Designing and building real environmental decision support ...ia/articulos/MPochetal-EMS.pdf · a...

ARTICLE IN PRESS

Environmental Modelling & Software XX (2003) XXX–XXXwww.elsevier.com/locate/envsoft

Designing and building real environmental decision supportsystems

Manel Pocha,∗, Joaquim Comasa, Ignasi Rodrı´guez-Rodaa, Miquel Sanchez-Marre` b,Ulises Corte´s b

a Laboratori d’Enginyeria Quı´mica i Ambiental (LEQUIA), Universitat de Girona, Campus de Montilivi s/n, Girona, Catalonia 17071Spainb Knowledge Engineering and Machine Learning Group, Universitat Polite`cnica de Catalunya, Campus Nord, Edifici C5, Jordi Girona 1-3,

08034 Barcelona, Spain

Received 31 October 2002; received in revised form 3 February 2003; accepted 5 March 2003

Abstract

The complexity of environmental problems makes necessary the development and application of new tools capable of processingnot only numerical aspects, but also experience from experts and wide public participation, which are all needed in decision-makingprocesses. Environmental decision support systems (EDSSs) are among the most promising approaches to confront this complexity.The fact that different tools (artificial intelligence techniques, statistical/numerical methods, geographical information systems, andenvironmental ontologies) can be integrated under different architectures confers EDSSs the ability to confront complex problems,and the capability to support learning and decision-making processes. In this paper, we present our experience, obtained over thelast 10 years, in designing and building two real EDSSs, one for wastewater plant supervision, and one for the selection of wastewatertreatment systems for communities with less than 2000 inhabitants. The flow diagram followed to build the EDSS is presented foreach of the systems, together with a discussion of the tasks involved in each step (problem analysis, data collection and knowledgeacquisition, model selection, model implementation, and EDSS validation). In addition, the architecture used is presented, showinghow the five levels on which it is based (data gathering, diagnosis, decision support, plans, and actions) have been implemented.Finally, we present our opinion on the research issues that need to be addressed in order to improve the ability of EDSSs to copewith complexity in environmental problems (integration of data and knowledge, improvement of knowledge acquisition methods,new protocols to share and reuse knowledge, development of benchmarks, involvement of end-users), thus increasing our understand-ing of the environment and contributing to the sustainable development of society. 2003 Elsevier Ltd. All rights reserved.

Keywords:Environmental decision support systems; Artificial intelligence; Wastewater treatment; Knowledge management

1. Introduction

1.1. Statement of the problem

The increasing rhythm of industrialisation, urbanis-ation and population growth that our planet has faced forthe last few hundred years has forced society to considerwhether human beings are changing the very conditionsthat are essential to life on Earth. Environmental pol-

∗ Corresponding author. Tel.:+34-972-418804; fax:+34-972-418150.

E-mail addresses:[email protected] (M. Poch); [email protected] (J. Comas); [email protected] (I. Rodrı´guez-Roda);[email protected] (M. Sa`nchez-Marre`); [email protected] (U. Corte´s).

1364-8152/$ - see front matter 2003 Elsevier Ltd. All rights reserved.doi:10.1016/j.envsoft.2003.03.007

lution affects the quality of water, air, and soil nega-tively, and hence plant, animal and human life (Sydowet al., 1998; El-Swaify and Yakowitz, 1998).

Whenever we attempt to tackle these issues, we areimmediately confronted with complexity. There are atleast two important reasons for this:

– Uncertainty, or approximate knowledge. Some of thesources of this uncertainty can be tamed withadditional data or further investigation. Such is thecase of uncertainty arising from random processes orfrom deficiencies in knowledge (lack of data, unsuit-able datasets, etc.). But in other cases, uncertainty isinsurmountable. This is the case for chaotic behaviour,or for self-organisation processes. It is also typical of

ARTICLE IN PRESS2 M. Poch et al. / Environmental Modelling & Software XX (2003) XXX–XXX

socio-ecological systems, which involve numerousplayers, each with their own goals.

– Multiplicity of scales. Environmental problems havebeen associated traditionally with distinct spatialscales (i.e. local, national, global), each associatedwith specific timescales. However, interactions amongthese scales are becoming increasingly clear. There-fore, advocating a single perspective that encompasseseverything in a system is becoming increasingly dif-ficult—plus ineffective.

The consensus is developing that, in order to accountfor these caveats, environmental issues must be con-sidered in terms of complex systems. But not all environ-mental systems present the same level of complexity interms of both the degree of uncertainty and the risk asso-ciated with decisions. If the degree of complexity is rep-resented as a function of uncertainty, on one hand, andthe magnitude or importance of the decision, on theother, then we might distinguish three levels of com-plexity (Funtowicz and Ravetz, 1993, 1999):

– The first level of complexity would correspond to sim-ple, low uncertainty systems where the issue at handhas limited scope. A single perspective and simplemodels would suffice to provide satisfactory descrip-tions of the system. With regard to water issues, thislevel corresponds, for example, to the evolution ofoxygen in a pristine stream after a pulse input ofassimilable organic matter. In the context of industrialprocesses, an example is the design of a single treat-ment operation where the input is perfectly defined.In these cases, the information arising from analysismay be used for more wide-reaching purposes beyondthe scope of the particular researcher.

– The second level would correspond to systems withenough uncertainty that simple models, applicable todifferent situations and manageable by any competentpractitioner, can no longer provide satisfactorydescriptions. Acquired experience becomes then moreand more important, and the need to involve expertsin problem solving becomes advisable. In the case ofwater issues, this level would correspond to a generalmodel of water quality, where the need arises to estab-lish which factors are the most important. In the caseof an industrial process, this level would correspondto the installation of a wastewater treatment plant(WWTP), where goals for the quality of the outputare well established but these can be reached throughdifferent schemes, and it is the responsibility of thedesigner to choose the most appropriate configuration.

– The third level would correspond to truly complexsystems, where much epistemological or ethicaluncertainty exists, where uncertainty is not necessarilyassociated with a higher number of elements orrelationships within the system, and where the issues

at stake reflect conflicting goals. It is then crucial toconsider the need to account for a plurality of viewsor perspectives. In the case of water issues, anexample would be the problem of water quality in astream catchment. Here, a variety of factors(economical, technical, ecological, etc.) are at play,and associated with each factor is a different set ofgoals. Thus, different kinds of expertise need to betaken into account. In the case of an industrial process,this level of complexity is associated, for instance,with the environmental aspects of wastewater treat-ments, which are discussed at the level of the com-pany’s policy. Thus, the problem is not the design ofend of pipe installations for the treatment of specificoutputs, but a more global view on the problem thatwould contemplate, for example, the installation ofcleaner technologies in the production process itself.

