximum raw score for the jth attribute

The maximum score transformation has the advantage of being proportional,

with the most desirable score always equal to unity. However interpretation of

the least desirable transformed score is difficult (Malczewski 1999).

Score Range Transformation:

(benefit criterion) (2.3) minmax

min

'jj

jijij

b

xxxx

x−

−=

max'j

ijij

b

xx

x =

max1'j

ijij

c

xx

x −=


(cost criterion) (2.4)

The score range transformation has the advantage of giving a complete range of

values from 0 (the worst score) to 1 (the best), but has the disadvantage of not

being proportional.

Another method for obtaining a standardised scale is through value or utility

functions. A value function is a utility function used in a deterministic decision

situation (Keeney and Raiffa 1976). It yields a standardised scale by

transforming the raw score of the attribute in question to a value between 0 and 1

via subjective value judgements (Hepner 1984). One of the most widely used

procedures for generating a value function is the midvalue method, which is

carried out in the following steps (Bodily 1985).

1. Determine the range over which the attribute is to be assessed, and assign the

values 0 and 1 to the end points.

2. Find the midvalue point (the point having a value (utility) half way between

the end points) from decision-maker input and assign the value of 0.5 to that

point.

3. Find the midvalue points either side of 0.5 to yield the points with a value of

0.25 and 0.75.

4. Repeat step 3 to obtain as many points as needed.

5. Draw the value curve through the previously obtained points and fit an

analytical expression to the curve.

The term utility function describes both deterministic value functions, and

functions obtained using a probabilistic approach (Keeney and Raiffa 1976).

Among the techniques applied in the latter situation, the indifference technique,

otherwise described as the 50-50-lottery method (Bodily 1985), is analogous to

the midvalue method. The major steps involved in the two procedures are the

minmax

max

'jj

ijjij

c

xxxx

x−

−=


same. However the indifference technique requires the decision-maker to obtain

midvalues by assessing an outcome that has the same utility as a 50-50 gambling

of two other outcomes with established utility values, as described in equation

2.5.

N.B. In equations 2.5 and 2.8 the term ‘p’ is used to denote two different

variables. The usage has been kept consistent with the referenced work in each

case and the reader should treat each case independently and not assume that a

given meaning applies in other sections of the text.

(2.5)

Where: )( jj xu is the utility function of the jth attribute

)( +jj xu is the utility of the best outcome (= 1)

)( −jj xu is the utility of the worst outcome (= 0)

p is probability

Construction of the curve is accomplished by varying p in increments until an

adequate number of points have been established.

2.4.2.2 Criterion weighting

The evaluation criteria specified for a given MCLP often vary in terms of the

relative importance decision-makers place on them. This creates the problem of

quantifying the disparity of importance amongst evaluation criteria, which may

be solved by a number of different procedures. This section describes three

procedures for weighting criteria in location problems: ranking, rating, and

pairwise comparison. The choice of which procedure to use will generally

depend upon the competing requirements of accuracy and ease of use.

The simplest method of weighting criteria is to arrange them in rank order using

units on an ordinal scale, according to the decision-makers preferences. Weights

are then assigned according to the following formula (Stillwell, Seaver et al.

1981):

pxupxupxu jjjjjj =−+== −+ )()1())((?)(


(2.6)

Where: wj = The weight of the jth criterion

Rj = The rank of the jth criterion

n = The number of criteria

The ranking approach is attractive in practice due to its simplicity (Voogd 1983).

Rating offers a slightly more comprehensive approach by either allocating a pre-

determined number of points across the evaluation criteria, or assigning a value

to the criteria of maximum importance and rating each subsequent criteria in

relation to it (Stillwell, Seaver et al. 1981). Weights are then normalised to a

suitable interval scale.

One of the better techniques for developing criteria weights is the pairwise

comparison method developed by Saaty (1980) within the context of his

Analytical Hierarchy Process (AHP) (Saaty 1980; Eastman, Jin et al. 1995). In

AHP pairwise comparisons of criterion importance are recorded in a square

reciprocal matrix. The AHP offers particularly fine resolution but can be time

consuming.

2.4.2.3 Alternatives and the decision matrix

After an analysis of constraints, the set of feasible decision alternatives may be

specified. Constraints in this context are Boolean criteria that serve to reduce the

number of available alternatives under consideration. This process may be either

compensatory or noncompensatory, where a compensatory analysis allows

weaknesses on one criterion to be traded off against strengths on another, and

noncompensatory techniques enable the elimination of an alternative based

( )1

1

1+−∑

+−=

∑=

k

n

k

jj

rn

rnw


solely on the poor performance of a single criterion (Jankowski 1995). Feasible

alternatives are those that survive this elimination process.

The set of feasible alternatives may then be combined with the remaining criteria

attributes and weights in a decision matrix, which is a construct used to visualise

the computational heart of the multicriteria evaluation process. It enables the

comparison of decision alternatives in relation to their set of criteria outcomes as

shown in Figure 2.3.


Figure 2.3: Decision Matrix

modified from: (Malczewski 1999)

Generation of the decision matrix provides the necessary inputs for an

aggregation to be performed. Aggregation provides the means by which

alternatives may be quantitatively compared, and is implemented according to

the decision rules.

2.4.2.4 Aggregation via decision rules

Decision rules are procedures that allow for ordering alternatives (Starr and

Zeleney 1977). They typically combine criteria into a single composite index,

and a statement of how alternatives are to be compared using that index

(Eastman, Jin et al. 1995). This section describes three common aggregation

AAttttrriibbuuttee11 AAttttrriibbuuttee22 AAttttrriibbuuttee33 ……………….... AAttttrriibbuutteenn

Alternative1 Outcome

1,1

Outcome

1,2

Outcome

1,3

……….

.

Outcome

1,n

Alternative2 Outcome

2,1

Outcome

2,2

Outcome

2,3

……….

.

Outcome

2,n

………. ………. ………. ………. ………. ………. Alternativem Outcome

m,1

Outcome

m,2

Outcome

m,3

……….

.

Outcome

m,n

Preferences Weight1 Weight2 Weight3 ……….. Weightn

State of decision

environment

GGOOAALL

Objective 1 Objective 2

Sub-objective Sub-objective Sub-objective


procedures: Linear weighted combination, compromise programming and

ordered weighted averaging.

The most prevalent procedure in spatial MCE is linear weighted combination

(Eastman, Jin et al. 1995). The underlying mathematical theory of the linear

weighting process is presented in the following equation, which brings together

alternative outcomes and the criterion weights in a linear equation.

(2.7)

Where:

S = outcome score for alternative j

w = weight of criterion i (on a 0 to 1 scale)

x = score of criterion i for alternative j (on an arbitrary but common

evaluation scale)

c = constraints on alternative j (0 or 1)

⊗ = Multiplication

(Eastman, Jin et al. 1995)

Compromise programming is based on the displaced ideal concept, which

considers that there is an ideal solution and attempts to minimise the distance

from it (Zeleny 1982). Locating the solution closest to the ideal solution is

accomplished using weighted Lp norms or metrics as shown in equation 2.8.

(2.8)

subject to

X∈x , q,1,2,...... kfor 0 =≥kw

where

Lp is the distance metric

wk is the weight of the kth objective function

jiij cxwS ∏⊗= ∑ )(

( ) ( )( ) ( )

pp

kk

kkk

pkp xfxf

xfxfwL

1

min⎥⎥⎦

⎤

⎢⎢⎣

⎡⎥⎦

⎤⎢⎣

⎡−−

=−+

+∑


fk+(x) is the ideal solution of the kth objective function

fk-(x) is the nadir or anti-ideal value of the kth objective function

p is a power parameter ranging from 1 to infinity

Ordered Weighted Averaging (OWA) is a variation from linear weighted

combination. Whereas a linear weighted combination is a compensatory

technique, meaning that bad outcomes on one criterion can be compensated for

by better outcomes on another, OWA offers the ability to control the level of

compensation. The strict OWA operator weights criteria on the basis of their

rank suitability order rather than being weighted on their inherent qualities

(Yager 1988). Thus, if at a certain location three criteria (X,Y,Z) are ranked in

terms of suitability ZYX (highest to lowest) and it has been decided to apply

weights of 0.5, 0.3, 0.2, then the weighted combination would look like equation

2.9.

0.5Z + 0.3Y + 0.2X (2.9)

A hybrid approach based on the OWA operation also includes the original

criterion weights with order weights. This approach provides flexibility in the

degree of trade-off applied at a given location. In linear weighted combination

criteria weights determine how factors trade-off relative to one another. However

the level of trade-off is not adjustable. Extension via OWA offers the ability to

adjust the level of trade-off and risk aversion. Risk aversion is measured on a

scale between a totally AND aggregation where the minimum suitability term

has total dominance and a weight of 1 (representing total risk aversion) and a

totally OR aggregation where the maximum term has total dominance and a

weight of 1 (representing a risk-taking attitude). The level of trade-off

approaches zero as order weights approach either end of an ANDORness scale.

The terms ANDness ORness and TRADEOFF are quantified using the following

equations (Jiang and Eastman 2000).

ANDness = (1/(n-1))Σ((n-i)Worder i) (2.10)

ORness = 1 - ANDness (2.11)


TRADEOFF = 1

)/11

2

iorder (−

−−

∑n

nn W (2.12)

Where

n is the total number of factors

i is the order of factors

Worder i is the order weight for the factor of the ith order

N.B. It should be noted that these AND / OR operations are dissimilar to that

employed in Approximate Reasoning. The fuzzy sets used in Approximate

Reasoning are subject to AND / OR operations similar to those used in

traditional set theory.

2.4.2.5 Sensitivity analysis

Once alternatives have been analysed and rated, a sensitivity analysis provides a

means to test the robustness of favourable solutions in the light of uncertainty.

This process involves examining how a change of one criterion score will affect

the final result (Voogd 1983). The two most important elements to consider in

sensitivity analysis are criterion weights and criterion (attribute) values, with the

objective being to find the set of nondominated solutions. Nondominated

solutions are described as those which are feasible and no other feasible solution

exists that improves the performance on a single criterion outcome without

worsening another (Malczewski, Pazner et al. 1997).

2.4.2.6 Limitations of MCE

There are limitations within current MCE theory and its application to spatial

problems as MCE is still a maturing field. The main limitations of MCE based

spatial decision support may be listed as follows:

• The process of standardising criteria continues to provide a major challenge

(Eastman, Jin et al. 1995; Jiang and Eastman 2000)


• Deriving criterion weights particularly when multiple stakeholders are

involved is difficult, as any given group will usually exhibit some variance of

opinion (Herrera, Herrera-Viedma et al. 1996)

• A robust methodology for accepting inputs from multiple stakeholders is

required (Jankowski 1995; Malczewski 1996)

• A method for dealing with uncertainty in spatial datasets is required (Beard

1994; Hunter and Goodchild 1995; Kyriakidis and Goodchild 1999)

• The final step of deciding which particular solution provides the absolute best

location for the facility in question when several solutions have a similar

rating has proved difficult (Eastman, Jin et al. 1995)

• Different MCE decision rules generate different outputs and there is no

accurate way of choosing the best method (Carver 1991; Heywood, Oliver et

al. 1995).

• The size and shape of the proposed site is not explicitly included in an

analysis (Brookes 1997)

• The use of MCE in computer-based decision support systems is limited by

the fact that highly capable analytical systems are often used as simple

visualisation tools, primarily due to difficulties in use and understanding of

the systems by strategic decision-makers (Klosterman 2000)

2.4.3 Artificial intelligence and soft computing

Artificial intelligence (AI) is an attempt to reproduce aspects of human

intelligence via a computer algorithm. The most promising AI techniques for

spatial analysis are contained within the area of soft computing, and are useful

for handling the many types of uncertainty and ambiguity inherent in complex

systems. Real world decision-making is subject to uncertainty in the sense that

the goals, constraints and consequences of possible actions are not always

precisely known (Bellman and Zadeh 1970). It is this uncertainty that is the

foundation of many current difficulties with the analysis of spatial problems.

Three soft computing techniques that are becoming more common in spatial

analysis are fuzzy logic, artificial neural networks and genetic algorithms. This

section discusses these three methods.


2.4.3.1 Fuzzy logic

A common flaw in spatial decision-making is the unreasonable level of

resolution implied when screening attribute values, whereby an artificial cut-off

value is specified. For example, if it is specified an acceptable site must be within

5km of a river, why is a site 4.99km away acceptable and a site 5.01km away

unacceptable (Malczewski 2002)? There is usually ambiguity and imprecision in

defining such cut-off values, which may be represented using fuzzy set theory.

In classical set theory a set is a collection of definite and distinct objects. Objects

contained within the set are its members and any member of the collection of

objects (universe of discourse) from which the set is drawn (e.g. the set of real

numbers) is either a member or non-member of the set. Fuzzy set theory applies

a degree of membership instead of considering an object as strictly in or out.

Limitations in fuzzy logic approaches to site selection are mainly based on the

uncertainty involved in choosing membership functions to represent the

linguistic terms. As yet there appears to be no definite method to do this, and as

the meaning of words tends to vary with the person using them, it may be

unfeasible to develop a completely reliable way to solve the problem. A detailed

introduction to the specifics of fuzzy logic as they apply to decision-making is

provided in Chapter 3.

2.4.3.2 Neural networks

Neural networks are a simulation of the way a human brain functions at a cellular

level. A collection of processing units acting as neurons, are connected by a

weighted network representing the role of synapses. When presented with a set

of inputs and a set of outputs during a training phase, the network stores the

relationships between inputs and outputs in the structure of connections. In this

way the network can ‘learn’ to perform new tasks.


Neural networks have been found to be an effective tool for spatial decision-

making (Sui 1993; Zhou and Civko 1996) and have the advantage of being

independent of the problem domain, thereby allowing a user to focus on the

problem rather than a set of rules for its solution. They have been applied to

spatial problems such as land valuation (Almond, Jenkins et al. 1997) and urban

development (Feng and Xu 1999), as well as traditional multicriteria problems

(Zhou and Civko 1996). A major drawback of using neural networks in site

selection problems is that the way a network adapts its structure to solve

problems is largely hidden from the user. Such a ‘black box’ approach to

decision-making is unlikely to be fully embraced by decision-makers and interest

groups (O'Sullivan and Unwin 2003), as there is a sense of loss of control.

2.4.3.3 Genetic algorithms (GA’s)

Genetic algorithms (GAs) were devised by John Holland in the 1960s, and

developed with the aid of his students and colleagues throughout the 60s and 70s.

In his 1975 book, Holland presented the GA as an abstraction of biological

evolution, providing a theoretical framework for natural and artificial adaptive

systems (Holland 1975). GA’s have two primary advantages. Firstly they are

efficient in complex search spaces, and are capable of avoiding the problem of

finding only local optima. Secondly they are independent of the problem domain,

providing a method capable of application to a multitude of complex real world

problems.

While there is no rigorous definition of a GA, most GA’s have four elements in

common (Mitchell 1999).

1. A population of chromosomes: A chromosome represents a point on the

search space, i.e. a candidate solution. Chromosomes are typically coded as

bit strings, e.g. 111001, whereby each locus (gene) in the chromosome has

two possible alleles: 0 and 1. (Chromosomes consisting of strings made up of

non-binary alphabets with more possible alleles are also allowable).


2. Selection via fitness: A fitness function is used to assess how well each

chromosome solves the problem at hand by assigning it a score.

Chromosomes with high scores are more likely to have the chance to breed.

3. Breeding via crossover: The crossover operation randomly chooses a locus

and exchanges the subsequences before and after that locus between two

chromosomes to create two offspring. For example crossing over 000111 and

111000 after the third locus creates 000000 and 111111.

4. Mutation: The mutation operation randomly flips bits in a chromosome.

The probability of this occurring is usually very small, often in the vicinity of

0.001.

A simple GA starts with a randomly generated population, selects individuals for

crossover using an increasing function of fitness, replaces the existing population

with the offspring created from crossovers, randomly mutates some genes in

some individuals, then iterates the process from the point of selection. The run

lasts for a specified number of generations or until a stopping criterion (e.g. a

specified level of fitness) is satisfied.

Although the exact mechanisms underpinning GA’s have not been fully defined,

it is generally accepted that GA’s work by retaining the building blocks of good

solutions. These building blocks are referred to as schema (Holland 1975). A

schema is a partial template for a solution, for example in a six bit chromosome

one schema may be *10***, which represent the set of all bit strings with a one

at the second locus and a zero at the third.

GA’s have been applied to such spatial problems as motorway routing (Pereira

1996) defining the size and shape of selected sites (Brookes 1997), and multi-

objective site selection problems such as transmission tower siting (Krzanowski

and Raper 1999). While the fundamental process is relatively simple in concept

there are some quite challenging aspects to implementing a GA, most

predominantly in terms of problem representation. From a decision-maker point

of view they are may also be perceived as a ‘black box’ decision aid.


2.5 Technology platforms

The practical implementation of a decision science method requires a suitable

technology platform. This is provided by the mature technology of Geographical

Information Systems (GIS). A GIS provides a customisable software

environment with a set of tools for spatial data storage, manipulation and display.

Use of decision-making models in GIS results in a hybrid system commonly

referred to as a Spatial Decision Support System (SDSS). SDSS and GIS are

inextricably linked whereby a SDSS offers support in relation to a particular

spatial problem and GIS provides the highly evolved technical toolbox with

which to implement the SDSS (Crossland, Wynne et al. 1995).

Due to recent advances in computer hardware and software, and the growth of

information generation and distribution technologies such as remote sensing and

the Internet, SDSSs have arrived in both concept and application (Brail 2000).

There is much published literature on SDSSs and new applications will continue

to emerge as technology improves, and becomes more widely accepted.

Published applications cover a wide range of spatial problems from land use

planning to vehicle routing. A detailed description of GIS in spatial decision-

making is provided in Chapter 4.

2.6 Discussion

GMCLP’s are a demanding type of multicriteria decision problem that are

commonplace in infrastructure planning and environmental management, as well

as other more routine activities such as real estate investment. A particularly

challenging aspect of GMCLP’s is the inherent uncertainty involved. The

uncertainty most prominent in these problems is not easily classified by a

probability distribution, and is derived from questions such as; ‘whose opinion is

most important?’; ‘which criteria are most important?’; ‘how reliable are our

predictions of future scenarios?’; ‘how can we best classify qualitative

attributes?’; and more generally ‘what is the exact relationship between raw data


and site suitability?’. Having no precise answer to these questions makes it

impossible to consistently and precisely identify the best alternative(s). This

leads to an inability to accurately measure the quality of an algorithmically

derived solution. How can we ever be certain that an Airport has been placed in

the best position, or that an industrial facility has a negligible effect on

surrounding ecosystems, or that we have made the best choice when buying a

new home? In reality the utility of a selected site may not be known with any real

precision until some years after the decision has been made, and even then the

performance metrics used are subject to uncertainty and disagreement. It is little

wonder that solving such problems is described as a ‘surprisingly difficult task’

(Carlsson and Fuller 1996). However the significant impacts and expense

inherent in many site selection decisions, particularly those involving large-scale

infrastructure, demands that the methods employed in their solution be the best

available.

Decision Science has provided several formal methods for multicriteria decision

problems that are applicable to GMCLP’s. These range from simple Boolean

operations to advanced artificial intelligence and soft computing techniques. The

most widely applied methods come from the field of multicriteria evaluation, but

several shortcomings have been noted. Most important of these are the inability

to deal with uncertainty, inability to deal with a group environment, and the

perception by decision-makers that the methods are not user friendly. The more

advanced AI and soft computing techniques offer an ability to overcome some of

the shortcomings of MCE, but it is necessary to deploy them in a user friendly

way that avoids the ‘black box’ scenario. Until this is accomplished, the potential

of advanced techniques will only be tapped in specialised highly technical

applications.

An important characteristic of the many decision science techniques available is

their heterogeneous nature. The problem of choosing the best method for a

particular problem has largely been overlooked in current literature on spatial

decision-making, despite the fact that applying different techniques to the same

problem often results in different answers. Different methods tend to seek

different characteristics in a solution. For example some decision situations


require an answer with minimal risk of a bad outcome on any criteria, and non-

compensatory methods or an OWA approach best serves this objective. Other

problems may be best solved using a compensatory method or, if there is a group

environment, by looking for the best level of consensus between parties. As yet

there appears to be no hybrid method, whereby decision-makers directly choose

their desired set of characteristics, and the algorithm adjusts outcomes to fulfil

this requirement.

2.7 Conclusions

The initial review found that MCE is currently the dominant analytical technique

for the solution of multicriteria location problems. However several

shortcomings were noted. Most important of these are the inability to deal with

uncertainty, inability to deal with a group environment, and the perception by

decision-makers that current methods are not user friendly. The universally

accepted technology platform for the analysis of location problems was found to

be a GIS, coupled or fully integrated with decision-making models. Advanced

artificial intelligence and soft computing techniques offer an ability to overcome

some of the shortcomings of MCE, but it is necessary to deploy them in a user

friendly way in order to avoid the perception of a ‘black box’ scenario.

47Chapter 3 Approximate Reasoning

Chapter 3

AAPPPPRROOXXIIMMAATTEE RREEAASSOONNIINNGG

3.1 Introduction

Approximate Reasoning (AR) is a fuzzy logic based technique that can be useful

in decision-making. AR utilises fuzzy set methods to characterise and operate

upon imprecise inputs. In approximate reasoning linguistic inputs are quantified

as fuzzy numbers and manipulated with specialised fuzzy computation

techniques.

Utilising AR and linguistic variables enables users to overcome some difficulties

encountered with MCE analysis. Fuzzy numbers provide a convenient way to

represent linguistic uncertainty, and procedures for criteria standardisation can

benefit from a universal linguistic suitability scale.

This Chapter provides an introduction to the fundamentals of AR and it’s use in a

decision-making context. Existing AR methods for site selection are then

highlighted, and conclusions drawn.

3.2 Fuzzy logic

Approximate reasoning is based on the concepts of fuzzy logic introduced by

Lotfi A Zadeh, a professor of Electrical Engineering at the University of

California at Berkely (Zadeh 1965). The key insight of fuzzy logic is that an

emphasis on precise and detailed modelling leads to models that are hard to

understand due to their complexity. This led to the principle of incompatibility


(Zadeh 1973), which challenges the ability of conventional analysis techniques to

deal with complex systems:

‘Stated informally the essence of this principle is that as the complexity of a

system increases, our ability to make precise and yet significant statements about

its behaviour diminishes until a threshold is reached beyond which precision and

significance (or relevance) become mutually exclusive characteristics.’

The fundamental numerical structure of fuzzy logic was derived by reflecting

upon the fact that the cognitive skills possessed by human beings were able to

grasp the essential nature and characteristics of systems that proved too complex

to model successfully. Humans seemed to do this by using vague or fuzzy

linguistic expressions to describe the states and relationships inherent in the

system, and this led to the development of fuzzy logic methods to quantify these

approximate relationships. A simple way of characterising fuzzy logic is

therefore to say that it is logic of approximate reasoning (Zadeh 1975), and the

use of fuzzy logic is referred to by a number of terms including fuzzy systems,

fuzzy computation, and Approximate Reasoning. The specific unit used to

accomplish fuzzy logic operations is a fuzzy set.

3.2.1 Fuzzy sets

Fuzzy sets are an extension of classical set theory, based on the concept that

linguistic descriptions often have vague rather than sharp boundaries. All

concepts of classical set theory have their counterpart in fuzzy set theory,

however there are some concepts unique to fuzzy sets. Fuzzy sets reject the

requirement of classical sets that each object be either a member or non-member

of any given set. Thus, if X is a collection of objects denoted generically by x, a

fuzzy set A in X is a set of ordered pairs:

A = {(x, μA(x))|x∈X} (3.1)


Where: μA(x) is the membership function of x in A

The membership function, μA(x), maps x from the universe of discourse

consisting of all possible values of x, to the membership space [0,1] The

membership function may assume any value from 0 to 1 inclusive. If μA(x)

assumes only the values 0 and 1, A is crisp set, not a fuzzy set. Many different

types of membership functions are possible. Figure 3.1 shows a typical

membership function for a fuzzy set representing the linguistic expression

approximately three, where the universe of discourse is the set of all real

numbers.

0

1

0 1 2 3 4 5 6

Figure 3.1: Fuzzy membership function for the term ‘approximately three’

The fuzzy membership function shown in Figure 3 describes the possibility that

each real number fulfils the description ‘approximately three’. It is important to

note that the concept of possibility employed in fuzzy logic is conceptually

different from the more commonly recognised concept of probability, although

both assume values between zero and one (Ruspini and Mamdani 1998). The

relationship between these two areas has been the subject of much debate,

however an understanding of the complex issues raised by comparing fuzzy logic

to probability is not a pre-requisite for using and understanding fuzzy systems,

and consequently is not explored any further here.

μA(x)

x


3.2.2 Fuzzy numbers

An extension of the concept of a fuzzy set is that of a fuzzy number. A fuzzy

number M is a convex normalised fuzzy set. It is piecewise continuous and has a

peak value of 1, which occurs at least once. There is an infinite set of fuzzy

numbers.

There are several types of membership functions suitable to represent fuzzy

numbers. These range from gaussian and sigmoidal functions to linear parameter

based representations such as triangular or trapezoidal fuzzy numbers. A

standard trapezoidal fuzzy number T can be represented completely by a

quadruplet Tpz(a,b,α,β) where the interval [a,b] is the core, α and β are the left

and right bandwidths respectively, and [a-α, b+β] is the support, as shown in

Figure 3.2.

Support

μ

xa - α βb +a b

Figure 3.2: Trapezoidal Fuzzy Number Tpz(a,b,α,β)

(Bonissone 1982)

hoshiko

Rectangle

halla

This figure is not available online. Please consult the hardcopy thesis available from the QUT Library


Several authors consider that linear trapezoidal fuzzy numbers are suitable to

capture the vagueness of linguistic assessments, since it may be impossible and

unnecessary to use more complex representations e.g. (Tong and Bonissone

1980; Bonissone and Decker 1986; Delgado, Verdegay et al. 1992). This

representation also has the advantage of being able to represent crisp numbers or

sets (α = β = 0), as well as triangular fuzzy numbers (a = b). The support may

also be used as a measure of how vague or uncertain the term is in relation to the

base variable.

3.3 Approximate reasoning in multicriteria decision-making

Approximate reasoning has been introduced to decision-making largely because

of the ability of fuzzy sets to capture the vagueness of linguistic terms in

statements of natural language, that are often used by decision-makers. As

human reasoning is approximate rather than precise in nature, problems where

human knowledge is required as an input are described as being approximate

reasoning problems. In general terms an approximate reasoning problem is one

where information is inadequate to make precise, categorical statements about a

system (Ruspini and Mamdani 1998). The solution of such a problem takes the

form of a set of possibilities. An example of approximate reasoning is (Ruspini

and Mamdani 1998):

A1: Most engineers are clever

A2: Bill is an engineer

A3: It is very likely that Bill is clever

Where the fuzzy terms most and very likely could be represented by using fuzzy

numbers. Use of fuzzy numbers in this way forms the basis for the approximate

reasoning approach to decision analysis, which has also been termed the

‘linguistic approach’ (Zadeh 1976).


The linguistic approach to decision analysis relies on a systematic use of words

to characterize the values of variables, probabilities, relations, and truth-values of

assertions. The central concept is that of a linguistic variable whose values are

words or sentences, which serve as names of fuzzy subsets of a universe of

discourse. The linguistic approach represents a blend of quantitative and

qualitative analysis by using numbers to make the meaning of words more

precise.

A linguistic variable as defined by Zadeh (1976) is characterised by a quintuple

(X, T(X), U, G, M) where:-

X is the name of the variable. (e.g. Age)

T(X) is the term set which gives x it’s linguistic values. (e.g. Young, Not

Young,…Old, etc)

U is the universe of discourse. (e.g 0-150)

G is a syntactic rule which generates the terms in T(X).

M is a semantic rule which associates with each term, x, in T(X) its meaning,

M(X). The meaning is defined by a membership function μx which associates

each member of U with a degree of compatibility in X, within the interval [0,1].

It is possible to distinguish three main types of approximate reasoning methods

for multicriteria decision-making from the literature:

1. Fuzzy MCE

2. Fuzzy inference systems

3. Pairwise comparison methods

The three methods are described in the following sections.

3.3.1 Fuzzy MCE

Fuzzy MCE is an extension of the multiattribute decision-making process

described in Section 2.4.2. MCE in its simplest form reduces to the linear


weighted combination of an alternative’s criteria outcomes with the weighting

coefficient of each criterion as shown in Equation 2.7. Fuzzy approaches

generally process fuzzy data then dissolve fuzziness at the end of the procedure

to produce a set or sets of crisp data (Perny and Roy 1992).

The general case is well represented by the Baas and Kwakernaak (1977)

approach, where both an alternatives criterion outcome and criterion weighting

criteria are treated as fuzzy quantities. In order to determine the fuzzy evaluation

of alternative Ai based on fuzzy ratings and weights we first consider the

normalised version of the linear weighted combination function g mapping 2n

into (where is the set of real numbers and 2n represents 2n tuples of the set in n-dimensional Euclidian space) (Baas and Kwakernaak 1977).

Ignoring constraints:

(3.2)

where: z = (w1, w2, …., wn, r1, r2, …., rn)

Multiplication is carried out according to the minimum rule, so on the product

space 2n we define a membership function μzi, given by:

(3.3)

Where: )( jwj wμ is the membership function of criteria weights

)( kRik rμ is the membership function of criteria outcomes

Λ is the minimum operator

Through the mapping g: 2n → the fuzzy set z = ( 2n, μzi) induces a fuzzy set _R i = ( μR), with membership function:

∑∑

=

jj

jjj

w

rwzg

)()(

⎥⎦⎤

⎢⎣⎡ΛΛ⎥⎦

⎤⎢⎣⎡Λ=

==)()(

11 kRik

n

kjwj

n

jzi rw μμμ


)()( sup_

_

)(:

_zr zi

rzgziR

μμ=

= , ∈_r (3.4)

To simplify calculation Bonissone (1982) devised a parameter-based approach

using trapezoidal fuzzy numbers that simplifies computation and understanding.

The algebraic operations required in a weighted combination are performed as

shown in equations 3.2 and 3.3 (Bonissone 1982).

Addition:

(3.5)

Multiplication:

( ) ( )∏=

⊗=2

12222211111 ,,,,,,

ii baTpzbaTpzTpz βαβα

= ( )2112212112212121 ,,, ββββαααα ++−+ bbaabbaaTpz (3.6)

Where ⊗ is an approximated algebraic operator.

This approach enables rapid computation of weighted combinations of fuzzy

quantities, and gives a fuzzy set as its final output, and the fuzzy output is then

ranked or linguistically rated. There is a vast amount of literature on the ranking

of fuzzy numbers, and this may be partly due to the fact that no method seems to

give a satisfactory solution to every situation, and many methods produce

different rankings for the same problem (Fodor, Perny et al. 1998). Ribeiro

(1996) asserts the simple measure of taking the highest support as the best is a

sufficient ranking method for fuzzy multiple attribute decision making (Ribeiro

1996), and contends that using a crisp scale for criteria weights creates a simpler

set of outputs to rank and is more efficient in terms of computation without loss

of meaning. Bonissone (1982) suggested linguistic approximation as a useful

rating method. His procedure associates a label with a membership distribution

( ) ( )2222211111

2

1

,,,,,, βαβα baTpzbaTpzTpzi

i ⊕=∑=

( )21212121 ,,, ββαα ++++= bbaaTpz


on the basis of semantic similarity, via feature extraction and pattern recognition

techniques.

3.3.2 Fuzzy inference systems

Fuzzy inference systems are the most commonly used form of approximate

reasoning. The key element of a fuzzy inference system is a set of linguistic rules

derived from expert knowledge of the problem, which form the basis for the

inference engine to assess alternatives. Consequently, fuzzy inference systems

are also commonly referred to as fuzzy rule-based systems, and fuzzy expert

systems. The rules usually take the form of an if-then statement such as ‘if the

slope is steep then the site suitability is poor’. Where the linguistic terms ‘steep’

and ‘poor’ are defined by fuzzy sets on the universe of discourse consisting of all

possible slope and suitability values respectively.

In a standard Mamdani type inference system (Mamdani and Assilian 1975)

there are four steps involved in the process, as shown in Figure 3.3, which

illustrates an inference system for site selection with two criteria.


0

0.2

0.4

0.6

0.8

1

0 2 4 6 8 10 12

Slope (%)

Steep

0.7

Figure 3.3: Fuzzy inference system

The inference process starts with obtaining crisp values for the required inputs.

The next step is to take the inputs and determine the degree to which they belong

to each of the appropriate fuzzy sets via membership functions. This value is then

used to create a new output fuzzy set for each rule via implication, and these sets

are then aggregated into a single fuzzy set. The last step is to derive a crisp value

from the fuzzy set, which is commonly the centroid of the set. Different

inference systems often use different methods for implication and aggregation.

Building an inference system for a particular task is made easier by the fact that

the tools for their construction are available in off the shelf software packages

such as MATLAB. The major effort involved is in tailoring the if-then rules to

produce consistently accurate outputs.

IInnppuutt 11:: SSllooppee

IInnppuutt 22:: EElleevvaattiioonn

Rule 1: If slope is steep then

suitability is poor

Rule 2: If elevation is high then

suitability is good

Σ OOuuttppuutt:: SSuuiittaabbiilliittyy

Step 1:

Obtain inputs as

crisp numbers

Step 2:

Evaluate If Then

rules using fuzzy

reasoning

Step 3:

Aggregate outputs

Step 4:

De-fuzzify to

obtain a crisp

answer


3.3.3 Pairwise comparison methods

Pairwise comparison methods are based on the aggregation of fuzzy binary

preference relations. The preference analysis is carried out on the Cartesian

product A x A, where A = {a1,….,an}, the set of feasible alternatives. Preferences

are assessed for each criterion, and the resulting binary relations are aggregated

to obtain the best overall alternative (Perny and Roy 1992). Pairwise comparison

involves much computational effort when dealing with large numbers of

alternatives.

