paperceus

8/9/2019 paperceus

1/24

A method to analyse neighbourhoodcharacteristics of land use patternsPeter H. Verburg a,b, * , Ton C.M. de Nijs c,

Jan Ritsema van Eck a,d , Hans Visser c, Kor de Jong aa Faculty of Geographical Sciences, Utrecht University, P.O. Box 80115, 3508 TC,

Utrecht, The Netherlandsb Department of Environmental Sciences, Wageningen University, P.O. Box 37, 6700 AA,

Wageningen, The Netherlandsc National Institute of Public Health and the Environment, P.O. Box 1, 3720 BA,

Bilthoven, The Netherlandsd Netherlands Institute of Spatial Research, P.O. Box 30314, 2500 GH, The Hague, The Netherlands

Accepted 11 July 2003

Abstract

Neighbourhood interactions between land use types are often included in the spatiallyexplicit analysis of land use change. Especially in the context of urban growth, neighbourhoodinteractions are often addressed both in theories for urban development and in dynamicmodels of (urban) land use change. Neighbourhood interactions are one of the main drivingfactors in a large group of land use change models based on cellular automata (CA).

This paper introduces a method to analyse the neighbourhood characteristics of land use.For every location in a rectangular grid the enrichment of the neighbourhood by specic landuse types is studied. An application of the method for the Netherlands indicates that differentland use types have clearly distinct neighbourhood characteristics. Land use conversions can

be explained, for a large part, by the occurrence of land uses in the neighbourhood.The neighbourhood characterization introduced in this paper can help to further unravel

the processes of land use change allocation and assist in the denition of transition rules forcellular automata and other land use change models. 2003 Elsevier Ltd. All rights reserved.

Keywords: Land use change; Cellular automata; Modelling; Neighbourhood interaction; Urbanization;The Netherlands

* Corresponding author. Address: Department of Environmental Sciences, Wageningen University,P.O. Box 37, 6700 AA, Wageningen, The Netherlands. Tel.: +31-317-485208; fax: +31-317-482419.

E-mail address: [email protected] (P.H. Verburg).

0198-9715/$ - see front matter 2003 Elsevier Ltd. All rights reserved.doi:10.1016/j.compenvurbsys.2003.07.001

Computers, Environment and Urban Systems28 (2004) 667690

www.elsevier.com/locate/compenvurbsys
http://mail%20to:%[email protected]/http://mail%20to:%[email protected]/

8/9/2019 paperceus

2/24

1. Introduction

Changes in land cover and land use are among the most important human in-

duced changes that have impact on the functioning of the earth system (Lambinet al., 2001; Turner II et al., 1990). Apart from impacts on biodiversity, climatechange and global warming (Tyson, Steffen, Mitra, Fu, & Lebel, 2001), land coverand land use change can also, indirectly, inuence the vulnerability of places andpeople to climatic, economic or socio-political perturbations (Kasperson, Kasper-son, & Turner, 1995; Kasperson & Kasperson, 2001). For the consequences at localand regional levels, the spatial patterns of land use change are therefore as relevantas the aggregate volume of change.

Researchers from different scientic disciplines have recently addressed land usechange issues to better understand the causes and consequences of land use changeand explore the extent and location of future land use changes. A unifying hypo-thesis that links researchers from different disciplines is that humans respond to cuesfrom both the physical environment and the sociocultural context. Land use changeis therefore often seen as a function of socio-economic and biophysical factors thatare referred to as driving factors of land use change (Turner II, Ross, & Skole,1993). Driving factors that inuence the magnitude and extent of land use change areoften related to the functioning of local and national markets, policy and demo-graphic conditions.