In this sense, it is important to realise that environ-mental problems are characterised by dynamics andinteractions that do not allow for an easy divisionbetween social and biogeophysical phenomena. Manyecological theories have been developed in systemswhere humans were absent, or in systems where humanswere considered an exogenous, simple, and detrimentaldisturbance. The intricate ways in which humans interactwith ecological systems have been rarely considered(Kinzig, 2001). Embracing a socio-economical perspec-tive implies accepting that all decisions related toenvironmental management are characterised by mul-tiple, usually conflicting objectives, and by multiple cri-teria (Ostrom, 1991). Thus, in addition to the role ofexperts, it becomes increasingly important to considerthe role of wide public participation in the decision-mak-ing processes. Experts are consulted by policy makers,the media, and the public at large to explain and adviseon numerous issues. Nonetheless, many recent caseshave shown, rather paradoxically, that while expertise isincreasingly sought after, it is also increasingly contested(Ludwig, 2001).

In our opinion, this third level cannot be tackled withthe traditional tools of mathematical modelling. To con-front this complexity, a new paradigm is needed. Adopt-ing it will require that we deal with new intellectual chal-lenges.

1.2. New tools for a new paradigm

In the last few decades, mathematical/statistical mod-els, numerical algorithms and computer simulations havebeen used as the appropriate means to gain insight intoenvironmental management problems and provide usefulinformation to decision-makers. To this end, a wide setof scientific techniques have been applied to environ-mental management problems for a long time and withgood results.

ARTICLE IN PRESS3M. Poch et al. / Environmental Modelling & Software XX (2003) XXX–XXX

But most of these efforts were focused on problemsthat we could assign to the first level of complexity.Consequently, many complex environmental problemshave not been effectively addressed by the scientificcommunity. Recently, however, the effort to integratenew tools to deal with more complex systems has led tothe development of the so-called environmental decisionsupport systems (EDSSs) (Guariso and Werthner, 1989;Rizzoli and Young, 1997).

EDSSs have generated high expectations as a tool totackle problems belonging to the second and third levelsof complexity. Thus, in a recent review of the relevantliterature in the topic, more than 600 references werefound (including journal articles, conference papers, andtechnical reports) during the 90s, with only 10 referencesin 1992 and more than 150 references per year towardsthe end of the decade (Cortes et al., 2002). The rangeof environmental problems to which EDSSs have beenapplied is wide and varied, with water management atthe top (25% of references), followed by aspects of riskassessment (11.5%) and forest management (11.0%).Equally varied are the tasks to which EDSSs have beenapplied, ranging from monitoring and data storage toprediction, decision analysis, control planning, reme-diation, management, and communication with society.It is not surprising then that three of the top 10 mostdownloaded articles published in Environmental Model-ling and Software in January–December 2001 dealwith EDSSs.

This review, together with the work of other authors,also revealed that there is a wide range of opinions onwhat constitutes an EDSS. The fact that this approachis relatively recent and integrates multiple tools meansthat there is not a single, consensual definition of EDSS.However, even though one may argue that a databasemanagement system could be used as a decision supportsystem, today’s consensus is that EDSSs must adopt aknowledge-based approach, which includes the steps ofknowledge acquisition, representation, and management.

The fact that different tools can be integrated underdifferent architectures makes EDSSs difficult to define.It also means that different approaches to design andimplementation coexist.

In this context, we present our experience with thedesign and implementation of two EDSSs in the domainof water management. We explicitly describe theirdevelopment and the architecture used for the appli-cations.

1.3. EDSS development

According to Fox and Das (2000), a decision supportsystem is a computer system that assists decision-makersin choosing between alternative beliefs or actions byapplying knowledge about the decision domain to arriveat recommendations for the various options. It incorpor-

ates an explicit decision procedure based on a set oftheoretical principles that justify the “ rationality” ofthis procedure.

Thus, an EDSS is an intelligent information systemthat reduces the time in which decisions are made in anenvironmental domain, and improves the consistencyand quality of those decisions (Haagsma and Johanns,1994; Cortes et al., 2001). Decisions are made when adeviation from an expected, desired state of a system isobserved or predicted. This implies a problem awarenessthat in turn must be based on information, experienceand knowledge about the process. Those systems arebuilt by integrating several artificial intelligencemethods, geographical information system components,mathematical or statistical techniques, and environmen-tal ontologies (Fig. 1).

How a particular EDSS is constructed will varydepending on the type of environmental problem and thetype of information and knowledge that can be acquired.With these constraints in mind, and after an analysis ofthe available information, a set of tools can be selected.This applies not only to numerical models, but also toartificial intelligence (AI) methodologies, such as knowl-edge management tools. The use of AI tools and modelsprovides direct access to expertise, and their flexibilitymakes them capable of supporting learning and decision-making processes. Their integration with numericaland/or statistical models in a single system provideshigher accuracy, reliability and utility (Cortes et al.,2000).

This confers EDSSs the ability to confront complexproblems, in which the experience of experts providesvaluable help for finding a solution to the problem. Italso provides ways to accelerate identification of theproblem and to focus the attention of decision-makerson its evaluation. Once implemented, an EDSS, like anyknowledge-based system, has to be evaluated for whatit knows, for how it uses what it knows, for how fast it

Fig. 1. EDSS conceptual components.


can learn something new, and, last but not least, for itsoverall performance. Fig. 2 shows schematically themethodology used to develop the two study casespresented here.

Both the proposed EDSS development procedure andthe EDSS architecture are general enough to be intendedto cope with any kind of EDSS deployment.

We propose an EDSS architecture based on five levels(Fig. 3):

� The first level of the EDSS (data gathering)encompasses the tasks involved in data gathering andregistration into databases. Original raw data are oftendefective, requiring a number of pre-processing pro-cedures before they can be registered in an under-standable and interpretable way. Missing data anduncertainty must be also considered in this level.

� The second level, diagnosis level, includes the reason-ing models that are used to infer the state of the pro-cess so that a reasonable proposal of actuation can bereached. This is accomplished with the help of statisti-cal, numerical and artificial intelligence models.

� The third level, decision support level, establishes a

Fig. 2. Flow diagram for development of an EDSS.

supervisory task that entails gathering and mergingthe conclusions derived from knowledge-based andnumerical techniques. This level also raises the inter-action of the users with the computer system throughan interactive and graphical user–machine interface.When a clear and single conclusion cannot bereached, a set of decisions ordered by their probabilityshould be presented to the user.

� In the fourth level, plans are formulated and presentedto managers as a list of general actions suggested tosolve a specific problem.

� The set of actions to be performed to solve problemsin the domain considered are in the fifth level. Thesystem recommends not only the action, or a sequenceof actions (a plan), but also a value that has to beaccepted by the decision-maker. This is the last levelin the architecture that closes the loop.