3.4 Use of approximate reasoning in location problems

Fuzzy sets have most prominently been applied to spatial decision-making as a

substitute for a utility function when standardising criteria. Use of fuzzy sets in

this way has been instrumental in designing many decision-making tools

(Stefanakis, Vazirgiannis et al. 1996; Jiang and Eastman 2000). Jiang and

Eastman (2000) argue that all standardised criteria belong to a general class

termed fuzzy measures. This approach provides a function for mapping attribute

values to a [0,1] utility scale as shown in Figure 3.4, producing a fuzzy set on the

universe of discourse derived from all possible attribute values. While fuzzy

measures may be used to generate inputs to multicriteria evaluation procedures,

the use of fuzzy sets in this way does not technically constitute an approximate

reasoning system. Also, the utility value given for each attribute value is crisp

not fuzzy, perhaps leading to a false sense of precision. There is also no rule

specifying that this function be continuous or convex, so not all such fuzzy sets

are fuzzy numbers.


00.10.20.30.40.50.60.70.80.9

1

0 2 4 6 8 10

Slope (%)

Util

ity

Figure 3.4: The variable slope as a fuzzy measure

Although fuzzy logic has been shown to have particular advantages in site

selection due to the continuous nature of many spatial variables (Hall, Wang et

al. 1992; Burrough and McDonnell 1998; Fisher 2000; Jiang and Eastman 2000),

there has been relatively little literature on fuzzy site selection methods in GIS.

Existing methods are based on either inference systems or fuzzy MCE, as

pairwise comparisons are unfeasible with the large numbers of alternatives

usually inherent in site selection.

Zeng (2001) uses fuzzy sets to codify expert knowledge and fuzzy selection

criteria such as distance from schools and other facilities in REGIS, a Real Estate

Geographical Information System. The fuzzy inference system used in REGIS

requires the specification of multiple If-Then statements such as ‘IF the SLOPE

is good OR excellent AND the aspect is good AND proximity to a school is good

THEN the location is excellent’ (Zeng and Zhou 2001). While inference systems

have been well proven in many applications, the need for such a comprehensive

set of statements is a burden when preparing for each new problem, making

REGIS a problem specific system.

A simple fuzzy MCE approach is to process linguistic labels directly using a

symbolic approach (Malczewski 2002). This method is based on the premise that

the set of linguistic terms is an ordered structure uniformly distributed on a scale.

Symbolic MAX and MIN operators are used, which simply choose the lowest or


highest ordered term in a comparison, rather than operate on membership

functions. The screening procedure first takes the MAX of the criterion outcome

and criterion weight and then takes the MIN of each criterion’s score from this

operation for each alternative. The final step is a binary screening, in which an

alternative scoring below a given level is assigned zero and all others are

assigned unity. This approach exhibits a loss of information, as no membership

functions are used, which is compensated for by the advantage of allowing

decision-makers to use linguistic terms instead of defining artificial cut-off

points.

Fuzzy MCE using membership functions has generally been used in site

selection outside GIS, where there are a small number of feasible alternatives to

consider. Liang and Wang (1991) use parameter based fuzzy numbers and

arithmetic operations in their method of facility site selection, however this

method has not been used in a Spatial Decision Support System. It has been

noted, however, that the method is suitable for application in a computerised

system (Liang and Wang 1991).

3.5 Conclusions

Most fuzzy methods used in spatial problems process crisp values obtained from

fuzzy membership functions and not the functions themselves, thereby losing the

information value of a fuzzy quantity. A fuzzy number possesses both a mean

value and a spread (support) that may be used to indicate the uncertainty of an

answer. However there is currently no robust way for decision-makers to input

their level of confidence in applying a particular linguistic label. The inflexibility

of an inference system once the rules are generated, and the processing power

required for pairwise comparison fuzzy methods, seem to indicate that a fuzzy

MCE approach to facility site selection is most appropriate.

Chapter 4 Spatial Decision Support Systems 61

Chapter 4

SSPPAATTIIAALL DDEECCIISSIIOONN SSUUPPPPOORRTT SSYYSSTTEEMMSS

4.1 Introduction

Spatial Decision Support Systems (SDSSs) are a kind of Decision Support

System (DSS) concerned with spatial problems such as site selection. SDSSs

offer support to decision-makers by transforming data into information via a

model. The technology platform upon which to operate is provided by the mature

technology of Geographical Information Systems (GIS). SDSS and GIS are

inextricably linked whereby a SDSS offers support in relation to a particular

spatial problem and GIS provides the highly evolved technical toolbox with

which to implement the SDSS.

This chapter provides an overview of SDSS fundamentals, the ways in which

SDSSs are used and developed, and a discussion of the main components of a

SDSS, before assessing the limitations of current systems. When discussing the

various components of a SDSS, the Dialog, Data, and Models (DDM) paradigm

is used as a means of classification. DDM specifies three main areas for

consideration; the dialog that provides an interface with the user, the data upon

which the system is based, and the models that transforms data into usable

information.


4.2 Overview

4.2.1 Decision support and expert systems

Although the term Decision Support System (DSS) has no universally accepted

definition, and is often relegated to the status of a buzzword as it has a low

communication value. There is a vast literature on DSSs and due to the sheer

magnitude of published work this research has focused on the components of

DSSs relevant to the problem at hand. Specific systems have been researched in a

discipline specific way, and all systems referenced in detail are spatial in nature

and the techniques that form their foundation are essential to the exposition of

the current research. A complete survey of the thousands of systems detailed in

the literature is beyond the scope of this review.

The basic concepts involved in DSS’ are simple, and were published in the early

1970s when such systems were defined as ‘interactive computer-based systems,

which help decision-makers utilize data and models to solve unstructured

problems’ (Scott-Morton 1971). In an unstructured problem human intuition is

frequently the basis for decision-making (Turban 1995), whereas structured

problems are solved using a set of strictly quantifiable procedures. Thus DSS’ do

not actually make decisions, but transform raw data into usable information

(Malczewski 1999). They enable an evaluation of technical aspects of choices,

providing an enhancement to a human decision-makers’ intuitive process, instead

of replacing it. In general a DSS does this by either optimising factors or

satisfying constraints as a complement to the limitations of human memory (Lu,

Yu et al. 2001). A more contemporary definition of a DSS would include other

attributes such as flexibility, adaptability, and a user-friendly interface, but the

fundamental purpose of a DSS is still to transform data into information via a

model.

Most DSS’, like most decision problems, will be unique, and instead of framing

another definition or attempting taxonomy, a comprehensive description of an


individual system may be obtained by examining its dimensions as measured on

appropriate scales. This approach is illustrated in Table 4.1.

The dimensions shown in Table 4.1 reflect possible qualities of a decision-

making situation. If a decision situation exhibits several of these dimensions then

it is a suitable candidate for a DSS (Pullar and Pettit 2000). Thus, spatial

decisions benefiting from a SDSS are usually not trivial, as they involve multiple

facets and require the agreement of several parties whose interests do not always

converge (Laaribi, Chevallier et al. 1996). Due to the cost of constructing a

SDSS, suitable decision situations will also have a high cost of failure

(Jankowski, Andrienko et al. 2001).

Another type of support system often confused with a DSS is an Expert System

(ES). Expert systems are based on applied artificial intelligence, and attempt to

reproduce or exceed the level of performance of a human expert in a specialised

problem area. The basic premise of an expert system is to transfer expertise from

the expert to a computer system. Expert systems tend to have a narrow focus, and

they are generally too inflexible for the multi-disciplinary nature of site selection

problems (Turban 1995).


Table 4.1: Dimensions of a DSS

modified from: (Pullar and Pettit 2000)

DDiimmeennssiioonn DDeessccrriippttiioonn SSccaalleess Data The nature of the data, is it

objective or subjective? • Physical • Statistical • Subjective

Type The type of data eg. Monetary value, habitat value.

• Quantitative • Indicative • Qualitative

Spatial Spatial data plus influences on neighbouring sites and patterns of distribution.

• Static • Neighbouring • Distributed

Analysis Type of analysis. • Query • Formulation • Simulation

Temporal Change based on predictions from the past or dynamic modelling.

• Static • Predictive • Dynamic

Model Classes of models. • Descriptive • Empirical • Deterministic

Reliability Techniques to asses sensitivity and risk.

• Certainty • Uncertainty • Risk

Objectives A single ultimate objective may be multi-faceted or there may be multiple competing objectives.

• Single • Multi-faceted • Multiple

Congruency Compatibility of objectives. • Complementary • Reconcilable • Conflicting

Involvement Level of collaboration and negotiation required in the decision-making process.

• Single • Consultative • Diverse group

4.2.2 Basic concepts of spatial decision support systems

Spatial Decision Support Systems (SDSSs) are a type of DSS focused on spatial

problems such as site selection. SDSSs require a technology platform upon

which to operate, and this is provided by the mature technology of Geographical

Information Systems (GIS). Due to recent advances in computer hardware and


GIS software, and the growth of information generation and distribution

technologies such as remote sensing and the Internet, SDSSs have arrived in both

concept and application (Brail 2000). There is much published literature on

SDSS applications, with new applications continually emerging as technology

improves and becomes more widely accepted. A small sample of published

applications includes:

• Urban land use planning (Bell, Dean et al. 2000; Quattrochi, Luvall et al.

2000)

• Rural land use planning (Matthews, Sibbald et al. 1999)

• City planning (Shiffer 1994)

• Transport infrastructure planning (Affum 1997)

• Resource management (Ghermay, Rochon et al. 2000)

• Hazardous waste siting (Pullar and Pettit 2000)

• Ecosystems management (Ji 1996; Bellamy and Lowes 1999)

• Environmental impact assessment (Klungboonkrong and Taylor 1998;

Colorni, Laniado et al. 1999)

• Air pollution monitoring (Flassak, Witt et al. 1995)

• Water pollution monitoring (Leon, Lam et al. 2000)

• Vehicle routing (Keenan 1998)

• Spatial allocation of educational resources (Malczewski and Jackson

2000)

• Telecommunications transmission tower siting (Krzanowski and Raper

1999)

SDSSs have been proven effective in practice as well as in theory. Laboratory

experiments conducted to investigate the effects of using a SDSS on decision-

maker performance found significant differences between solutions to a site

selection task developed by SDSS users and those developed by non-users. SDSS

users experienced shorter solution times and fewer errors (Crossland, Wynne et

al. 1995). However, despite the large number of published applications, there are

still many limitations to current systems. One of the major hurdles yet to be

crossed by SDSS designers is improving ease of use. It is common for highly


capable SDSSs to be used as simple visualisation tools, primarily due to

difficulties in use and understanding of the systems by strategic decision-makers

(Klosterman 2000). Consequently, one of the main objectives in SDSS design

should be to increase willingness to use a SDSS as many studies reveal that

millions of dollars have been wasted on unused SDSSs (Lu, Yu et al. 2001).

In summary it can be stated that a SDSS is a software system reliant on GIS

technology, built to help solve a non-trivial spatial problem. A SDSS does not

solve problems by itself, but will transform raw data into usable information. The

analytical capabilities of a SDSS should ideally be made accessible to users via a

simple intuitive interface, and effort should be made to increase the perceived

benefit of using the system, or there is a real risk that it will end up as an

expensive data visualisation tool. When properly designed and implemented

SDSSs have been proven to be an effective aid to spatial decision-making.

4.3 Components of a SDSS

When discussing the various components of a SDSS, the Dialog, Data, and

Models (DDM) paradigm is useful as a means of classification (Sprague and

Watson 1993). According to this approach the components of any DSS can be

broken down into three parts where the dialog is the interface between the user

and the system, data is the database that supports the system, and the models

provide the intelligence necessary for analysis. In a SDSS all components may

reside within a GIS.

4.3.1 Geographical Information Systems

As a SDSS is constructed to assist decision-makers with spatial decision

problems, it must be spatially explicit. To perform the operations involved in

spatial data storage, display and manipulation, a SDSS draws upon GIS

technology. SDSS and GIS are inextricably linked whereby a SDSS offers


support in relation to a particular spatial problem and GIS provides the highly

evolved technical toolbox with which to implement the SDSS (Crossland,

Wynne et al. 1995).

In the broadest possible sense a GIS is a set of software tools that allow for the

processing of spatial data into information, generally information tied explicitly

to, and used to make decisions about, some portion of the earth (Demers 2000).

GIS provides an ideal technology platform for development of Spatial Decision

Support Systems, as the purpose of most GIS is ultimately of a decision support

nature (Kraak 1999). There is no precise, absolutely agreed definition of a GIS.

However a GIS will generally consist of the following four subsystems.

1. A data input subsystem that collects and pre-processes spatial data from

various sources. This subsystem is also largely responsible for the

transformation of different types of spatial data.

2. A data storage and retrieval subsystem that organises the spatial data in a

manner that allows retrieval, updating, and editing.

3. A data manipulation and analysis subsystem that performs tasks on the data,

aggregates and disaggregates, estimates parameters and constraints, and

performs modelling functions.

4. A reporting subsystem that displays all or part of the database in tabular,

graphic, or map form.

Figure 4.1 shows the ESRI ArcGIS interface. Map layers are listed on the left

and a map screen illustrating the spatial characteristics of the variable are

displayed in the main window. The non-spatial data or ‘attributes’ of each spatial

entity are listed in a separate table, as shown at the bottom of the screen for

zoning regions.


Figure 4.1: ESRI ArcGIS

GIS has now matured to the point that Sui and Goodchild (2001) argue that the

rapid development of GIS technology renders traditionally instrumental views of

GIS inadequate to capture the essence of the technology. GIS may now be

thought of as a new media, communicating geographical information in digital

form by conveying aspects of the real world to a wide variety of non-technical

users, and thereby playing a part in their everyday lives (Sui and Goodchild

2001). In fact one of the fundamental challenges currently faced by GIS

developers is to find new applications that do justice to the available technology

(Hunter 1997).

4.3.2 Dialog

The most important components of any SDSS are those that involve interaction

with the user (Pereira and Duckstein 1993). These interface components are

referred to here as the dialog, and due to the situation specific nature of a SDSS,

the standard interface of most commercially available GIS lacks the necessary


tools for an effective dialog (Malczewski 1999). This leaves the SDSS developer

with the task of providing a suitable dialog to access the specific modelling

capabilities offered by the system.

The purpose of the dialog is to provide a simple and effective means for users to

enter and extract information, and this task is not limited to the creation of

colourful mapping outputs. It has been observed that the quality of decisions is

dependent on the quality and amount of relevant information available

throughout the entire decision making process (Shiffer 1992), and that achieving

consensus in a group environment requires that complete information is available

to all parties (Sharifi, Toorn et al. 2002). Creating a dialog that can be used in a

consistent manner by multiple parties throughout the decision process involves

two core requirements; simplicity and flexibility (Sprague and Watson 1993).

Dialogs employing tabular information display have been widely used in multi-

criteria DSS’ (see Voogd 1983; Massam 1988) but the large number of

alternatives involved in site selection decisions seems best communicated

through graphic outputs (Tomlin 1990). Visualisation is a critical aspect of a

multicriteria location analysis, and should be considered an integral part of

decision support approaches. SDSSs should allow the decision-maker to analyse

outcomes in an interactive, exploratory way to satisfy their individual priorities

(Malczewski, Pazner et al. 1997; Klosterman 2000). It may therefore be stated in

summary that a successful SDSS dialog requires the combination of simplicity

and flexibility with interactive visualisation techniques.

4.3.3 Data

The basis of any SDSS is its data. In fact the outputs from all SDSS operations

are reliant on the comprehensiveness and accuracy of the base data set. It is

therefore prudent to address certain fundamental issues before embarking on a

data collection exercise. These issues include (ESRI 2001):

• What is the intended use of the data?


• What are the specific geographic features needed?

• What attributes of (i.e. information about) these features are needed?

• What is the geographic extent of the area of interest?

• How current should the data be?

• Will periodic updates be needed?

• Is the data compatible with the software and hardware available to

end-users?

• What are the data licensing requirements?

After working through the fundamental issues, more technical considerations

should be addressed, and uncertainty is generally considered to play the decisive

role in the technical fitness of a dataset. The bottom line here is that the data used

should be of a high enough quality to be useful in the decision making process.

DeBruin and Bregt (2001) suggest the value of a given data set is proportional to

the reduction in uncertainty it brings to a chance node in a decision tree (DeBruin

and Bregt 2001). Their method of data evaluation requires a loss function

representing the cost of an incorrect judgement about the target outcome, and

probabilistic accuracy measures for the spatial data. Many authors have noted

that there is a need for these accuracy measures to be incorporated in metadata

accompanying spatial datasets. They should specify such things as estimates of

error variance, confidence intervals or probability distributions (Beard 1994;

Hunter and Goodchild 1995; Kyriakidis and Goodchild 1999). However these

measures are not always available, and in the absence of such information, an

uncertainty assessment may be carried out by comparing samples of the datasets

in consideration with reference data, although obtaining samples of the dataset

before purchase may also prove difficult.

4.3.3.1 Spatial data representation

SDSS data will be either spatial or non-spatial in nature, where spatial data

includes all data with an explicit spatial component, such as the location of a

house, or an ownership boundary. GIS offers three basic ways of representing


spatial data, either as points, lines or polygons. The data structure underlying

these entities can be either raster or vector. The raster system uses a grid of

equally sized cells, each with a corresponding attribute value. Vector structures

utilise points of zero area described by coordinate pairs. The points may be

joined to form a line or polygon, as shown in Table 4.2. Vector structures

generally offer greater positional accuracy whereas the raster format is more

suited to analysis operations such as MCE (Eastman, Jin et al. 1995). Most

modern GIS packages are capable of conversion between the two.

Table 4.2: GIS Spatial Entities in Vector and Raster

VVeeccttoorr RRaasstteerr

Point

Line

Polygon

4.3.3.2 Raster data and cell size

When using the raster format for modelling, a decision must be made about the

size of the raster cells. It is an intuitive and well-documented fact that the

accuracy of raster maps will decrease as cell size increases (Congalton 1997).

Wehde (1982) found that as cell size increased beyond twice the size of the

minimum mapping unit, mapping error would reach 100 percent for minimum

sized polygons. However as cell size becomes smaller other problems, such as

increase in data storage requirements, and a decrease in processing speed become

apparent. Over-estimation of the perimeter of polygons is another consequence

of increased resolution (Theobald 2000). Accuracy of the original data should be


the prime consideration when selecting cell size (Goonetilleke and Jenkins

1995).

4.3.3.3 Non-spatial data (Attribute Data)

Non-spatial data is usually data directly related to the spatial entity, such as

whether a house is one or two level, who owns the land, or the lot number.

Nonspatial data may be structured in several ways, ranging from a simple record

structure to more complex relational approaches. The flat-file, or record based

data model is the simplest data structure available, and consists of a spreadsheet

or rectangular database consisting of a row for each geographical entity, broken

into a fixed set of attribute fields for non spatial entries. These simple record

based models have largely been superseded by the relational model, which is a

set of tabular relations, where rows or ‘tuples’ for each entity have associated

attribute fields whose values are drawn from their specific domains. Each table in

a relational database contains at least one column similar to a column in other

tables, which acts as the relational join. Relational databases support traditional

set operations such as union, intersection and difference, as well as other specific

relational operations. Interaction with the database is usually performed via a

specialised language such as the Structured Query language (SQL), which has

become a de facto standard (Worboys 1999). A further development is that of

object based GIS data models, of which the fundamental concept is that of

encapsulation, which places a wrapper around an identifiable collection of data

(referred to as instance variables) and the code that operates upon it. The state of

an object is then determined by the value of the instance variables in its wrapper.

An example of a geographical object may be a country, defined by its border as a

polygon. The instance variables in this case may include: Name, Population,

Capital, etc.

While advanced data structures are invaluable in complex information storage,

querying and display, the flat file format is usually adequate for storage of inputs

to most multi-criteria decision-making processes. The values being stored and

processed in multicriteria evaluation tend to be numeric suitability scores that are


adequately represented by the flat file approach, and are easily accessed for the

necessary arithmetic operations.

4.3.4 Models

A model is a simplified abstraction of reality, used to simulate real actions. In a

SDSS models act as the intelligence of the system, providing a means to examine

the outcomes of potential courses of action, and implement decision analysis. In

decision analysis one distinguishes three features of the situation, the decision to

be made, events which can affect the result, and the result itself. Logical and

mathematical models are constructed to represent the relations between these

three features of the decision situation.

Although individual models will vary, multicriteria decision analysis in a SDSS

will generally fit within a well-established framework. Multi-criteria spatial

decision-making utilises the ability of GIS to represent a geographical area of

interest in multiple dimensions. Each dimension is represented as a map layer,

detailing the spatial variation of some variable (see figure 4.1). These map layers

are then processed to provide information on how that particular variable

(criterion) effects the overall decision. Each of the processed map layers is

referred to as a ‘suitability map’, and the process of standardising criteria to

create suitability maps is one of the major modelling challenges (Eastman, Jin et

al. 1995).

Most spatial multicriteria decision-making processes utilise raster data, whereby

every pixel represents an alternative in the site selection process, and

implementation of numerical techniques is straightforward. Suitability maps are

combined with decision-maker preferences according to the rules of the decision

model on a cell-by-cell basis, and an aggregated output generated. The resulting

processed map may be thought of as a locational decision matrix, with each

locational alternative displayed according to its calculated suitability score, as

illustrated in Figure 4.2. Note that in this example a single solution is shown

however multiple solutions may be found.


Suitability Maps

Criteria Standardisation

Weights Aggregation

C1 Wx

Σ

C2 Wy

C3 Wz

Locational Decision Matrix

Figure 4.2: Multi-criteria decision-making in GIS

Figure 4.2 shows a relatively simple MCE process, entirely implemented within

GIS, however some decision-making and data pre-processing models cannot be

implemented directly within the GIS environment. They require mechanisms to

link them with the facilities offered by the GIS. A well-referenced methodology,

based on an extensive body of literature, has been developed to describe the

different ways in which model linkages, or ‘coupling’ may be achieved. An

extensive review was conducted by Lilburne (1996), who noted that integration

has generally been classified in terms of data transfer and consistency in the

interface as summarised in Table 4.2.


OOnn ee

ff uull ll

ss yyss tt

ee mm

fully

in

tegr

ated

inte

grat

ed

inte

grat

ed

full

inte

grat

ion

inte

grat

ed

GGII SS

ee nn

hh aann cc

ee mmee nn

tt

parti

al li

nkag

e

Mod

ellin

g

stro

ng

clos

e co

uplin

g

stro

ngly

cou

pled

TT rraa nn

ss ppaa rr

ee nntt

func

tiona

l

SS aamm

ee uu ss

ee rr

ii nntt ee

rr ffaa cc

ee

Inte

grat

ed

DDii ff ff

ee rree nn

tt uu ss

ee rr

ii nntt ee

rr ffaa cc

ee

ad-h

oc

Shift

ing

SS aamm

ee dd aa

tt aa

ff ii llee ss

data

cen

tred

tight

tight

data

shar

ing

oper

atio

nal

loos

e co

uplin

g

tight

tight

DDii ff ff

ee rree nn

tt dd aa

tt aa ff ii

ll eess

appl

icat

ion

cent

red

ad-h

oc

loos

e

loos

e

data

tran

sfer

rudi

men

tary

loos

e

loos

e co

uplin

g

loos

e co

uplin

g

loos

e

loos

e

NNoo

ii nntt ee

gg rraa tt

ii oonn

isol

ated

isol

ated

AAuu tt

hh oorr

(Ow

en 1

993)

(Tim

, Miln

er e

t al.

1992

)

(Fed

ra 1

993)

(Stra

nd 1

991)

(Cho

u an

d D

ing

1992

)

(Hol

tfrer

ich

and

Yew

Kua

n 19

93)

(Bat

ty a

nd X

ie

1994

)

(Goo

dchi

ld,

Hai

ning

et a

l. 19

92)

(San

dhu

and

Trel

eave

n 19

96)

(Nye

rges

199

2)

(Stu

art a

nd S

tock

s 19

93)

Tab

le 4

.3: C

oupl

ing

Met

hods

mod

ified

from

: (Li

lbur

ne, 1

996)


A more succinct way of classifying coupling methods is offered by Scholten

(1997), who states that the problem of linking new spatial analytical functions to

a standard GIS can be facilitated using three major approaches.

1. Full integration of spatial analytic procedures within the GIS

2. Close coupling between statistical spatial data analysis software and GIS

3. Loose coupling where an independent spatial data analysis module relies on

a GIS for its input data, and for functions such as such as graphic display, via

the import and export of data in a common format.

The idea of coupling external models to a GIS is steadily losing favour, as it is

now commonly proposed that the ideal SDSS would be a fully integrated system

(Klosterman 2000). Due to recent improvements GIS software is now

sufficiently robust to develop discipline specific applications in a fully integrated

way (Pettit and Pullar 1999), and this paves the way for a widely recognised

trend within GIS applications to progress from an operational support system to a

fully integrated strategic decision support system (Scholten and LoCascio 1997).

New decision support applications may now be fully developed within the GIS

environment without the need for external coupling by utilising inbuilt

customisation tools. An example of such tools is ArcObjects, provided with the

ESRI ArcGIS software packages (ESRI 2001). ArcObjects provides an

infrastructure for application customisation and incorporates software

components that expose the full range of functionality in ArcGIS to developers.

ArcObjects enables elements such as complete maps (the map object) to be

simply customised and integrated with other elements of a model in a seamless

fashion. These new customisable GIS packages may pave the way for fully

integrated systems to become the norm.


4.4 Development and implementation

The construction of a DSS, spatial or otherwise, is a complicated process, and

there is no single best approach. A complex DSS requires a group of people to

build and maintain it, therefore suitable decision problems should also have a

high ‘decision equity’ (i.e. the cost of making a poor decision is high)

(Jankowski, Andrienko et al. 2001). Additionally they should be non trivial

(above common sense reasoning), perhaps involving multiple facets and

requiring the agreement of several parties whose interests do not always

converge (Laaribi, Chevallier et al. 1996). Turban (1995) recommends an 8-step

development process, consisting of:

1. Planning: Needs assessment, problem diagnosis & definition of objectives

2. Research: Review relevant information

3. Analysis: Develop a conceptual design, define resources & normative models

4. Design: Develop specifications for the four components, Interface, GIS

Database, Knowledge base (relational database), & model base.

5. Construction: Technical implementation of the design

6. Implementation: Testing, evaluation, demonstration, orientation, &

deployment

7. Maintenance: Ongoing support & documentation

8. Adaptation : Continually repeat the process to improve the system

When planning for a SDSS early and continuous involvement with management

and end users is advisable, as proper fit in an institutional context is essential for

successful DSS use. It is also prudent to address the following institutional issues

in the planning stage (Brown 1998):

• Should any kind of decision aiding procedure be used and if so what kind?

• Is the DSS’ role to enhance or justify decisions?

• Where in the organisation should the DSS be located?

• Who should be involved and how?


• Should use of the DSS be required / optional / reviewable?

• Should it lead to a preferred external action or predict option consequences?

• Should the preferred action be mandated / for information only / disclosed

publicly?

Once the decision to implement a SDSS has been made and institutional issues

have been addressed, a development strategy must be chosen and implemented.

A fundamental assumption in the development of many software systems is that

there is complete foreknowledge of requirements. This is often an unrealistic

assumption for SDSS design, where the problems analysed by system will be

constantly changing, and users will find new ways to utilise the SDSS. An

iterative development process, which involves close co-operation with the end

users of the system, is usually the most successful strategy. The process may be

termed evolutionary prototyping and is carried out in four steps (Turban 1995).

1. Selection of a core sub-problem.

2. Development of a functional system for that sub-problem using the eight

steps above.

3. Evaluation of the system, its structure and outcomes.

4. Implementation of refinements, modifications and expansion.

A prototyping strategy like the one above allows the system to grow with users

requirements, and provides the necessary flexibility to incorporate new

functionality.

4.5 Conclusions

SDSSs are a type of DSS that integrate GIS technology with decision-making

models to aid in the solution of spatial problems. The ideal SDSS would be both

flexible and user friendly, be fully integrated within a GIS software package,

provide real-time graphical interactivity and cater for group decision-making.


Decision-makers and DSS designers generally experience difficulty in predicting

exactly how a SDSS will be utilised, which makes the development of a SDSS a

challenging task. Early and continuous involvement of end users is advisable, as

is designating exactly how the system will fit in an institutional context, although

this may also change over time. Evolutionary strategies whereby a prototype

system is developed to focus on a sub-problem, and tailored from experience to

be more fully functional is an intuitively sound approach and one that is

consistently recommended in the literature. This approach tends towards creating

a more generic package that will function as a living system and continue to

evolve over time.

As the foundation of a SDSS GIS technology is a vital ingredient in the

development process. GIS is now a mature enough technology to develop fully

integrated systems using inbuilt customisation tools, as many off the shelf GIS

packages now provide access to spatial and analytical functions at a fine enough

level of granularity to be useful to SDSS developers. This trend may have

adverse consequences, as spatial models should ideally be based on hard science

and remain independent of particular software packages. It is also important that

SDSS design should not be driven by technology, but by problems and users,

however the ease with which new models can now be implemented by GIS

customisation, and the professional look of the result, make the fully integrated

system hard to resist.

The major hurdle facing developers is how to make systems that are simple and

easy to use. There is a general tendency towards ‘shallow use’ of SDSSs by real

world planners and decision-makers, which is largely the result of real or

perceived difficulty in using such systems. The need for simplicity should also

extend to the decision-makers understanding of how the system works.

Experience has shown that even if a system is user friendly, decision-makers are

unlikely to accept outputs that are generated by a ‘black-box’ that is beyond their

understanding and control. There are also few applications that fully cater for the

group environments that are the norm in infrastructure site selection. There is

also a real void of systems capable of accepting uncertainty assessments directly

from decision-makers. Allowing decision-makers to specify their uncertainty


levels when making judgements may lead to more honest assessments and

ultimately to better decisions.

79Chapter 5 Problem analysis and conceptual system design

Chapter 5

PPRROOBBLLEEMM AANNAALLYYSSIISS AANNDD

CCOONNCCEEPPTTUUAALL SSYYSSTTEEMM DDEESSIIGGNN

5.1 Introduction

The planning and research phase produced a set of objectives for system design,

and reviewed current literature to identify the obstacles and opportunities

relevant to achieving those objectives. Analysis and Conceptual System Design

entailed a critical analysis of this information to draw out the primary reasons for

limitations on current approaches to site selection. New approaches to mitigating

these limitations were then proposed at a conceptual level. The outputs from this

process were a set of desired system characteristics and the underlying concepts

that were proposed to deliver those characteristics.

5.2 Causes of current limitations on SDSSs for site selection

Literature has shown that current approaches to site selection employed in

Spatial Decision Support Systems (SDSSs) have exhibited practical limitations,

and it is postulated here that these limitations may be classified into two groups:

1. Limitations on effectiveness: Limitations on effectiveness stem from the

analytical inability of a SDSS to accurately model the problem.


2. Limitations on use: Limitations on use stem from real or perceived

difficulties in using or understanding the SDSS that prevent users from

utilising its full potential.

Both classes of limitation are capable of undermining the ability of a SDSS to aid

the decision process, and tend to have a circular relationship with each other.

Thus if a system is ineffective at modelling the characteristics of the decision

situation it becomes difficult to use, and if a system is difficult to use it is less

likely that decision-makers will employ it to model the situation effectively.

Looking further into the problem reveals that each of the two classes of

limitations exhibits a set of root causes. The main causes of the limitations on

current SDSSs identified in this research are illustrated in Figure 5.1.


Figure 5.1: Sources of current limitations on SDSSs

It follows that four primary capabilities are desirable in any new SDSS. Firstly a

simple and robust method to accept inputs from multiple decision-makers,

secondly an ability to handle uncertainty, thirdly an overall sense of simplicity in

ease of use and understanding, and lastly a sense of control. These four areas are

discussed more fully in the remainder of Section 5.2, and conceptual ideas to

mitigate these limitations are introduced in Section 5.3.

5.2.1 Multiple decision-makers

Most large-scale site selection decisions are made with the interests of multiple

stakeholder groups in mind, and groups may differ in their opinions in several

ways. It is common for decision-makers to differ in their weighting of the various

criteria, however they may also assess the same criterion differently or disagree

on an aspect of the decision-making process itself. A decision may be broken

down into a hierarchy of objectives and attributes (Saaty 1980), and exactly

which objective each attribute serves may also be a source of conflict. A method

is needed whereby each decision-makers opinion on weighting of criteria,

criterion assessment and uncertainty is considered.

LLiimmiittaattiioonnss oonn SSppaattiiaall DDeecciissiioonn SSuuppppoorrtt

SSyysstteemmss

Limitations on effectiveness

Limitations on use

Inability to deal with a

heterogeneous group

Inability to deal with

uncertainty

Lack of

simplicity

Lack of control


The difficulty in solving group problems lies in how to fully accept and combine

differing assessments from each member of the group. Deriving an aggregated

criterion weighting alone is inadequate if criterion assessments vary, and

assessing each criterion is complicated by conflicting opinions about which

attributes best represent which objective. As aggregation procedures tend to

produce a single measure of suitability, there is also a need to keep conflicts

visible after an aggregation, so they may be fully explored.

5.2.2 Uncertainty

Uncertainty in spatial decision-making has traditionally been considered in terms

of the physical processes and variables that form the basis of imprecise datasets

upon which decisions are made. Keeney and Raiffa (1976) define two basic

sources of uncertainty, the first being uncertainty about the source data used to

make a decision, and the second is uncertainty about future events which may

effect the decision outcome. These two sources of uncertainty can be further

classified into two types of uncertainty, being either probabilistic or fuzzy

(Malczewski 1999). Probabilistic uncertainty is described by a probability

distribution and fuzziness by fuzzy set theory. However these distinctions may

mean little to a strategic decision-maker making subjective value judgements,

who has little or no training in mathematical analysis.