One of the factors that are often included in spatially explicit analysis of landuse change is the interaction between neighbouring land use types. Especially inthe context of urban growth, neighbourhood interactions are often addressedbased on the notion that urban development can be conceived as a self-organizingsystem in which natural constraints and institutional controls (land-use policies)temper the way in which local decision-making processes produce macroscopicurban form. Different processes can explain the importance of neighbourhoodinteractions. Simple mechanisms for economic interaction between locations areprovided by the central place theory (Christaller, 1933) that describes the uniformpattern of towns and cities in space as a function of the distance that consumers inthe surrounding region travel to the nearest facilities. Spatial interaction betweenthe location of facilities, residential areas and industries has been given more at-tention in the work of Krugman (Fujita, Krugman, & Mori, 1999; Krugman,1999). The spatial interactions are explained by a number of factors that eithercause concentration of urban functions (centripetal forces: economies of scale,localized knowledge spill-overs, thick labour markets) and others that lead to aspatial spread of urban functions (centrifugal forces: congestion, land rents, factorimmobility etc.).

This paper introduces an empirical method to analyse neighbourhood interac-tions. This method should be helpful to test and validate our hypotheses concerningthe neighbourhood interactions. Applied to changes in land use patterns it gives

insight in the neighbourhood interactions over specic period of time, which mayassist modellers in the implementation and quantication of neighbourhood inter-actions in land use models.

668 P.H. Verburg et al. / Comput., Environ. and Urban Systems 28 (2004) 667690

8/9/2019 paperceus

3/24

Neighbourhood interactions are an important component of many land usemodels. The most common method to implement neighbourhood interactions inland use change models are cellular automata. Cellular automata were originally

conceived by Ulam and Von Neumann in the 1940s to provide a formal frameworkfor investigating the behaviour of complex, extended systems (von Neumann, 1966).In land use models cellular automata typically model the transition of a cell from oneland use to another depending on the land use of cells within the neighbourhood of the cell. Cellular automata are used in almost all land use change models for urbanenvironments (Candau, 2000; Jenerette & Wu, 2001; Sui & Zeng, 2001; Torrens &OSullivan, 2001; Ward, Murray, & Phinn, 2000; White, Engelen, & Uijee, 1997; Wu,1998). Besides urbanization, CA-based models now also simulate other processes of land use change, e.g., Messina and Walsh (2001) study land use and land coverdynamics in the Ecuadorian Amazon, an area where tropical forest is converted intoagricultural land. Applications of CA for land use change modelling in which bothurban and rural land uses are considered are provided by Engelen, White, Uljee, andDrazan (1995) and White and Engelen (2000).

The denition of the transition rules of a CA model is the most essential part toobtain realistic simulations of land use and land cover change. Land use change isthe result of a complicated decision-making process; however, the transition rules of CA models are often dened on an ad hoc basis. Methods to derive the transitionrules are lacking. In a recent editorial on research priorities for CA and urbansimulation Torrens and OSullivan (2001) argue that urban CA models are nowmostly technology driven instead of really informing theories through the explora-tion of hypothetical ideas about urban dynamics.

Recently, different approaches have evolved to better match the transition rule setwith reality. Sui and Zeng (2001) use historic conversions of land use to deriveempirical evidence for the importance of the different factors and use multiple re-gression techniques to quantify the weights of the different factors within the tran-sition rules. Other authors use advanced calibration methods for the model as awhole to ne-tune the coefficients of the transition rules based on a number of pattern and quantity measures (Clarke, Hoppen, & Gaydos, 1996; Messina & Walsh,2001; Silva & Clarke, 2002; Straatman, Engelen, & White, in press).

Calibration of CA transition rules is complex due to the many interacting coef-cients that do not necessarily yield unique solutions: different processes (rule sets)may lead to identical patterns. Calibration, therefore, does not always lead to newunderstandings of the relative importance of the different coefficients and is inap-propriate for testing hypothesis concerning the underlying factors of urban devel-opment. The same argument holds for other methods that calibrate the transitionrule set without explicating the relations used. Li and Yeh (2001, 2002) propose amethod that overcomes the denition problem of the transition rules of a CA modelby training articial neural networks. However, neural networks do not give insightin the relations actually used in modelling, leaving the user uninformed about the

possible lack of causality in the relations that are used in the model. Also the methodof Yang and Billings (2000a, 2000b) that solves this inverse problem of cellularautomata based on genetic algorithms has a number of drawbacks. This method is,

P.H. Verburg et al. / Comput., Environ. and Urban Systems 28 (2004) 667690 669

8/9/2019 paperceus

4/24

at present, only operational for simple, binary patterns. Land use patterns withmultiple different land use types are much more difficult to unravel.