2. Two case studies

In this section, two case studies are presented wherethe proposed methodology has been applied. The two


Fig. 3. EDSS architecture.

case studies correspond to two different situations withdifferent forms of complexity. Although specific detailsof the corresponding databases cannot be published(proprietary information of Consorci per a la Defensa dela Conca del riu Besos and Agencia Catalana del’Aigua), the available information included in theexamples is clear enough to follow the process ofdesigning and building the EDSS.

The first case corresponds to the application of anEDSS to the supervision of a WWTP. Here, both quanti-tative information (obtained on-line and off-line) andqualitative information are used, with the important par-ticipation of experts. While discrepancies among expertsmay arise, there are no conflicts of interest. Of the fourconceptual components of the EDSS as stated above, thegeographic component is not relevant in this case, whilethe numeric component is the only one traditionally usedfor tackling the problem. In the scheme of complexityand risk, it lies between the second and third levels.

The second case corresponds to the selection of waste-water treatment and disposal systems in Catalonia. It isa planning problem in which the temporal componenthas little relevance, since on-line responses are notneeded. The importance of numeric methods is lowerthan in the first case, while the importance of the GISand expert experience components increases. Conflicts

of interest among experts may arise, and the interactionsbetween social and biogeophysical phenomena becomerelevant. In the scheme of complexity, it would lie inthe third level.

2.1. Wastewater treatment plant supervision

2.1.1. EDSS building2.1.1.1. Problem analysis A typical WWTP usuallyincludes a physical and/or chemical primary treatmentand a biological secondary treatment to remove organicmatter and suspended solids from wastewater. Primarytreatment is designed to physically remove solid materialfrom the incoming wastewater. The wastewater flowingto the secondary treatment is called the primary effluent.Secondary treatment usually consists of a biological con-version of dissolved and colloidal organic compoundsinto stabilised, low-energy compounds and new biomasscells, caused by a very diversified group of micro-organisms, in the presence of oxygen. The mixture ofmicroorganisms and particles has the ability to settle andseparate from treated water in the clarifier. A biologicalreactor followed by a secondary settler or clarifier consti-tutes the activated sludge process, which is the mostwell-known process of secondary treatment because it isalso the most widely used.


Like other environmental and biotechnological pro-cesses, WWTPs are complex systems involving manyinteractions between physical, chemical and biologicalprocesses, e.g. chemical and biological reactions, kin-etics, catalysis, transport phenomena, separations, etc.The successful management of these systems requiresmultidisciplinary approaches and expertise from differ-ent scientific fields. Some of the special and problematicfeatures of these processes are:

� Intrinsic instability: most of the chemical and physicalproperties as well as the size and species diversity ofthe population of microorganisms involved inenvironmental processes do not remain constantover time.

� Many of the facts and principles underlying thedomain cannot be characterised precisely solely interms of a mathematical theory or a deterministicmodel with clearly understood properties.

� Uncertainty and imprecision of data or approximateknowledge and vagueness: these processes generate aconsiderable amount of qualitative information.

� Huge quantity of data/information: the application ofcurrent computer technology to the control and super-vision of these environmental systems has led to asignificant increase in the amount of data acquired.

� Heterogeneity and scale: because the media in whichenvironmental processes take place are not homo-geneous and cannot easily be characterised bymeasurable parameters, data are often heterogeneous.Also, the different scale times inherent to differentmeasures in the process have to be properly integratedand managed.

Due to the complexity of wastewater treatment pro-cess control, even the most advanced conventional hardcontrol systems have encountered limitations when deal-ing with problem situations that require qualitative infor-mation and heuristic reasoning for their resolution(Olsson et al., 1998). Indeed, to describe these qualitat-ive phenomena or to evaluate circumstances that mightcall for a change in the control action, some kind oflinguistic representation built on the concepts andmethods of human reasoning, such as intelligent sys-tems, has been necessary. This is also the reason whyhuman operators have, until now, constituted the finalstep in closed-loop plant control. A deeper approach isnecessary to overcome the limited capabilities of con-ventional automatic control techniques when dealingwith abnormal situations in complex systems, and toprovide the level and quality of control necessary to con-sistently meet environmental specifications.

For these reasons, the use of EDSSs began to lookpromising in terms of solutions to these problems. Areasonable, distributed proposal outlines the scope forthe integration of AI tools such as pattern recognition,

knowledge-based systems, fuzzy logic, artificial neuralnetworks, case-based systems, or inductive decisiontrees, which handle the particular characteristics of com-plex processes with numerical and conventional compu-tational techniques, such as statistical methods, advancedand robust control algorithms and system identificationtechniques.

The WWTP selected to develop and apply our pro-posed Supervisory System prototype is located in Gran-ollers, in the Besos river basin (Catalonia). Nowadays,this facility provides preliminary, primary and secondarytreatment to remove the organic matter, suspended solidsand, under some conditions, nitrogen contained in theraw water of about 130,000 inhabitant-equivalents. TheGranollers WWTP has several particular characteristicsthat increase the potential advantages of the developmentand application of an intelligent supervisory system tocontrol and supervise the wastewater treatment process.Among these characteristics, we would like to emphasisethe following:

� Availability of a significant amount of historical rec-ords describing plant operation.

� This plant has a high level of automation centralisedin a computer that collects on-line data and controlsmost of the plant operations.

� The Granollers WWTP is a highly variable system.A wide range of different situations take placethroughout the year causing significant changes in theinfluent characteristics (storms, overloading, nitrifi-cation in hot periods, uncontrolled industrial spillsetc.), which affect standard process operation.

� High level of specialisation of plant experts who havebeen working in the plant from the beginning of itsoperation. They are perfectly acquainted with all sortsof details that make up the heuristic knowledge ofthe plant.

2.1.1.2. Data collection and knowledge acquisition Avariety of methods were used for the development of aknowledge base for this study. Conventional knowledgeacquisition methods (literature review, interviews, etc.)were used first. To overcome the limitations of conven-tional methods, these were supplemented with the useof different automatic knowledge acquisition methods.These latter methods can be supervised, mainly inductivelearning techniques such as CN2, C4.5 and k-NN orunsupervised, such as some clustering method (R.-Rodaet al., 2001). Fig. 4 illustrates the main sources andmethods that were used to acquire both general and spe-cific knowledge on the wastewater treatment processes.

2.1.1.3. Model selection Two types of models wereselected: rule-based reasoning models (expert system)and case-based reasoning models.

Rule-based systems (RBS) offer a number of advan-


Fig. 4. Methods used to acquire knowledge.

tages that overcome some of the limitations of othertechniques: they facilitate the inclusion and retention ofheuristic knowledge from experts and allow the pro-cessing of qualitative information; knowledge is rep-resented in an easily understandable form (rules); a well-validated expert system offers potentially optimalanswers because action plans are systematised for eachproblematic situation; and, finally, expert systems makethe acquisition of a large general knowledge base poss-ible with somewhat rigid use for any WWTP.