Uncertainty in site selection is often due to uncertainty in a value judgement that

may be hard to directly associate with a physical process. In an unstructured

problem such as site selection, human intuition is frequently the basis for

decision-making (Turban 1995). The uncertainty inherent in intuitive value

judgements is quantified in the minds of decision-makers, and may bear no

measurable relationship to the stochastic uncertainty in source data or future

events, although it may be partially or wholly based on these factors. This highly

elusive and difficult to represent type of uncertainty plays a major role in human

reasoning. It is proposed here that this type of uncertainty be defined by the term

‘decision-maker uncertainty’ as the intuitive reasoning of the decision-maker is

the physical process most responsible for it. Decision-maker uncertainty arises

when the decision-makers themselves provide the only measure of the


relationship between source data and suitability by making statements such as ‘A

location less than fifty metres from the main road would be good’. It is extremely

common in decision problems with qualitative variables, and has been largely

overlooked in the literature on site selection.

The distinction between decision-maker uncertainty and data uncertainty is

important as data uncertainty such as known inaccuracies in distance

measurement or temporal fluctuations in demand has often been represented

successfully through the use of stochastic modelling techniques (Murray 2003).

Decision-maker uncertainty is somewhat different in nature. In the mind of a

decision-maker if a suitability assessment is uncertain they might simply lower

their assessment to compensate. If asked for an estimate of their confidence in

the assessment they will most probably answer with a linguistic term such as

‘very certain’ or ‘uncertain’. Underneath this perceived level of uncertainty there

is also the inherent vagueness of the assessment itself. The simplest way to

provide a suitability assessment is via a set of linguistic terms such as ‘good’,

‘bad’, ‘very bad’, etc, but inherent in these terms is an element of linguistic

uncertainty.

One may therefore postulate that when dealing with subjective linguistic

suitability assessments from decision-makers the overall level of decision-maker

uncertainty comprises two elements. Firstly the uncertainty quantified by the

decision-maker, for which the term ‘quantitative uncertainty’ is suggested.

Secondly there is the imprecision of the suitability term used by the decision-

maker, which is defined as ‘linguistic uncertainty’ in the literature eg. (Zadeh

1975; Herrera and Herrera-Viedma 2000). In keeping with the requirement of

simplicity it is therefore necessary to devise a method to directly incorporate

these two elements into an analysis. The problem is how to turn an assessment

such as ‘I am certain that location A is good with respect to Criterion 1’ into a

mathematical quantity that can be manipulated by an analytical model, whilst

retaining its information value.


5.2.3 Simplicity

The need for simplicity in Spatial Decision Support Systems is paramount, as has

been noted by many authors eg. (Crossland, Wynne et al. 1995; Brail 2000; Lu,

Yu et al. 2001). There are two kinds of simplicity to consider here. Firstly the

semantics of the analytical method used should be simple enough for users to

easily interact with the system. Secondly the mathematics of the technique

should be simple enough to be implemented in an algorithm capable of analysing

millions of discrete alternatives in a realistic timeframe. These two requirements

are often conflicting, as simplifying user interaction generally requires more

effort behind the scenes.

Simplicity in use and interaction has often been noted as a requirement in

Decision Support Systems (Turban 1995), and one of the main objectives in DSS

design should be to increase willingness to use DSS as many studies reveal that

millions of dollars have been wasted on unused DSS’ (Lu, Yu et al. 2001). In

fact while spatial decision support systems have been proven to increase

decision-maker effectiveness (Crossland, Wynne et al. 1995), few applications

are actually in use to support decision-makers in siting decisions (Maniezzo,

Mendes et al. 1998), and highly capable analytical systems are often used as

simple visualisation tools, primarily due to difficulties in use and understanding

of the systems by strategic decision-makers (Klosterman 2000).

Most GIS have very limited inbuilt capabilities for the simple integration of

decision-maker preferences with spatial data, and the use of MCE within GIS

provides a platform for this integration (Malczewski 1999). However there are

many stages in the MCE process that are complex or cumbersome to implement.

Among these are criteria rating and standardisation, selection of an appropriate

aggregation procedure, and differentiating amongst alternatives with a similar

overall rating.


5.2.4 Control

A key requirement of any decision support methodology is to deliver a sense of

control of the decision-making process to the decision-makers themselves.

Methods that operate as a ‘black box’, where users have little understanding or

control over outputs are unlikely to be fully embraced by decision-makers and

interest groups (O'Sullivan and Unwin 2003). However delivering a sense of

control to decision-makers requires that they be able to ‘look inside’ the

analytical processes employed and choose among the various analysis options

available. The choices made at this level are crucial to the overall results

obtained, as it is a well-noted fact that different decision-making methods often

produce different results. The most dominant influence on outputs is exerted by

the choice of aggregation procedure (Carver 1991; Heywood, Oliver et al. 1995).

5.3 A conceptual framework

The conceptual framework presented in this section is a blueprint for a new

Spatial Decision Support System for infrastructure site selection, at a conceptual

level. It proposes key methods and concepts for the new system, whilst leaving

the details of algorithm design to Chapter 6, and construction of the actual

system to Chapter 7.

5.3.1 Why use Approximate Reasoning?

It is a core hypothesis of this research that AR is a suitable basis for an

infrastructure site selection algorithm. AR techniques offer two immediate

advantages over crisp (non-fuzzy) methods. Firstly they enable uncertainty to be

factored into an analysis, as is discussed in more detail in Section 5.3.3. Secondly

they simplify that analysis as is described in Section 5.3.4. The use of AR in

decision-making is also backed up by a vast amount of literature and practical

experience, as was touched on in Chapter 3. But perhaps the most potent


argument for AR can be made at a more conceptual level. The fundamental

advantage of AR in decision-making is that it is a bridging technology that

enables human beings to more effectively interact with an analytical model.

The advent of powerful modelling capabilities, made possible by the digital

computer, has brought about an enormous increase in our ability to precisely

simulate complex real world systems. Engineers and scientists value this

precision in data, however the human mind has a finite ability to resolve detail

and store information. It uses words as labels for imprecise bundles, also termed

fuzzy granules, as a means to cope with complex problems. This mismatch of

precision between human and computer produces a decrease in our ability to

make precise and significant statements about models, as they grow more

complex.

In general there may be distinguished three distinct entities related to modelling

the physical world:

1. The physical process to be modelled.

2. The abstract (usually mathematical) representation of that process, termed the

model.

3. Human understanding of both the physical process and the mathematical

model, which makes construction and application of the model possible.

AR enhances the bridge between mathematical models and the associated

physical reality by facilitating a better human understanding of the modelling

process. Fuzzy methods are capable of capturing the vagueness of linguistic

terms in statements of natural language. This in turn provides greater capability

to model systems through human commonsense (approximate) reasoning and

creates a more useful aid to decision-making (Klir and Yuan 1999).


5.3.2 Catering for multiple decision-makers

It is proposed here that to adequately deal with the inevitable conflicts that will

arise from a heterogeneous group of decision-makers requires three elements.

1. A criterion weighting approach that applies to attributes and not objectives.

2. A method to accept differences of opinion in both criteria weighting and

rating.

3. The ability to identify conflicts in the post aggregation data exploration phase

of the decision-making process.

The first step in overcoming the three hurdles facing a heterogeneous group is to

ask decision-makers to assess and weight attributes directly. In this way the

attributes can be weighted and assessed with respect to the objective foremost in

the mind of each decision-maker, thereby bypassing the conflict that may arise in

trying to reach a consensus on which attributes best represent each objective.

To allow the processing of differences of opinion in weighting and rating, each

alternatives criterion outcomes and criterion weights are weighted and combined

using a Relevance Matrix (RM). The format for a RM is shown in Figure 5.2.

R = ⎥⎥⎥

⎦

⎤

⎢⎢⎢

⎣

⎡

JKJ

K

RR

RR

..........

..........

1

111

M

Figure 5.2: Relevance matrix

Deriving the relevance matrix is ideally achieved via consensus, and should be

based on the competency of a decision-maker to make assessments relating to

each criterion. However it may also be derived via a non-weighted averaging of

each decision-maker’s assessments of the competencies of others in the group.

Each criterions relevance values are normalised so that a criterion does not gain

extra importance based on solely relevance values. The values defined in the

The relevance matrix describes the relevance of the kth decision-makers opinion with respect to attribute j. (values are scaled after input so each criterions relevance values sum to 1)


relevance matrix are then used in a double weighted MCE aggregation of fuzzy

suitability scores.

Double weighting has been used previously to add extra weight to lower criterion

outcomes in hybrid Ordered Weighted Averaging techniques (Jiang and Eastman

2000). It is proposed that a criterion assessment from each decision-maker is

weighted according to the decision-makers preference and relevance as shown in

Equation 5.1.

∑∑= =

××=J

j

K

kjkjkijki WR

1 1

OS | i = 1…I (5.1)

N.B. Fuzzy quantities are shown in bold type

Where:

iS is the suitability of alternative i.

ijkO is the criteria outcome for alternative i with relation to criterion j and

decision-maker k, including quantitative and linguistic uncertainty.

jkR is the relevance of decision-maker k’s opinion with respect to criterion j.

jkW is the weight assigned to criterion j by decision-maker k

The aggregation output should be a fuzzy number representative of each

alternative’s overall compensatory suitability and uncertainty.

The task now remains to extract conflicts in the post aggregation data exploration

phase. To accomplish this it is proposed to extract an extra parameter

representative of conflict. The complete set of parameters proposed is discussed

fully in Section 5.3.5.


5.3.3 Handling Uncertainty

Although the use of fuzzy numbers to model linguistic uncertainty is common,

there is no universal method to derive the fuzzy numbers, or adjust the fuzzy

number to include the extra dimension of the quantitative uncertainty level

placed on the linguistic assessment. The problem of matching a fuzzy number

with a linguistic label dates back to the genesis of the linguistic approach, and is

beyond the scope of this thesis. However the ability to adjust a fuzzy number in

line with a decision-maker uncertainty assessment is a simpler task.

Incorporating decision-maker uncertainty into an analysis in this way offers real

advantages, as this type of uncertainty assessment can always be obtained

regardless of the level of knowledge about source data.

Linguistic uncertainty has long been represented using fuzzy set theory. Usually

the fuzzy set is reduced to a parametric form such as a triangular or trapezoidal

fuzzy number. Figure 5.3 shows a triangular fuzzy number (TFN) which

represents a linguistic term in three parameters (a,b,c).

0

0.2

0.4

0.6

0.8

1

0 0.2 0.4 0.6 0.8 1

a c

b

Suitability

μ 'OK'

Figure 5.3: The suitability term ‘OK’ as TFN(0.3,0.5,0.7)


A method is now needed to encapsulate the level of quantitative uncertainty

expressed when the term is used in the form ‘I am very certain that this

alternative is OK with respect to criterion 3’. This may be accomplished using

the relatively new concept of a type-2 fuzzy set and its footprint of uncertainty

(FOU). A type-1 fuzzy set has a crisp membership function where each point on

the universe of discourse (x-axis) has a crisp membership value μ on the y-axis.

A type-2 fuzzy set possesses a secondary membership function (2MF) drawn

along a third axis that describes the relationship between the universe of

discourse and the primary membership function (Mendell and John 2002). The

2MF exists within a footprint of uncertainty (FOU). An example of a FOU is


0

0.2

0.4

0.6

0.8

1

Suitability

μ

Primary Term FOU

Figure 5.4: Footprint of uncertainty

In this case the FOU of the suitability term is defined by moving vertices a and c

of the primary TFN outwards to the boundary of [0,1]. This provides a means to

superimpose quantitative uncertainty on the primary TFN by varying the

expected value of the 2MF with the quantitative uncertainty assessment. In this

case primary vertices a and c are reallocated to a different point for different

quantitative uncertainty assessments as shown in Figure 5.5.


0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Suitability

μ

Totally certain

Certain

Moderately certain

Uncertain

Totally uncertain

'OK'

Figure 5.5: How a quantitative uncertainty assessment affects the primary

MF

It is proposed that by using TFN’s and the type-2 concepts described above it is

possible to represent linguistic suitability and uncertainty assessments as a

simple, robust mathematical quantity, capable of being manipulated by an

analytical model whilst retaining the original information value. The scaled TFN

will provide the basic unit to be manipulated in the algorithm. All that remains is

to devise a mathematical method of deriving the new a and c values from the

original primary MF and a linguistic uncertainty assessment. The specifics are

left for Chapter 6.

5.3.4 Creating simplicity

Simplicity in use and interaction is largely a product of choosing an AR

technique. Decision-makers benefit from a universal linguistic suitability scale

that greatly simplifies criteria standardisation. AR enables the use of words as the

basis for interaction with the system, both in terms of input and feedback.

Another way to increase ease of use is to produce a fully integrated system,

utilising a standard GIS interface. GIS is a mature technology and the methods

standard GIS packages employ to view and interact with spatial information are

the result of an ongoing process of refinement dating back to the 1960s. Instead


of trying to reinvent the wheel and create an entirely new interface, it is more

efficient to create a set of tools to enhance existing GIS functionality where

necessary, whilst retaining the highly effective aspects of the standard interface.

This may be achieved by incorporating tools into a toolbar that integrates

seamlessly with existing functionality.

The mathematical simplicity required to enable the analysis of large numbers of

alternatives in real-time is provided by adhering to three constraints:

1. Utilising parameter-based fuzzy numbers, thereby avoiding the extra burden

of more complex membership functions.

2. Manipulation of the TFN’s by arithmetic operations, easily performed by GIS

software.

3. Use of a scoring function to de-fuzzify outputs, thereby avoiding the use of

pairwise comparisons to rank alternatives, as the number of calculations

required to do this becomes unwieldy with large numbers of alternatives.

Details of the fuzzy algorithm are given in Chapter 6.

5.3.5 Giving decision-makers control

It is proposed here that in order to successfully deliver control to decision-makers

an easily understandable method to choose between different aggregation

procedures is required. This may be achieved by generating four descriptive

parameters for each alternative that are indicative of the qualities sought by

differing aggregation procedures, and independent of the problem domain.

Decision-makers then decide which of these parameters are most important to the

problem at hand. The parameters are as follows:

1. Utility: Utility is a measure of an alternatives fulfilment of all evaluation

criteria in a compensatory way. It is calculated via a weighted summation of

all criterion outcomes. A good solution requires good utility.

2. Certainty: Certainty is a measure of how predictable the outcome for a

particular alternative is. A good solution is one with a high level of certainty.


3. Safety: Safety is a measure proportional to the lowest criterion outcomes.

Alternatives with poor outcomes on some criteria may rate well in terms of

utility but will be unsafe or ‘risky’. A good solution is a safe solution.

4. Consensus: Consensus requires that all parties agree on the various aspects

of an alternative. Alternatives that are rated similarly on all criteria by all

decision-makers in the group exhibit a high level of consensus. A good

solution requires consensus.

None of the four parameters are sufficient to guarantee a good solution in

isolation. However by weighting and combining them decision-makers take

control of the process, and find solutions that best satisfy the dynamics of each

problem. Moreover by breaking down each solution into easily understandable

quantities, the mystery of what happens during analysis is lessened in the eyes of

users.

The four parameters should constitute an integral part of the interactive data

exploration and visualisation phase of the decision-making process. Using a

graphical point and click interface, decision-makers should be able to explore

each alternative site by receiving linguistic feed back on the four parameters,

plus an overall aggregated rating derived from combining them.

5.4 Conclusions

It was proposed that limitations on current SDSSs are derived from an inability

to deal with multiple conflicting parties, an inability to handle uncertainty, a lack

of simplicity in use and interaction and not delivering enough control to decision-

makers. This chapter has provided a conceptual blueprint for algorithm design

and system construction by outlining the desired characteristics of the system. It

was found that the system should possess the following characteristics:

• The ability to accept inputs from a heterogeneous group of decision-

makers, independently weighting and rating multiple attributes.


• An approximate reasoning algorithm based on a fuzzy MCE aggregation

of parameter-based fuzzy numbers that encapsulate linguistic suitability

and uncertainty assessments.

• The algorithm should utilise arithmetic operators for aggregation and a

scoring function for de-fuzzification to minimise calculation time and

enable real-time interactivity.

• The system should be fully integrated into existing GIS software.

• Linguistic outputs should be a set of descriptive parameters that give

decision-makers the ability to choose the characteristics of a solution that

are most appropriate to their specific problem, thereby enabling them to

gain control over the properties maximised during aggregation.

Chapter 6 Algorithm design 95

Chapter 6

AALLGGOORRIITTHHMM DDEESSIIGGNN

6.1 Introduction

The previous chapter has highlighted several limitations on current approaches to

site selection, and provided a conceptual blueprint for mitigating those

limitations. Algorithm design consisted of the formal implementation of those

conceptual ideas that specifically relate to the decision-making model. The

implementation takes the form of a new Approximate Reasoning Algorithm for

Infrastructure Site Selection (ARAISS). While it is not the contention of this

research that it is possible to develop a perfect analytical model for the solution

of all Infrastructure Site Selection Problems, ARAISS implements several

concepts that offer an improvement over current methodologies.

The core capabilities of ARAISS are its use of approximate reasoning to handle

uncertainty, its multiple decision-maker capability, its simplicity, and the way it

hands over control to decision-makers. ARAISS is described in detail in Section

6.2, and the results from MATLAB testing of the algorithm are given in Section

6.3. Conclusions are then drawn.

6.2 ARAISS

One of the principal outcomes of this research is the ARAISS algorithm

described in this section. ARAISS is a new and unique approach to infrastructure

site selection, which is loosely based on fuzzy multiattribute utility theory

(Ribeiro 1996). ARAISS is specifically targeted to an audience of strategic


decision-makers locating a new facility. It was designed to accommodate

qualitative and quantitative variables, and offers a means to perform an initial

analysis based on the issues of foremost importance in the minds of stakeholders.

As such it is a generic Spatial Decision Support algorithm suitable for the first

stage of a site selection process. A more comprehensive follow up assessment

incorporating a more detailed analysis is envisaged as a means to further validate

recommendations made from the ARAISS process.

6.2.1 Framework

Figure 6.1 shows the general framework for ARAISS. It is a two-phase

procedure where the final location is sought via an iterative process of reducing

alternatives. In Phase 1, decision-makers first define the problem, and then a

constraint analysis is performed to exclude totally unfeasible alternatives. A set

of linguistic suitability terms to be used when rating the various criteria is then

defined. Each decision-maker then contributes their preferences for criterion

weighting and rating, and this information is combined with decision-maker

relevance values in an aggregation. The aggregation derives output parameters

for the Utility, Certainty, Safety and Consensus of each alternative.

Phase 2 involves exploration and reduction of alternatives. Decision-maker

preferences for minimum acceptable parameter values, and parameter weights

are sought. They provide a means to rate and rank alternatives in terms of their

overall suitability, and thereby reduce the number of alternatives under

consideration by consensus. The desired outcome of this process is the selection

of a site or sites, which conform to the strategic needs of all decision-makers.

Once the strategic analysis has been performed, it may be necessary to analyse

tactical and operational issues using a more specific modelling procedure, and to

consider micro-placement issues such as footprint and orientation before making

a comprehensive decision. This last non-strategic phase is beyond the scope of

this thesis.


Figure 6.1: ARAISS Framework

Define problem

Define decision-maker relevance values

Perform compensatory aggregation function

Identify feasible alternative sites for analysis

Choose Decision-makers

Identify Utility, Certainty, Safety and Consensus

Final site selection

Define and rate criteria (factors)

Start

Define linguistic terms

Tactical and operational assessment

Micro-placement

Identify constraints

Explore and reduce alternatives

Iterate

Phase 2

Perform adjusted aggregation

Define criterion weights

Review inputs

Phase 1


6.2.2 Notation

Notation and terminology for site selection analysis has been inconsistent in the

literature. For example compare the notation of Malczewski (1995) to that of

Eastman (1995). To avoid confusion the following notation is used consistently

throughout this thesis:

A = {A1,A2,……AI} The set of I feasible alternatives

C = {C1,C2,……CJ} The set of J criteria (factors)

D = {D1,D2,……DK} The set of K decision-makers

W = ⎥⎥⎥

⎦

⎤

⎢⎢⎢

⎣

⎡

JKJ

K

WW

WW

..........

..........

1

111

M

R = ⎥⎥⎥

⎦

⎤

⎢⎢⎢

⎣

⎡

JKJ

K

RR

RR

..........

..........

1

111

M

Ok =

⎥⎥⎥

⎦

⎤

⎢⎢⎢

⎣

⎡

JIkkJ

Ikk

OO

OO

..........

..........

1

111

M

The overall suitability of alternative i (Si) is some function of each decision-

makers preferences for criterion outcomes and weights, combined with their

relevance to each criterion.

6.2.3 Linguistic term sets

The natural language approach to decision analysis relies on a systematic use of

words to characterize the values of variables, probabilities, relations, and truth-

values of assertions. The central concept is that of a linguistic variable whose

values are words or sentences, which serve as the names of fuzzy subsets of a

universe of discourse. The linguistic approach represents a blend of quantitative

The matrix of criterion outcomes for alternative i and criterion j, based on decision-maker k’s suitability and uncertainty assessments.

The matrix of relevance values specifying the relevance of the kth decision-makers opinion with respect to criterion j. (values are scaled after input so each criterions relevance values sum to 1)

The matrix of criterion weights specifying the kth decision-makers opinion of the weighting of criterion j.


and qualitative analysis by using numbers to make the meaning of words more

precise (Zadeh 1976).

A linguistic variable is generally characterised by the quintuple (X, T(X), D, Y,

M) where:-

X is the name of the variable. (e.g. Age)

T(X) is the term set which gives x it’s linguistic values. (e.g. Young, Not

Young,…Old, etc)

D is the universe of discourse. (e.g 0-150)

Y is a syntactic rule which generates the terms in T(X).

M is a semantic rule which associates with each term, x, in T(X) its

meaning, M(X). The meaning is defined by a membership function μ(x)

that associates each member of D with a degree of compatibility in x,

within the interval [0,1].

ARAISS uses four term sets:

T(S) site suitability terms

T(W) terms for weighting of criteria and decision-maker relevance

T(U) terms describing the level of uncertainty

T(G) terms for generating new suitability terms in T(S)

Generation of linguistic term sets involves two primary considerations. The first

is the selection of a grammar, i.e. the cardinality of the term set and syntactic

labelling as defined by a syntactic rule. The second is how to define a semantic

for each term, which in this case will take the form of a triangular fuzzy number

(TFN) or crisp number, via a semantic rule.

On the issue of grammar the first consideration is cardinality i.e. the number of

terms in the set. The term set should be small enough to be manageable and not

impose unnecessary precision, yet comprehensively cover adequate

discrimination of assessments. Typical values of cardinality are odd numbers,


usually close to 7, with the middle term centred on a utility of 0.5 (Herrera and

Herrera-Viedma 2000). The labels themselves can be generated using either a

context-free grammar consisting of pre-existing primary terms, which are

expanded upon using the syntactic rule G, or by means of an ordered structure.

However both approaches imply limitations on flexibility and do not allow

decision-makers to interactively generate new terms. For example, when

considering the importance of a criterion, where does one place the word

‘significant’ within a set such as {…., unimportant, moderately important,

important, …..}.

ARAISS uses a hybrid method, whereby a set of five primary suitability terms

P(S) based on an ordered structure, may be enhanced by the addition of a

maximum of four new user specified terms N(S). The term generation term set

T(G) enables users to interactively generate additional suitability terms via a

linguistic comparison to existing terms. All term sets other than suitability are

composed only of primary terms, and are not subject to additions. Primary terms

are as follows:

Primary suitability terms: P(S) = {s0 = totally unsuitable, s1 = bad, s2 =

indifferent, s3 = good, s4 = perfect}

Primary weighting terms: P(W) = T(W) = {w0 = irrelevant, w1 = unimportant,

w2 = moderately important, w3 = important, w4 = critical}

Primary uncertainty terms: P(U) = T(U) = {u0 = very certain, u1 = certain, u2 =

moderately certain, u3 = uncertain, u4 = very uncertain}

Generation terms: P(G) = T(G) = {g0 = zero, g1 = very small, g2 = small, g3 =

medium, g4 = large, g5 = very large}

The advantages of approaching the issue of term set generation in this way are

threefold. Firstly the cardinality of the suitability term set is limited to

manageable values between five and nine, secondly it provides a solid foundation

structure that broadly covers all possible values, and lastly it enables decision-


makers to include terms with a specific contextual meaning. A consequence is

that semantic definition is a little more complex, although this is transparent to

the user.

6.2.3.1 Semantic definition

There are a number of options for semantic definition of a linguistic term.

Approaches used range from assuming a symmetrical distribution of terms on the

given universe of discourse and allocating each term a subdomain within it

(Yager 1995), to more complex approaches whereby the term set can be non-

symmetrical and subdomains are further characterized by fuzzy membership

functions (Herrera, Herrera-Viedma et al. 1996).

ARAISS utilizes two types of semantic definitions. Suitability terms are

characterized here as triangular fuzzy numbers, whereas all other terms are

allocated crisp values. The advantage sought in combining fuzzy and non-fuzzy

semantic values is to preserve the extra information conveyed by a fuzzy

number, without falling victim to the problems of multiplication and division of

two TFN’s. The multiplication and division operators proposed for use on TFN’s

are only approximations (Bonissone 1982), leading to, among others, the

problem of an unwarranted increase in the support. Proposed semantic definition

of primary terms is described in Table 6.1 where:

T(S) are triangular fuzzy numbers on [0,1]

T(W) are crisp numbers on [0,1]

T(U) are integers 0..N-1 where N is the number of uniformly distributed,

ordinal uncertainty terms

T(G) are crisp numbers on [0,1]


Table 6.1: Semantic definition of primary terms

Suitability (as a TFN)

Weighting Uncertainty Term generation

Totally unsuitable

(0,0,0)

Irrelevant

.1

Very Certain

0

Zero**

Very Small 0 .1

Bad (0,.2,.4) Unimportant .3 Certain 1 Small .3 Indifferent (.3,.5,.7) Moderately

Important .5 Moderately

Certain 2

Moderate .5

Good (.6,.8,1) Important .7 Uncertain 3 Large .7 Perfect (1,1,1) Critical 1* Very

Uncertain 4 Very Large .9

* This is the static value & may also be dynamic - see section 6.2.4

** The term g0 = zero allows re-labeling of any term.

Two operations are performed on the linguistic suitability terms prior to

aggregation. The first is the generation of new suitability terms to enable

decision-makers to utilize context specific words and increase the resolution of

the set. Secondly, uncertainty scaling of suitability terms provides a means to

represent quantitative uncertainty separate from the linguistic uncertainty of the

suitability term.

6.2.3.2 Term generation

Term generation is necessary to give decision-makers the use of context specific

words, and to increase the resolution of the term set. It is facilitated by a hedging

procedure that enables the addition of up to four new suitability terms to a set of

around five primary terms, whilst still preserving the ordinal quality of the set.

The first step in this process is choosing the term that will immediately precede

the new term in utility. The semantic value of the new term will take the form of

a TFN, with its centre of gravity situated between this term and the next term

above. Equation 6.1 defines the breakpoints of the new term:

(6.1)

Where:

gxxxx )(' −+− −+=


'x is the value of the new breakpoint

−x is the value of the breakpoint in the lower term

+x is the value of the corresponding breakpoint in the higher term

g is the term generation term

Figure 6.2 illustrates the new suitability term ‘Acceptable’ which has been

generated as a ‘large’ increase upon ‘Indifferent’.

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Suitability

μ

Totally UnsuitableBadIndifferentGoodPerfectAcceptable

Figure 6.2: Term generation

6.2.3.3 Uncertainty scaling

It was proposed in Chapter 5 that there are two types of uncertainty inherent in

decision-maker suitability assessments, linguistic and quantitative. In ARAISS

linguistic uncertainty is represented by the fuzziness of the primary suitability

term (TFN(a,b,c)), whereas quantitative uncertainty is represented using the

concept of a type-2 fuzzy set and its footprint of uncertainty (FOU). The FOU

has also been used to model other types of uncertainties such as ambiguity,

nonspecificity or strife (Mendell and John 2002). The FOU of the suitability term

is defined here by moving vertices a and c of the primary TFN outwards to the

boundary of [0,1] as shown in Figure 5.4. Primary vertices a and c are


reallocated according to the uncertainty assessment as shown in Figure 5.5, and

these points are defined as follows:

(6.2)

(6.3)

(6.4)

Where:

Supp is the width of the support of the new primary membership

n is the term number chosen by the decision-maker from a set of N-1

uniformly distributed uncertainty terms

The scaled term now envelops both suitability and quantitative uncertainty

information in a type-1 fuzzy number, enabling the use of relatively simple type-

1 processing procedures whilst increasing information value.

acnN

acSupp −+⎟⎠⎞

⎜⎝⎛

−+−

=1

1

⎪⎪⎪

⎩

⎪⎪⎪

⎨

⎧

−

<−−

<

=

Otherwise 2

2 1 if 1

2b if 0

Suppb

SuppbSupp

Supp

a

⎪⎪⎪

⎩

⎪⎪⎪

⎨

⎧

+

<−

<

=

Otherwise 2

2 1 if 1

2 if

Suppb

Suppb

SuppbSupp

c


6.2.4 Dynamic weighting

The representation of importance assessments via linguistic terms with static

semantic definitions has inherent limitations. These stem from the fact that some

classifications of importance are dependant on other values of importance in the

set. For example, a decision-maker may wish a criterion to be weighted so

heavily as to outweigh all other criteria. This is defined here as critical

importance, and it’s semantic definition requires a dynamic weighting approach.

When using critical importance it is necessary to generate the semantic values of

weights as a function of the other weighting terms in the set, as shown in

Equation 6.5. Critical importance is not equivalent to evaluating an alternative on

the basis of that criterion alone, as alternatives with similar ratings for the critical

criterion are further classified according to other criterion outcomes. ARAISS

utilises the concept of critical importance in criterion weighting (Wjk j=1..J),

decision-maker relevance values (Rjk k=1..K), and in weighting the output

parameters defined in Section 6.2.6. Critical importance can only be

implemented once in any set of weighting factors.

(6.5)

Where:

cW is the critical weighting coefficient

nW is the static weight of factor n (the static weight of critical is 1)

N is the number of weights in the set

The final step is to normalize all weights in the set by dividing by Wc.

∑=

=N

nnc WNW

1

2


6.2.5 Generating suitability values

Site selection procedures in GIS require the creation of a set of suitability maps,

which were introduced in Chapter 4. A suitability map is a representation of how

a given criterion varies over space. For the purposes of algorithm design, two

classes of suitability maps must be considered: discrete and continuous. Discrete

maps are those based on categorical variables such as land use or zoning. Such

variables are relatively simple for decision-makers to directly translate into

linguistic suitability values. Continuous variables such as slope, proximity or

elevation may also be discretely categorized using cut-off values, however this

approach loses valuable detail and there is usually ambiguity and imprecision in

defining such cut-off values (Malczewski 2002). A better approach is to

represent continuous criteria by using a utility function (Jiang and Eastman

2000).

However whilst utility functions may be used to generate inputs to fuzzy data

processing procedures such as fuzzy inference systems, the utility value given for

each attribute value is crisp, not fuzzy. Although crisp numbers may be

successfully processed by ARAISS it is more realistic to represent utility values

as a fuzzy number, similar to those defined in the suitability term set. However

there is now a dichotomy between the discrete number of terms and the

continuous variation of the attribute value.

Both the continuous nature of the variable and the fuzziness of its utility value

are preserved in ARAISS, using a fuzzification method. Decision-makers

classify points on the domain of the source variable according to their suitability

and uncertainty, and values that lie at a point x, in between the classified points,

are given a fuzzy rating as follows:

(6.6)

( )( )lh

llhx xx

xxSuppSuppSupp−

−−=


(6.7)

(6.8)

(6.9)

Where:

Suppx is the width of the support of the suitability TFN at point x

Supph is the width of the support of TFN (ah,bh,ch) at the next highest rated point

Suppl is the width of the support of TFN (al,bl,cl) at next lowest rated point

xh is the next highest rated point

xl is the next lowest rated point

N.B. the term ‘Rating’ is used here to signify a numeric evaluation of an

alternative that is not dependant on the value of other alternatives. When

alternatives are evaluated directly against one another the term ‘Ranking’ is

used.

( )( )lh

llh

xxxxbbb

−−−

=

⎪⎪⎪

⎩

⎪⎪⎪

⎨

⎧

−

<−−

<

=

Otherwise 2

21 if 1

2 if

x

xx

x

Suppb

SuppbSupp

Suppbo

a

⎪⎪⎪

⎩

⎪⎪⎪

⎨

⎧

+

<−

<

=

Otherwise 2

21 if 1

2 if

x

x

xx

Suppb

Suppb

SuppbSupp

c


6.2.6 Aggregation and output parameters

An aggregation procedure brings together all the decision variables to produce an

overall evaluation of each alternative. Many possible aggregation methods were

discussed in Chapter 2, and the choice of aggregation procedure for ARAISS was

made with three requirements in mind:

1. The procedure should yield syntactic and semantic definitions of utility,

certainty, safety and consensus.

2. The procedure should have a high resolution, making it easier to isolate the

best alternative(s). i.e. the procedure should serve to limit the number of

alternatives with the same rating.

3. The procedure should not be too computationally intensive, enabling the user

to interact with the system in real time.

The first aggregation in ARAISS is based on fuzzy multiattribute decision-

making theory as described by (Ribeiro 1996). In order to process linguistic

variables, procedures for performing arithmetic operations on the trapezoidal

fuzzy numbers are needed. A comprehensive set of operations was developed by

Bonissone (1982) and is used in ARAISS. The fundamental operations used are

addition, subtraction, multiplication and division as shown below.

N.B Notation is in the bandwidth format as shown in Chapter 3.