The main drawback of all these calibration techniques forms the huge set of

parameters to be calibrated and consequently, the large amount of computing time.A good initial set of transition rules would be of great help to get these procedures ontheir way.

The importance of neighbourhood interactions in land use modelling and thedrawbacks of calibration and automatic procedures to derive transition rules forneighbourhood relations in land use change models call for new approaches to ex-plore neighbourhood relations in land use change. In addition to the calibrationmethods mentioned above, we aim to provide a method to explore the neighbour-hood relations empirically based on land use datasets. Such an exploration of theneighbourhood relations is a rst step to narrow down the solution space for im-plementing neighbourhood relations in land use change models.

The primary objective of this paper is to introduce a method for the analysis of neighbourhood interactions based upon an empirical analysis of changes in land usepattern. This method should be helpful to test and validate our hypotheses con-cerning the neighbourhood interactions. Furthermore, the method aims to assistmodellers in the implementation and quantication of neighbourhood interactions inland use models in general and the Environment Explorer (White & Engelen, 2000)specically.

The paper rst describes the method, applies it to land use data of the Netherlandsfor 1989 and explores the regional variation and scale dependency. Next the resultsof changes in land use between 1989 and 1996 will be discussed. In the last section of this paper we will discuss the possible use of the method for land use changemodellers.

2. Methods

2.1. Characterization of neighbourhood characteristics

Neighbourhood characteristics are calculated when a situation requires theanalysis of relationships between locations, rather than interpret the characteristicsat individual locations. In raster-based geographic analysis, neighbourhood opera-tions are used to compute a new value for every location as a function of itsneighbourhood. A neighbourhood is any set of one or more locations that bear aspecied distance and/or directional relationship to a particular location, theneighbourhood focus (Tomlin, 1990, p. 96). The operations that are used to calculateneighbourhood characteristics are called convolution, spatial ltering, or focalfunctions (Bonham-Carter, 1994; Burrough & McDonell, 1998). Various statisticscan be used to characterise the neighbourhood of a location.

In order to provide a generic method we will not limit ourselves to specic sizes of the neighbourhood. In many studies the size of the neighbourhood is chosen arbi-trarily and only the direct neighbourhood of a location is taken into account (e.g.


8/9/2019 paperceus

5/24

Von Neumann or Moore neighbourhoods). Others have argued that human activi-ties are inuenced by a wider space making a exible denition of the neighbourhoodessential (White & Engelen, 2000). We will follow this approach.

To characterize the neighbourhood of a location in a land use map we have de-ned a measure that is based on the over- or under representation of different landuse types in the neighbourhood of a location. This measure, the enrichment factor( F ), is dened by the occurrence of a land use type in the neighbourhood of a lo-cation relative to the occurrence of this land use type in the study area as a wholefollowing:

F i;k ;d nk ;d ;i=nd ;i N k = N

1

F i;k ;d characterizes the enrichment of neighbourhood d of location i with land usetype k . The shape of the neighbourhood and the distance of the neighbourhood fromthe central grid-cell i is identied by d . Fig. 1 shows the shape of the neighbourhoodsused in this study. nk ;d ;i is the number of cells of land use type k in the neighbourhoodd of cell i, nd ;i the total number of cells in the neighbourhood while N k is the numberof cells with land use type k in the whole raster and N all cells in the raster. So, if theneighbourhood of a certain grid cell contains 50% grass whereas the proportion grassin the country as a whole is 25% we characterize the neighbourhood by an enrich-ment factor 2 for grassland. When the proportion of a land use type in the neigh-bourhood equals the national average, the neighbourhood is characterized by afactor 1 for that land use type. An under representation of a certain land use type inthe neighbourhood results in an enrichment factor between 0 and 1.

7x7 neighborhood (d=3)




Fig. 1. Conguration of neighbourhoods used in this study.