A Case-Based Reasoning System (CBS) is based onthe idea that solving a problem for the second time isusually easier than solving it for the first time becausewe remember and repeat the previous solution or recallour mistakes and try to avoid them. The basic idea is toadapt solutions applied in the past to particular problemsaffecting process performance and apply them to newproblems that are similar in nature with less effort thanwith other methods that start from scratch. A case isdescribed as a conceptualised piece of knowledge rep-resenting an experience that teaches a fundamental les-son on how to achieve the reasoner’s goals. In the waste-water treatment domain, the case is a codified descriptionof a specific state or experience of the studied WWTP.Codification should be in computer storable form inorder to be easily retrieved in the future. This paradigmsupplies a flexible and dynamic model allowing theEDSS to be adapted and used for any WWTP with asimilar technology.

2.1.1.4. Model implementation Among the differentpossibilities (tables, decision trees, or knowledge dia-grams and frames) for the representation of the elicitedknowledge, decision trees were selected as the most suit-able representation (Sanchez-Marre et al., 1996; Comaset al., 2003). All the symptoms, facts, procedures andrelationships used for problem diagnosis can be cast intoa set of decision trees. The translation of the knowledgecontained in a branch of decision trees into a productionrule is direct. The resulting trees, which avoid contradic-tions and redundancies, comprise, in our case, diagnosis,cause identification, and action strategies for a widerange of problems in WWTP operation. Logic treesserve as a record of the expert’s step-by-step informationprocessing and decision-making activity. Some branchesare specific and contain peculiarities of the plant, whileothers are more general and can be applied to any plant.

The set of specific cases is stored in a structured mem-ory in a case-base (the case-library) and initialised witha set of typical cases in the plant. The CBS developmentincludes the case definition, the case-library structuredefinition, and the selection of the initial seed. CBSsrequire a case-library to broadly cover the set of potentialproblems. These cases are indexed in memory so as tobe retrieved whenever the experiences they encapsulatecan contribute to achieving the goals of the process. Bothsuccesses and failures must be included in the case-library. It is advisable to initialise the library with a setof common situations (or cases) obtained from real dataor provided by experts on the process. Thus, the CBSwill be from the very start ready to propose solutions toproblems that are similar to those considered in theinitial “seed” . The initial seed at Granollers included 74real cases from the historical database, which covered abroad range of situations covering the main problems inthe process as well as normal situations. The library isupdated with new cases as the knowledge about the pro-cess progresses; so the CBS evolves into a better rea-soner and system accuracy benefits from these newacquisitions. However, because large amounts of infor-mation can overcrowd the library, only the most relevantcases are included.

Fig. 5 shows our intelligent approach to improve thesupervision of WWTPs. The integration of different AItechnologies (RBS, CBS and Artificial Neural Networks)with numerical methods (classical control systems ormodels) leads to a hybrid knowledge-based system cap-able of overcoming the limitations found when solvingcomplex problems with a sole classical or AI technique.Thus, this multidisciplinary approach appears to be themost optimal solution to guarantee the successful controlof complex processes like WWTP.

2.1.1.5. Validation of the EDSS Field testing was con-sidered to be the most effective validity test. The mainobjective of field validation was to test the use of theoverall EDSS in situ with real cases. We wanted to testthe system in its real environment and identify needs forfurther modifications. The system performance wastested for more than 10 months in its actual operatingenvironment, where it is working as a real-time decisionsupport system. During the period of exhaustive vali-dation, the EDSS coped with 123 different problem situ-


Fig. 5. Our hybrid intelligent approach to supervise WWTP.

ations and suggested suitable action strategies, in mostof the cases.

The WWTP problem situations detected includedfoaming, rising, filamentous bulking, underloading,overloading, deflocculation (including possible toxicshock), hydraulic shock, mechanical fault, poor primarysettler performance, non-biological origin problem onclarifier (sudden oscillation of up-flow velocity and badperformance of the clarifier due to an excess of biomassconcentration), and influent nitrogen/organic mattershock. From those, 79.7% were successfully identified(98 situations, about one-third in advance and two-thirdthe same day), and 8.1% were wrong identified (10situations), while 12.2% were not identified (15situations). Table 1 lists the number of correct situationsdetected in this period of time, specifying the problemand whether they were detected in advance or they weredetected the same day. Nowadays, the EDSS is used asa complementary tool of diagnosis for the everyday man-agement of the activated sludge process (Rodrıguez-Roda et al., 2002).

2.1.2. EDSS operationThe different tasks of the five levels of the EDSS for

WWTP control and supervision are performed cycli-cally, using a supervisory cycle. Fig. 6 outlines theEDSS supervisory cycle.

Each cycle is composed of five steps and severaltasks: data gathering (with the data acquisition andupdating tasks), diagnosis, decision support (with theuser-validation and action tasks) and plans and actions(with the supervision, prediction and evaluation tasks).

Table 1Situations successfully identified during the period of validation

Situations Number of Number ofsituations situationsdetected in detected theadvance same day

Filamentous bulking 5 4Foaming 7 10Rising – 2Underloading 8 8Organic overloading 3 1High influent organic concentration – 3High influent nitrogen concentration – 7Hydraulic overloading 3 7Deflocculation problems 4 4Primary settler problems – 5Mechanical or electrical problems – 15Non-biological problems on clarifier – 2Total 30 68

The operation of the EDSS supervisory cycle can besummarised as follows.

2.1.2.1. Data gathering Every time the supervisorycycle is launched, the main task to be performed is datagathering and updating current data for the inferenceprocess. Data gathering is accomplished through on-linedata acquisition systems (sensors and equipment) andoff-line data acquisition systems (biological, chemicaland physical water and sludge analyses and other quali-tative observations of the process). Moreover, this levelof operation implements data filtering, validation andmanagement processes on the temporally evolving (real-time) database where on-line data, off-line data and datacalculated by the system are stored.

According to the manager of the plant, there is a mini-mum set of variables—the basic information—that mustbe updated in order to make a reliable diagnosis of thecurrent state of the process. In the Granollers WWTP,these are the influent flow rate and the chemical oxygendemand of the biological influent. However, the diag-nosis mechanisms cannot rely on these two values, andthe same conclusion can be inferred from different vari-ables, avoiding the interruption of the process becauseof the lack of information.

2.1.2.2. Diagnosis Once the data have been collected,they are sent to the diagnosis module where the knowl-edge-based systems (ES and CBS) are executed concur-rently without any kind of interaction between them.Thus, the current state of the process will be diagnosedthrough a reasoning task based on both the expert rulesand the most similar cases retrieved. If a problem isdetected or suspected, the diagnosis module will also tryto identify the specific cause. The solution to the most


Fig. 6. Supervisory cycle.

similar case is modified so as to adapt it to the new situ-ation.