( ) ( )2222211111 ,,,,,,: βαβα baTpzbaTpzAddition +

(6.10)

( ) ( )2222211111 ,,,,,,: βαβα baTpzbaTpznSubtractio −

(6.11)

( ) ( )2222211111 ,,,,,,: βαβα baTpzbaTpztionMultiplica ×

(6.12)

( )21212121 ,,, ββαα ++++= bbaaTpz

( )21212121 ,,, αββα −−−−= abbaTpz

( )2112212112212121 ,,, ββββαααα ++−+= bbaabbaaTpz


( )βα ,,,1:baTpz

Inverse

(6.13)

2

1:TpzTpzDivision

(6.14)

6.2.6.1 Utility

A measure for utility is derived using a double weighted fuzzy combination, as

shown Equation 6.15. Each alternative is given a compensatory outcome based

on suitability assessments from the set of decision-makers. The assessments are

weighted by criterion weights and the relevance of each decision-makers

opinion.

∑∑= =

××=J

j

K

kjkjkijki WROS

1 1| i = 1…I (6.15)

Where:

iS is the compensatory suitability of alternative i (as a TFN)

ijkO is the criteria outcome for alternative i with relation to criterion j and

decision-maker k, including quantitative uncertainty (as a TFN).

jkR is the relevance of decision-maker k’s opinion with respect to criterion j.

jkW is the weight assigned to criterion j by decision-maker k

⎟⎟⎠

⎞⎜⎜⎝

⎛+

−−

−=

)(,

)(,1,1

ββ

αα

bbaabaTpz

21

1Tpz

Tpz ×=


The output of Equation 6.15 is a fuzzy number representative of each

alternative’s overall compensatory suitability and uncertainty. To enable the

derivation of a linguistic rating for each alternative it is first necessary to carry

out a simple score range normalisation using Equation 6.16.

(6.16)

Normalisation of TFN’s is accomplished in ARAISS by using crisp numbers for

xmax and xmin, with xmin set to 0 and xmax set to cmax, (the third breakpoint of the

highest possible score using the maximum suitability term and the user defined

weighting and relevance parameters in an aggregation).

The next step is to rank and rate the normalised outputs. There is a vast amount

of literature on the ranking of fuzzy numbers (Fodor, Perny et al. 1998). Some

ranking procedures such as dominance relation require a series of pairwise

comparisons that can create a substantial computational burden, particularly if a

large number of alternatives exist. The simplest and most computationally

efficient ranking methods are scoring functions that assign a crisp value to each

set independently. ARAISS ranks fuzzy numbers using a scoring function that

measures a TFN’s centre of gravity along the x-axis. The score is calculated

using Equation 6.17.

Rs(i) = Rs(N(Si)) = Rs(TFN(a,b,c)) =

(6.17)

Where:

Rs(i) is the utility score for alternative i

minmax

min))((xx

xxxTFN−

−=Ν

22

2)(

2)(

211

22

bcab

abbc

b−+−

⎟⎟⎠

⎞⎜⎜⎝

⎛ −−−⎟⎠⎞

⎜⎝⎛ −

+


Rating a fuzzy number via a linguistic approximation is essentially a pattern

recognition problem, solved by extracting a set of features for comparison. As

the suitability term set is ordinal, a single feature can be used to ascertain the

position of the closest term. ARAISS uses the score from Equation 6.17, as

shown in Equation 6.18.

If

(6.18)

Where:

sl is the linguistic suitability term approximation

operator

ns is the nth term in a set of N-1 suitability terms

iS is the overall suitability of alternative i as a TFN

6.2.6.2 Safety

Safety is gained by making a decision that satisfies all criteria according to some

minimum standard. Safety is assured by eliminating risk, which is apparent in

alternatives with at least one poor criterion outcome. A very bad score on even a

minimally weighted criterion may, in reality, affect the overall rating of an

alternative much more than it’s weight suggests. Ordered weighted averaging has

been proposed as a countermeasure to this situation (Yager 1988), and may be

incorporated into a weighted aggregation procedure to provide control over the

level of compensation (Jiang and Eastman 2000). However to utilise this

approach decision-makers need to specify a precise value for the level of

compensation, which may not always be possible, and the aim here is to avoid

the need for non-linguistic input. ARAISS generates a linguistic assessment of

nis sS =)(l

( ) ( ))()()()(1

0 isns

N

nisns SRsRSRsR −Λ=−−

=


the risk inherent in each alternative using the risk score gained from Equation

6.19 in Equation 6.20:

(6.19)

If

(6.20)

Where:

Rr(i) is the risk score for alternative i

MinO is the minimum outcome required to eliminate risk (specified

linguistically by decision-makers)

rl is the linguistic risk term approximation operator

nr is the nth element of a set of N-1 risk terms (we use the term generation

term set here)

∧ is the minimum operator

6.2.6.3 Consensus

Consensus is reached when all parties are in agreement on criterion weights and

outcomes. Consensus is achieved by eliminating conflict, which occurs when an

alternative is rated poorly and weighted highly on a criterion by one decision-

maker, and is rated well, or weighted poorly on the same criterion by another

decision-maker. Risk is a necessary but not sufficient condition for conflict, so

the analysis is limited to those alternatives with a risk measure greater than zero.

Conflict is assessed using Equation 6.21, and a linguistic assessment of the level

nrr riR =))((l

( )

( )

( )⎪⎪⎪

⎩

⎪⎪⎪

⎨

⎧

⎟⎠⎞

⎜⎝⎛ ∧∧−

≥⎟⎠⎞

⎜⎝⎛ ∧∧

===

==

Otherwise

For 0

)(11

11

Mins

ijk

K

k

J

jsMins

Minsijk

K

k

J

js

r

OR

OROR

OROR

iR

( ) ( ))()(1

0iRriRr rn

N

nrn −∧=−−

=


of conflict is obtained in an identical way to that of Risk. Again the term

generation terms are used.

(6.21)

Where:

Rc(i) is the conflict score for alternative i

∨ is the maximum operator

6.2.6.4 Certainty

Certainty is the level of confidence placed in an outcome. Certainty is achieved

by eliminating uncertainty, and the uncertainty score Ru(i) is the width of the

support of the aggregated output. Uncertainty is rated linguistically by Equation

6.22.

nu ui =)(l

If

(6.22)

Where:

ul (i) is the linguistic uncertainty approximation for alternative I

U((ls(Si),un) is the TFN of the linguistic suitability term approximation for Si,

scaled for uncertainty using un

nu is the nth term in a set of N-1 uncertainty terms

2

)()()(

111⎟⎠⎞

⎜⎝⎛ −∧−−∨∨

==== jkijks

K

kjkijks

K

k

J

j

c

wORwORiR

( )( ) ( )

( )( ) ( )inis

N

n

inis

SSuppuSlUSupp

SSuppuSlUSupp

−∧

=−−

=),(

),(1

0


6.2.7 Adjusted aggregation and alternative exploration

Decision-makers can now decide which parameters are most important as they

explore and reduce the set of feasible alternatives in an interactive, iterative

process. Alternatives are reduced by selecting minimum standards for each of the

four parameters or creating an overall adjusted suitability value via Equation

6.23. The adjusted suitability score is then used to generate an adjusted linguistic

suitability rating using Equation 3.19. Weighting of the four parameters is via

consensus, or a non-weighted averaging of each decision-maker’s preferences,

which enables a variety of non-compensatory outcomes to be generated.

(6.23)

Where:

A(i) is the adjusted suitability value of alternative i

ws is the weighting of the suitability score

wu is the weighting of the uncertainty score

wr is the weighting of the risk score

wc is the weighting of the conflict score

6.3 ARAISS simulation exercises

Before comprehensive testing implementation in a SDSS it was necessary to test

and debug the algorithm in a way that would allow obvious flaws in the

methodology to be found. The MATLAB mathematical analysis software was

used to trial the algorithm in a number of simulated site selection problems. The

problems were based on a relatively small number of alternatives, which allowed

criterion ratings to be manually specified without spatial functions. The author

simulated all input data with the aim of creating a rich enough decision-making

environment for all the features of the algorithm to be tested. An example of the

crus

ccrruuss

wwwwwiRwiRwiRwiRiA

+++−+−+−+

=))(1())(1())(1()()(


procedure used is given in Section 6.3.2, and the full MATLAB code is provided

in Appendix D.

6.3.1 Validating ARAISS

A fundamental problem in designing an algorithm to solve infrastructure site

selection problems is that there is often no perfect solution to find, and it is not

always possible to derive the best compromise from initial assessments. Using a

pre-determined optimization algorithm is standard procedure in many areas of

problem solving, and works particularly well when the exact utility of a solution

can be precisely measured and used as feedback to improve performance.

However the exact utility of a solution in site selection is seldom known.

Multiple, conflicting criteria, and the added human element of conflicting

opinions of measurement and importance create an ill-structured problem that is

often dynamic, in that assessments may change as the solution space is

examined. It is also relevant to note that problem-solving strategies vary from

person to person, making the group situation a particularly dynamic environment.

In such a climate the traditional model of testing a new algorithm against known

others using standard test data and set benchmarks becomes obsolete.

In the case of ARAISS a second major hurdle is the absence of a standard dataset

with which to generate results and compare those results to known solutions.

Datasets used in other published work on multi-criteria site selection either lacks

multiple decision-maker inputs, or uncertainty data. In fact due to the unique

approach of ARAISS, which requires decision-makers to weight output

parameters not generated by other methods, comparative testing is challenging

from the outset.

It is proposed in this research that the best method of assessment for a site

selection algorithm is to use real world examples, where the algorithm can be

utilized in an actual decision-making situation, and the decision-makers

themselves can assess its performance and usefulness. This approach was taken

with ARAISS after its implementation in a GIS based SDSS, and is detailed in

Chapter 8.


6.3.2 Example simulation

Several MATLAB simulations were conducted to test the common sense validity

of the algorithm, and the following example is typical of the process used. The

problems were all based on three decision-makers rating five alternatives with

respect to three criteria. The term set used was shown in Table 6.1, and Figure

6.2.

Abbreviations used in the example problem are as follows:

Suitability

Totally Unsuitable (TU), Bad (B), Indifferent (In), Good (G), Perfect (P).

Uncertainty

Very Certain (VC), Certain (C), Moderately Certain (MC), Uncertain (U), Very

Uncertain (VU).

Weights

Irrelevant (Ir), Unimportant (U), Moderately Important (MI), Important (Im),

Critical (C).

Term Generation and Feedback

Zero (Z), Very Small (VS), Small (S), Moderate (M), Large (L), Very Large

(VL).

Other Abbreviations

Decision-maker-n (DMn), Criterion-n (Cn), Alternative-n (An) (n= 1,2,3,…).


6.3.2.1 Data inputs

Decision-maker relevance values for the example problem are given in Table 6.2,

and decision-maker preferences for weighting and rating are provided in Tables

6.3-5.

Table 6.2: Decision-maker relevance

C1 C2 C3 DM1 U C U DM2 Im Im Im DM3 Im Im MI

Table 6.3: Decision-maker 1 inputs

Alternative C1 C2 C3 A1 In, C G, VC In, MC A2 In, VU G, U In, VU A3 P, C P, C B, MC A4 P, C B, VC P, MC A5 TU, MC B, MC B, MC

Wgts MI C Im


Alternative C1 C2 C3 A1 G, VC In, C G, MC A2 G, U In, U In, MC A3 P, MC P, C B, MC A4 P, VU P, U P, C A5 B, MC TU, MC B, MC

Wgts I I I



Alternative C1 C2 C3 A1 G, VC G, VC In, C A2 In, VU In, VU G, VU A3 P, C P, VC TU, MC A4 P, VC G, C P, C A5 TU, MC TU, MC B, MC

Wgts C Ir MI

6.3.2.2 Results

Inputs were aggregated according to the ARAISS decision rules given in Section

6.2.6. Figure 6.3 shows outputs from the fuzzy weighted combination of

Equation 6.15 after normalisation. Table 6.6 shows output parameter values and

the results of an adjusted aggregation in linguistic form.

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

μ

Suitability

A1A2A3A4A5

Figure 6.3: Fuzzy outputs from Equation 6.15


Table 6.6: Final outputs in linguistic form

Alternative Utility Uncertainty Risk Conflict Overall Rank

A1 G MC Z M G 2 A2 I VU Z Z G 1 A3 G VC VL M I 3 A4 G VC M L I 4 A5 B MC VL M B 5

Wgts MI VU U I

6.3.2.3 Interpretation

Results from the example problem show that the algorithm performs as expected.

Alternative 1 scores well in terms of Utility Uncertainty and Risk but exhibits a

‘Moderate’ level of conflict, which brings it’s overall rank back to second, as

consensus is considered ‘Important’ in a solution. Alternative 2 scores well on all

counts but uncertainty, but as uncertainty is considered ‘very unimportant’ in this

example Alternative 2 is ranked as the best option. Alternative 3 suffers from a

‘Very Large’ risk level, Alternative 4 from conflict, and Alternative 5 is rates

poorly on all counts.

6.4 Conclusions

This chapter has described the ARAISS site selection algorithm and

demonstrated its use in an example problem. The algorithm works by accepting

linguistic inputs from a set of decision-makers, which are brought together in a

fuzzy double-weighted aggregation. Four parameters are then extracted for each

alternative, indicating levels of Utility, Safety, Consensus, and Certainty.

Weighting and combining the four parameters enables decision-makers to decide

which aspects of the solution are most important to their specific problem, and

thereby delivers a real means of control over algorithm performance.


In a mathematical sense ARAISS is a hybrid algorithm, combining elements of

fuzzy set theory and multicriteria evaluation in a new way. ARAISS can generate

compensatory and non-compensatory solutions, in a multi-decision-maker

framework, but remains computationally efficient enough to use in problems

with large numbers of alternatives.

The algorithm performed as expected in example problems, delivering sound

results in a five alternative, three decision-maker problem with simulated inputs.

Potential solutions were successfully differentiated on the basis of utility,

certainty, risk, and conflict, and the ultimate solution was chosen based on a

weighted summation of these parameters.

Chapter 7 InfraPlanner 121

Chapter 7

IINNFFRRAAPPLLAANNNNEERR

7.1 Introduction

InfraPlanner is a prototype Spatial Decision Support System (SDSS) designed to

aid decision-makers with Group Multicriteria Location Problems. The system

was designed primarily to implement a spatial decision-making model based on

the ARAISS algorithm described in Chapter 6, however the choice of model is

only one aspect of a SDSS. The construction of a SDSS is a complex task

involving ongoing consultation with end users and other experts, and there are

many possible alternatives for interface design and model integration, as well as

data format and handling issues. Facing these problems has previously been

identified as an important area for ongoing research (Goodchild, Haining et al.

1992), and will continue to provide challenges for some time, to those adapting

new methods to a GIS environment (Scholten and LoCascio 1997). Pressing

spatial problems cannot wait for new technologies to arrive, and developers

should adapt technologies available now, to suit their particular situation (Lam

and Swayne 2001).

The InfraPlanner prototype SDSS is a ‘Fully Integrated’ system that consists of a

set of tools integrated into ArcGIS software. ArcGIS is an umbrella term for a

rage of GIS applications furnished by the Environmental Systems Research

Institute (ESRI). The primary application used in development of InfraPlanner

was the ArcMap module, which is a vector and raster GIS package. ArcMap is

provided in a subset of ArcGIS products collectively referred to as ArcView. The

system was created by developing new tools in ArcMap using Visual Basic for

Applications (VBA) customisation.

InfraPlanner is a ‘Knowledge Based’ system, in that it provides both access to

data, and facilities for manipulating that data via a set of rules. It was designed to


allow the input and analysis of data and expert knowledge during a group

multicriteria site selection process, and it is possible to utilise the system in the

following ways within that context:

• Data storage and visualisation via the standard GIS database and interface

• Information sharing and consultation via generation of linguistic suitability

maps

• As a scenario analysis tool for use by planners and decision-makers in

isolation via a fuzzy multicriteria evaluation analysis

• Group site selection analysis via a group fuzzy multicriteria evaluation

analysis

This Chapter describes InfraPlanner in its entirety. Firstly a general system

overview is provided, the development process followed is then described, and

finally the use and functionality of the system are illustrated.

7.2 Overview of the prototype system

The core attributes of the prototype system fall into five categories as shown

below. These five aspects of the system are described in the following five

Sections.

1. Target application

2. Target audience

3. Dialog design

4. Database

5. Model

7.2.1 Target application

The term ‘target application’ describes the problem or types of problems the

system was designed for. Definition of a target application provides focus for the

development process, and makes designing evaluation problems possible.


The InfraPlanner system was intended as a generic site selection tool capable of

providing support for Group Multicriteria Location Problems (GMCLP’s) of a

strategic nature. These site selection problems are characterised by multiple

decision-makers, multiple criteria, a large number of spatial alternatives and

uncertainty. The target application of the prototype system can be stated more

specifically by the following sentence:

‘To provide a tool to aid in the solution of strategic GMCLP’s within the 2700 ha

Brisbane Airport site’

7.2.2 Target audience

Experience has shown that a SDSS needs to be targeted to a specific audience to

be successful, as the system will eventually become redundant if there is no

potential or actual user in mind (Lam and Swayne 2001). Targeting specific users

was also useful for identifying the most important individuals for ongoing

consultation.

The InfraPlanner system is targeted at strategic decision-makers from either

technical or non-technical backgrounds. The prototype system, focused on

Brisbane Airport, has a target audience consisting of:

Those working at Brisbane Airport in Management, Planning, Infrastructure

Development, and Environmental management.

7.2.3 Dialog design

The dialog, or interface, provides users with the means to interact with the

system, and needs to be as user-friendly as possible. Experience has shown that a

user-friendly interface should possess both simplicity and flexibility.


Simplicity has been consistently cited as a desirable quality in a SDSS interface,

and there are several basic qualities a simple user interface should possess,

including (Sparkman 1999):

• Consistent, clearly understood screen layout

• Readable font size

• Shallow menu hierarchy

• A plan for programmatically guiding the user

• Limited information display at one time

• Status updates

• Online help

Specific strategies beyond the basic rules of good software design are seldom

given in research literature. Two further strategies employed in the development

of InfraPlanner to enhance the ease and simplicity of user interaction were:

1. Linguistic interaction

As discussed in Chapter 3, it is natural and intuitive for humans to interact using

words. InfraPlanner was designed to be a fully linguistic system where inputs

and feedback are given in natural language. The interfaces were specifically

designed to accommodate vague linguistic statements such as ‘I am moderately

certain that 100 metres from the waterway is good’ and thereby avoid asking

decision-makers to undertake complex mathematical procedures.

2. Heavy use of the existing GIS interface

GIS is a mature technology and the methods standard GIS packages employ to

view and interact with spatial information are the result of an ongoing process of

refinement dating back to the 1960s. Instead of expending resources producing

an entirely new interface, InfraPlanner was conceived as a set of tools that

enhance existing GIS functionality where necessary, whilst retaining the highly

effective aspects of the standard interface. This was achieved by incorporating


InfraPlanner tools into a toolbar that integrates seamlessly with existing

functionality.

Flexibility was also a requirement of the user interface. This implies that the

system should be suitable for the full range of site selection decisions possible,

with respect to the target audience. InfraPlanner incorporates two strategies to

increase flexibility:

1. Provision of a comprehensive generic database covering a range of spatial

information within the primary area of interest

2. Use of a generic multicriteria evaluation model

7.2.4 Database

The foundation of any SDSS is its data, and a comprehensive database was

compiled for the prototype system. The geographical area of interest was defined

as all areas within the Airport boundary plus significant sites bordering on the

Airport. The accumulated data provided the basis for defining criteria falling into

three main categories:

1. Environmental Impact

2. Cultural impact

3. Operational issues (general airport planning rules)

Whilst it is unrealistic to plan for all possible data needs, the following themes

provided a sound basis for many site selection problems.

• Flora

• Fauna

• Habitat value

• Topography

• Land use and zoning

• Cultural heritage sites


• Contaminated sites

• Airport facilities including all buildings, roads, taxiways and runways

• Environmentally sensitive areas such as waterways

• Nearby residential communities

The real value of data is made apparent when it is combined with expert

knowledge and processed into useful information. This is accomplished by a

decision-making model.

7.2.5 Model

The Approximate Reasoning Algorithm for Infrastructure Site Selection

(ARAISS) described in Chapter 6 provides the ability to combine raw data

themes in accordance with decision-maker knowledge. It is a generic multi-

criteria group decision-making model capable of application to a wide variety of

strategic site selection decisions. The model is accessed via a set of user forms

designed to aid the various stages of the decision-making process.

The model accepts two types of inputs:

1. Data: in the form of pre-processed raster layers indicative of the spatial

variation of a raw attribute value.

2. Knowledge: in the form of linguistic assessments from decision-makers.

The model outputs raster layers of the same resolution as input layers, attributed

and colour coded to match terms in the linguistic term set used in decision-maker

input. Two basic types of raster layers may be created.

1. Criterion Maps: Criterion Maps are the InfraPlanner equivalent of a

suitability map, with the added element of uncertainty. They are a raster map

identifying the spatial variation of the suitability and uncertainty of one

attribute, according to one decision-maker. Criterion Maps are created from a

data layer indicative of the spatial variation of the raw value of an attribute


(eg. a zoning map) by associating the raw attribute value with the suitability

and uncertainty values provided by the decision-maker. Criterion Maps can

be either discrete (based on a categorical variable such as land ownership) or

continuous (based on a variable that covers a continuous range of values such

as proximity from a feature).

2. Decision maps: Decision Maps are created by combining criterion maps with

decision-maker preferences for criterion weights, decision-maker relevance,

and output parameter weights. Decision maps describe the spatial variation of

aggregated site parameters: Suitability, Uncertainty, Risk, and Conflict. The

output of an adjusted aggregation (whereby Suitability, Uncertainty, Risk,

and Conflict are weighted and combined) is also a type of decision map.

The specifics of how InfraPlanner accomplishes this are given in Section 7.4.

7.3 Development process

Development of the InfraPlanner system followed a simplified evolutionary

prototyping structure, conducted in close consultation with end users. The

process consisted of a logical sequence of activities, which was documented via a

logic model.

Logic modelling is a resource management tool used to document the underlying

reasons and goals behind a program of activities. In a logic model the program is

divided into six elements.

1. Resources are the raw materials available

2. Activities make use of the available resources

3. Outputs are the tangible results of an activity

4. Customers are those who receive the outputs

5. Outcomes (Short, medium, or long term) are the reason for undertaking the

activities

6. External influences are those influences that are beyond the scope and control

of the program


The logic model detailing the sequence of events involved in the InfraPlanner

development process was shown in Figure 1.2. The main activities in the model

are described in Sections 7.3.1 – 7.3.5.

7.3.1 Planning

The planning phase of system development was conducted in conjunction with

end users from BAC. A needs assessment and problem diagnosis was conducted

to define the goals of the system, and to determine the types of decisions the

system would provide assistance with. Goals defined in the planning phase

included:

• The system should support decision-makers with facilities placement

decisions within the Brisbane Airport grounds

• The system should be able to accommodate qualitative variables such as

socio-economic and environmental impacts

• The system should accommodate multiple criteria and points of view

• Outputs from the system should be graphical where possible, preferably in a

mapping format

• The modelling capabilities of the system should be transparent and easily

understandable

These basic statements of desired functionality were then used as the focus for a

state of the art literature review.


7.3.2 Research

The research phase consisted of a state of the art review of published literature on

the techniques and technology involved in spatial decision-making. Specifically,

the review was conducted to answer the following research questions:

1. What are the analytical techniques used in the solution of multicriteria

location problems?

2. What are the major limitations of these techniques?

3. What technology platforms are used in the analysis of spatial problems and

what are their major characteristics?

4. What are the most promising methods for advancing current techniques and

technologies?

The review clearly showed that Multi-criteria evaluation (MCE) is the most

suitable analytical technique for the solution of multicriteria location problems.

However several shortcomings were noted. Most important of these are the

inability to deal with uncertainty, inability to deal with a group environment, and

the perception by decision-makers that current methods are not user friendly. The

universally accepted technology platform for the analysis of location problems

was found to be a GIS, coupled or fully integrated with decision-making models.

Advanced Artificial Intelligence and soft computing techniques offered an ability

to overcome some of the shortcomings of MCE, but it was necessary to deploy

them in a user friendly way in order to avoid the perception of a ‘black box’

scenario.

7.3.3 Analysis and design

The primary objective of the design phase was to produce a clear conceptual

system design specification based on outputs from the planning and research

phases, and to provide input from technology experts. This phase produced two

major outputs.


1. A new fuzzy model for the type of location problems encountered by

decision-makers at Brisbane Airport and strategic decision-makers in

general. The model is described in Chapter 2.

2. A design specification for the prototype system to implement the new

model. The key objectives contained in the design specification are

summarised in Section 7.2, and the working system is fully described in

Section 7.4.

7.3.4 Construction

Construction covered the technical implementation of the design. In the case of

InfraPlanner construction consisted of integrating the new fuzzy decision-making

model into the selected GIS package. Technology selection was a vital aspect of

the design process as the capabilities of the chosen GIS package have a

significant impact on functionality, compatibility and development time. The

three key aspects considered when choosing among the many commercially

available systems were:

1. Level of raster analysis functionality

2. Customisation capabilities

3. File format compatibility

ArcGIS was eventually chosen, as it possessed comprehensive raster analysis

functionality, offered an inbuilt Visual Basic for Applications (VBA)

customisation environment, and was compatible with the existing MicroStation

CAD software employed by Brisbane Airport.

ArcGIS customisation is based around the manipulation of a set of

programmatically controllable software objects collectively referred to as

ArcObjects. ArcObjects offered access to the objects that make up ArcGIS

software at a high enough level of granularity to be a flexible and effective

development tool. Construction mainly focused on development of the set of user

forms described in Section 7.4. The forms offer an intuitive visual way to


implement the algorithm described in Chapter 6. Source code is provided in the

Appendix.

7.3.5 Implementation

Implementation consisted of testing and evaluation of the system. InfraPlanner

was tested and evaluated using a real world site selection problem faced by the

Brisbane Airport. Brisbane Airport planners, regulators, and external consultants

were involved in the validation problem, which involved the location of a new

recycling facility on the Airport site. The validation problem is fully described in

Chapter 8.

7.4 The InfraPlanner prototype

InfraPlanner consists of a set of tools designed to implement ARAISS in a GIS

environment. The tools are accessed in ArcMap via the InfraPlanner Toolbar


Figure 7.1: The InfraPlanner toolbar

Tools available from the InfraPlanner Toolbar are structured as follows:

Project Tools:

Select project: A user form to select a stored decision project

Create new project: A user form to create a new decision project

View project Information: A user form listing current project options

Create Maps:


Criterion map: User forms to create discrete or continuous

suitability maps from a raster map

Decision maps: A user form to bring suitability maps together in an

aggregation and create output parameter maps

Format Map:

Tools to format existing raster maps with numeric attribute values into a

suitability, utility, risk, uncertainty or conflict map. The transformation is purely

visual not analytical, and is not discussed further here.

Explore Maps:

A point and click tool to be used to interactively explore selected locations and

all their outcomes, or to perform an adjusted aggregation.

InfraPlanner tools are used to follow the general framework of ARAISS, as

shown in Figure 7.2. Specific tools are described in the following Sections.


Figure 7.2: How InfraPlanner tools fit into the decision-making framework

Define problem

Define decision-maker relevance values

Perform compensatory aggregation function

Identify feasible alternative sites for analysis

Choose Decision-makers

Identify Utility, Certainty, Safety and Consensus

Final site selection

Define and rate criteria (factors)

Start

Define linguistic terms

Tactical and operational assessment

Micro-placement

Identify constraints

Explore and reduce alternatives

Iterate

Perform adjusted aggregation

Define criterion weights

Review inputs


7.4.1 Project tools

The project tools are used to specify and view the type of decision, decision-

makers involved, criteria, and the linguistic term set used for input and feedback.

Setting the project information is shown in Figure 7.3.

Figure 7.3: Setting project information


Inherent in the input of project information is choosing a linguistic suitability

term set to be used for decision-maker input and feedback. InfraPlanner creates

new term sets by using Equation 6.1 to add new terms to a set of primary

suitability terms as shown in Figure 7.4. The user first chooses the primary term

set to build on and then uses the ‘Create Term Set’ user form to follow the

process described in Section 6.2.3.

Figure 7.4: Creating a new term set

7.4.2 Creating maps

InfraPlanner provides the ability to create several types of maps used in the

decision-making process. In most cases some pre-processing is required to

provide the system with suitably classified raster input maps. Pre-processing is

performed using standard Map Algebra techniques, such as those described in


Chapter 2. Pre-processing usually consists of converting a vector map to raster

format, or performing a proximity function to create a raster map indicative of

distance from some feature. Boolean constraint maps are also created using

standard map algebra techniques.

7.4.2.1 Suitability Maps

Suitability maps are either based on discrete (categorical) variables such as

regional zoning, or a variable that takes a continuous range of values, such as

elevation. Discrete suitability maps are created using the form shown in Figure

7.5. Users specify a source theme containing the pre-processed baseline data for

the suitability map and classify the categories it contains using linguistic

suitability and uncertainty assessments.

Figure 7.5: The discrete criterion map user form

Users interact with the discrete criterion map user form as shown in Figure 7.6.


Figure 7.6: Creating a discrete criterion map.

Name the new Criterion Map to be created

Load the Discrete Criterion Map form from the InfraPlanner Toolbar

Input a description of the map

Select a source theme from the pre-screened list of discrete source maps

Choose the attribute field within the map to linguistically classify.

Rate each attribute category in terms of suitability and uncertainty

Click the ‘Create’ button to create the new map

The new criterion map is created and displayed


Continuous criterion maps are created using the form shown in Figure 7.7. Users

specify points along the domain of the source variable and rate them with

linguistic suitability and uncertainty values. Points whose values lie between the

rated points are classified according to equations 6.6 – 6.9, which is essentially a

linear extrapolation of the centre point, and support of the TFN.

Figure 7.7: The continuous criterion map user form

Users interact with the continuous criterion map user form as shown in Figure

7.8.


Figure 7.8: Creating a continuous criterion map

Name the new Criterion Map to be created

Load the Continuous Criterion Map form from the InfraPlanner

Toolbar

Input a description of the map

Select a source theme from the pre-screened list of continuous source maps

Rate a minimum of 3 points in terms of suitability and uncertainty

Click the ‘Create’ button to create the new map

The new criterion map is created and displayed


7.4.2.2 Decision Maps

The term ‘Decision Maps’ is a generic term used within InfraPlanner to describe

the four aggregated parameter maps (Suitability, Uncertainty, Risk and Conflict).

Using the ‘Create Decision Maps’ user form, users associate a previously created

criterion map with each criterion and decision-maker in the chosen decision

project. They also input the decision-maker relevance and criterion weighting for

each criterion in the decision project, with respect to each decision-maker. The

output parameter maps are created by an aggregation of the criterion maps, using

equations 6.15 – 6.22. The ‘Create Decision Maps’ user form is shown in figure

7.9, and user interaction is illustrated in Figure 7.10.

Figure 7.9: The decision maps user form


Figure 7.10: Creating decision maps

Name the new Decision Maps to be created.

Load the Decision Map user form from the InfraPlanner

Toolbar

Choose a constraint map to limit the area under consideration

Enter the Decision-maker relevance value for the displayed decision-maker

Click the ‘Add‘ button when the inputs are correct to cycle to the next set of inputs and repeat the previous step.

Click the ‘Create’ button when all inputs have been entered to create the new maps

The new parameter maps are created and

displayed

Enter the weight of the displayed criterion according to the displayed decision-maker.

Choose the previously created criterion map that represents the displayed criterion according to the displayed decision-maker.


7.4.3 Exploring maps

Map exploration is facilitated using a point and click tool that allows users to

examine any feasible alternative site in all its dimensions. An interactive report is

displayed which provides information on the four output parameters plus

individual criterion outcomes and provides an opportunity to set the weighting

parameters for an adjusted aggregation. Map exploration is shown in Figure 7.11.

Figure 7.11: Map exploration


7.5 Validating InfraPlanner

Validation of the working prototype was conducted in three ways. The first and

most valuable means of validation was the case study presented in Chapter 8.

Results showed that users found InfraPlanner simple to use and understand, and

selected sites that they deemed acceptable. Secondly a peer reviewed paper was

presented at ANZIIS 2003 as shown in Appendix A. Lastly a focus group was

created at ANZIIS 2003, consisting of five researchers and academics from the

fields of AI and soft computing. A combination of discussion paper and

questionnaire was created for the group, and is reproduced in Appendix E. The

focus group was given the discussion paper after the presentation of the

conference paper, and asked for their feedback on the algorithm and the

methodology used to create it. The group exercise quickly took the form of a

vigorous discussion, in which feedback was positive, with all present agreeing

that both the development process and the model derived from it was valid. Some

researchers noted that the use of a software design flowchart would be a good

way to represent the model, as they found the logic model used difficult to

follow. Other specific comments included:

• Documentation of the model should be in a commonly accepted, and easily

understandable format

• Users need to be able to understand the impacts of preferences given during

the data input phase

• Use of fuzzy numbers to represent words is valid, but needs more work in

terms a rigorous and repeatable way of defining the membership function

• Users should understand how the model works

• The concept of weighting inputs based on relevance of opinion was valid but

a rigorous way of generating the relevance matrix was needed

7.6 Discussion

InfraPlanner is a working prototype of a generic SDSS for GMCLP’s of a

strategic nature. The system demonstrates that approximate reasoning techniques

are suitable for use in SDSSs, although designing and building the InfraPlanner


Spatial Decision Support System proved to be extremely challenging.

Constructing a DSS is generally considered to be a complex, time consuming

task, requiring a group of skilled individuals, and this was proven in practice.

There are many small issues that are not generic enough to be mentioned in

publications on SDSSs but nonetheless proved problematic. Among these were

choosing a GIS package from the myriad of options available, and dealing with

the organisational changes that occurred during the development process.