8/9/2019 paperceus

6/24

This neighbourhood characteristic results for each grid cell i in a series of en-richment factors for the different land use types ( k ). The procedure is repeated fordifferent neighbourhoods located at different distances ( d ) from the grid cell to study

the inuence of distance on the relation between land use types. In this study we haveused square rings at a distance d from the central cell as neighbourhoods (based onthe Moore neighbourhood, see Fig. 1). A C++ program was used to perform allcalculations.

The average neighbourhood characteristic for a particular land use type l ( F l ;k ;d ) iscalculated by taking the average of the enrichment factors for all grid cells belongingto a certain land use type l , following:

F l ;k ;d 1

N Xi2 L F i;k ;d 2where L is the set of all locations with land use type l and N l the total number of grid-cells belonging to this set.

The enrichment factor dened above is comparable to the location quotient that isoften used in economic geography (Smith, 1975). The average factors are symmet-rical for land use in the central cell ( l ) and land use in the neighbourhood ( k ), i.e., thevalue of the average enrichment factor is equal for the enrichment of the neigh-bourhood of land use l with land use k and the enrichment of the neighbourhood of land use k with land use l . Small deviations can occur through edge effects.

Condence intervals for testing the signicance of this characteristic cannot be

calculated following standard procedures. Spatial autocorrelation causes all statis-tical tests to be biased and is therefore inappropriate for determining the signicanceof the calculated characteristics (Anselin, 1988). Therefore, we have characterized thevariability by the standard deviation and studied regional variability in the charac-teristics by comparing aggregated results with region-specic results. The standarddeviation is dened by

S l ;k ;d ffiffiffiffiffiffiffiffiffiffiffiffiffi1 N l 1Xi2 L F i;k ;d F l ;k ;d 2s 3

Regional variability was determined by comparing the average characteristics for thecountry as a whole with the characteristics specically derived for a number of biophysical and administrative regions.

In this study the neighbourhood characteristics for land use were determined forthe land use pattern of the Netherlands in 1989. The characteristics of the locationswhere land use changed between 1989 and 1996 were specically selected and theaverage neighbourhood characteristics of these locations were determined sepa-rately.

2.2. Logistic regression

An analysis of the (average) neighbourhood characteristics alone does notindicate to what extent the spatial pattern of land use can be explained by the


8/9/2019 paperceus

7/24

neighbourhood characteristics. An analysis of the explanatory capacity of the en-richment factors is helpful for determining the relevance of neighbourhood inter-actions for the particular case study. To assess the explanatory capacity of the

enrichment factors we have used them in a logistic regression model to relate thelocation of changed land use with the calculated enrichment factors. The goodness of t of the logistic regression model is evaluated by the ROC (Relative OperatingCharacteristic; Pontius & Schneider, 2001; Swets, 1986). The ROC is based on acurve relating the true-positive proportion and the false-positive proportion for thecomplete range of cut-off values in classifying the probability. The ROC statisticmeasures the area beneath the curve and varies between 0.5 (completely random) and1 (perfect discrimination).

Other probabilistic methods such as those based on Bayes theory and the relatedweights of evidence approach could also have been used for this purpose. Logisticregression is however more suitable because the

enrichment factor has continuousvalues instead of the binary values used as independent variables in the weights of evidence approach (Bonham-Carter, 1994). In logistic regression, the probability of conversion of a grid cell ( P i) is described as a function of a set of enrichment factorsfollowing:

LogP i

1 P i b0 bgrass ;d 1 F i;grass ;d 1 b forest ;d 1 F i;forest ;d 1 b arable ;d 1 F i;arable ;d 2 bk ;d F i;k ;d 4

where the independent variables ( F i;k ;d ) are the enrichment factors of the individualgrid-cells i of the neighbourhood d with land use k ; and bk ;d are the coefficients to beestimated with a maximum likelihood estimation. The number of independentvariables included in the equation can be very large when many interactions atdifferent distances are included in the specication. The selection of interactionsincluded depends on the theoretical considerations from the researcher and thecomplexity envisaged. The resulting probabilities can be compared with the locationsthat actually changed to determine the goodness of t of the regression model. Thegoodness of t is a measure of the amount of spatial variation of land use changethat can be explained by the neighbourhood characteristics. In case large variabilityin neighbourhood characteristics is prevalent the level of explanation of the re-gression equation will be low. High levels of explanation indicate that neighbour-hood interactions should be taken into account in the further analysis or modellingefforts.