2.1.2.3. Decision support The conclusions reached inthe diagnosis phase are sent to the decision support mod-ule. This upper module infers the global situation of theWWTP. The final result is sent through the computerinterface to the operator, who will finally decide on theaction to be taken (user-validation and action) (Fig. 7).The expert can use a dynamic mechanistic or a black-box model implemented to support the selection processof an action plan by simulating the possible conse-quences of applying different alternatives (Belanche etal., 1999, 2000).

2.1.2.4. Plans and actions The EDSS suggests anaction plan resulting from the supervision and predictiontasks, and integrating the expert recommendations sentby the RBS and the experience retrieved by the CBS. Incase of conflict, the user acts as the expert of the process,evaluating the suggestions of both systems, checkingtheir validity and deciding which is the best strategy todeal with the situation. The evaluation of the results ofthe application of the action plan to solve the problemallows the system to close the CBS working cycle (seeFig. 8), to learn from successful and failed past experi-ences, and to update the case-library.

These features can be detected by the EDSS itself(unless a manual operation is carried out), but it is essen-

tial that confirmation be provided by the plant manager,who will have the opportunity to change misleadinginformation or add missing information. In addition, theEDSS can extend the knowledge bases by acquiring newknowledge from new sources.

2.2. Selection of wastewater treatment and disposalsystems for communities with less than 2000inhabitants

2.2.1. EDSS building2.2.1.1. Problem analysis WWTPs, and especiallybiological plants, based on several variants of the acti-vated sludge process (suspended growth biomass), arecurrently the predominant system for urban wastewatertreatment and disposal in Catalonia. In accordance withthe goals established in the First Urban WastewaterTreatment Programme of Catalonia (PSARU I), WWTPshave been built to serve every town in Catalonia withmore than 2000 inhabitants. In communities with lessthan 2000 inhabitants, however, the situation is different.Few of them have wastewater treatment systems in placetoday (2001), but all should have them by 2005. Thenumber of communities lower than 2000 inhabitant-equivalents (i.e.) involved in the Small CommunitiesWastewater Treatment Plan of Catalonia (PSARU II) isabout 3500 agglomerations, affecting to more than 800municipalities. It means planning the wastewater treat-ment for 200,000 inhabitants (currently censed),


Fig. 7. User-interface showing a summary of the EDSS for WWTP supervision.

Fig. 8. Case-Based Reasoning System working cycle.

approximately 5% of the Catalan population. The rest ofthe 95% were already considered within the plan alreadydeveloped to treat wastewater from populations higherthan 2000 i.e. The distribution of population lower than2000 i.e. shows that the main part of this population livesin rural communities (73.8% of the agglomerations haslower than 200 i.e.).

Other important remarks of the small communities in

Catalonia concern the relevant contribution of the per-centage of seasonal population (45.1%) with respect tothe permanent population and the proportion of indus-trial wastewater in some villages (globally, 27.2%),especially when it cannot be assimilable to urban waste-water.

While the European Water Directive 91/271/EECspecifies the type of treatment to implement in towns


with more than 2000 inhabitant-equivalents, for smallercommunities this directive only states that the type oftreatment should be appropriate. This significantlychanges the decision process of selecting the optimaltreatment. In this context, ‘appropriate treatment’ isdefined as one that fulfils the quality standards set forthe receiving waters. This suggests new dimensions ofanalysis where the landscape and the affected environ-ment together with the characteristics of the wastewatertreatment technologies for small communities are to betaken into account. Thus, in order to make sound rec-ommendations based on the available technologies andthe characteristics of the receiving environment and thelandscape, it becomes necessary to acquire and integrateexpertise from diverse disciplines. This change of per-spective on the problem with respect to PSARU I sug-gests that we should move towards a paradigm thatallows dealing with complexity.

In view of this complexity, three dimensions of analy-sis must be taken into account during the decision-mak-ing process:

1. The characteristics of the small community itself.This is an aspect of evident importance given thelarge number of rural communities in Spain and thevariety of climatic, geomorphologic and populationdynamics conditions that should be taken into con-sideration when selecting the best option. It is alsoimportant to consider that, unlike larger communitiesand towns, rural communities directly experience theimplementation of the sewage treatment system, withrespect to both perceived benefits and perceivedimpacts on their environment.

2. The receiving environment which should improve sig-nificantly once the Small Communities WastewaterTreatment Plan of Catalonia is implemented. Protec-tion of the receiving environment is of the highestimportance, as endorsed by the recent 2000/60/ECDirective. In order to improve on the current state, anassessment of the current ecological quality of the siteis needed. Significantly, many of the sites where treat-ment systems are to be implemented are in protectedareas or in rural areas with high actual or potentialecological quality that deserves to be preserved orrestored.

3. The wastewater treatment systems appropriate forsmall communities. These differ broadly in terms ofboth technology and operation, and need to be accom-modated to each particular situation. Thus all theadvantages, disadvantages, and any factors that mightaffect the final decision must be taken into consider-ation.

For each particular case, the integration of these threetypes of information—the rural community, the receiv-ing environment, and the type of treatment—will suggest

optimal and multidisciplinary scenarios to support thedecision-making process, since they will have taken intoaccount not just technical aspects of treatment optimis-ation, but also environmental, economical and social fac-tors. Reflecting the will to face the problem in all itscomplexity, the Catalan Water Agency (ACA, “AgenciaCatalana de l’Aigua” ) decided to design an EDSS. Aconsortium formed by four universities (Universitat deGirona, Universitat Politecnica de Catalunya, Universitatde Barcelona and Universitat Autonoma de Barcelona)and the Spanish Scientific Council (CSIC) was com-missioned to develop a system that would attempt toreproduce the reasoning process followed by a group ofexperts facing the highly complex situation at stake. Thegoal of embracing complexity implied that we shouldnot limit ourselves to ‘ formal’ knowledge, but shouldattempt to incorporate ‘non-formal’ knowledge. The lat-ter derives both from the ‘subjective’ reasoning pro-cesses of experts in different disciplines and from theknowledge accumulated by persons or social groupssensitive to the problems and involved in finding sol-utions to them. This allowed us to recognise the multipleviews and interests that are involved in the decision-making process: financial cost, social and environmentalbenefits, technical criteria, and so on.

An EDSS was chosen as the most suitable tool to sup-port the identification of the appropriate wastewatertreatment for small communities because it integratesexpert knowledge and encourages a multidisciplinaryapproach—with respect to the affected land and environ-ment—that incorporates knowledge from affected per-sons and social groups. It is then possible to reach aconsensus among disparate views that approaches anoptimal solution. Furthermore, since an EDSS is a com-puter system, it not only allows the management andanalysis of large volumes of numerical data but also thatof symbolic and, sometimes, uncertain and inexact data.