Chapter 8 A case study using InfraPlanner 145

Chapter 8

AA CCAASSEE SSTTUUDDYY UUSSIINNGG IINNFFRRAAPPLLAANNNNEERR

8.1 Introduction

Validation of any proposed algorithm requires a practical implementation to test

assumptions made during the design process. In many cases there exists a set or

sets of standard real world data and solutions upon which to compare the

accuracy of a given algorithm. In the case of site selection decisions under the

types of uncertainty discussed in Chapter 5, there appears to be no standard

dataset that incorporates all the variables used as inputs to the InfraPlanner

algorithm. Specifically there is no dataset that includes subjective uncertainty

assessments from multiple decision-makers, and preferences for decision-maker

relevance, or the priorities placed on the four output parameters; Utility,

Certainty, Consensus, and Safety. To overcome this data shortage problem an

experiment was conducted using a real world site selection decision at Brisbane

Airport, where the desired inputs and outputs could be generated and commented

upon by decision-makers themselves.

Inputs were generated for three stakeholder groups using actual decision-makers

or representatives chosen by the experimenter for their knowledge of the

situation. The problem used was real, and the objective was to choose the best

location for a recycling facility on the 2700 ha Brisbane Airport site. This chapter

details the problem and all inputs, the process used to implement InfraPlanner in

deriving solutions, the results generated, and a discussion of their relevance.


8.2 The problem

The problem worked through here concerns the location of a new recycling

facility on the Brisbane Airport grounds. The Airport occupies 2700ha of land,

located 13km North East of the Queensland State Capital, Brisbane, and

adjoining Moreton Bay. The site is flat and low lying, occupying part of the

original Brisbane river delta, which has undergone extensive changes since the

1830s, with most of the original network of tidal waterways being replaced with

constructed drains. Much of the vegetation on the site has been planted in the last

15 years, and was chosen to reduce the attraction of birds. There are, however,

some environmentally sensitive areas to consider when locating new

developments, as well as issues associated with airport facilities, Government

legislation and the effects of airport operations on local communities. Figure 8.1

shows the general layout of the Brisbane Airport site.

The facility to be located inputs masonry from demolished buildings and, via

crushing and grinding, turns out various grades of landfill material. The main

impacts of such an operation on its immediate vicinity are noise and dust

emissions. There are three separate groups with an interest in the outcome. The

Brisbane Airport Corporation (BAC), as represented by their Infrastructure

Planning Manager. The Commonwealth Government, as represented by an

independent contractor responsible for ensuring regulatory compliance, and a

local residential community adjoining the Airport, whose inputs were provided

by an Airport representative with knowledge of their concerns. The groups differ

considerably in their priorities and suitability assessments, creating a rich

decision-making environment.


Figure 8.1: Brisbane Airport Layout


8.3 Procedure

The experiment was structured to follow the decision-making framework shown

in Figure 6.1, with the first steps involving problem definition and definition of a

linguistic term set. The linguistic terms used are shown in Table 8.1.

Table 8.1: Linguistic terms

Suitability (as a TFN) Weighting Uncertainty Term generation

Totally unsuitable

(0,0,0) Irrelevant 0

Very Certain 0 Zero 0

Bad (0,.2,.4) Unimportant .3 Certain 1 Very Small .1 Indifferent (.3,.5,.7) Moderately

Important .5 Moderately

Certain 2 Small .3

Good (.6,.8,1) Important .7 Uncertain 3 Medium

.5

Perfect (1,1,1) Very Important

.9 Very Uncertain

4 Large .7

Probably Good*

(.51,.71,.91) Critical 1* Very Large .9

• The suitability term ‘probably good’ was included as a ‘large’ increase on ‘indifferent’ at the request of decision-makers..

With the problem defined as ‘selecting the best site for the recycling facility’, the

next step was to identify the constraints (Boolean criteria) that would limit the

sites under consideration. During an initial consultation a set of five constraints

was derived:

1. Airport Boundary: The site must lie within the existing airport

boundary to avoid the cost of land acquisition.

2. Existing Buildings: Sites already occupied should be excluded to avoid

the loss of existing facilities.

3. Road access: The site must be within 200m of selected access

roads to avoid the cost of building new access.

4. Zoning: The site must lie in a zone designated ‘General

Industry’ or ‘Light Industry’ as defined by the

BAC 1998 Master Plan to comply with

Government planning requirements.


5. Conservation: The site must not occupy an area of high

conservation value, thereby preserving the

sensitive areas on the Airport grounds.

The map of unconstrained alternatives (available sites) was derived using map

algebra techniques, and is shown in Figure 8.2.


Figure 8.2: Unconstrained Alternatives


The next step in the process involved the definition and linguistic assessment of

criteria that vary on a suitability scale from ‘Totally Unsuitable’ to ‘Perfect’.

Decision-makers directly defined six criteria as important to the site selection

process. They are described in Table 8.2:

Table 8.2: Criteria definition

Criterion

Name

Type Description

Environmental value

Continuous As all areas of high conservation value are excluded, this criterion defines on a continuous scale how the distance from sensitive areas affects suitability

Zoning Discrete The facility may be placed in either a ‘General Industry’ or ‘Light Industry’ zone, and this criterion describes how that decision affects suitability.

Tenant Amenity Continuous Defines how the distance from sensitive tenants affects suitability

Community Impact

Continuous Defines how distance from the closest residential community affects suitability

Landfill Discrete It would be desirable to locate the facility close to areas which are more in need of the fill material generated by the facility.

Traffic Impact Discrete To regulate traffic flow and trucking noise, the use of some access roads is more desirable than others.

These criteria are represented as a set of suitability maps, created using

InfraPlanner interfaces to convert linguistic inputs from each decision-maker to a

spatially explicit format, as shown in Figures 8.3 and 8.4.


Figure 8.3: Creating a continuous suitability map for community impact

Figure 8.4: Creating a discrete suitability map for zoning


Each decision-maker will generate a map for each criterion for which their

opinion is deemed to be sufficiently relevant to include, and which they feel to be

relevant to the decision environment. Thus a decision-maker may opt out of

generating a map on the grounds of lack of expertise or if they feel that a

particular criterion should have no effect upon overall suitability. The inputs

required to generate the maps take the form of sentences, from which the relevant

information is input to the suitability map generation interface. Inputs were as

follows:

BAC Inputs:

Environmental value is ‘important’: It is ‘moderately certain’ that sites

of moderate conservation value are ‘good’ whilst it is ‘very certain’ that

all others are ‘perfect’.

Zoning is ‘very important’: It is ‘very certain’ that general industry zones

are ‘perfect’ whilst it is ‘moderately certain’ that light industry zones are

‘good’.

Tenant Amenity is ‘important’: It is ‘very certain’ that sites less than

50m from sensitive tenants are ‘totally unsuitable’. It is ‘moderately

certain’ that sites 100m from sensitive tenants are ‘good’. It is ‘certain’

that sites 500m from sensitive tenants are ‘perfect’.

Community Impact is ‘important’: It is ‘very certain’ that sites less than

500m from Pinkenba are ‘totally unsuitable’. It is ‘uncertain’ that sites

1000m from Pinkenba are ‘good’. It is ‘very uncertain’ that sites 2000m

from Pinkenba are ‘perfect’, and ‘certain’ that sites 4000m from Pinkenba

are ‘perfect’.

Landfill is ‘moderately important’: It is ‘very certain’ that sites on

Lomandra Dr are ‘perfect’. It is ‘moderately certain’ that sites on Randle

Rd, Sugarmill Rd and Viola Pl are ‘good’. It is ‘moderately certain’ that

sites on Airport Dr are ‘indifferent’.


Traffic impact is ‘important’: It is ‘very certain’ that sites on Airport

Drive are ‘bad’. It is ‘moderately certain’ that sites on Lomandra Drive

and Viola Pl are ‘good’. It is ‘certain’ that sites on Randle Road and

Sugarmill Rd are ‘perfect’.

Community Inputs:

Environmental value is ‘very important’: It is ‘moderately certain’ that

sites of moderate conservation value are ‘probably good’ whilst it is

‘certain’ that all others are ‘perfect’.

Zoning is ‘irrelevant’:

Tenant Amenity is ‘irrelevant’:

Community Impact is ‘critical’: It is ‘very certain’ that sites less than


3000m from Pinkenba are ‘good’. It is ‘certain’ that sites 4000m from

Pinkenba are ‘perfect’.

Landfill is ‘irrelevant’:


Drive are ‘perfect’. It is ‘certain’ that sites on Lomandra Drive and Viola

Pl are ‘bad’. It is ‘very certain’ that sites on Randle Road and Sugarmill

Rd are ‘bad’.

Government Inputs:

Environmental value is ‘important’: It is ‘moderately certain’ that sites

of moderate conservation value are ‘good’ whilst it is ‘very certain’ that

all others are ‘perfect’.


Zoning is ‘very important’: It is ‘very certain’ that general industry zones

are ‘perfect’ whilst it is ‘moderately certain’ that light industry zones are

‘good’.

Tenant Amenity is ‘important’: It is ‘very certain’ that sites less than

50m from sensitive tenants are ‘totally unsuitable’. It is ‘moderately

certain’ that sites 100m from sensitive tenants are ‘good’. It is ‘certain’

that sites 500m from sensitive tenants are ‘perfect’.

Community Impact is ‘important’: It is ‘very certain’ that sites less than


1000m from Pinkenba are ‘good’. It is ‘very uncertain’ that sites 2000m

from Pinkenba are ‘perfect’, and ‘certain’ that sites 4000m from Pinkenba

are ‘perfect’.

Landfill is ‘moderately important’: It is ‘very certain’ that sites on

Lomandra Dr are ‘perfect’. It is ‘moderately certain’ that sites on Randle

Rd, Sugarmill Rd and Viola Pl are ‘good’. It is ‘moderately certain’ that

sites on Airport Dr are ‘indifferent’.


Drive are ‘bad’. It is ‘moderately certain’ that sites on Lomandra Drive

and Viola Pl are ‘good’. It is ‘certain’ that sites on Randle Road and

Sugarmill Rd are ‘perfect’.

As illustrative examples, the suitability maps created by BAC and the residential

community representative for Traffic Impact and Community Impact are shown

in Figures 8.5, 8.6, 8.7 and 8.8.


Figure 8.5: BAC Traffic Impact Suitability Map


Figure 8.6: BAC Community Impact Suitability Map


Figure 8.7: Traffic Impact Suitability Map for the Residential Community


Figure 8.8: Community Impact Suitability Map for the Residential

Community


It was then necessary to perform an aggregation using the criterion weightings,

relevance weights, and suitability maps previously created. Inputs are

summarised in Tables 8.3 and 8.4. The interface used is shown in Figure 8.9.

Table 8.3: Criterion weighting

Criterion Weights

BAC Pinkenba Commonwealth

Environmental

Important

Very

Important Important

Zoning Very Important Irrelevant Very Important

Tenant amenity Important Irrelevant Important

Pinkenba Important Critical Important

Landfill Moderately

Important Irrelevant

Moderately

Important

Traffic

Important Important

Moderately

Important


Table 8.4: Decision-maker Relevance values

Criterion DM Relevance

BAC Pinkenba Commonwealth

Environmental Important

Moderately

Important Important

Zoning

Very

Important Irrelevant Very Important

Tenant amenity Important Irrelevant Important

Pinkenba Important

Very

Important Important

Landfill Important Irrelevant

Moderately

Important

Traffic Important Important

Moderately

Important


Figure 8.9: Performing an aggregation


8.4 Results

After all data was input the InfraPlanner aggregation interface was used to create

the following four maps:

1. A compensatory double-weighted aggregation (Utility)

2. Conflict assessment (Consensus)

3. Risk assessment (Safety)

4. Uncertainty assessment (Certainty)

The four parameter maps are shown in Figures 8.10 – 8.13.


Figure 8.10: Utility


Figure 8.11: Uncertainty


Figure 8.12: Risk


Figure 8.13: Conflict


An adjusted aggregation based on decision-maker preferences for the importance

of minimizing conflicts risks and uncertainty, or maximizing compensatory

suitability was then performed to enable an adjusted overall suitability estimate

to be derived. To perform an adjusted aggregation it is necessary to weight the

four output parameters, and the following weightings were used to derive the

map shown in Figure 8.14:

Utility is ‘Very Important’

Risk is ‘Very Important’

Conflict is ‘Important’

Uncertainty is ‘Unimportant’

Using the output maps and some further analysis it was possible to identify the

sites of interest (in this case based on individual cells) as shown in Figure 8.15.

Examining the sites exposes the main difficulties behind the site selection task.

Decision-makers disagreed on the best site for the facility and there was also a

difference between the site with the best utility and the site with the best safety.

The sites identified as having the best consensus and certainty, were not viable

solutions in this case as decision-makers agreed with certainty that these sites

were unsuitable. This illustrated that the parameters are not suitable for use in

isolation and must be combined effectively to generate valid solutions. The

adjusted aggregation leaned towards the site with a good combination of all

factors.

Using the interactive exploration interface it was then possible to examine each

alternative comprehensively, and narrow down possible solutions by setting

minimum thresholds for any parameter. The interface offers the ability to view

the decision area as a regular map or use any of the derived raster maps. Clicking

on a particular location produces a natural language analysis in real time as



Figure 8.14: Adjusted aggregation


Figure 8.15: Sites of interest


Figure 8.16: Alternative exploration

Unfortunately no location was completely satisfactory to all, and the primary

benefit gained from the system was the clear identification of the source of

conflict, which has become the subject of negotiation between parties.

8.5 Discussion

The nature of the site selection problem presented here is typical of many real

world situations. A fundamental problem in designing systems to solve such

problems is that there is often no universally accepted solution to find, and it is

not always possible to derive the best compromise from initial assessments. Most

GIS based decision-making methods assume that crisp numerical suitability

assessments can be processed according to a pre-determined algorithm to derive

a solution. However the complex nature of many site selection decisions make

such assumptions unrealistic. It was noted during the selection process that

decision-makers were reluctant to place their faith in a derived solution without

fully understanding how that solution was obtained. This creates a significant


hurdle for system designers whose aim is to replicate, and by default replace, the

decision-making process.

Using a pre-determined optimisation algorithm is standard procedure in many

areas of problem solving, and works particularly well when the exact utility of a

solution can be precisely measured and used as feedback to improve

performance. However the exact utility of a solution in site selection is seldom

known. Multiple, conflicting criteria, and the added human element of

conflicting opinions of measurement and importance create an ill-structured

problem that is often dynamic, in that assessments may change as the solution

space is examined. It is also relevant to note that problem-solving strategies vary

from person to person, making the group situation a particularly dynamic

environment.

InfraPlanner was designed as an intelligent spatial decision support system to

provide decision-makers with relevant, understandable processed information,

whilst leaving them in control of the decision-making process. To this end it was

noted that decision-makers expressed satisfaction with outputs, as they enabled

the group to find the core elements behind their conflicting assessments. In a real

world situation, where political issues can dominate operational concerns, it is

often most beneficial to identify these core areas as they may be traded off for

concessions outside the sphere of the site selection task.

Giving decision-makers the ability to generate a variety of solutions that

maximized aggregated suitability or minimized risk, conflict and uncertainty

provided an easily understandable way for decision-makers to take more control

of the analysis, rather than accepting imposed heuristics. Moreover, whilst the

system makes computationally deriving a solution from input data possible, its

major strength was the high information value of outputs. The experiment

confirmed that a focus on a meaningful, interactive exploration of alternative

outcomes, as opposed to attempting to derive a solution from initial inputs, is a

valid way to support decision-makers in their task. Further specific feedback was

limited due to data privacy issues.


There are some important limitations of the current ‘InfraPlanner’ system:

Firstly, the method used is limited to analysing problems with a single objective,

which makes it unsuitable for situations where multiple facilities are to be

located simultaneously or multiple land uses considered. Secondly, the use of

single cells as alternatives does not accurately represent the true size and spatial

configuration of a proposed development, which has been surprisingly seldom

noted (Brookes 1997). Lastly, utilizing linguistic terms for data input may

unnecessarily limit the accuracy of results in those cases where hard quantitative

data is available.

Another difficulty noted in the group situation was in the definition of criteria.

As an example, some decision-makers noted overlap in their perception of

community impact versus environmental impact. Some authors have described

multicriteria decisions, particularly those with multiple objectives, in terms of a

hierarchical structure, whereby some criteria encompass others, eg (Saaty 1980).

In a group situation this provides another area for disagreement and/or

misunderstanding.

The experience gained from the example at Brisbane Airport proved the validity

of an approximate reasoning approach to group site selection problems under

uncertainty. The InfraPlanner system enabled decision-makers to express their

assessments linguistically and receive meaningful linguistic feedback, whilst

taking more control of the process than other methods allow, and satisfaction

with outputs was expressed. The results also indicated a definite benefit from

utilizing a multi-decision-maker framework, as consensus was unattainable. An

emphasis on providing meaningful processed information, rather than offering a

heuristically derived solution was also found to be beneficial.

175Chapter 9 Conclusions

Chapter 9

CCOONNCCLLUUSSIIOONNSS

9.1 Introduction

This research has focused on the use of Approximate Reasoning to improve the

techniques and technology of spatial decision support in Infrastructure Site

Selection. A new Approximate Reasoning Algorithm for Infrastructure Site

Selection (ARAISS) was developed and implemented in a new Spatial Decision

Support System (InfraPlanner). The algorithm was then tested and validated in a

real world site selection problem at Australia’s Brisbane Airport.

This concluding chapter presents a final overview of the research presented in

this thesis. The activities conducted during the research program are summarised,

and the conclusions drawn are highlighted. Directions for future research are

suggested before finishing with a set of concluding remarks.

9.2 Summary of Results

The project was a combination of theory-focused research consisting of the

theoretical development of a new fuzzy site selection algorithm (ARAISS) and

design-focused research consisting of the practical application of the theory in a

new SDSS (InfraPlanner). Results are summarised in the following sections.

9.2.1 Answer to the Research Question

As stated in Chapter 1, the fundamental question behind this research was:


“Can Approximate Reasoning (AR) be integrated into a GIS based SDSS to

mitigate current difficulties with SDSSs utilised for Infrastructure Site

Selection?”

Results from hypothetical trial problems and a real world case study have clearly

shown that it is both possible and beneficial to integrate approximate reasoning

into a GIS based SDSS. As detailed in the following sections, ARAISS

performed well in both simulated and actual problems, in providing valid

solutions and supplementary information in a format that was easy for decision-

makers to use and understand.

9.2.2 Achievement of the Research Aims

The aim of this research was to create new knowledge at the intersection of

Physical Planning, Decision Science, Soft Computing, Decision Support and

Expert Systems, Geographical Information Systems, and Software Design.

This research has contributed to knowledge by showing that the integration of

AR and SDSSs is possible and putting forth one practical way to achieve it. A

new AR algorithm for site selection in GIS was devised, tested and implemented

in a real world case study. Both the algorithm and the outcomes of the case study

were separately peer reviewed and published at international conferences as

shown in Appendix A.

9.2.3 Achievement of the Research Objectives

Two specific research objectives were defined for this research:

1. Develop a practical infrastructure site selection algorithm based on an

Approximate Reasoning ‘linguistic’ approach.

2. Develop a new spatial decision support system based on the algorithm

developed in objective 1.


Objective 1 resulted in the creation of an Approximate Reasoning Algorithm for

Infrastructure Site Selection (ARAISS). ARAISS implements several concepts

that offer an improvement over current methodologies. The core capabilities of

ARAISS are its use of approximate reasoning to handle uncertainty, its multiple

decision-maker capability, its simplicity, and the way it hands over control to

decision-makers.

Objective 2 resulted in the creation of the InfraPlanner Spatial Decision Support

System. InfraPlanner is a prototype Spatial Decision Support System (SDSS)

designed to aid decision-makers with Group Multicriteria Location Problems. It

was created in ArcView GIS and is based on ARAISS.

9.3 Research Overview

The research was conducted in four phases as illustrated below:

1. Planning and Research:

• Needs assessment, problem diagnosis & definition of system

objectives.

• Review relevant literature and gather other information.

2. Analysis & Design:

• Conceptual design of the InfraPlanner system.

• Development of the decision-making algorithm.

3. Construction:

• Coding and debugging of the InfraPlanner prototype.

4. Implementation and Feedback:

• Testing and evaluation of InfraPlanner in a real world validation

problem.

• Critical assessment of the prototype and suggestions for future

improvements to the system.


The activities conducted and conclusions drawn from each of the phases are

summarised in the following sections.

9.3.1 Planning and research

Planning and research consisted of a set of increasingly targeted critical literature

reviews. The initial review found that Multicriteria Evaluation (MCE) was

currently the dominant analytical technique used in the solution of infrastructure

site selection problems. Several shortcomings were noted with current MCE

techniques, and most important of these were the inability to deal with

uncertainty, inability to deal with a group environment, and the perception by

decision-makers that current methods are not user friendly.

The universally accepted technology platform for the analysis of location

problems was found to be a Geographical Information System (GIS), coupled or

fully integrated with decision-making models. Advanced artificial intelligence

(AI) and soft computing techniques offered an ability to overcome some of the

shortcomings of MCE, but it was necessary to deploy them in a user friendly way

in order to avoid the perception of a ‘black box’ system. A ‘black box’ system

occurs when users have little or no understanding or control of an analysis

beyond the input of data and knowledge, and it was found that systems based on

current advanced AI techniques often fall within this category.

Approximate reasoning methods based on the use of fuzzy sets, were then

investigated. It was found that most fuzzy methods used in spatial problems

process crisp values obtained from simplifying fuzzy membership functions, and

not the functions themselves, thereby losing the information value of a fuzzy

quantity. A fuzzy number possesses both a mean value and a spread (support)

that may be used to indicate the uncertainty of an answer, however it was found

that there was currently no robust way for decision-makers to input their level of

confidence in applying a particular linguistic label.

The inflexibility of a fuzzy inference system once the rules were generated, and

the extra processing power required for fuzzy pairwise comparison methods left


fuzzy MCE as the most appropriate approach to facility site selection. A method

was needed to incorporate approximate reasoning in an MCE analysis suitable

for site selection problems.

The planning and research phase concluded with a review of current software

systems used to solve site selection problems. The umbrella term used for these

systems is Spatial Decision Support Systems (SDSSs). SDSSs are a type of

Decision Support System (DSS) that integrates GIS technology with decision-

making models to aid in the solution of spatial problems. It was found that the

ideal SDSS would be both flexible and user friendly, be fully integrated within a

single GIS software package, provide real-time graphical interactivity and cater

for group decision-making.

The literature consistently noted that the major hurdle facing developers was how

to make systems that are simple and easy to use. There was found to be a general

tendency towards ‘shallow use’ of SDSSs by real world planners and decision-

makers, which was largely the result of real or perceived difficulty in using such

systems. There was also found to be a void of systems capable of accepting

uncertainty assessments directly from decision-makers.

9.3.2 Analysis and design

It was noted during the analysis stage that limitations on current SDSSs are

derived from an inability to deal with multiple conflicting parties, an inability to

handle uncertainty, a lack of simplicity in use and interaction and not delivering

enough control to decision-makers. A conceptual blueprint for the design of a

new algorithm and its implementation in a SDSS was created, and it was

proposed that the new system should possess the following characteristics:

• The ability to accept inputs from a heterogeneous group of decision-makers,

independently weighting and rating multiple attributes.


• An approximate reasoning algorithm based on a fuzzy MCE aggregation of

parameter-based fuzzy numbers that encapsulate linguistic suitability and

uncertainty assessments.

• The algorithm should utilise arithmetic operators for aggregation and a

scoring function for de-fuzzification to minimise calculation time and enable

real-time interactivity.

• The system should be fully integrated into existing GIS software.

• Linguistic outputs should be a set of descriptive parameters that give

decision-makers the ability to choose the characteristics of a solution that are

most appropriate to their specific problem, thereby enabling them to gain

control over the properties maximised during aggregation.

The ARAISS site selection algorithm was designed to achieve the goals outlined

during conceptual design. The algorithm works by extracting four parameters

inherent in each alternative that indicate levels of Utility, Safety, Consensus, and

Certainty. Weighting of the four parameters enables decision-makers to decide

which aspects of the solution are most important to their specific problem, and

thereby delivers a real means of control over algorithm performance. The

algorithm performed as expected in example problems, delivering sound results

in a five alternative, three decision-maker problem with simulated inputs.

9.3.3 Construction

InfraPlanner was created as a working prototype of a generic SDSS for site

selection problems of a strategic nature. The system demonstrated that

approximate reasoning techniques are suitable for use in SDSSs, although

designing and building the InfraPlanner Spatial Decision Support System proved

to be extremely challenging. Constructing a DSS is generally considered to be a

complex, time consuming task, requiring a group of skilled individuals, and this

was proven in practice. There are many small issues that are not generic enough

to be mentioned in publications on SDSSs but nonetheless proved problematic.

Among these were choosing a GIS package from the myriad of options available,


and dealing with the organisational changes that occurred during the

development process.

9.3.4 Implementation and feedback

An experiment was conducted using a real world site selection decision at

Brisbane Airport, where the desired inputs and outputs could be generated and

commented upon by actual decision-makers. Inputs were generated for three

stakeholder groups using actual decision-makers or representatives chosen by the

experimenter for their knowledge of the situation. The problem used was real,

and the objective was to choose the best location for a recycling facility on the

2700 ha Brisbane Airport site.

The experiment confirmed that a focus on a meaningful, interactive exploration

of alternative outcomes, as opposed to attempting to derive a solution from initial

inputs, is a valid way to support decision-makers in their task. The results

generated by the system were found to be sound, and corresponded well with the

real sites preferred by the decision-making group. Decision-makers found the

method easy to use and the outputs were perceived as helpful.

It is important to note that ARAISS was designed to analyse qualitative site

selection problems with a single objective, and may need to be augmented to

cater for at least three other common types of site selection problems. Firstly in

situations where multiple facilities are to be located simultaneously or multiple

land uses considered. Secondly where the use of single cells as alternatives does

not effectively represent the spatial configuration of a proposed development.

Thirdly where using linguistic terms for data input does not offer the best means

of information input, for example where hard quantitative data is available.

9.4 Validation

A fundamental problem in designing an algorithm to solve infrastructure site

selection problems is that there is often no perfect solution to find, and it is not


always possible to derive the best compromise from initial assessments. Using a

pre-determined optimization algorithm is standard procedure in many areas of

problem solving, and works particularly well when the exact utility of a solution

can be precisely measured and used as feedback to improve performance.

However the exact utility of a solution in site selection is seldom known.

Multiple, conflicting criteria, and the added human element of conflicting

opinions of measurement and importance create an ill-structured problem that is

often dynamic, in that assessments may change as the solution space is

examined. It is also relevant to note that problem-solving strategies vary from

person to person, making the group situation particularly dynamic. In such a

climate the traditional model of testing a new algorithm against others using

standard test data and set benchmarks becomes obsolete.

In the case of ARAISS a second major hurdle is the absence of a standard dataset

with which to generate results and compare those results to known solutions.

Datasets used in other published work on multi-criteria site selection either lacks

multiple decision-maker inputs, or uncertainty data. In fact due to the unique

approach of ARAISS, which requires decision-makers to weight output

parameters not generated by other methods, validation is challenging from the

outset.

In the absence of an existing dataset with all the necessary inputs and outputs to

test ARAISS and InfraPlanner, there were three practical means of validation

applicable to this research:

1. Using custom made sample datasets based on hypothetical problems to

evaluate the success of the algorithm.

2. Using a real problem to gain feedback on the suitability of results

produced by ARAISS, and the benefits gained by using InfraPlanner.

3. Peer review of the ARAISS model and the process used to develop it.

Use of the first method was described Section 6.3, which gives the results of

simulation exercises conducted using MATLAB. Several MATLAB simulations

were conducted to test the common sense validity of the algorithm. The problems


were all based on three decision-makers rating five alternatives with respect to

three criteria. Results showed that ARAISS performed as expected, producing

commonsense results and successfully extracting the four output parameters of

Utility, Certainty, Risk and Conflict.

Chapter 8 describes the use of Infra Planner in a real site selection problem at

Australia’s Brisbane Airport. The problem was based on six criteria with inputs

coming from three separate decision-maker groups. Once again sound results

were produced, with the algorithm selecting the same site as had been previously

earmarked. Decision-makers found it easy to provide their preferences

linguistically, and the output information provided by InfraPlanner was found to

be useful and easily interpretable.

Peer review was facilitated first and foremost by publication of the ARAISS

algorithm and its implementation in InfraPlanner as shown in Appendix A, and

secondly via a focus group conducted at the ANZIIS 2003 conference. Feedback

was positive, with all present agreeing that both the development process and the

model derived from it was valid. Some researchers noted that the use of a

software design flowchart would also be a good way to represent the model, as

they found the logic model used difficult to follow.

9.5 Key Findings

The following summary points are presented here as the key findings of this

research.

1. The key limitations of existing MCE techniques used for site selection were

found to be:

a) An inability to handle uncertainty.

b) An inability to handle a group decision-making environment.

c) Real or perceived difficulty of use, and a limited sense of control.


2. It was found that approximate reasoning could be used to mitigate these

difficulties in the following ways:

a) Uncertainty in spatial decision-making may be modelled using fuzzy

numbers to quantify linguistic suitability assessments. The fuzzy numbers

may be scaled for varying levels of uncertainty using the concept of type-

2 fuzzy sets.

b) Inputs from a heterogeneous group may be brought together by the use of

a relevance matrix, which is a device to weight a decision-makers ability

to judge a particular criterion.

c) The use of linguistic inputs and outputs, coupled with an emphasis on

providing useful information rather than direct solutions was found to

provide a simple way to interact with a SDSS, that delivered a greater

sense of control. In particular the identification of the overall utility,

uncertainty, risk, and conflict inherent in each solution provides greater

information value than a single numerical score.

9.6 Directions for future research

Expanding on the working prototype opens up several possibilities, and further

work is recommended to expand ARAISS and InfraPlanner to be capable of

handling multiple facility problems, and explicitly include the size and spatial

configuration of the required land parcels. Potential also exists to include existing

philosophies that have proven effective in this type of problem such as factor

analysis, approaches based on the triple bottom line, the use of key performance

indicators, and data envelopment analysis.

Genetic algorithms also offer a promising method to explore feasible alternatives

without resorting to the massive number of calculations required to fully examine

the solution space of such problems. Research on other artificial intelligence

techniques such as neural networks should produce benefits in complex spatial

decisions.


At a practical level, this new functionality may be added to InfraPlanner in three

basic ways:

1. By enhancement of the suitability map generation capability to include

extra parameters

2. By enhancement of the aggregation capability to process extra

information

3. By enhancement of the interactive feedback capability to display and

optimise based on the extra data required in the approaches outlined

above

There also exist several fundamental difficulties with multi-criteria decision-

making not addressed in this thesis, that offer promising direction for future

work. These include:

• Selection of criteria and criterion overlap

• Development of more accurate means of semantically representing decision-

maker preferences

• Methods for generating consensus in a group environment

• Methods for choosing a suitable decision-maker group

• Methods to quickly process raw data into the format necessary for use in a

SDSS

9.7 Concluding remarks

This research has produced a new fuzzy algorithm for the selection of sites for

large-scale infrastructure, and implemented it in a new Spatial Decision Support

System. The algorithm performed well in both hypothetical and real site selection

problems, however hard empirical validation is difficult to perform when there is

no completely accurate way to rate solutions to complex site selection tasks.


The construction of the algorithm, based on type-2 fuzzy set concepts, proved

practical, and produced results consistent with those chosen by real world

planners and decision-makers. Calculation times were sufficiently short to enable

seamless integration into a real-time GIS based analysis, and the format of inputs

and outputs proved simple and easy for users to understand.

Further validation of the methods developed in this research is recommended, as

are the integration of artificial intelligence techniques and other decision-making

philosophies. The confluence of Physical Planning, Decision Science, Fuzzy

Logic, Soft Computing, Decision Support and Expert Systems, Geographical

Information Systems, and Software Design should prove to be fertile ground for

innovation for many years to come.

References 187

RREEFFEERREENNCCEESS

Affum, J. K. (1997). “Integration and development of GIS-based tools for

transportation planning applications.” Transportation Research Part A:

Policy and Practice 31(1): 58.

Almond, N., D. Jenkins, et al. (1997). Development of a prototype residential

valuation system. 5th Conference on computers in urban planning and

urban management, Bombay.

Author, A. (1999). “Introduction to paper and commentaries on the Delphi

technique.” International Journal of Forecasting 15(4): 351-352.

Baas, S. M. and H. Kwakernaak (1977). “Rating and ranking of multiple-aspect

alternatives using fuzzy sets.” Automatica 13(1): 47-58.

Batty, M. and Y. Xie (1994). “Research Article. Modelling inside GIS: Part 1.

Model structures, exploratory data analysis and aggregation.” Int. J.

Geographical Information Systems 8(3): 291.

Beard, M. K. (1994). Accommodating uncertainty in query response. 6th

International Symposium on Spatial Data Handling, London, Taylor &

Francis.

Beckman, M. (1968). Location Theory. New York, Random House.

Bell, M., C. Dean, et al. (2000). “Forecasting the pattern of urban growth with

PUP: a web-based model interfaced with GIS and 3D animation.”

Computers, Environment and Urban Systems 24(6): 559-581.

Bellamy, J. A. and D. Lowes (1999). “Modelling Change in State of Complex

Ecological Systems in Space and Time: An Application to Sustainable

Grazing Management.” Environment International 25(6-7): 701-712.

Bellman, R. E. and L. A. Zadeh (1970). “Decision-making in a fuzzy

environment.” Management Science 17(4): 141-64.

Bodily, S. E. (1985). Modern decision-making: a guide to modelling with

decision support systems. New york, McGraw-Hill.

Bogetoft, P. and P. M. Pruzan (1997). Planning with multiple criteria :

investigation, communication, and choice. Copenhagen, Copenhagen

Business School Press.

References 188

Bonissone, P. and K. S. Decker (1986). Selecting uncertainty calculi and

granularity: an experiment in trading off precision and complexity.