2.3. Data

In this study we used the 1989 and 1996 maps of land use, similar to those appliedin the Environment Explorer, produced by the Central Bureau for Statistics (CBS

Bodemstatistiek). Since 1989 these maps are based on aerial photographs (scale1:18,000), distinguishing 33 different land use types. These maps were checked forinconsistencies by the National Institute of Public Health and the Environment,


8/9/2019 paperceus

8/24

rasterized to 25 25 m (Raziei & Evers, 2001) and combined with the LGN database(Landelijk Grondgebruiksbestand Nederland), another series of high resolutionland use maps produced by Wageningen University and Research Center (Alterra

Green World Research) based on the interpretation of Landsat_TM remote sensingimages (de Wit, van der Heijden, & Thunnissen, 1999). The integration of both mapsmade it possible to sub-divide the agricultural land use type (de Nijs, de Niet, deHollander, Filius, & Groen, 2001). For this study the data set was aggregated toa 500 500 m resolution. This aggregation was based on the majority-rule. Such aprocedure could yield a bias, which would especially cause land use types with arelative small coverage to disappear (He, Ventura, & Mladenoff, 2002; Milne &Johnson, 1993; Moody & Woodcock, 1994). Therefore, we constrained the aggre-gation by requiring the total areas of each land use type to correspond in both thehigh and low-resolution maps.

Based on this procedure consistent land use maps at 500 500 m were producedfor 1989 and 1996 and reclassied into ten land use types relevant to the analysispresented in this paper (Fig. 2). Table 1 gives a description of the land use types.

Fig. 2. Land use in the Netherlands in 1989 and 1996 at a resolution of 500 500 m; dark grey shadesindicate greenhouses, recreational area and industry/commercial area; black indicates residential area andairports.

Table 1Land use classication used in this analysis

Land use type DescriptionOther agriculture Agricultural land not belonging to greenhouses, grassland or arable land,

includes horticulture, orchards etc.Grassland Grasslands, incl. semi-natural grasslandsArable land All arable landsGreenhouses GreenhousesResidential areas Residential areas and social-cultural facilities, incl. houses, roads within

residential areas, schools, hospitals, churches etc.Industrial/commercial Mining areas, industries, harbours, shopping malls, prisons and all

service industriesForest/nature Forests and natural areas, incl. peat areas, swamps, heather etc.

Recreation Parks, sport elds, kitchen garden complexes, camp sites etc.Airports AirportsWater Water


8/9/2019 paperceus

9/24

3. Results

3.1. Neighbourhood characteristics of land use conguration in 1989

The characteristics of the neighbourhood of the different land use types in 1989for the smallest neighbourhood ( d 1; 3 3 neighbourhood) are listed in Table 2. If the value for a land use type exceeds 1 this indicates an enrichment of the neigh-bourhood. If the value is between 0 and 1, a less than average occurrence of the landuse type is present in the neighbourhood. The large values at the diagonal of thetable represent the positive spatial autocorrelation of all land use types. Based onthese results a broad classication of the spatial clustering of land use types can bemade. Residential, industrial and recreational areas occur in clusters, whereas forest/nature and recreational areas also show a slightly positive neighbourhood relation.Arable land is often in the neighbourhood of other agricultural land and grasslandonly has a positive neighbourhood relation with other grassland. Greenhouses arefound in the neighbourhood of the residential areas, especially in the denselypopulated western part of the Netherlands. However, the high standard deviations of the characteristics for greenhouses indicate the high variability in the neighbourhoodcharacteristics of this land use type.