2.2.1.2. Collecting data and knowledge acquisitionThree different sources of knowledge were pooled

together to build a knowledge base as comprehensive aspossible (Fig. 9). These knowledge sources were:

� Interviews with experts in water management andwastewater treatment, as well as with experts in thequality of the receiving environment.

� Reviews from scientific and technical literature aswell as knowledge drawn from visits to relevantregions where this type of wastewater treatment pro-gramme has already been implemented.

� Analysis of the available historical data for the receiv-ing environment as well as from data on the smallcommunities themselves.

The first source of knowledge we turned to was agroup of experts in the wastewater treatment process.


Fig. 9. Knowledge sources and knowledge bases involved in the knowledge acquisition phase.

The knowledge we were seeking was extracted from aseries of interviews or conversations. Specifically, we setup a series of interviews with experts in wastewatermanagement and environmental experts from the admin-istration, research centres, and engineering consultingfirms. From this series of interviews, we gathered heuris-tic knowledge specific to Catalonia. This knowledge,accrued during years of work and experience in the samefield, is essential for the successful development of theEDSS. This was supplemented with knowledge derivedfrom specialised journals and books, and from notestaken during the field visits. Finally, the analysis of his-torical databases of permanent and temporary streams aswell as of targeted WWTPs (where these existed) anddata from the small communities obtained through adetailed survey distributed to each one of the munici-palities, formed the third source of knowledge.

In order to produce a knowledge base as comprehen-sive and accurate as possible, we organised the knowl-edge acquired from the three sources described aboveinto three distinct type of knowledge:

1. Knowledge for the (quantitative) assessment of thecharacteristics and state of the receiving environment,such as quantity of water in the stream, presence ofaquifers, sensible zones, groundwater nitrate pollutionvulnerability, and protected areas. This knowledge,acquired through conversations with experts from theCatalan Water Agency (ACA), allowed us to deter-mine the minimum treatment level for each case con-sistent with the current state of the receiving environ-ment, such as primary treatment, secondarytreatment—only carbon removal—, secondary withnitrification, secondary with N/D, nutrient removal ornutrient removal plus disinfection.

2. Knowledge for the identification of disposal sites andcharacteristics for each community, such as numberof inhabitants, surface available, climatic, geologicaland hydrological conditions, future prospects, benefits

and impacts of the new WWTP, and other economic,social and environmental aspects. This was obtainedthrough a survey of municipalities conducted by anengineering firm. One caveat of this type of surveyis that the answers given to the questionnaire bymunicipal officers may be subjective, and hencequalitative and vague. It is nonetheless a valuable toolsince it provides information on the territory and theenvironment that can be obtained only from localknowledge. Moreover, the views of local officers onthe selection of treatment often differ from those ofexperts, and should be included in the decision-mak-ing process.

3. Knowledge about the treatment alternatives for smallcommunities, with information about removalefficiencies, space requirements, climatic constraints,geological and hydrographical features, such as alti-tude, slope, presence of aquifers, groundwater nitratepollution vulnerability, investment and operatingcosts, manpower, social aspects and any advantageand disadvantage that must be considered to beimplemented.

2.2.1.3. Model selection Among the several types ofknowledge-based systems, we chose a rule-based system(RBS) because it allowed the best representation of theknowledge needed to select the optimal wastewater treat-ment system, with due consideration to the receivingenvironment and to the characteristics of the rural com-munity. We developed the rule-based expert system intwo main parts. In the first one, the RBS assists in theselection of the treatment level adequate to fulfil the tar-get quality standards for the receiving environment. Inthe second one, the RBS is subsequently used to selectthe specific type of treatment.

2.2.1.4. Model implementation Once the knowledgeacquisition process was completed, we proceeded tostructure the acquired knowledge or, in other words,


transform it into a graphical representation that is easyto understand and amend by experts. For instance, theknowledge about the receiving environment was organ-ised and documented in the form of decision trees as aprior step to developing the part of the rule-based systemfor the selection of the level of wastewater treatment asa function of the receiving environment (Fig. 10).

In addition, the knowledge acquired about treatmentalternatives allowed us to construct two useful matrices.One allows the qualitative comparison of the alternativetreatments based on economic, environmental, techno-logical, and other criteria. The other matrix associatesthe level of treatment established for the receivingenvironments with the optimal treatment system for eachcase. These two matrices formed the basis for a hier-archical discriminant table, which, after many modifi-cations aimed at removing redundancies and contradic-tions, became the core of the rule-based system tosupport the specific type of treatment.

The function of this table is to assess the value of fourkey variables (inhabitant-equivalents, level of treatmentrequired, water flow of the receiving media and availablesurface of the site to build the WWTP) for the selectionof treatment and propose one or more alternatives forwastewater treatment for each of the communities. Inaddition to these four key variables, we organised theremaining considerations for each type of treatment asa series of the so-called safety rules:

� Discarding rules include criteria for discarding aparticular treatment proposed as an alternative.

Fig. 10. Decision tree for the selection of the level of treatment.

� Favouring rules evaluate criteria for favouring cer-tain treatments.

� Disadvantaging rules evaluate criteria that lower thevalue of certain treatments in certain situations.

The integration of the RBS with geographical infor-mation and a numerical model, for economical esti-mations of each alternative, was accomplished with anEDSS built on a hierarchical multilevel architecture(data gathering, diagnosis, decision support, plans andactions levels). In addition, the EDSS developed pro-vides easy connectivity with external applications,including database operations. Finally, a menu-basedinterface provides a simple and transparent way to com-municate with end-users.

2.2.1.5. EDSS validation The execution of a series ofexperiments with preliminary real data collected fromthe receiving media and small communities enabled usto validate the accuracy, correctness, consistency, andusability of the acquired knowledge. When necessary,the knowledge base was confronted against experts andthe rules were refined, adjusted, corrected and/orextended.

Once validated, the EDSS was applied to all the dif-ferent agglomerations comprised in the Small Communi-ties Wastewater Treatment Plan of Catalonia. The 3482small communities were grouped by river basins andprocessed by the EDSS to obtain a proposal with themost suitable treatment for each community. Alterna-tively, each small community can be processed individu-


ally by the EDSS. This option is better if we want toidentify the reasoning path followed by the EDSS. Theresults of the EDSS application to about 3500 real caseswere proved satisfactory because the EDSS proposed theoptimal solution according to the pool of experts. Someof the WWTPs proposed by the EDSS are already underthe building’s project step.

2.2.2. EDSS operationFig. 11 offers a schematic representation of how the

EDSS proceeds to provide the optimal treatment alterna-tive for a particular community (or for all the communi-ties within a particular catchment). For each community,the level of treatment is first established in order to main-tain or improve the current ecological state of the receiv-ing media and then, according to the treatment and com-munity features, a set of possible alternatives oftreatment are proposed (not only one). Each one of thesealternatives setting to the level of treatment can be laterfavoured, disadvantaged or discarded, according to thefeatures of the community and receiving media. Thesteps followed during the EDSS operation may be sum-marised as follows.