Uncertainty in artificial intelligence. L. H. Kanal and J. F. Lemmer.

Amsterdam, North Holland: 217-247.

Bonissone, P. P. (1982). A fuzzy sets based linguistic approach: Theory and

applications. Approximate reasoning in decision analysis. M. Gupta and

E. Sanchez. Amsterdam, North Holland Publishing Company: 329-339.

Brail, R. (2000). Introduction. Planning support systems: integrating geographic

information systems, models and visualisation tools. R. Brail and R.

Klosterman. Redlands, California, ESRI Press.

Brookes, C. J. (1997). “A genetic algorithm for locating optimal sites on raster

suitability maps.” Transactions in GIS 2: 201-212.

Brown, R. V. (1998). Fitting Decisison Aids to an Institutional Context:

Methodology for Organizational Design, Decision Analysis Society.

2001.

Burrough, P. A. and R. A. McDonnell (1998). Principles of geographical

information systems. Oxford, Oxford University Press.

Carlsson, C. and R. Fuller (1996). “Fuzzy multiple criteria decision making:

Recent developments.” Fuzzy Sets and Systems 78(2): 139-153.

Carver, S. (1991). “Integrating multi-criteria evaluation with geographical

information systems.” Int. J. Geographical Information Systems 5(3):

321-339.

Chou, H. and Y. Ding (1992). Methodology of integrating spatial

analysis/modelling and GIS. 5th international symposium on spatial data

handling, Charleston, SC.

Chu-Carrol, J. and S. Carberry (2000). “Conflict resolution in collaborative

planning dialogs.” Int. J. Human-Computer Studies(53): 969-1015.

Colorni, A., E. Laniado, et al. (1999). “Decision support systems for

environmental impact assessment of transport infrastructures.”

Transportation Research Part D: Transport and Environment 4(1): 1-11.

Congalton, R. G. (1997). “Exploring and evaluating the consequences of vector-

to-raster and raster-to-vector conversion.” Photogrammetric Engineering

and Remote Sensing 63(4): 425-434.

References 189

Crossland, M. D., B. E. Wynne, et al. (1995). “Spatial decision support systems:

An overview of technology and a test of efficacy.” Decision Support

Systems 14(3): 219-235.

DeBruin, S. and A. Bregt (2001). “Assessing fitness for use: the expected value

of spatial data sets.” Int. J. Geographical Information Systems 15(5): 457-

471.

Delgado, M., J. L. Verdegay, et al. (1992). “Linguistic decision-making models.”

International Journal of Intelligent Systems(7): 351-370.

Demers, M. (2000). Fundamentals of Geographic Information Systems. New

York, John Wiley & Sons.

Eastman, J. R., W. Jin, et al. (1995). “Raster procedures for multi-criteria/multi-

objective decisions.” Photogrammetric Engineering and Remote Sensing

61(5): 539-547.

ESRI (2001). Building custom solutions (developing with COM) ArcObjects.

OZRI 2001, Sydney, ESRI Press.

ESRI (2001). Data for your GIS. 2001.

Fedra, K. (1993). GIS and environmental modelling. Environmental modelling

with GIS. M. Goodchild, B. Parks and L. Steyaert. New York, Oxford

University Press.

Feng, S. and L. Xu (1999). “An intelligent decision support system for fuzzy

comprehensive evaluation of urban development.” Expert Systems with

Applications 16(1): 21-32.

Fisher, P. (2000). Fuzzy modelling. GeoComputation. S. Openshaw and R. J.

Abrahart. London, Tayler & Francis.

Flassak, T., H. Witt, et al. (1995). “Surveillance system for air pollutants by

combination of the decision support system COMPAS and optical remote

sensing systems.” Proceedings of SPIE - The International Society for

Optical Engineering 2506: 266-273.

Fodor, J., P. Perny, et al. (1998). Decision-making models. Handbook of fuzzy

computation. E. H. Ruspini, P. P. Bonissone and W. Pedrycz. London,

Institute of Physics Publishing.

Ghermay, L., H. S. Rochon, et al. (2000). “Utilizing archival remotely sensed

data in support of wetland resource management.” International

Geoscience and Remote Sensing Symposium (IGARSS) 5: 1921-1923.

References 190

Goodchild, M., R. Haining, et al. (1992). “Integrating GIS and spatial data

analysis: problems and possibilities.” Int. J. Geographical Information

Systems 6(5): 407-423.

Goonetilleke, A. and A. Jenkins (1995). GIS And Hydrologic Modelling: An

experimental investigation of the influence of area estimation error on

surface modelling. Brisbane, Physical Infrastructure Centre - Queensland

University of Technology.

Haastrup, P., V. Maniezzo, et al. (1998). “A decision support system for urban

waste management.” European Journal of Operational Research 109(2):

330-341.

Hall, G. B., F. Wang, et al. (1992). “Comparison of boolean and fuzzy

classification methods in land suitability analysis by using a geographical

information system.” Environment and Planning A 24: 497-516.

Hepner, G. F. (1984). “Using value functions as a possible suitability scaling

procedure in automated composition mapping.” Professional geographers

36(4): 468-472.

Herrera, F. and E. Herrera-Viedma (2000). “Linguistic decision analysis: steps

for solving decision problems under linguistic information.” Fuzzy Sets

and Systems 115(1): 67-82.

Herrera, F., E. Herrera-Viedma, et al. (1996). “A model of consensus in group

decision making under linguistic assessments.” Fuzzy Sets and Systems

78(1): 73-87.

Heywood, I., J. Oliver, et al. (1995). “Building an exploratory multi-criteria

modelling environment for spatial decision support.” Innovations in GIS

2: 127-136.

Hobbs, B., F (1980). “A comparison of weighting methods in powerr plant

siting.” Decision Sciences 11: 725-737.

Holland, J. (1975). Adaptation in natural and artificial systems. Michigan,

University of Michigan press.

Holtfrerich, D. R. and C. Yew Kuan (1993). Integration of a commercial GIS in

INFORMS-TX. GIS 93, Vancouver.

Hopkins, L. (1977). “Methods for generating land suitability maps: a

comparative evaluation.” Journal for American institute of planners

34(1): 19-29.

References 191

Hunter, G. (1997). Socio-economic and Environmental Applications Research:

Overview and Future Prospects II - Invited Response. Geographic

Information Research at the Millennium, Le Bischenberg, France.

Hunter, G. J. and M. Goodchild (1995). “Dealing with error in spatial data sets: a

simple case study.” Photogrammetric Engineering and Remote

Sensing(61): 529-537.

Hwang, C. L. and K. Yoon (1981). Multiple attribute decision making methods

and applications: A state of the art survey. Berlin, Springer-Verlag.

Jankowski, P. (1995). “Integrating geographical information systems and

multiple criteria decision-making methods.” Int. J. Geographical

Information Systems 9(3): 251-273.

Jankowski, P., N. Andrienko, et al. (2001). “Map-centred exploratory approach

to multiple criteria spatial decision making.” Int. J. Geographical

Information Science 15(2): 101-127.

Ji, W. (1996). “Ecosystem management: a decision support GIS approach.”

International Geoscience and Remote Sensing Symposium (IGARSS) 4:

2225-2227.

Jiang, H. and J. R. Eastman (2000). “Application of fuzzy measures in multi-

criteria evaluation.” Int. J. Geographical Information Science 14(2): 173-

184.

Keenan, P. B. (1998). “Spatial decision support systems for vehicle routing.”

Decision Support Systems 22(1): 65-71.

Keeney, R. L. (1992). Value-focused thinking: a path to creative decision-

making. Cambridge, MA, Harvard University Press.

Keeney, R. L. and H. Raiffa (1976). Decisions with multiple objectives:

preferences and value trade-offs. New York, Wiley.

Klir, G. and B. Yuan (1999). Basic concepts and history of fuzzy set theory and

fuzzy logic. Handbook of fuzzy computation. E. H. Ruspini, P. P.

Bonissone and W. Pedrycz. London, Institute of physics publishing.

Klosterman, R. (2000). Planing support systems: a new perspective on computer

aided planning. Planning support systems: integrating geographic

information systems, models and visualisation tools. R. Brail and R.

Klosterman. Redlands, California, ESRI Press.

References 192

Klungboonkrong, P. and M. Taylor (1998). “A microcomputer-based- system for

multicriteria environmental impacts evaluation of urban road networks.”

Computers, Environment and Urban Systems 22(5): 425-446.

Kraak, M. J. (1999). Visualising spatial distributions. Geographical information

systems: principles and technical issues. P. Longley, M. Goodchild, D.

Maguire and D. Rhind. New York, John Wiley & Sons Inc.

Krzanowski, R. M. and J. Raper (1999). “Hybrid genetic algorithm for

transmitter location in wireless networks.” Computers, Environment and

Urban Systems 23: 359-382.

Kwak, N. K. and L. Changwon (1998). “A multicriteria decision-making

approach to university resource allocations and information infrastructure

planning.” European Journal of Operational Research 110(2): 234-242.

Kyriakidis, P. C. and M. Goodchild (1999). “Geostatistics for conflation and

accuracy assessment of digital elevation models.” Int. J. Geographical

Information Science(13): 677-707.

Laaribi, A., J. J. Chevallier, et al. (1996). “A spatial decision aid: a multicriterion

evaluation approach.” Computers, Environment and Urban Systems

20(6): 351-366.

Lam, D. and D. Swayne (2001). “Issues of EIS software design: some lessons

learned in the past decade.” Environmental Modelling and Software 16:

419-425.

Leon, L. F., D. C. Lam, et al. (2000). “Integration of a nonpoint source pollution

model with a decision support system.” Environmental Modelling and

Software 15(3): 249-255.

Liang, G.-S. and M.-J. J. Wang (1991). “A fuzzy multicriteria decision-making

method for facility site selection.” Int. J. Production Research 29(11):

2313-2330.

Lu, H.-P., H.-J. Yu, et al. (2001). “The effects of cognitive style and model type

on DSS acceptance: An empirical study.” European Journal of

Operational Research 131(3): 649-663.

Malczewski, J. (1996). “A GIS-based approach to multiple criteria group

decision-making.” Int. J. Geographical Information Systems 10(8): 955-

971.

References 193

Malczewski, J. (1999). GIS and multicriteria decision analysis. New York, John

Wiley.

Malczewski, J. (2002). “Fuzzy screening for land suitability analysis.” Graphical

& Environmental Modelling 6(1): 27-39.

Malczewski, J. and M. Jackson (2000). “Multicriteria spatial allocation of

educational resources: an overview.” Socio-Economic Planning Sciences

34(3): 219-235.

Malczewski, J., M. Pazner, et al. (1997). “Visualization of multicriteria location

analysis using raster GIS: A case study.” Cartography and Geographic

Information System 24(2): 80-90.

Mamdani, E. H. and S. Assilian (1975). “An experiment in linguistic synthesis

with a fuzzy logic

controller.” International Journal of Man-Machine Studies 7(1): 1-13.

Maniezzo, V., I. Mendes, et al. (1998). “Decision support for siting problems.”

Decision Support Systems 23(3): 273-284.

Massam, B. H. (1988). “Multicriteria decision making (MCDM) techniques in

planning.” Progress in planning 30(1): 1-82.

Matthews, K. B., A. R. Sibbald, et al. (1999). “Implementation of a spatial

decision support system for rural land use planning: integrating

geographic information system and environmental models with search

and optimisation algorithms.” Computers and Electronics in Agriculture

23(1): 9-26.

McHarg, I. (1969) Design with nature. New York, Natural History Press.

Mendell, J. M. and R. I. John (2002). Footprint of uncertainty and its importance

to type-2 fuzzy sets. Sixth international conference on artificial

intelligence and soft computing, Banff, Canada, IASTED.

Mitchell, M. (1999). An introduction to genetic algorithms. Massachusetts, MIT

Press.

Morrill, R. and J. Symons (1977). “Efficiency and equity aspects of optimum

location.” Geographical Analysis(9): 215-225.

Murray, A. T. (2003). “Site placement uncertainty in location analysis.”

Computers Environment and Urban Systems 27: 205-221.

Nijkamp, P. and M. Giaoutzi (1993.). Decision support models for regional

sustainable development : an application of geographic information

References 194

systems and evaluation models to the Greek Sporades Islands. Aldershot

:, Avebury,.

Nijkamp, P., P. Rietveld, et al. (1990). Multicriteria evaluation in physical

planning. New York, Elsevier Science.

Nyerges, T. L. (1992). Coupling GIS and spatial analytical models. 5th

international symposium on spatial data handling, Charleston, SC.

O'Sullivan, D. and D. J. Unwin (2003). Geographic information analysis.

Hoboken, NJ., Wiley.

Owen, P. (1993). The right tool for the task: integrating spatial analysis

techniques. GIS/LIS 93.

Pereira, A. G. (1996). “Generating alternative routes by multicriteria evaluation

and genetic algorithm.” Environment And Planning B: planning and

Design 23: 711-720.

Pereira, J. and L. Duckstein (1993). “A multiple criteria decision-making

approach to GIS-based land suitability evaluation.” Int. J. Geographical

Information Systems 7(5): 407-424.

Perny, P. and B. Roy (1992). “The use of fuzzy outranking relations in

preference modelling.” Fuzzy Sets and Systems(49): 33-53.

Pettit, C. and D. Pullar (1999). “An integrated planning tool based upon multiple

criteria evaluation of spatial information.” Computers, Environment and

Urban Systems 23(5): 339-357.

Prato, T. (1999). “Multiple attribute decision analysis for ecosystem

management.” Ecological Economics 30(2): 207-222.

Pullar, D. and C. Pettit (2000). Use of Decision Support Systems: Decision

Alchemy or Science. AURISA 2000, Coolum, Australia.

Quattrochi, D. A., J. C. Luvall, et al. (2000). “A decision support information

system for urban landscape management using thermal infrared data.”

Photogrammetric Engineering and Remote Sensing 66(10): 1195-1207.

Ribeiro, R. A. (1996). “Fuzzy multiple attribute decision making: A review and

new preference elicitation techniques.” Fuzzy Sets and Systems 78(2):

155-181.

Ruspini, E. H. and E. H. Mamdani (1998). Approximate reasoning. Handbook of

fuzzy computation. E. H. Ruspini, P. P. Bonissone and W. Pedrycz.

London, Institute of Physics publishing.

References 195

Saaty, T. L. (1980). The Analytic Hierarchy Process. New York, McGraw Hill.

Saaty, T. L. and K. P. Kearns (1985). Analytical Planning - The Organization of

Systems. Sydney, Pergamon Press.

Sandhu, R. and P. Treleaven (1996). Client-server approaches to model

integration within GIS. 3rd international conference on integrating GIS

and environmental modelling, Santa Fe.

Scholten, H. J. and A. LoCascio (1997). GIS Application Research: History,

Trends and Developments. Geographic Information Research at the

Millennium, Le Bischenberg, France.

Scott-Morton, M. S. (1971). Management Decision Systems: Computer-Based

Support for Decision Making. Cambridge, MA, Division of Research,

Harvard university.

Sharifi, M. A., W. V. D. Toorn, et al. (2002). “Application of GIS and

multicriteria evaluation in locating sustainable boundary between the

Tunari National Park and Cochabamba City (Bolivia).” Journal of Multi-

Criteria Decision Analysis 11: 151-164.

Shiffer, M. J. (1992). “Towards a Collaborative Planning System.” Environment

And Planning B: planning and Design 19: 709-722.

Shiffer, M. J. (1994). A geographically Based Multimedia Approach to City

Planning. Human Factors in Computing Systems. C. Plaisant. New York,

New York: Association for Computing Machinery: 265-266.

Simon, H. A. (1960). The new science of management decision. New York,

Harper & Row.

Smith, P. (1980). A Review of Some Methods for Weighting Criteria in the

Evaluation of Multi-Dimensional Alternatives. Planning Research Papers.

Brisbane, University of Queensland. 4.

Sparkman, T. G. (1999). Lessons learned applying software engineering

principles to visual programming language application development.

IEEE Computer Society's International Computer Software and

Applications Conference, Phoenix Arizona.

Spradlin, T. (1997). A Lexicon of Decision Making, Decision Analysis Society.

2001.

Sprague, R. H. and H. J. Watson (1993). Decision Support Systems: Putting

Theory Into Practice. Englewood Cliffs, New Jersey, Prentice Hall.

References 196

Starr, M. K. and M. Zeleney (1977). Multiple Criteria Decision Making.

Amsterdam, North Holland.

Stefanakis, E., M. Vazirgiannis, et al. (1996). Spatial decision-making based on

fuzzy set methodologies. International Archives of Photogrammetry and

Remote Sensing, Vienna.

Stillwell, W. G., D. A. Seaver, et al. (1981). “A comparison of weight

approximation techniques in multiattribute utility decision making.”

Organizational Behaviour and Human Performance 28(1): 62-77.

Strand, E. J. (1991). “The system integrator's shell game: where's the GIS.” GIS

World (Oct,91): 26-28.

Stuart, N. and C. Stocks (1993). Hydrological modelling within GIS: an

integrated approach. HydroGIS 93: Application of geographical

information systems in hydrology and water resources, Vienna.

Sui, D. (1993). “Integrating neural networks with GIS for spatial decision-

making.” Operational Geographer 11(2): 13-20.

Sui, D. and M. Goodchild (2001). “GIS as media?” Int. J. Geographical

Information Science 15(5): 387-390.

Theobald, D. M. (2000). “Reducing linear and perimeter measurement errors in

raster-based data.” Cartography and Geographic Information Science

27(2): 111-116.

Tim, S. U., M. Milner, et al. (1992). GIS/simulation model linkage: processes,

problems and opportunities. American society of agricultural engineers.

Winter meeting.

Tomlin, C. D. (1990). Geographic information systems and cartographic

modelling. New Jersey, Prentice Hall.

Tong, R. and P. Bonissone (1980). “A linguistic approach to decisionmaking

with fuzzy sets.” IEEE Transactions on Systems, Man and Cybernetics

10(11): 716-723.

Turban, E. (1995). Decision Support and Expert Systems. New Jersey, Prentice-

Hall.

Voogd, H. (1983). Multicriteria evaluation for urban and regional planning.

London, Pion.

References 197

Worboys, M. (1999). Relational databases and beyond. Geographical information

systems: principles and technical issues. P. Longley, M. Goodchild, D.

Maguire and D. Rhind. New York, John Wiley & Sons Inc.

Yager, R. R. (1988). “On ordered weighted averaging aggregation operators in

multi-criteria decision making.” IEEE Transactions on Systems, Man and

Cybernetics 18(1): 183-190.

Yager, R. R. (1995). “An Approach to Ordinal Decision Making.” International

Journal of Approximate Reasoning 12(3-4): 237-261.

Zadeh, L. A. (1965). “Fuzzy Sets.” Information Control 8: 338-53.

Zadeh, L. A. (1973). “Outline of a new approach to the analysis of complex

systems and decision processes.” IEEE Transactions on Systems, Man

and Cybernetics SMC(3): 28-44.

Zadeh, L. A. (1975). “The concept of a linguistic variable and its application to

approximate reasoning.” Inf Sci 8(Part I): 199-249.

Zadeh, L. A. (1975). Fuzzy logic and approximate reasoning. Fuzzy sets fuzzy

logic and fuzzy systems: selected papers by Lotfi A. Zadeh. G. Klir and

B. Yuan. Singapore, World Scientific.

Zadeh, L. A. (1976). The linguistic approach and its application to decision

analysis. Fuzzy sets, fuzzy logic, and fuzzy systems: Selected papers by

Lotfi A.Zadeh. G. Klir and B. Yuan. Singapore, World Scientific.

Zeleny, M. (1982). Multiple Criteria decision Making. New York, Mcgraw Hill.

Zeng, T. and Q. Zhou (2001). “Optimal spatial decision making using GIS: a

prototype of a real estate geographical information system (REGIS).” Int.

J. Geographical Information Science 15(4): 307-321.

Zhou, J. and D. L. Civko (1996). “Using genetic learning neural networks for

spatial decision-making in GIS.” Photogrammetric engineering and

remote sensing 11: 1287-1295.

References 198


Appendix A Publications 199

Appendix A

PPUUBBLLIICCAATTIIOONNSS


The following peer reviewed original publications were the direct result of the

research presented in this thesis. They are reprinted here in Appendix A.

1. Bailey, D., A. Goonetilleke, and M. Deriche. A decision support system for site selection of large-scale infrastructure facilities using natural language. in Operations Research into the 21st Century. 2003. Noosa, Australia: The Australian Society for Operations Research (ASOR).

2. Bailey, D., A. Goonetilleke, and D. Campbell. Information analysis and dissemination for site selection decisions using a fuzzy algorithm in GIS. in Information and Knowledge Sharing. 2003. Scottsdale, Arizona: ACTA Press.

3. Bailey, D., D. Campbell, and A. Goonetilleke. An experiment with approximate reasoning using 'InfraPlanner'. in ANZIIS 2003. 2003. Sydney: Queensland University of Technology.

halla

These articles are not available online. Please consult the hardcopy thesis available from the QUT Library

Appendix B The Brisbane Airport Environment 203

Appendix B

TTHHEE BBRRIISSBBAANNEE AAIIRRPPOORRTT

EENNVVIIRROONNMMEENNTT


Brisbane Airport occupies a site of 2700 ha, located 13km north east of the

Brisbane CBD, adjoining Moreton bay. The flat and low lying site occupies part

of the original Brisbane river delta, which has undergone extensive changes since

the 1830s, with most of the original network of tidal waterways being replaced

with constructed drains. Much of the vegetation on the site has been planted in

the last 15 years, and was chosen to reduce the attraction of birds.

The major infrastructure currently on the Airport site as shown in Figure 8.1

consists of:

• 3560m long main runway (01/19) and associated taxiways

• 1760m long cross runway (14/32) and associated taxiways

• Domestic terminal building and apron

• International terminal building and apron

• General aviation buildings and apron

• Old international terminal building and apron

• Several Maintenance and support facilities

• Two major freight facilities plus several smaller facilities

• Three major catering facilities

• Private and rental vehicle parking areas

• Administration offices and control facilities

• Refuelling facilities and depot

• Lighting to runways, taxiways, aprons roads and car parks

Ten habitats have been classified on the site, as shown in the table below:

HHaabbiittaatt DDeessccrriippttiioonn SSttaattuuttoorryy

CCoonnssiiddeerraattiioonnss

CCoonnsseerrvvaattiioonn

VVaalluuee

Casuarina Plantation

Monoculture of casuarina glauca, originally providing a relatively poor habitat, but well-established areas infested with weeds and may become attractive.

Low


Open Grassland

Closely mown grass surrounding the airports core facilities, which provides a poor habitat.

Low

Remnant Mangrove Communities (general)

Avicennia marina, the grey mangrove, is the dominant species. These are native mangrove communities, which are in good condition, provide a complex and diverse habitat, and contribute to the productivity of adjoining fisheries.

A,E,I,C,W,R,J,Q,F,N

High

Serpentine creek mouth mangrove community

A closed scrub community of Ceriops taga,l which is uncommon within the bay. It is contiguous with the luggage point to jubilee creek mangroves and provides a significant habitat for mammals, reptiles, amphibians and avifauna.

A,E,I,C,W,R,J,Q,F,N

Very high

Channel Mangrove Communities

Numerous drainage channels have been colonised by, or planted with grey or river mangroves.

A,E,I,W,R,J,Q,F,N

Moderate

Remnant Saltmarsh Communities

These communities fringe areas of remnant mangrove, and also occur in freshwater wetland sedges adjacent to Kedron Brook Floodway. The saltcouch Sporobolous virginicus) dominates, although patches of samphire are common.

A,E,I,W,R,J,Q,F,N

High

Freshwater Wetlands and Sedge Communities

These communities are recently colonised in poorly drained and inadequately filled former sandmining areas. The dominant species is Phragmites australis, although much diversity is supported by wetland areas in general.

A,E,I,W,R,J,Q,F,N

High

Coastal Dunes and Foreshore

May form important habitats for migratory (and other) birds.

A,E,I,C,W,R,J,Q,F,N

Moderate

Remnant and Engineered Creeks and Channels

It is likely that Serpentine and Jackson creeks support significant communities of flathead, whiting and bream. Engineered creeks and channels are likely to serve as significant nursery grounds for many species that subsequently migrate to the open waters of Moreton Bay, as well as invertebrates.

A,E,I,W,R,J,Q,F,N

Moderate

Remnant Bushland

A small (5ha) isolated bushland site at Pinkenba

A,I,Q Low

Abbreviations for statutory considerations:


A - Airports act 1996 and the Airports (Environment Protection) Regulations E - Endangered Species Protection Act 1992 I - Environmental Protection (Impact Proposals) Act 1974 C - Coastal Protection and Management Act 1995 W - Wetlands Policy of the Commonwealth of Australia 1997 R - Ramsar Convention J - JAMBA & CAMBA treaties Q - QLD Environmental Protection Act 1994 F - Fisheries Act 1994 N - Nature Conservation Act 1994 The airport is also generally bounded to the east, north and west by ecologically

significant habitats. At least thirteen rare, endangered or vulnerable fauna species

may be associated with the sites tidal or swampy areas as shown in the table

below.

Scientific Name Common Name Conservation Status Siting Commonwealth State

Birds

Esacus neglectus beach thick knee - V P Anas castanea chestnut teal - R O Numenius Madagascariensis eastern curlew - R O Sterna albifrons little tern E V O Ephippiorhynchus asiaticus jabiru - R P Rostratula benghalensis painted snipe - R P Dryolimnas pectoralis Lewin’s rail - R P Insects

Acrodipsas illidgei Illidge’s blue butterfly - E P

Marine Reptiles

Caretta caretta loggerhead turtle E E M Chelonia mydas green turtle V V M

Marine Mammals

Dugong dugon dugong - V M Sousa chinensis ndo-Pacific humpback - R M Dolphin

Terrestrial Mammals

Xeromys myoides alse water rat V R P Commonwealth: Commonwealth Endangered Species Protection Act 1992


State: Queensland Nature Conservation Act 1992 and Nature Conservation (wildlife) Regulations 1994

E Endangered: in danger of extinction, and survival is unlikely if threats continue V Vulnerable: likely to become Endangered in the near future if threats continue R Rare: not considered Endangered or Vulnerable and may be abundant in restricted areas O Observed on site P Possibly on site M Marine animal probably occurring near site

The foreshore, intertidal and freshwater wetlands of the airport site may also

support a number of bird species protected by international treaty. These birds

are generally associated with the Moreton Bay area, which is arguably the most

important feeding ground for migratory waders along the east Australian coast

(Driscoll 1992).

The most significant communities on, and adjacent to the site are all associated

with wetland habitats: both intertidal (mangrove and saltmarsh) and freshwater.

Each of these habitats could be detrimentally affected by a range of activities

including:

Reclamation

Changes to the drainage patterns and hydrology of the site

Alteration to tidal inundation patterns of the site and flushing of the waterways

Increase in sediment loads

Dredging and maintenance of the channels

Discharge of contaminated water or fuel/chemical spillage

Increase in the nutrient levels of the water

Disturbance of acid sulphate soils and the consequent acidification of the water

Feral animals

Control of mosquito and biting midges

Proliferation of exotic weeds

Increases in noise and activity levels

In general, the effects include the following:

Increases in noise and activity could detrimentally affect bird life


Fragmentation and development could lead to the further introduction and

proliferation of exotic weeds and feral animals

Decreases in water quality are likely to affect populations of turtle, dugong and

dolphins in the area

RUST PPK, in their 1996 review also states that air pollution may pose

significant environmental concerns, in the areas immediately surrounding the

airport. This is not addressed in the AES, as the Airports (Environmental

Protection) Regulations do not apply to pollution generated by an aircraft. It is

dealt with under Commonwealth legislation namely the Air Services Act 1995

and Air Navigation (Aircraft Engine Emissions) Regulations. The affects of

aircraft noise on the environment may also be considerable.

The area upon which the Brisbane Airport is situated is claimed as the traditional

country of the Turrbal corporation, which has indicated that there were special

places within the vicinity of the airport site, including an unrecorded bora ring

destroyed during runway construction.

Since European settlement, a range of land-uses and events occurred on the area

now known as Brisbane Airport, most of which have left little trace. The only

physical items listed on the register of the National estate lie outside the airport

boundary, however significant archaeological sites connected with the convict

era and the WWII history of Brisbane, may exist within the site.

211Appendix C MATLAB Code

Appendix C

MMAATTLLAABB CCOODDEE


The following MATLAB code was used to perform a simulation of the ARAISS

algorithm on a three decision-maker, three criteria, five alternative problem. Code for the

main functions only has been included. Full code is available upon request.

Tripleanalyse.M (Main Loop)

% performs a fuzzy analysis and normalisation on the 3

datasets & termset & relevance mtx in the workspace

decode;

aggregateandnormalise;

rankoutputs;

showmatches;

plotoutputs;

clear;

Decode.M

% decodes the 3 coded decision matrices (dm1 2 & 3) using

the suitability & uncertainty termsets in the workspace

saves the resulting 3d fuzzy matrices

% NEED TO ADJUST THE WAY WEIGHTS ARE INPUT AS SOME ARE NOW

>= 1

load termset termset;

load dm1 dm1;

load dm2 dm2;

load dm3 dm3;

load rm rm;

[rows,cols] = size(dm1);

%weightsum is for use in critical weight

dm1weightsum = 0.0;

dm2weightsum = 0.0;


dm3weightsum = 0.0;

for i = 1:cols

dm1weightsum = dm1weightsum + dm1(rows, i);



end

fuzzydm1 = zeros(rows,cols,4);

for i = 1:rows

for j = 1:cols

for k = 1:4

if i < rows

fuzzydm1(i,j,k) = termset(dm1(i,j),k);

elseif dm1(i,j) < 1

fuzzydm1(i,j,k) = dm1(i,j);

else

fuzzydm1(i,j,k) = 2 * cols *

(dm1weightsum) + .1; %critical

dm1critscore = fuzzydm1(i,j,k)

end

end

end

end


for i = 1:rows

for j = 1:cols


for k = 1:4

if i < rows


elseif dm2(i,j) < 1


else




end

end

end

end


for i = 1:rows

for j = 1:cols

for k = 1:4

if i < rows


elseif dm3(i,j) < 1


else




end


end

end

end

[rows,cols] = size(rm);

fuzzyrm = zeros(rows,cols,4);

%normalise the crit rm values

rmcrit = zeros(1,cols);

rmsum = zeros(1,cols);

for i = 1:rows

for j = 1:cols

if rm(i,j) == 1

rmcrit(j) = 1

rmsum(j) = rm(1,j) + rm(2,j) + rm(3,j);

end

end

end

for j = 1:cols

if rmcrit(j) == 1

for i = 1:rows

if rm(i,j) < 1

rm(i,j) = rm(i,j) / (2 * rows * (rmsum(j))

+ .1)

end

end

end

end


for i = 1:rows

for j = 1:cols

for k = 1:4

fuzzyrm(i,j,k) = rm(i,j);

end

end

end

% now scale the fuzzy dm matrices for uncertainty

load dm1uncert dm1uncert;



for i = 1:rows-1 % no need to scale weights

for j = 1:cols

fuzzydm1(i,j,:) =

uncertscale(fuzzydm1(i,j,:),dm1uncert(i,j));

fuzzydm2(i,j,:) =


fuzzydm3(i,j,:) =


end

end

% Normalise weights with critical component - uncertainty

not a factor in weight

dm1weightsum = 0.0;

dm2weightsum = 0.0;


dm3weightsum = 0.0;

[rows,cols] = size(dm1);

dm1crit = 0.0;

dm2crit = 0.0;

dm3crit = 0.0;

for i = 1:cols

if dm1(rows, i) == 1

dm1crit = 1

end

end

for i = 1:cols


dm2crit = 1

end

end

for i = 1:cols


dm3crit = 1

dm3(rows, i)

end

end

if dm1crit == 1

for i = 1:cols


for j = 1:4

fuzzydm1(rows, i, j) = fuzzydm1(rows, i, j)/

dm1critscore;

end

end

end

if dm2crit == 1

for i = 1:cols

for j = 1:4


dm2critscore;

end

end

end

if dm3crit == 1

for i = 1:cols

for j = 1:4


dm3critscore;

end

end

end

save fuzzyrm fuzzyrm;

save fuzzydm1 fuzzydm1;




Aggregateandnormalise.M

function f = aggregateandnormalise()

% aggregates & normalises the 3 decisionmatrices stored in

the workspace to obtain alternative ratings (last row of

dm1

% 2 & 3 is weights)

getmax;

load maxscore maxscore;

markerpoint = maxscore(2);

load fuzzydm1 fuzzydm1;



load fuzzyrm fuzzyrm;

[rows,cols,dims] = size(fuzzydm1);

aggdm1 = zeros(rows-1,dims);

for i = 1:rows-1

rating = zeros(1,dims);

for j = 1:cols

for k = 1:dims


outcome(k) = fuzzydm1(i,j,k);

weight(k) = fuzzydm1(rows,j,k);

relevance(k) = fuzzyrm(1,j,k);

end

rating =

trapadd(rating,trapmult(outcome,trapmult(weight,relevance))

);

end

aggdm1(i,:) = rating;

end


for i = 1:rows-1


for j = 1:cols

for k = 1:dims




end

rating =


);

end



end


for i = 1:rows-1


for j = 1:cols

for k = 1:dims




end

rating =


);

end


end

for i = 1:rows-1

finaloutcome =

trapadd(aggdm1(i,:),trapadd(aggdm2(i,:),aggdm3(i,:)));


outputmatrix(i,:) =

trapnormalise(finaloutcome,markerpoint);

end

save outputmatrix outputmatrix;

Rankoutputs.M

function f = rankoutputs()

% ranks the outputmatrix from best to worst result using a

weighted addition of breakpoints

load outputmatrix outputmatrix;

[rows,cols] = size(outputmatrix);

x = zeros(1,rows);

for i = 1:rows

%TFN = outputmatrix(i,:);

x(i) = 0.1 * outputmatrix(i,1) + .4 *

outputmatrix(i,2) + .4 * outputmatrix(i,3) + 0.1 *

outputmatrix(i,4);

end

f = zeros(1,rows);

for i = 1:rows

best = 0;

for j = 1:rows


if x(j) > best

best = x(j);

f(i) = j;

y = j;

end

end

x(y) = 0;

end

rankorder = f

Showmatches.M

% matches the TFN's in the outputmatrix to the closest

terms in the suitability, uncertainty, risk & conflict

termsets



matches = zeros(rows,6);

for i = 1:rows

matches(i,:) = lingapprox(i);

end

MatchOrder = ['Suitability ', 'Uncertainty ', 'Risk ',

'Conflict ', 'Overall']

Matches


Plotoutputs.M

% plot the alternative aggregated outcomes



y = [0 1 1 0];

scalevector = [0,1,0,1];

for i = 1:rows

switch i

case{1}, linecolor = 'r';

case{2}, linecolor = 'g';

case{3}, linecolor = 'b';

case{4}, linecolor = 'c';

case{5}, linecolor = 'm';

case{6}, linecolor = 'k';

otherwise, linecolor = 'y';

end

set(gca,'NextPlot','add');

plot(outputmatrix(i,:),y,linecolor);

labels(i,1) = int2str(i);

end

axis(scalevector);

legend(labels);

% getmaxscore;

% getminscore;

% load maxscore maxscore;


% load minscore minscore;

% plot(trapnormalise(maxscore,maxscore(2)),y,linecolor);

% plot(trapnormalise(minscore,maxscore(2)),y,linecolor);

Lingapprox.M

function f = lingapprox(a)

% chooses a term from the term set that is the closest to

the TFN by comparing center of gravity score

% returns the index of the term and the index of

uncertainty measure

% suitability

load outputmatrix outputmatrix

tfn = outputmatrix(a,:);


[rows,cols] = size(termset);

displacementset = zeros(1,rows);

for i = 1:rows

displacementset(i) = abs(score(termset(i,:)) - score(tfn));

end

match = 1;


min = displacementset(1);

for i = 1:rows

if displacementset(i) < min

min = displacementset(i);

match = i;

end

end

suitabilityterm = match;

suitabilityvalue = score(tfn);

%uncertainty

load uncertaintyset uncertaintyset

[rows,cols] = size(uncertaintyset);


for i = 1:rows

uncertterm =

uncertscale(termset(suitabilityterm,:),uncertaintyset(i));

uncerttermsupport = uncertterm(4) - uncertterm(1);

tfnsupport = tfn(4) - tfn(1);

displacementset(i) = abs(uncerttermsupport - tfnsupport);

end

match = 1;


for i = 1:rows




match = i;

end

end

uncertaintyterm = match ;

uncertaintyvalue = uncertaintyset(match);

% Risk

load termgenset termgenset;

[rows,cols] = size(termgenset);


for i = 1:rows

displacementset(i) = abs(termgenset(i) - riskscore(a));

end

match = 1;


for i = 1:rows



match = i;

end

end

riskterm = match;


riskvalue = riskscore(a);

% conflict

load termgenset termgenset;

[rows,cols] = size(termgenset);


for i = 1:rows

displacementset(i) = abs(termgenset(i) - conflictscore(a));

end

match = 1;


for i = 1:rows



match = i;

end

end

conflictterm = match;

conflictvalue = conflictscore(a);

load parameterweights parameterweights;

for i = 1:4

if parameterweights(i) == 1 % critical


parameterweights(i) = 8 * (parameterweights(1) +

parameterweights(2) + parameterweights(3) +

parameterweights(4)) + .1;

end

end


[rows,cols] = size(termset);

overallscore = parameterweights(1)* suitabilityvalue +

parameterweights(2) * (1-uncertaintyvalue) +

parameterweights(3) * (1-riskvalue) + parameterweights(4) *

(1-conflictvalue);

normaloverallscore = overallscore/(parameterweights(1) +

parameterweights(2) + parameterweights(3) +

parameterweights(4));

for i = 1:rows

displacementset(i) = abs(score(termset(i,:)) -

normaloverallscore);

end

match = 1;


for i = 1:rows



match = i;


end

end

overallsuitabilityterm = match;

f = [suitabilityterm uncertaintyterm riskterm conflictterm

overallsuitabilityterm normaloverallscore];

233Appendix D ArcObjects VBA Code

Appendix D

AARRCCOOBBJJEECCTTSS VVBBAA CCOODDEE


The following code controlled the functionality of the InfraPlanner interfaces

described in Chapter 7. Full code is available upon request.

CREATE DISCRETE CRITERION MAP

Option Explicit

Private m_pMxDoc As IMxDocument

Private m_pMaps As IMaps

Private m_pMap As IMap

Private m_pLayer As ILayer

Private m_pEnumLayers As IEnumLayer

Private strCriteriaName As String

Private strCriteriaDesc As String

Private m_pDecisionMap As IMap

Private m_pMapFrame As IMapFrame

Private m_pPageLayout As IPageLayout

Private m_pActiveView As IActiveView

Private m_bClearFlag As Boolean

Private m_iNumCategories As Integer

Private m_iCategoryNumber As Integer

Private m_pTable As ITable

Private m_lRatingArray() As Long 'The new rating

Private m_dValueArray() As Double 'Raw value of the

category

Private Sub btnAddRating_Click()

If cboRating.ListIndex < 0 Then


MsgBox "Select a rating"

Exit Sub

End If

lboCategories.Selected(m_iCategoryNumber) = False

Dim pFields As IFields

Set pFields = m_pTable.Fields

Dim pRow As IRow

If m_iCategoryNumber <= m_iNumCategories - 1 Then

lboRating.AddItem cboRating.Text

Set pRow = m_pTable.GetRow(m_iCategoryNumber)

m_lRatingArray(m_iCategoryNumber) =

DecodeRating(cboRating.ListIndex + 1)

m_dValueArray(m_iCategoryNumber) =

pRow.Value(pFields.FindField("Value"))

'uncomment these lines to check the values going

in

'MsgBox "Rating: " & m_iCategoryNumber & ", " &

m_lRatingArray(m_iCategoryNumber) & vbLf + _

'"Value: " & m_dValueArray(m_iCategoryNumber)

m_iCategoryNumber = m_iCategoryNumber + 1

Else: MsgBox "Box Full"

End If


If m_iCategoryNumber <= m_iNumCategories - 1 Then

lboCategories.Selected(m_iCategoryNumber) =

True

lblCategory.Caption = lboCategories.Value

End If

End Sub

Private Sub btnCreate_Click()

If Not m_iCategoryNumber = m_iNumCategories Then

MsgBox "Need to rate all categories"

Exit Sub

End If

Create

Unload Me

End Sub

Private Sub btnExit_Click()

Unload Me

End Sub

Private Sub cboField_Change()

lboCategories.Clear


lboRating.Clear

If Not m_bClearFlag Then

Dim pRLayer As IRasterLayer

Set pRLayer = m_pLayer

Dim pRasterBandCollection As IRasterBandCollection

Set pRasterBandCollection = pRLayer.Raster

Dim pRasterBand As IRasterBand

Set pRasterBand = pRasterBandCollection.Item(0)

Set m_pTable = pRasterBand

Dim pQueryFilter As IQueryFilter

Set pQueryFilter = New QueryFilter

pQueryFilter.SubFields = cboField.Text

'pQueryFilter.WhereClause = "STATE_NAME =

'California'"

m_iNumCategories = m_pTable.RowCount(pQueryFilter)

If m_iNumCategories <= 20 Then


Set pFields = m_pTable.Fields

Dim pRow As IRow

Dim i As Integer

For i = 0 To m_pTable.RowCount(pQueryFilter) -

1


Set pRow = m_pTable.GetRow(i)

lboCategories.AddItem

pRow.Value(pFields.FindField(cboField.Text))

Next i

ReDim m_lRatingArray(m_iNumCategories) As Long

ReDim m_dValueArray(m_iNumCategories) As

Double

Else: MsgBox "More than 20 categories, this layer

may be unsuitable"

End If

m_iCategoryNumber = 0

lboCategories.Selected(m_iCategoryNumber) = True

lblCategory.ZOrder (0)



End If

End Sub

Private Sub cboRating_Change()


End Sub

Private Sub cboSourceTheme_Change()

MapControl1.ClearLayers

m_bClearFlag = True

cboField.Clear

m_bClearFlag = False

lboCategories.Clear

'need to clear the chart too

Dim i As Integer

For i = 0 To m_pMap.LayerCount - 1

Set m_pLayer = m_pMap.Layer(i)

If m_pLayer.Name = cboSourceTheme.Text Then

Exit For

End If

Next i

MapControl1.AddLayer m_pLayer

' Rotate the map to -62.9 deg

Dim pRotateOperation As IRotateOperation


Set pRotateOperation = New RotateOperation

pRotateOperation.ActiveView = MapControl1.ActiveView

pRotateOperation.Rotation = -62.9

pRotateOperation.Do

'get the field names







Dim pTable As ITable

Set pTable = pRasterBand


Set pFields = pTable.Fields

For i = 0 To pFields.FieldCount - 1

cboField.AddItem pFields.Field(i).Name

Next i

End Sub

Private Sub Frame1_Click()


End Sub

Private Sub Label10_Click()

End Sub

Private Sub lblCategory_Click()

End Sub

Private Sub UserForm_Click()

End Sub

Private Function DecodeRating(ByVal Rating) As Long

If (Not ((Rating = 1) Or (Rating = 2) Or (Rating = 3)

Or (Rating = 4) Or (Rating = 5) Or (Rating = 6) Or

(Rating = 7))) Then

MsgBox "Invalid rating: " & Rating

Exit Function

End If

Select Case Rating

Case 1

DecodeRating = 0

Case 2

DecodeRating = 1

Case 3

DecodeRating = 3


Case 4

DecodeRating = 5

Case 5

DecodeRating = 7

Case 6

DecodeRating = 9

Case 7

DecodeRating = 10

End Select

End Function

Private Sub UserForm_Initialize()

Set m_pMxDoc = ThisDocument

Set m_pMaps = m_pMxDoc.Maps

Set m_pMap = m_pMaps.Item(0)

lblCategory.Visible = True

'Get the rasterlayers

Dim i As Integer

i = 0

Do Until m_pMap.Name = "Source Rasters"


Set m_pMap = m_pMaps.Item(i)

i = i + 1

Loop

'set up the source theme combo box

Dim j As Integer

For j = 0 To m_pMap.LayerCount - 1

Set m_pLayer = m_pMap.Layer(j)

If TypeOf m_pLayer Is IRasterLayer Then







Dim pRasterProps As IRasterProps

Set pRasterProps = pRasterBand

If pRasterProps.IsInteger = True Then

cboSourceTheme.AddItem m_pMap.Layer(j).Name

m_pMap.Layer(j).Visible = True

End If

End If


Next j

AddTerms

End Sub

Private Sub AddTerms()

cboRating.AddItem "Totally Unsuitable"

cboRating.AddItem "Very Bad"

cboRating.AddItem "Bad"

cboRating.AddItem "Indifferent"

cboRating.AddItem "Good"

cboRating.AddItem "Very Good"

cboRating.AddItem "Perfect"

End Sub

Private Sub CreateDiscreteCriteria()

'This sub uses a raster model

' can use AlgbOp for single operation

' Get raster from layer



Dim pInRaster As IRaster

Set pInRaster = pRLayer.Raster


' Create a RasterModel object

Dim pRModel As IRasterModel

Set pRModel = New RasterModel

' Create spatial analysis environment

Dim pEnv As IRasterAnalysisEnvironment

Set pEnv = pRModel

' Set output workspace

Dim pWS As IWorkspace

Dim pWSF As IWorkspaceFactory

Set pWSF = New RasterWorkspaceFactory

Set pWS = pWSF.OpenFromFile("c:\temp", 0)

Set pEnv.OutWorkspace = pWS

' & vbLf + _ is used to seperate equations

' N.B. get an error if use + vbLf + _ as shown in

samples

' N.B. can't reuse created rasters, once created

they are set & trying to

' change them creates an error

' N.B leave a space before closing quotes

'just copy the raster & then edit it's table to

re-classify

pRModel.Script = "[CriteriaMap] = [input1] / 10 "


' "[seg2] = ([input2] = " &

m_dValueArray(2) & ") * " & m_lRatingArray(2) & vbLf +

_

' "[seg3] = ([input3] = " &


_

' "[seg4] = ([input4] = " &


_

' "[seg5] = ([input5] = " &


_

' "[seg6] = ([input6] = " &


_

' "[seg7] = ([input7] = " &


_

' "[CriteriaMap] = [Seg1] +

[Seg2] + [seg3] + [seg4] + [seg5] + [seg6] + [seg7]"

'

' Bind to raster

pRModel.BindRaster pInRaster, "input1"

' Run the model

pRModel.Execute

' Get outputs

Dim pRaster1 As IRaster

Set pRaster1 = pRModel.BoundRaster("CriteriaMap")


' Unbind raster

pRModel.UnbindSymbol "input1"

' attempt to do the reclass via the table

' Dim pRasterBandCollection As

IRasterBandCollection

' Set pRasterBandCollection = pRaster1

'

' Dim pRasterBand As IRasterBand

' Set pRasterBand = pRasterBandCollection.Item(0)

'

' Set m_pTable = pRasterBand.AttributeTable

'

'

' ' Get field index

' Dim FieldIndex As Integer

' FieldIndex = m_pTable.FindField("Value")

'

' Dim pQueryFilter As IQueryFilter

' Set pQueryFilter = New QueryFilter

' pQueryFilter.SubFields = "Value"

'

' Dim pCursor As IRasterCursor

' Set pCursor = m_pTable.Update(pQueryFilter, True)

'

' Dim pRow As IRow

' Dim i As Integer

'

' For i = 0 To m_pTable.RowCount(pQueryFilter) - 1

'

' Set pRow = pCursor.Next

'


'

' 'Set pRow = m_pTable.GetRow(i)

'

' MsgBox "Old value " & i & " = " &

pRow.Value(FieldIndex)

'

' pRow.Value(FieldIndex) = m_lRatingArray(i)

'

' MsgBox "New value " & i & " = " &

pRow.Value(FieldIndex) & vbLf + _

' "Array value: " & m_lRatingArray(i)

'

'

'

'

'

' Next i

'

'

' Add the results into Map

i = 0

Do Until m_pMap.Name = "Criteria Rasters"


i = i + 1

Loop

Dim pRLayer1 As IRasterLayer

Set pRLayer1 = New RasterLayer


pRLayer1.CreateFromRaster pRaster1

Set m_pLayer = pRLayer

m_pMap.AddLayer pRLayer1

pRLayer1.Name = TxtName.Text

m_pLayer.Name = pRLayer.Name

Set m_pMxDoc.ActiveView = m_pMap

m_pMxDoc.ActiveView.Refresh

m_pMxDoc.UpdateContents

End Sub

Private Sub Create()

' Get the input raster from the first layer in ArcMap

Dim pRasterLy As IRasterLayer

Dim pLy As ILayer

Set pLy = m_pLayer

If Not TypeOf pLy Is IRasterLayer Then Exit Sub

Set pRasterLy = pLy

Dim pGeoDs As IGeoDataset

Set pGeoDs = pRasterLy.Raster

' Create a raster descriptor and select the field

used for reclassify

Dim pRD As IRasterDescriptor

Set pRD = New RasterDescriptor

pRD.Create pGeoDs, New QueryFilter, "Value"

Set pGeoDs = pRD

' Create a Spatial operator

Dim pReclassOp As IReclassOp


Set pReclassOp = New RasterReclassOp



Set pEnv = pReclassOp




Set pWS =

pWSF.OpenFromFile("c:\temp\temporaryrasters", 0)


' Set the Remap

Dim pRemap As IRemap

Dim pNRemap As INumberRemap

Set pNRemap = New NumberRemap

Dim i As Integer

For i = 0 To m_iNumCategories - 1

pNRemap.MapValue m_dValueArray(i),

m_lRatingArray(i)

Next i

Set pRemap = pNRemap

' Perform Spatial operation

Dim pOutRaster As IRaster

Set pOutRaster = pReclassOp.ReclassByRemap(pGeoDs,

pRemap, False)







Set pEnv = pRModel



samples





'remap limited to integers & this creates a

double value

pRModel.Script = "[CriteriaMap] = [input1] * 0.1

"

' Bind to raster

pRModel.BindRaster pOutRaster, "input1"

' Run the model


pRModel.Execute

' Get outputs



' Unbind raster


' Add it into ArcMap

Set pRasterLy = New RasterLayer

pRasterLy.CreateFromRaster pRaster1

i = 0



i = i + 1

Loop

Set m_pLayer = pRasterLy

m_pMap.AddLayer pRasterLy

pRasterLy.Name = TxtName.Text

m_pLayer.Name = pRasterLy.Name




SetUpLayer


End Sub

Sub SetUpLayer()

'Classifies the layer in m_pLayer linguistically

' Get raster input from layer



Dim pRaster As IRaster

Set pRaster = pRLayer.Raster

' Create classfy renderer and QI RasterRenderer

interface

Dim pClassRen As IRasterClassifyColorRampRenderer

Set pClassRen = New

RasterClassifyColorRampRenderer

Dim pRasRen As IRasterRenderer

Set pRasRen = pClassRen

Dim pProps As IRasterClassifyUIProperties

Set pProps = pClassRen

'pProps.ShowClassGaps = True

' Set raster for the render and update

Set pRasRen.Raster = pRaster

pClassRen.ClassCount = 5


pRasRen.Update

'Make the start & end colors

Dim StartColor As IColor

Set StartColor = New RgbColor

StartColor.RGB = RGB(255, 0, 0)

Dim EndColor As IColor

Set EndColor = New RgbColor

EndColor.RGB = RGB(0, 255, 0)

' Create a color ramp to use

Dim pRamp As IAlgorithmicColorRamp

Set pRamp = New AlgorithmicColorRamp

pRamp.Size = 5

pRamp.FromColor = StartColor

pRamp.ToColor = EndColor

pRamp.CreateRamp True

' Create symbol for the classes

Dim pFSymbol As IFillSymbol

Set pFSymbol = New SimpleFillSymbol

' loop through the classes and apply the color and

label


Dim i As Integer

For i = 0 To pClassRen.ClassCount - 1

pFSymbol.Color = pRamp.Color(i)

pClassRen.Symbol(i) = pFSymbol


Select Case i

Case 0

pClassRen.Label(i) = "Totally Unsuitable"

Case 1

pClassRen.Label(i) = "Bad"

Case 2

pClassRen.Label(i) = "Indifferent"

Case 3

pClassRen.Label(i) = "Good"

Case 4

pClassRen.Label(i) = "Perfect"

End Select

Next i

' attempt at setting breaks N.B. there may be a

problem

' if the raster does not contain values that

extend to these breaks!

pClassRen.Break(0) = 0

pClassRen.Break(1) = 0.09






' Update the renderer and plug into layer

pRasRen.Update

Set pRLayer.Renderer = pClassRen



End Sub

CONTINUOUS CRITERION

Option Explicit












Private m_dMaxDistance As Double

Private m_dMinDistance As Double

' Rated points

Private m_dDistance1 As Double






Private m_iRating1 As Integer





Private m_iPointCount As Integer

Private Function DecodeRating(ByVal Rating) As Double



(Rating = 7))) Then

MsgBox "Invalid rating: " & Rating

Exit Function

End If

Select Case Rating

Case 1

DecodeRating = 0

Case 2

DecodeRating = 0.1

Case 3

DecodeRating = 0.3

Case 4

DecodeRating = 0.5


Case 5

DecodeRating = 0.7

Case 6

DecodeRating = 0.9

Case 7

DecodeRating = 1

End Select

End Function

Private Sub CreateProximityCriteria()

'Get the number of line segments & gradients & y-

intercepts

' only works for 3 points as a test

If Not (m_iPointCount = 3) Then

MsgBox "Must be 3 points"

Exit Sub

End If

Dim m1 As Double

Dim m2 As Double

Dim yint1 As Double

Dim yint2 As Double

Dim y1 As Double


Dim y2 As Double

Dim x1 As Double

Dim x2 As Double

' segment 1

y1 = DecodeRating(m_iRating1)


x1 = m_dDistance1

x2 = m_dDistance2

If Not ((x2 - x1) = 0) Then

m1 = (y2 - y1) / (x2 - x1)

yint1 = y1 - ((y2 - y1) / (x2 - x1)) * x1

Else: MsgBox "Divide by zero error"

Exit Sub

End If

' segment 2



x1 = m_dDistance2

x2 = m_dDistance3

If Not ((x2 - x1) = 0) Then

m2 = (y2 - y1) / (x2 - x1)

yint2 = y1 - ((y2 - y1) / (x2 - x1)) * x1

Else: MsgBox "Divide by zero error: Operation

terminated"

Exit Sub

End If




' Get raster from layer



Dim pInRaster As IRaster

Set pInRaster = pRLayer.Raster






Set pEnv = pRModel





Set pWS =



' Set model, & vbLf + _ is used to seperate

equations


samples






pRModel.Script = "[Bool1] = [input1] <= " &

m_dDistance2 & vbLf + _

"[Bool2] = [input1] > " &

m_dDistance2 & vbLf + _

"[Seg1] = (([input1] * " & m1 &

") + " & yint1 & ") * [Bool1]" & vbLf + _

"[Seg2] = (([input1] * " & m2 &

") + " & yint2 & ") * [Bool2]" & vbLf + _

"[Complete] = [Seg1] + [Seg2] "

& vbLf + _

"[Under1] = [Complete] <= 1 " &

vbLf + _

"[Over0] = [Complete] > 0 " &

vbLf + _

"[Over1] = [Complete] > 1 " &

vbLf + _

"[Over0Under1] = [Under1] *

[Over0] " & vbLf + _

"[CriteriaMap] = ([Complete] *

[Over0Under1]) + [Over1] "

' to avoid the crap below try sizing the segments

using x intercepts

' then let the end segments = 0 or 1 if necessary


' & vbLf + _

' "[CriteriaMap] = [Complete]*

([Complete] > 0.0) "

' & vbLf =

' "[Under1] = [Complete1] <= 1 "

& vbLf + _

' "[Over1] = [Complete1] > 1 " &

vbLf + _

' "[CriteriaMap] = [Complete1] *

[Under1] "

' & vbLf + _

' "[CriteriaMap] = [Over1] * 1.0

"

' & vbLf + _

' "[CriteriaMap] = [Complete2] +

[1Map] "

'doesn't work yet

' Bind to raster

pRModel.BindRaster pInRaster, "input1"

' Run the model

pRModel.Execute

' Get outputs



' Unbind raster




Dim i As Integer

i = 0



i = i + 1

Loop





pRLayer1.Name = TxtName.Text




Set m_pLayer = pRLayer1

SetUpLayer

End Sub


Unload Me

End Sub

Private Sub cboRating_Change()


End Sub

Private Sub cboSourceTheme_Change()


'need to clear the chart too

Dim i As Integer



If m_pLayer.Name = cboSourceTheme.Text Then

Exit For

End If

Next i







pRotateOperation.Do

'get the max & min values for the label








Dim pRasterStats As IRasterStatistics

Set pRasterStats = pRasterBand.Statistics

m_dMaxDistance = pRasterStats.Maximum

m_dMinDistance = pRasterStats.Minimum

lblMaxDistance.Caption = m_dMaxDistance

lblMinDistance.Caption = m_dMinDistance

' Set the distance cbo box up

cboDistance.Clear

Dim iBoxNumber As Integer

iBoxNumber = 0

cboDistance.AddItem iBoxNumber

For i = 1 To 20

iBoxNumber = iBoxNumber + m_dMaxDistance / 20


Next i

iBoxNumber = m_dMaxDistance + 1


lboRatedPoints.Clear

m_iPointCount = 0


End Sub

Private Sub ComboBox2_Change()

End Sub


CreateProximityCriteria

Unload Me

End Sub

Private Sub btnAddPoint_Click()

'Add a rated point to the list

'check if there is a distance & rating

If ((Not IsNumeric(cboDistance.Text)) Or

(cboRating.ListIndex = -1)) Then

MsgBox "Must have a valid distance and rating

selected"

Exit Sub

End If

'Get the point number & update variables

'Need to change this to an array


Select Case m_iPointCount

Case 0

m_dDistance1 = cboDistance.Text

m_iRating1 = cboRating.ListIndex + 1

Case 1

If (cboDistance.Text <= m_dDistance1) Then

MsgBox "Enter points from lowest distance to highest"

Exit Sub

End If



Case 2



Exit Sub

End If



Case 3



Exit Sub

End If




Case 4



Exit Sub

End If



Case Else

MsgBox "Can't add any more points"

Exit Sub

End Select

lboRatedPoints.AddItem m_iPointCount + 1 & ": " &

cboDistance.Text & " " & cboRating.Text

' Update the chart

m_iPointCount = m_iPointCount + 1

UpdateChart

End Sub

Private Sub UpdateChart()

Dim i As Integer

Dim XValue, YValue As Double


With chtUtility

.chartType = VtChChartType2dXY

.ShowLegend = False

With .Plot.Axis(VtChAxisIdY).AxisTitle

.VtFont.Size = 12

.Visible = True

.Text = "Utility"

End With

With .Plot.Axis(VtChAxisIdX).AxisTitle

.VtFont.Size = 12

.Visible = True

.Text = "Raw Value"

End With

.Title.VtFont.Size = 12

.Title = "Rated Points"

.Plot.Axis(VtChAxisIdY).AxisScale.Type =

VtChScaleTypeLinear

.Plot.Axis(VtChAxisIdX).AxisScale.Type =

VtChScaleTypeLinear

'Tip from KB article Q194221:

.Plot.UniformAxis = False

'.Footnote.Text = ""

End With

'PenColor = True 'Draw in color

'ShowMarker = True 'Show plot points

'PenColor = False 'Black and White


'ShowMarker = False 'Don't show plot points

'Create a new array of plot points for this Series

'We will redim the first subscript differently, to

show that each

'series can have a different # of plot points:

ReDim ChartPoints(1 To m_iPointCount, 1 To 2)

'Update the array data:

' doesn't work - deletes previous points

For i = 1 To m_iPointCount

' this would work better using an array!

Select Case i

Case 1

XValue = m_dDistance1

YValue = DecodeRating(m_iRating1)

Case 2



Case 3



Case 4




Case 5



End Select

ChartPoints(i, 1) = XValue

ChartPoints(i, 2) = YValue

Next i

'We need to increase the ColumnCount. For X-Y Scatter

graphs, we

'need 2 columns for each series.

chtUtility.ColumnCount = 2

With chtUtility

With .Plot

.Wall.Brush.Style = VtBrushStyleSolid

'Normally, you might want the Wall background

of the Chart

'to be in color, if you're using Color pens,

and to be white

'if using B&W pens, but, since we're drawing

both a color


'series *and* a B&W series on *one* chart,

we'll just make

'the wall color, for now. If you want White,

uncomment the

'line found about 10 lines down:

'.Wall.Brush.FillColor.Set 255, 255, 225

'You can set the individual Pen colors

here, or just use

'the defaults.

'Else 'Based on an article in the VB KB:

'Uncomment the next line if you want the

wall color to

'be white:

'.Wall.Brush.FillColor.Set 255, 255, 255

'Set the different patterns for Black and

White plotting.

'You need to set the Pen for only the 'X'

column:

End With

.columnLabelCount = 2

'If the current series has more plot points that

the previous

'one, we need to change .RowCount accordingly:

.RowCount = UBound(ChartPoints, 1)

'Both of the next 2 lines seem to do the same

thing:

.Plot.SeriesCollection(1).SeriesMarker.Show = True


.Plot.SeriesCollection.Item(1).SeriesMarker.Show =

False

'Create the plot points for this series from the

ChartPoints array:

For i = 1 To UBound(ChartPoints, 1)

.DataGrid.SetData i, 1, ChartPoints(i, 1),

False

.DataGrid.SetData i, 2, ChartPoints(i, 2),

False

Next i

' 'Remove null points from *this* series, if it has

*fewer*

' 'points than the prior ones. If you don't remove

null points,

' 'then the graph will add 0,0 points, erroneously.

See MS

' 'Knowledge Base article Q177685 for more info:

' For lRow2 = lRow To OldRowCount&

' .DataGrid.SetData lRow2, CurSeries * 2 - 1,

0, True

' .DataGrid.SetData lRow2, CurSeries * 2, 0,

True

' Next

'

' 'Remove null points from *prior* series, if this

series

' 'has *more* points than the prior ones:

' If CurSeries > 1 Then

' For lRow = OldRowCount& + 1 To .rowCount

' For lRow2 = 1 To CurSeries - 1

' .DataGrid.SetData lRow, lRow2 * 2 -

1, 0, True

' .DataGrid.SetData lRow, lRow2 * 2, 0,

True


' Next

' Next

' End If

.Column = 1

.ColumnLabel = "Series " & Str(1)

.Refresh

End With

End Sub

Private Sub CommandButton2_Click()

End Sub

Private Sub Graph1_HotHit(HitSet As Integer, HitPoint

As Integer)

End Sub


UpdateChart

End Sub

Private Sub chtUtility_OLEStartDrag(Data As

MSChart20Lib.DataObject, AllowedEffects As Long)


End Sub


End Sub

Private Sub MapControl1_OnMouseDown(ByVal button As

Long, ByVal shift As Long, ByVal x As Long, ByVal y As

Long, ByVal mapX As Double, ByVal mapY As Double)

End Sub

Private Sub TextBox1_Change()

End Sub


End Sub





m_iPointCount = 0

'Get the rasterlayers

Dim i As Integer

i = 0


Do Until m_pMap.Name = "Source Rasters"


i = i + 1

Loop

Dim j As Integer




cboSourceTheme.AddItem m_pMap.Layer(j).Name


End If

Next j

AddTerms

End Sub

Private Sub UserForm_Terminate()

End Sub

Private Sub UserForm_Zoom(Percent As Integer)

End Sub

Private Sub AddTerms()


cboRating.AddItem "Totally Unsuitable"

cboRating.AddItem "Very Bad"

cboRating.AddItem "Bad"

cboRating.AddItem "Indifferent"

cboRating.AddItem "Good"

cboRating.AddItem "Very Good"

cboRating.AddItem "Perfect"

cboUncertainty.AddItem "Totally Uncertain"

cboUncertainty.AddItem "Uncertain"

cboUncertainty.AddItem "Moderately Certain"

cboUncertainty.AddItem "Certain"

cboUncertainty.AddItem "Totally Certain"

End Sub

Sub SetUpLayer()








interface


Set pClassRen = New











pRasRen.Update











pRamp.Size = 5









label


Dim i As Integer




Select Case i

Case 0


Case 1


Case 2


Case 3


Case 4


End Select

Next i


problem











pRasRen.Update




End Sub

AGGREGATION

Option Explicit













Private m_iDMNumber As Integer ' the current DM

Private m_iCriterionNumber As Integer ' the current

criterion

Private m_iNumCriteria As Integer

Private m_dWeightArray() As Double

Private m_sCriteriaArray() As String

Private m_bInputCompleteFlag As Boolean

Private Sub btnAddCriteria_Click()

If (cboWeight.ListIndex < 0) Or (cboMap.ListIndex < 0)

Or (cboRelevance.ListIndex < 0) Then

MsgBox "Select relevance, weight and map"

Exit Sub

End If

If m_bInputCompleteFlag Then

MsgBox "Data input complete. Click create to make

a decision map"

Exit Sub


End If

If (m_iCriterionNumber =

ProjectOptions.g_iNumCriteria) And (m_iDMNumber =

ProjectOptions.g_iNumDMs) Then

m_bInputCompleteFlag = True

End If

lboCriteria.AddItem "It is " & cboRelevance.Text & "

that " & lblDM.Caption & " rates " &

lblCriterion.Caption & " as " & cboWeight.Text & "

Using Map: " & cboMap.Text

' now put variables in arrays for maps & weights etc

ProjectOptions.g_dWeightArray(m_iDMNumber,

m_iCriterionNumber) = DecodeRating(cboWeight.ListIndex

+ 1)

ProjectOptions.g_sMapArray(m_iDMNumber,

m_iCriterionNumber) = cboMap.Text

ProjectOptions.g_dRelevanceArray(m_iDMNumber,

m_iCriterionNumber) =

DecodeRating(cboRelevance.ListIndex + 1)

'if there is no map the weight must be zero

If ProjectOptions.g_sMapArray(m_iDMNumber,

m_iCriterionNumber) = "NONE" Then

ProjectOptions.g_dWeightArray(m_iDMNumber,

m_iCriterionNumber) = 0

End If

'now update criterion & dm number


If Not m_bInputCompleteFlag Then

If m_iCriterionNumber <

ProjectOptions.g_iNumCriteria Then

m_iCriterionNumber = m_iCriterionNumber + 1

lblCriterion.Caption =

ProjectOptions.g_strCriteriaArray(m_iCriterionNumber)

Exit Sub

End If

m_iCriterionNumber = 1



m_iDMNumber = m_iDMNumber + 1

lblDM.Caption =

ProjectOptions.g_strDMArray(m_iDMNumber)

End If

End Sub


If Not m_bInputCompleteFlag Then

MsgBox "Input not complete"

Exit Sub

End If

If cboConstraint.ListIndex < 0 Then

ProjectOptions.g_sConstraintMap = "NONE"


Else

ProjectOptions.g_sConstraintMap =

cboConstraint.Text

End If

SaveProjectData

CreateDecisionMap

CreateRiskMap

CreateConflictMap

'Dim Response

'

'Response = MsgBox("Do you want a risk map?", vbYesNo)

'

'If Response = vbYes Then

' CreateRiskMap

'End If

'

Unload Me

End Sub


Unload Me

End Sub

Private Sub cboCriteria_Change()

End Sub


Private Sub btnTest_Click()

'SaveProjectData

End Sub

Private Sub cboConstraint_Change()