2.2.2.1. Data gathering The user introduces the codeof the system or catchment for which wastewater treat-ment alternatives are required. The EDSS then reads the

Fig. 11. EDSS operation.

database that stores the information on the place gath-ered from the municipal survey or from GIS databases.The data contained in the knowledge base are sub-sequently filtered and abstracted. Filtering consists of anumber of operations aimed at discarding erroneous,foreign or missing data. Abstraction transforms quanti-tative variables into qualitative variables.

2.2.2.2. Diagnosis The decision support system acti-vates a set of rules that evaluates the number of inhabi-tant-equivalents, the level of treatment, the abundance ofwater in the environment, and the area of land availablefor treatment facilities. This step concludes with a shortl-ist of alternative treatments. Subsequently, the safetyrules for the treatment alternatives included in the shortl-ist are activated. These rules may invoke other rules orprocedures (subroutines of the expert system) until afinal list of possible treatments is obtained. For eachalternative, this list provides an economic evaluation ofthe investment and operational costs and indicateswhether the alternative has been discarded, favoured ordisadvantaged, and, if so, the reasons why.

2.2.2.3. Decision support For each community, theEDSS provides an economic evaluation of the cost ofconstruction and operation of each of the alternatives asa function of the number of inhabitant-equivalents to be


treated and the type of treatment selected. For eachsolved system (whether it is a community, a set of com-munities for a given catchment, or a set of neighbouringcommunities), a report is produced containing the fol-lowing results:

� Characteristics of the community used in the reason-ing process of the EDSS

� List of the selected treatment alternatives markingwhich have been discarded, favoured or disadvan-taged.

� Environmental technical justification for the selectedtreatments and the reasons for discarding, favouringor disadvantaging it.

� Economic evaluation of each alternative.

Fig. 12 illustrates one of the interfaces (in Catalan) toshow the results of the EDSS to the final users. In thiscase, six possible wastewater treatments are proposed.Some of these alternatives are favoured (√) and someare disadvantaged (!) for different circumstances. Forexample, waste stabilisation ponds (or lagoons) are fav-oured with respect to the other treatments because: (1)this community presents an important contribution of theseasonal population (seasonal population/permanentpopulation = 4.118) and (2), this community belongs toa region with suitable climatic conditions for optimalperformance of ponds. On the other hand, waste stabilis-ation ponds are disadvantaged with respect to the othertreatments because the site where the WWTP will be

Fig. 12. User-interface showing the results of the EDSS proposal for the Malavella Parc community.

constructed is situated at less than 200 m away from thecommunity, which means that, in windy days, odoursfrom ponds could arrive to the population. The systemcan be requested to offer explanations about the con-clusions reached and the deductive processes followed.In this example, none of the possible treatments is dis-carded (X).

2.2.2.4. Plans and actions In order to make a finaldecision on the optimal treatment alternative for a givencommunity, or on the optimal wastewater treatmentalternative for a system composed by more than onecommunity (e.g. to decide implementing separate orcombined treatment systems for a set of communities,or connect them to an already existing or plannedWWTP), a consensual function was developed amongexperts in wastewater treatments, those on the receivingenvironment, the administration, and engineering firms.

This function allows a numerical evaluation accordingto three factors: an energetic or economical factor (i.e.the operating costs), an ecological factor or impact tothe receiving media factor (the ratio of wastewater andreceiving media water flows corrected by the removalefficiency of each treatment) and an environmental fac-tor considering the detrimental impact on the receivingstream (i.e. the length of sewers). Considering this selec-tion function, the EDSS finally proposes a set of alterna-tives with a hierarchical order from the optimum to theless recommended option. The decision-maker shouldfinally decide among these alternatives.


3. Discussion and conclusions

Environmental problems are complex in the ecologi-cal domain, and usually controversial in the socio-econ-omic domain. The optimal solution to those problemsmay be more easily found by tight cooperation amongscientists from several research fields and decision-mak-ers. EDSSs are increasingly used as a basis for betterdecision-making in many real applications. In this paper,a methodology and a possible architecture for EDSSdevelopment has been proposed. Also, the description oftwo real case studies within the water domain has beendetailed to ensure the reliability of the approach.

From our experience in the development of EDSS dur-ing last 10 years, it can be foreseen that the future ofEDSS research will be focused on the following issues.

3.1. Integration of several sources of data andknowledge

Integration of various sources of knowledge, intelli-gent techniques and numerical tools is the key step todevelop successful EDSS for environmental problems.Intelligent decision-making requires, either implicitly orexplicitly, a model of the world that embodies both priorknowledge and measured data. At the level of data andbackground information, numerous and often incompat-ible bits of information from disparate sources have tobe brought together. At the level of tools, there are sev-eral levels of integration, ranging from simple file trans-fer between different methods and programs to fullyintegrated systems. Typical examples of differentmethods that lend themselves to integration include geo-graphical information systems and models as well asrule-based systems, models and databases, algorithmicmodels and intelligent reasoning systems, simulation andoptimisation models.

3.2. Improvement of knowledge acquisition methods

EDSSs use different knowledge sources and this usu-ally implies different ways to represent, extract and com-bine information. The nature of the problems that EDSSstry to solve makes the knowledge acquisition step a cru-cial one. For most of the problems, there exist hugequantities of data on the process itself, but the infor-mation on the causal or dependence relations amongvariables is not well known. In many cases, AI tools areused to discover those relationships. Future work in thisfield should also include the integration of knowledge(about the problem domain) in the data mining task toincrease meaningful knowledge extraction (supervisedknowledge acquisition).

A possible solution to integrate and share informationabout knowledge structures is to build and useontologies. This task is only starting to be generally

recognised as a key issue in environmental fields. There-fore, this is the appropriate moment to define the relevantentities. Ontologies could be used to assess and evaluatethe knowledge about a certain topic or situation with thegoal of informing decision-makers. Ontologies can giveanswers to some of the following questions: What isknown and with what degree of certainty? What is notknown? What is the relevance of that knowledge todecision-makers? Construction of specific ontologies orequivalent paradigms could represent better the know-how and know-what in environmental systems.

3.3. Elaborate protocols to facilitate sharing andreuse of knowledge

Once an EDSS has acquired information on a complexenvironmental process, what are the available ways toshare that information with other systems? If EDSSs aredesigned to be cooperative, under which conditions doesthis cooperation occur? What happens if cooperationfails? Who will assess the quality of the exchanged infor-mation? Who will harmonise indicators and exchangeprotocols?

Solutions for sharing knowledge in environmentalprocesses are far from being fully developed, but onehas to consider the great variety of data, and the strongdependencies of environmental processes to local con-straints, such as weather conditions, climatic aspects,geographical positions, environmental or health lawregulations, etc. If specific models are to be developedfor environmental problems, greater generality, precision(when possible) and realism will be required.

3.4. Involvement of end-users in EDSS development

In general, the role of the user in EDSS developmentis still poorly defined. These systems are developed tosupport users’ decision-making activities in highly com-plex problems.

The following questions are still to be answered: towhat extent can an EDSS be modified directly by anyuser? Who should decide that an EDSS has to start alearning process? Who has to validate the results of suchprocess? Why should an EDSS start a learning process?Who is legally responsible for the decisions made byan EDSS?

We propose, as a first approach, the creation of userprofiles with different privileges and responsibilities inthe interaction with the EDSS. This will lead to thedefinition of different levels of interaction between theuser and the EDSS.

On the other hand, the users must be involved in thewhole process of EDSS design and development toensure the usability of the final system. The degree towhich the users become involved in EDSS development


will determine their level of confidence in the final sys-tem. In the worst case, the system might remain unused.

3.5. Development of benchmarks for the validation ofEDSS

In our opinion, one of the most promising researchlines in EDSS development is the definition of bench-marks to assess and evaluate the performance of EDSSsin a set of well-defined circumstances, and their capacityto react to new situations. This will also allow the cre-ation of a better framework for comparison betweenEDSSs.

We are aware of no attempt to do this. This validationof an EDSS in the appropriate context may simplify thetuning tasks and help to enhance the system’s perform-ance.

3.6. Final conclusion

Environmental issues belong to a set of criticaldomains where wrong management decisions may havedisastrous social, economic and ecological conse-quences. Decision-making performed by EDSSs shouldbe collaborative, not adversarial, and decision-makersmust inform and involve those who must live with thedecisions. What an EDSS contributes is not only anefficient mechanism to find an optimal or suboptimal sol-ution, given any set of whimsical preferences, but alsoa mechanism to make the entire process more open andtransparent. In this context, EDSSs can play a key rolein the interaction of humans and ecosystems, as they aretools designed to cope with the multidisciplinary natureand high complexity of environmental problems.

Acknowledgements

We wish to thank Josep Bou and Marta Lacambra(Agencia Catalana de l’Aigua) and Angel Freixo andJosep Arraez (Consorci per a la Defensa de la Conca delriu Besos) for their confidence in the possibilities ofEDSS. We would also like to acknowledge the IST-2000-28385 Agentcities.NET and the IST-1999-10176A-TEAM projects. We also benefited from the partialsupport of CICyT TIC2000-1011 and CICyT DPI2000-0665-C02-01 grants. Finally, we express our gratitude tothe anonymous reviewers for providing helpful com-ments and suggestions that clarify the contribution ofour manuscript.

References

Belanche, L., Valdes, J.J., Comas, J., R-Roda, I., Poch, M., 1999.Towards a model of input–output behaviour of wastewater treat-ment plants using soft computing techniques. Environmental Mod-elling and Software 14 (5), 409–419.

Belanche, L.I, Valdes, J.J., Comas, J., R-Roda, I., Poch, M., 2000.Prediction of the bulking phenomenon in wastewater treatmentplants. Artificial Intelligence in Engineering 14 (4), 307–317.

Comas, J., Rodrıguez-Roda, I., Sanchez-Marre, M., Cortes, U., Freixo,A., Arraez, J., Poch, M., 2003. A knowldge-based approach to thedeflocculation problem: integrating on-line, off-line, and heuristicinformation. Water Research 37 (10), 2377–2387.

Cortes, U., Sanchez-Marre, M., Cecaronni, L., R.-Roda, I., Poch, M.,2000. Artificial intelligence and environmental decision supportsystems. Applied Intelligence 13 (1), 77–91.

Cortes, U., Sanchez-Marre, M., Sanguesa, R., Comas, J., R-Roda, I.,Poch, M., Riano, D., 2001. Knowledge management in environ-mental decision support systems. AI Communications 14 (1), 3–12.

Cortes, U., Sanchez-Marre, M., Comas, J., Rodrıguez-Roda, I., Poch,M., 2002. Unpublished results.

El-Swaify, S.A., Yakowitz, D.S. (Eds.), 1998. Multiple ObjectiveDecision Making for Land, Water, and Environmental Manage-ment. Lewis Publishers.

Fox, J., Das, S., 2000. Safe and sound. In Artificial Intelligence inHazardous Applications. AAAI Press/The MIT Press.

Funtowicz, S.O., Ravetz, J.R., 1993. Science for the post-normal age.Futures 25 (7), 739–755.

Funtowicz, S.O., Ravetz, J.R., 1999. Post-normal science—an insightnom maturing. Futures 31 (7), 641–646.

Guariso, G., Werthner, H., 1989. Environmental Decision Support Sys-tems. Ellis Horwood-Wiley, New York.

Haagsma, I.G., Johanns, R.D., 1994. Decision support systems: an inte-grated approach. In: Zannetti, P. (Ed.), Environmental Systems,vol. II., pp. 205–212.

Kinzig, A., 2001. Bridging disciplinary divides to address environmen-tal and intellectual challenges. Ecosystems 4, 709–715.

Ludwig, D., 2001. The era of management is over. Ecosystems 4,758–764.

Olsson, G., Aspegren, H., Nielsen, M.K., 1998. Operation and controlof wastewater treatment—a Scandinavian perspective over 20years. Water Science and Technology 37 (12), 1–13.

Ostrom, E., 1991. Governing the Commons: The Evolution of Insti-tutions for Collective Action. Cambridge University Press.

Rizzoli, A.E., Young, W.Y., 1997. Delivering environmental decisionsupport systems: software tools and techniques. EnvironmentalModelling and Software 12 (2-3), 237–249.

R.-Roda, I., Comas, J., Poch, M., Sanchez-Marre, M., Cortes, U., 2001.Automatic knowledge acquisition from complex processes for thedevelopment of knowledge-based systems. Industrial EngineeringChemistry Research 40, 3353–3360.

Rodrıguez-Roda, I., Comas, J., Colprim, J., Poch, M., Sanchez-Marre,M., Cortes, U., Baeza, J., Lafuente, J., 2002. A hybrid supervisorysystem to support wastewater treatment plant operation: implemen-tation and validation. Water Science and Technology 45 (4-5),289–297.

Sanchez-Marre, U., Cortes, J., Lafuente, I., R.-Roda, I., Poch, M.,1996. DAI-DEPUR: a distributed architecture for wastewater treat-ment plants supervision. Artificial Intelligence in Engineering 10(3), 275–285.

Sydow, A., Rose, H., Rufeger, W., 1998. Sustainable development andintegrated assessment. Ercim News (32), 34.

Designing and building real environmental decision support ...ia/articulos/MPochetal-EMS.pdf · a...

Documents

Transcript of Designing and building real environmental decision support ...ia/articulos/MPochetal-EMS.pdf · a...