End Sub

Private Sub cboMap_Change()


If GetLayer(cboMap.Text) = True Then


Else: MapControl1.ClearLayers

MapControl1.Refresh

Exit Sub

End If






pRotateOperation.Do

End Sub



MsgBox ProjectOptions.g_strSelectedProject & vbLf &

ProjectOptions.g_strProjectName & vbLf &

ProjectOptions.g_strDMArray(2)

End Sub


End Sub


End Sub


End Sub


End Sub

Private Sub ListBox1_Click()

End Sub

Private Sub MapControl1_OnMouseDown(ByVal button As

Long, ByVal shift As Long, ByVal x As Long, ByVal y As

Long, ByVal mapX As Double, ByVal mapY As Double)

End Sub

Private Sub SrceThm_Click()


End Sub


End Sub


lblProjectName.Caption =

ProjectOptions.g_strProjectName

TxtMapName.Text = ProjectOptions.g_strProjectName

' dimension the weight & map arrays

ReDim

ProjectOptions.g_sMapArray(ProjectOptions.g_iNumDMs,

ProjectOptions.g_iNumCriteria)

ReDim

ProjectOptions.g_dWeightArray(ProjectOptions.g_iNumDMs

, ProjectOptions.g_iNumCriteria)

ReDim

ProjectOptions.g_dRelevanceArray(ProjectOptions.g_iNum

DMs, ProjectOptions.g_iNumCriteria)

m_bInputCompleteFlag = False

m_iDMNumber = 1

m_iCriterionNumber = 1

lblDM.Caption =

ProjectOptions.g_strDMArray(m_iDMNumber)







m_iNumCriteria = 0

'Get the criteria rasterlayers

If GetMap("Criteria Rasters") = False Then

MsgBox "Problem locating map"

Exit Sub

End If

'set up the map combo box

Dim j As Integer




cboMap.AddItem m_pMap.Layer(j).Name

cboConstraint.AddItem m_pMap.Layer(j).Name


End If

Next j

cboMap.AddItem "NONE"

cboConstraint.AddItem "NONE"


AddWeights

End Sub

Private Sub AddWeights()

cboWeight.AddItem "Irrelevant"

cboWeight.AddItem "Very Unimportant"

cboWeight.AddItem "Unimportant"

cboWeight.AddItem "Moderately Important"

cboWeight.AddItem "Important"

cboWeight.AddItem "Very Important"

cboWeight.AddItem "Critically Important"

cboRelevance.AddItem "Irrelevant"

cboRelevance.AddItem "Very Unimportant"

cboRelevance.AddItem "Unimportant"

cboRelevance.AddItem "Moderately Important"

cboRelevance.AddItem "Important"

cboRelevance.AddItem "Very Important"

cboRelevance.AddItem "Critically Important"

End Sub

Private Function DecodeRating(ByVal Rating) As Double



(Rating = 7))) Then

MsgBox "Invalid weight: " & Rating

Exit Function

End If

Select Case Rating


Case 1

DecodeRating = 0

Case 2

DecodeRating = 0.1

Case 3

DecodeRating = 0.3

Case 4

DecodeRating = 0.5

Case 5

DecodeRating = 0.7

Case 6

DecodeRating = 0.9

Case 7

DecodeRating = 1

End Select

End Function

Private Sub SetUpLayer()









interface


Set pClassRen = New










pRasRen.Update












pRamp.Size = 5








label


Dim i As Integer




Select Case i

Case 0


Case 1


Case 2



Case 3


Case 4


End Select

Next i


problem










pRasRen.Update





End Sub

Private Sub CreateDecisionMap()



NormaliseArrays

'add all the coefficients & inverse

Dim i As Integer

Dim j As Integer

Dim bLastMap As Boolean

Dim bDecisionMapFound

Dim dCoefficientArray() As Double

Dim iLastCriterion As Integer

Dim dTotalWeight As Double ' total coefficient weights

to normalise final decision map

dTotalWeight = 0

'get the last map reference right in case the last map

is NONE

For i = 0 To ProjectOptions.g_iNumCriteria - 1

If


ProjectOptions.g_iNumCriteria - i) <> "NONE" Then

iLastCriterion = ProjectOptions.g_iNumCriteria

- i

Exit For

End If

Next i

ReDim dCoefficientArray(ProjectOptions.g_iNumDMs,

ProjectOptions.g_iNumCriteria) As Double


For i = 1 To ProjectOptions.g_iNumDMs

For j = 1 To ProjectOptions.g_iNumCriteria

dCoefficientArray(i, j) =

(ProjectOptions.g_dWeightArray(i, j) *

ProjectOptions.g_dRelevanceArray(i, j))

dTotalWeight = dTotalWeight +

dCoefficientArray(i, j)

Next j

Next i

' Main Loop

Dim iDM As Integer

Dim iCriterion As Integer

Dim pCriteriaMap As IRasterLayer

Dim pDecisionMap As IRasterLayer

If ProjectOptions.g_sConstraintMap <> "NONE" Then

Dim pConstraintMap As IRasterLayer


Set pConstraintMap = m_pMap.Layer(i)

If pConstraintMap.Name =

ProjectOptions.g_sConstraintMap Then

Exit For

End If

Next i

Dim pInRaster3 As IRaster

Set pInRaster3 = pConstraintMap.Raster

End If

For iDM = 1 To ProjectOptions.g_iNumDMs

For iCriterion = 1 To ProjectOptions.g_iNumCriteria

bDecisionMapFound = False


If (iCriterion = iLastCriterion) And (iDM =


bLastMap = True

End If

'get criteriamap

If ProjectOptions.g_sMapArray(iDM, iCriterion) <>

"NONE" Then


Set pCriteriaMap = m_pMap.Layer(i)

If pCriteriaMap.Name =

ProjectOptions.g_sMapArray(iDM, iCriterion) Then

Exit For

End If

Next i


Set pInRaster1 = pCriteriaMap.Raster

Else: GoTo Line1 'there is no map

End If

'Now get decisionmap

If (iDM > 1) Or (iCriterion > 1) Then


Set pDecisionMap = m_pMap.Layer(i)

If pDecisionMap.Name = "TempDecisionMap" Then

bDecisionMapFound = True

Exit For

End If

Next i



Set pInRaster2 = pDecisionMap.Raster

End If






Set pEnv = pRModel





Set pWS =





samples

' N.B. can't reuse created rasters, once created they

are set & trying to




If (bLastMap = True) And

(ProjectOptions.g_sConstraintMap <> "NONE") Then

pRModel.Script = "[Map1] = [CriteriaMap] * " &

dCoefficientArray(iDM, iCriterion) & vbLf + _

"[Map2] = ([DecisionMap] + [map1]) /

" & dTotalWeight & vbLf + _

"[Output] = [Map2] *

[ConstraintMap]"

pRModel.BindRaster pInRaster1, "CriteriaMap"

pRModel.BindRaster pInRaster2, "DecisionMap"

pRModel.BindRaster pInRaster3, "ConstraintMap"

ElseIf bLastMap = True Then



"[Output] = ([DecisionMap] + [map1])

/ " & dTotalWeight



ElseIf ((iDM > 1) Or (iCriterion > 1)) And

bDecisionMapFound = True Then



"[Output] = [DecisionMap] + [map1]"



Else ' first map


pRModel.Script = "[Output] = [CriteriaMap] * " &

dCoefficientArray(iDM, iCriterion)


End If

' Run the model

pRModel.Execute

' Get outputs


Set pRaster1 = pRModel.BoundRaster("Output")

' Unbind raster & delete layer

pRModel.UnbindSymbol "CriteriaMap"

If ((iDM <> 1) Or (iCriterion <> 1)) And

bDecisionMapFound Then

pRModel.UnbindSymbol "DecisionMap"

Dim pLayer As ILayer


Set pLayer = m_pMap.Layer(i)

If pLayer.Name = "TempDecisionMap" Then

m_pMap.DeleteLayer pLayer

Exit For

End If

Next i

End If



If bLastMap = True Then

If GetMap("Decision Rasters") = False Then

MsgBox "Problem locating map"

Exit Sub

End If

End If







pRLayer1.Name = TxtMapName.Text

Else: pRLayer1.Name = "TempDecisionMap"

End If

m_pLayer.Name = pRLayer1.Name

Line1: ' for goto




Next iCriterion

Next iDM

SetUpLayer


ProjectOptions.g_sDecisionMapName = TxtMapName.Text

End Sub

Private Sub NormaliseArrays()

' make each criterions relevance values sum to 1

Dim dCriterionTotal As Double

Dim c As Integer

Dim d As Integer

Dim DM As Integer

For c = 1 To ProjectOptions.g_iNumCriteria

dCriterionTotal = 0

For d = 1 To ProjectOptions.g_iNumDMs

dCriterionTotal = dCriterionTotal +

ProjectOptions.g_dRelevanceArray(d, c)

Next d

For DM = 1 To ProjectOptions.g_iNumDMs

ProjectOptions.g_dRelevanceArray(DM, c) =

ProjectOptions.g_dRelevanceArray(DM, c) /

dCriterionTotal

Next DM

Next c

End Sub

Private Sub CreateRiskMap()




'NormaliseArrays ' no need if decision map has been

created

Dim bRiskMapFound As Boolean

Dim dMinScore As Double 'The minimum acceptable score

to avoid risk

dMinScore = 0.5


bLastMap = False

Dim iLastCriterion As Integer

Dim i As Integer

'get the last map reference right in case the last map

is NONE

For i = 0 To ProjectOptions.g_iNumCriteria - 1

If


ProjectOptions.g_iNumCriteria - i) <> "NONE" Then

iLastCriterion = ProjectOptions.g_iNumCriteria

- i

Exit For

End If

Next i



i = i + 1

Loop








Exit For

End If

Next i



End If

' Main Loop

Dim iDM As Integer



Dim pRiskMap As IRasterLayer



If (iCriterion = iLastCriterion) And (iDM =


bLastMap = True

End If

'get criteriamap


"NONE" Then






Exit For

End If

Next i



Else ' no map

GoTo Line1

End If

'Now get risknmap

bRiskMapFound = False

If (iDM <> 1) Or (iCriterion <> 1) Then


Set pRiskMap = m_pMap.Layer(i)

If pRiskMap.Name = "TempRiskMap" Then

bRiskMapFound = True

Exit For

End If

Next i


Set pInRaster2 = pRiskMap.Raster

End If







Set pEnv = pRModel





Set pWS =





samples


are set & trying to



If (bLastMap = True) And


pRModel.Script = "[NoNewRisk] = [CriteriaMap] >=

[RiskMap]" & vbLf + _

"[NewRisk] = [CriteriaMap] <

[RiskMap] " & vbLf + _

"[PartialRiskMap] = [RiskMap] *

[NoNewRisk]" & vbLf + _

"[PartialCritMap] = [CriteriaMap] *

[NewRisk]" & vbLf + _


"[FinalMap] = [PartialRiskMap] +

[PartialCritMap]" & vbLf + _

"[ContainsRisk] = [FinalMap] < " &

dMinScore & vbLf + _

"[Map1] = ((" & dMinScore & " -

[FinalMap]) / " & dMinScore & ") * [ContainsRisk]" &

vbLf + _

"[Output] = [Map1] *

[ConstraintMap]"


pRModel.BindRaster pInRaster2, "RiskMap"


ElseIf bLastMap = True Then









"[FinalMap] = [PartialRiskMap] +

[PartialCritMap]" & vbLf + _

"[ContainsRisk] = [FinalMap] < " &

dMinScore & vbLf + _

"[Output] = ((" & dMinScore & " -

[FinalMap]) / " & dMinScore & ") * [ContainsRisk]"




ElseIf ((iDM > 1) Or (iCriterion > 1)) And

bRiskMapFound Then









"[Output] = [PartialRiskMap] +

[PartialCritMap]"



Else 'the first map in the array

pRModel.Script = "[Output] = [CriteriaMap]"


End If

' Run the model

pRModel.Execute

' Get outputs






If ((iDM <> 1) Or (iCriterion <> 1)) And bRiskMapFound

Then

pRModel.UnbindSymbol "RiskMap"




If pLayer.Name = "TempRiskMap" Then


Exit For

End If

Next i

End If



i = 0

Do Until m_pMap.Name = "Decision Rasters"


i = i + 1

Loop

End If








pRLayer1.Name = TxtMapName.Text & "_Risk"

Else: pRLayer1.Name = "TempRiskMap"

End If


Line1:




Next iCriterion

Next iDM

SetUpRiskOrConflictLayer

End Sub

Sub SetUpRiskOrConflictLayer()


for risk or conflict








interface


Set pClassRen = New










pRasRen.Update












pRamp.Size = 5








label


Dim i As Integer




Select Case i

Case 0

pClassRen.Label(i) = "Zero"

Case 1

pClassRen.Label(i) = "Small"

Case 2

pClassRen.Label(i) = "Medium"

Case 3

pClassRen.Label(i) = "Large"

Case 4

pClassRen.Label(i) = "Very Large"

End Select


Next i


problem










pRasRen.Update




End Sub

Private Sub CreateMinCriterionMap(iCriterion)


' cycles thru the criterion maps from each DM &

Calculates the min Oijk




created

Dim bMinMapFound As Boolean


bLastMap = False

Dim i As Integer



i = i + 1

Loop

Dim iDM As Integer


Dim pMinMap As IRasterLayer

' Main Loop


bMinMapFound = False

If (iDM = ProjectOptions.g_iNumDMs) Then

bLastMap = True

End If


'get criteriamap


"NONE" Then





Exit For

End If

Next i



ElseIf bLastMap = True Then ' there is no crit map &

the last temp map is it

If GetLayer("TempMinMap") = True Then

m_pLayer.Name = TxtMapName.Text & "_Min " &

ProjectOptions.g_strCriteriaArray(iCriterion)

GoTo Line1

End If

Else ' there is no crit map

GoTo Line1

End If

'Now get MinMap

If (iDM <> 1) Then



Set pMinMap = m_pMap.Layer(i)

If pMinMap.Name = "TempMinMap" Then

bMinMapFound = True

Exit For

End If

Next i


Set pInRaster2 = pMinMap.Raster

End If






Set pEnv = pRModel





Set pWS =






samples


are set & trying to



If (iDM > 1) And bMinMapFound Then

pRModel.Script = "[NoNewMin] = [CriteriaMap] >=

[MinMap]" & vbLf + _

"[NewMin] = [CriteriaMap] < [MinMap]

" & vbLf + _

"[PartialMinMap] = [MinMap] *

[NoNewMin]" & vbLf + _


[NewMin]" & vbLf + _

"[Output] = [PartialMinMap] +

[PartialCritMap]"


pRModel.BindRaster pInRaster2, "MinMap"




End If

' Run the model

pRModel.Execute


' Get outputs





If (iDM > 1) And bMinMapFound Then

pRModel.UnbindSymbol "MinMap"




If pLayer.Name = "TempMinMap" Then


Exit For

End If

Next i

End If









pRLayer1.Name = TxtMapName.Text & "_Min " &


Else: pRLayer1.Name = "TempMinMap"

End If


Line1:




Next iDM

SetUpLayer

End Sub

Private Sub CreateMaxCriterionMap(iCriterion)

' cycles thru the criterion maps from each DM &

Calculates the max Oijk




created

Dim bMaxMapFound As Boolean



bLastMap = False

Dim i As Integer



i = i + 1

Loop

' Main Loop

Dim iDM As Integer


Dim pMaxMap As IRasterLayer


bMaxMapFound = False

If (iDM = ProjectOptions.g_iNumDMs) Then

bLastMap = True

End If

'get criteriamap


"NONE" Then





Exit For

End If


Next i



ElseIf bLastMap = True Then ' there is no crit map &

the last temp map is it

If GetLayer("TempMaxMap") = True Then

m_pLayer.Name = TxtMapName.Text & "_Max " &


GoTo Line1

End If

Else ' there is no crit map

GoTo Line1

End If

'Now get MaxMap

If (iDM > 1) Then


Set pMaxMap = m_pMap.Layer(i)

If pMaxMap.Name = "TempMaxMap" Then

bMaxMapFound = True

Exit For

End If

Next i


Set pInRaster2 = pMaxMap.Raster

End If







Set pEnv = pRModel





Set pWS =





samples


are set & trying to



If (iDM > 1) And bMaxMapFound Then

pRModel.Script = "[NoNewMax] = [CriteriaMap] <=

[MaxMap]" & vbLf + _

"[NewMax] = [CriteriaMap] > [MaxMap]

" & vbLf + _


"[PartialMaxMap] = [MaxMap] *

[NoNewMax]" & vbLf + _


[NewMax]" & vbLf + _

"[Output] = [PartialMaxMap] +

[PartialCritMap]"


pRModel.BindRaster pInRaster2, "MaxMap"




End If

' Run the model

pRModel.Execute

' Get outputs





If (iDM > 1) And bMaxMapFound = True Then

pRModel.UnbindSymbol "MaxMap"





If pLayer.Name = "TempMaxMap" Then


Exit For

End If

Next i

End If








pRLayer1.Name = TxtMapName.Text & "_Max " &


Else: pRLayer1.Name = "TempMaxMap"

End If


Line1:





Next iDM

SetUpLayer

End Sub

Private Sub CreateConflictMap()

' Creates a map expressing conflict amongst DM's

' Just simple rating conflict for now




created


bLastMap = False

Dim i As Integer



i = i + 1

Loop

' Main Loop



Dim pMinCriteriaMap As IRasterLayer

Dim pMaxCriteriaMap As IRasterLayer

Dim pConflictMap As IRasterLayer







Exit For

End If

Next i



End If


'create the min & max maps for that criterion

CreateMinCriterionMap (iCriterion)

CreateMaxCriterionMap (iCriterion)

If (iCriterion = ProjectOptions.g_iNumCriteria) Then

bLastMap = True

End If

'get min & max criteria maps


Set pMinCriteriaMap = m_pMap.Layer(i)


If pMinCriteriaMap.Name = TxtMapName.Text &

"_Min " &

ProjectOptions.g_strCriteriaArray(iCriterion) Then

Exit For

End If

Next i


Set pInRaster1 = pMinCriteriaMap.Raster


Set pMaxCriteriaMap = m_pMap.Layer(i)

If pMaxCriteriaMap.Name = TxtMapName.Text &

"_Max " &


Exit For

End If

Next i


Set pInRaster2 = pMaxCriteriaMap.Raster

'Now get ConflictMap

If (iCriterion > 1) Then


Set pConflictMap = m_pMap.Layer(i)

If pConflictMap.Name = "TempConflictMap" Then

Exit For

End If

Next i


Set pInRaster3 = pConflictMap.Raster

End If







Set pEnv = pRModel





Set pWS =





samples


are set & trying to



If (iCriterion = ProjectOptions.g_iNumCriteria) And



pRModel.Script = "[CriterionConflict] = [MaxMap] -


"[NoNewConflict] = [ConflictMap] >=

[CriterionConflict]" & vbLf + _

"[NewConflict] = [ConflictMap] <


"[PartialConflictMap] =

[ConflictMap] * [NoNewConflict]" & vbLf + _

"[PartialCritConfMap] =

[CriterionConflict] * [NewConflict]" & vbLf + _

"[Output] = ([PartialConflictMap] +

[PartialCritConfMap]) * [ConstraintMap]"



pRModel.BindRaster pInRaster3, "ConflictMap"


ElseIf (iCriterion > 1) Then

pRModel.Script = "[CriterionConflict] = [MaxMap] -


"[NoNewConflict] = [ConflictMap] >=


"[NewConflict] = [ConflictMap] <


"[PartialConflictMap] =

[ConflictMap] * [NoNewConflict]" & vbLf + _

"[PartialCritConfMap] =

[CriterionConflict] * [NewConflict]" & vbLf + _

"[Output] = [PartialConflictMap] +

[PartialCritConfMap]"




pRModel.BindRaster pInRaster3, "ConflictMap"


pRModel.Script = "[Output] = [MaxMap] - [MinMap]"



End If

' Run the model

pRModel.Execute

' Get outputs



' Unbind raster & delete layers

pRModel.UnbindSymbol "MaxMap"

pRModel.UnbindSymbol "MinMap"




If pLayer.Name = TxtMapName.Text & "_Max " &



Exit For

End If

Next i




If pLayer.Name = TxtMapName.Text & "_Min " &



Exit For

End If

Next i

If (iCriterion > 1) Then

pRModel.UnbindSymbol "ConflictMap"



If pLayer.Name = "TempConflictMap" Then


Exit For

End If

Next i

End If



i = 0

Do Until m_pMap.Name = "Decision Rasters"


i = i + 1


Loop

End If







pRLayer1.Name = TxtMapName.Text & "_Conflict"

Else: pRLayer1.Name = "TempConflictMap"

End If





Next iCriterion

SetUpRiskOrConflictLayer

End Sub

Private Function GetLayer(ByVal LayerName As String)

As Boolean


' This function puts the layer with LayerName into

m_pLayer

' and returns true if all went OK

' N.B. the correct map must be in m_pMap

Dim ReturnValue As Boolean

ReturnValue = False

Dim i As Integer



If m_pLayer.Name = LayerName Then

ReturnValue = True

Exit For

End If

Next i

GetLayer = ReturnValue

End Function

Private Function GetMap(ByVal MapName As String) As

Boolean

' sets m_pMap to the named map & RETURNS TRUE IF ALL

IS ok



ReturnValue = False

Dim i As Integer


For i = 0 To m_pMaps.Count - 1


If m_pMap.Name = MapName Then

ReturnValue = True

Exit For

End If

Next i

GetMap = ReturnValue

End Function

Private Sub SaveProjectData()

'writes variables in the projectoptions module to

ProjectName_DATA

If ProjectOptions.g_strProjectName = "" Then

MsgBox "No project to save"

Exit Sub

End If

Dim sDataFileName As String

sDataFileName = "D:\ArcGIS\DecisionTools\ProjectData\"

& ProjectOptions.g_strProjectName & "_DATA"

Open sDataFileName For Random As #1

Put #1, 1, ProjectOptions.g_sDecisionMapName ' Write

record to file.

Put #1, 2, ProjectOptions.g_sConstraintMap

' Arrays

Dim DM As Integer


Dim Cr As Integer

Dim SaveCounter As Integer

SaveCounter = 3


For Cr = 1 To ProjectOptions.g_iNumCriteria

Put #1, SaveCounter,

ProjectOptions.g_dRelevanceArray(DM, Cr)

SaveCounter = SaveCounter + 1

Next Cr

Next DM




ProjectOptions.g_dWeightArray(DM, Cr)


Next Cr

Next DM




ProjectOptions.g_sMapArray(DM, Cr)


Next Cr

Next DM


Close #1

End Sub

EXPLORATION

Option Explicit

Private m_pMxApp As IMxApplication




Private m_pIdentify As IIdentify

Private m_pIDArray As IArray

Private m_pRasterIdObj As IRasterIdentifyObj

Private m_pIdObj As IIdentifyObj

Private m_RatingArray As Variant



Unload Me

End Sub

Private Sub cboDM_Change()

LboCriteriaOutcomes.Clear

If Not GetMap("Criteria Rasters") Then

MsgBox "Problem locating Criteria Rasters"

Exit Sub

End If


Dim DM As Integer

DM = cboDM.ListIndex + 1

Dim i As Integer

For i = 1 To ProjectOptions.g_iNumCriteria

'Get the criteria map

If ProjectOptions.g_sMapArray(DM, i) = "NONE" Then

LboCriteriaOutcomes.AddItem

(ProjectOptions.g_strCriteriaArray(i) & ": No Map")

GoTo Line1

End If

If Not GetLayer(ProjectOptions.g_sMapArray(DM, i))

Then


(ProjectOptions.g_strCriteriaArray(i) & ": Map (" &

ProjectOptions.g_sMapArray(DM, i) & ") not found")

GoTo Line1

End If

Set m_pIdentify = m_pLayer

'Convert x and y to map units

Set m_pIDArray =

m_pIdentify.Identify(ThisDocument.p_pPoint)

'Get the FeatureIdentifyObject

If Not m_pIDArray Is Nothing Then

Set m_pRasterIdObj = m_pIDArray.Element(0)

Set m_pIdObj = m_pRasterIdObj

'm_pIdObj.Flash m_pMxApp.Display


'Report info from FeatureIdentifyObject

'MsgBox "Layer:" & m_pIdObj.Layer.Name & vbNewLine

& "Feature:" & m_pIdObj.Name

Else

MsgBox "Nothing to identify for " & m_pLayer.Name

End If

If IsNumeric(m_pIdObj.Name) Then


(ProjectOptions.g_strCriteriaArray(i) & ": " &

LingApprox(m_pIdObj.Name))

Else: LboCriteriaOutcomes.AddItem

(ProjectOptions.g_strCriteriaArray(i) & ": Location

not rated")

End If

Line1:

Next i

End Sub

Private Sub Frame2_Click()

End Sub


End Sub


End Sub



End Sub

Private Sub TabStrip1_Change()

End Sub


End Sub


If ProjectOptions.g_strProjectName = "" Then

MsgBox "No project selected"

Exit Sub

End If

'need to change this so term sets can change

m_RatingArray = Array("Totally Unsuitable", "Bad",

"Indifferent", "Good", "Perfect")

Set m_pMxApp = Application

Set m_pMxDoc = Application.Document


Dim i As Integer

'set up the DM combo box

For i = 1 To ProjectOptions.g_iNumDMs

cboDM.AddItem ProjectOptions.g_strDMArray(i)

Next i

'get decision rasters & decision map

If Not GetMap("Decision Rasters") Then


MsgBox "Decision Rasters not found"

Exit Sub

End If

If Not GetLayer(ProjectOptions.g_sDecisionMapName)

Then

MsgBox "Decision Map for " &

ProjectOptions.g_strProjectName & " not found"

Exit Sub

End If


'Convert x and y to map units

Set m_pIDArray =


'Get the FeatureIdentifyObject

If Not m_pIDArray Is Nothing Then



If m_pIdObj.CanIdentify(m_pLayer) Then

m_pIdObj.Flash m_pMxApp.Display

'Report info from FeatureIdentifyObject

'MsgBox "Layer:" & m_pIdObj.Layer.Name &

vbNewLine & "Feature:" & m_pIdObj.Name

Else: MsgBox "Can't identify"

Exit Sub

End If

Else: MsgBox "Nothing to identify"


Exit Sub

End If

'check if the point is OK

If m_pIdObj.Name = "NoData" Or (Not

IsNumeric(m_pIdObj.Name)) Or m_pIdObj.Name > 1 Or

m_pIdObj.Name < 0 Then

MsgBox "Location not rated"

Exit Sub

End If

lblLocationx.Caption = "X: " & ThisDocument.p_pPoint.x

lblLocationy.Caption = "Y: " & ThisDocument.p_pPoint.y

lblProject = ProjectOptions.g_strProjectName

lblScore = m_pIdObj.Name 'the raw value

lblRating = LingApprox(m_pIdObj.Name)

If GetLayer(ProjectOptions.g_sDecisionMapName &

"_Conflict") Then


Set m_pIDArray =




lblConflict = m_pIdObj.Name

End If

If GetLayer(ProjectOptions.g_sDecisionMapName &

"_Risk") Then


Set m_pIDArray =





lblRisk = m_pIdObj.Name

End If

End Sub

Private Function GetMap(ByVal MapName As String) As

Boolean

' sets m_pMap to the named map & RETURNS TRUE IF ALL

IS ok



ReturnValue = False

Dim i As Integer

For i = 0 To m_pMaps.Count - 1


If m_pMap.Name = MapName Then

ReturnValue = True

Exit For

End If

Next i

GetMap = ReturnValue

End Function

Private Function LingApprox(ByVal Score As Double) As

String

If IsNumeric(Score) And Score >= 0 And Score <= 1 Then


If Score < 0.1 Then

LingApprox = m_RatingArray(0)

Exit Function

ElseIf Score < 0.3 Then


Exit Function



Exit Function



Exit Function

Else: LingApprox = m_RatingArray(4)

End If

Else: LingApprox = "Unscored or score out of range"

End If

End Function

Private Function GetLayer(ByVal LayerName As String)

As Boolean

' This function puts the layer with LayerName into

m_pLayer

' and returns true if all went OK

' N.B. the correct map must be in m_pMap


ReturnValue = False

Dim i As Integer




If m_pLayer.Name = LayerName Then

ReturnValue = True

Exit For

End If

Next i

GetLayer = ReturnValue

End Function

345Appendix E ANZIIS Questionaire

Appendix E

AANNZZIIIISS QQUUEESSTTIIOONNAAIIRREE


Stating the problem

Selecting sites for infrastructure developments is a complex task, involving multiple stakeholders

with conflicting interests and poorly defined or uncertain evaluation criteria. A software system is

needed to store, analyze and visualize data and information relevant to the site selection task. The

use of Geographical Information Systems (GIS) in site selection has a long history, with most

approaches being based on a multiple criteria evaluation (MCE) framework. However most GIS-

based MCE methods have inherent difficulties and limitations, generally stemming from the

following causes.

• The methods tend to require crisp numeric value judgments, whereas some criteria may only be

defined using qualitative measures.

• There is generally an assumption of consensus among decision-makers, which does not exist in

reality.

• Assessments by decision-makers are input without defining the level of certainty the decision-

maker places on the assessment.

• The overall aggregated suitability of an alternative is derived solely from a weighted summation

of suitability assessments, without consideration of conflicts, risks or uncertainty inherent in the

alternative.

‘InfraPlanner’ is a SDSS, developed using a fuzzy algorithm to mitigate these difficulties. Broadly

speaking InfraPlanner is an intelligent information system based on approximate reasoning that

offers the following capabilities:

• Linguistic interaction: Linguistic interaction is provided using primary term sets semantically

defined by parameter-based fuzzy numbers, which may be enhanced via a hedging procedure to

add more terms. Both input and feedback is accomplished linguistically.

• Multiple decision-maker capability: The system accepts linguistic inputs from each party

involved in the decision-making process. Conflict between parties is assessed based on differing

suitability and weighting judgments and factored into overall site suitability.

• Uncertainty assessment: There are two types of uncertainty inherent in decision-maker

suitability assessments: linguistic and quantitative. Linguistic uncertainty is represented by the

fuzziness of the primary suitability term, whereas quantitative uncertainty is represented using

the concept of a type-2 fuzzy set and its footprint of uncertainty (FOU). Quantitative uncertainty


is the term used here to represent uncertainty in the source data and/or its relationship with site

suitability.

• User controllable aggregation: Users have the ability to choose an aggregation that minimizes

uncertainty, risk or conflict, or maximizes compensatory suitability. A variety of compensatory

and non-compensatory linguistically defined outcomes may be delivered.

To illustrate how InfraPlanner works and the process by which it was developed, two Logic Models

have been created. A Logic Model presents a model of how a program/process works to solve a

specified problem. The Development Logic Model illustrates how the problem of developing

InfraPlanner was approached. The InfraPlanner Algorithm Logic Model illustrates how

InfraPlanner aids the solution of site selection problems.


InfraPlanner Development Process Logic Model Resources Activities Outputs Customers Short term outcomes Long term outcomes

Users / planners / other

experts User needs

outlined Development

team User needs documented

Literature Critical review

State of the art report

Development team/ other researchers

Technology consultant Technology

review Technology requirements

Ideas for improvement documented

Interviews / focus groups

Development team

Technology selection

Development team

Software development

‘Infraplanner’ SDSS

BAC users / other planners /

researchers

SDSS developed & deployed

Better site selection decisions

GIS technology

External Influences: BAC staff turnover, technological development, advancements in operations research, political environment.


‘Infraplanner’ Algorithm Logic Model Resources Activities Outputs Customers Short term outcomes Long term outcomes

Raw spatial data

Single variable raster maps

DM’s / Users Raw data pre-processed

Single variable raster maps

Create suitability maps

Raster suitability maps

DM’s / Users / stakeholders

Spatial data now represents suitability

GIS conversion function

Output maps Explore & reduce

alternatives

Suitable sites DM’s / users / stakeholders

Reduced set of alternatives

Final site selected

External Influences: Local economy, political environment, community participation.

Linguistic DM prefernces

Suitability map interface

Raster suitability maps

Create parameter

maps

Suitability, uncertainty, risk & conflict maps

DM’s / Users / stakeholders

Spatial data processed & ready

for exploration

Linguistic DM prefernces

Aggregation interface


Questions

Development process Logic Model

1. Are the resources / activities shown in the logic model sufficient to achieve the

desired outcome? Y / N

If not, what resources / activities should be added to the development process?

……………………………………………………………………………………

……………………………………………………………………………………

………………………………………………………………

2. Is the sequence of activities logical? Y / N

If not, how could the sequence be changed?

……………………………………………………………………………………

……………………………………………………………………………………

………………………………………………………………

3. What other external influences should be considered?

……………………………………………………………………………………

……………………………………………………………………………………

………………………………………………………………

Algorithm Logic Model

4. Are the resources / activities shown in the logic model sufficient to achieve the

desired outcome? Y / N

If not, what resources / activities should be added to the solution algorithm?

……………………………………………………………………………………

……………………………….

……………………………………………………………………………………

………………………………..

5. Is the sequence of activities logical? Y / N

If not, how could the sequence be changed?


……………………………………………………………………………………

……………………………………………………………………………………

………………………………………………………………

6. What other external influences should be considered?

……………………………………………………………………………………

……………………………………………………………………………………

………………………………………………………………

General decision-making issues

7. Is a fuzzy set a valid way to quantify a word? Y / N

8. Should the opinion of some participants in the decision-making process be

weighted more heavily than others in some cases? Y / N

9. If so in what situation:

a) If they have greater expertise in rating a particular criterion

b) If they have a greater stake in the outcome

c) If they carry more responsibility for the outcome

9. Would you rather:

a) A computer algorithm to make a site selection decision for you

b) A computer algorithm to give you processed information based on your

preferences but still make the decision yourself

c) A combination of a and b

General comments:

Development of an optimal spatial decision-making system...

Documents

Transcript of Development of an optimal spatial decision-making system...