Results for larger neighbourhoods are given for a number of land use types in Fig.3. These graphs present the average enrichment factor ( F l ;k ;d ) as a function of thedistance ( d ; see Fig. 1). The enrichment factor is presented at a logarithmic scale toobtain an equal scale for land use types that occur more than average in theneighbourhood (enrichment factor >1) and land use types that occur less thanaverage in the neighbourhood (enrichment factor

8/9/2019 paperceus

10/24

Table 2Enrichment factors ( F l ;k ;d 1) of the land use pattern in 1989 for a 3 3 Moore neighborhood ( d 1); standard devia

Land use type incentral cell ( l )

Land use in neighborhood ( k )

Otheragriculture

Grassland Arableland

Green-houses

Resi-dential

Industrial Forest/nature

Other agriculture 3.7 (2.4) 0.6 (1.3) 1.4 (1.8) 1.7 (1.9) 1.0 (1.5) 0.9 (1.4) 0.4 (1.0) 0.9Grassland 0.6 (0.7) 2.0 (0.8) 0.7 (0.8) 0.4 (0.6) 0.4 (0.6) 0.4 (0.6) 0.4 (0.6) 0.Arable land 1.4 (1.5) 0.7 (1.1) 3.3 (1.8) 0.4 (0.8) 0.4 (0.9) 0.4 (0.9) 0.4 (0.8) 0.Greenhouses 1.7 (12.2) 0.4 (5.4) 0.4 (4.7) 126.5 (8.8) 1.6 (12.8) 1.4 (12.9) 0.2 (3.8) 1.8Residential 1.0 (2.1) 0.5 (1.3) 0.4 (1.2) 1.6 (2.4) 7.9 (4.4) 3.1 (3.9) 0.4 (1.4) 3.Industrial 0.9 (3.2) 0.4 (2.0) 0.4 (2.1) 1.4 (3.2) 3.1 (5.5) 15.3 (12) 0.5 (2.3) 2.Forest/nature 0.4 (1.1) 0.4 (1.1) 0.4 (1.1) 0.2 (0.7) 0.4 (1.2) 0.5 (1.4) 6.5 (3.0) 1.1Recreation 0.9 (2.8) 0.6 (2.2) 0.5 (2.0) 1.8 (4.1) 3.8 (5.4) 2.3 (4.4) 1.1 (3.7) 9.Airports 2.0 (23.1) 0.3 (8.7) 0.8 (15.3) 0.5 (8.0) 0.3 (7.4) 1.2 (20.5) 0.7 (12.6) 0.Water 0.2 (0.5) 0.1 (0.5) 0.1 (0.4) 0.1 (0.5) 0.2 (0.5) 0.5 (0.9) 0.2 (0.7) 0

8/9/2019 paperceus

11/24

compounds on the livelihood. This is represented by the somewhat lower enrichmentof the neighbourhood of residential area by industrial area at a short distance

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1

1.2

0 5 10 15 20

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1

0 5 10 15 20

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

0 5 10 15 20-0.4

-0.2

0

0.2

0.4

0.6

0.8

1

0 5 10 15 20

f

g

d e a,cb

g

d

e

f

ab

c

-0.5

-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

0.3

0.4

0 5 10 15 20

b

dg

ea

f

c

d

g e

a

cb

f

e

d

g

a

fc

b

other agriculture (a)grassland (b)arable land (c)residential (d)

industrial (e)forest/nature (f)recreation (g)

l o g

( e n r i c

h m e n

t f a c

t o r )

l o g

( e n r i c

h m e n

t f a c

t o r )

l o g

( e n r i c

h m e n

t f a c

t o r )

l o g

( e n r i c

h m e n

t f a c

t o r )

l o g

( e n r i c

h m e n

t f a c

t o r )

neighbourhood size (d)





forest/nature recreation

grassland residential

industrial/commercial

Fig. 3. The logarithm of the enrichment factor (log F l ;k ;d ) as a function of the distance of the neigh-bourhood from the central cell (d; see Fig. 1). Each of the ve graphs indicates the neighbourhoodcharacteristics for a specic land use type.


8/9/2019 paperceus

12/24

compared to the enrichment at a larger distance from the residential area. Anotherstriking result, characteristic for the Dutch landscape pattern, is the negative relationbetween residential area and forest/nature at short distance that turns into a slightly

positive relation at larger distances. Historically, arable land and grassland sur-rounded settlements in the central and eastern part of the Netherlands, while, atsome more distance, wild lands were used for fuel wood and extensive grazing(heather elds). These patterns are still clearly visible in large parts of the Nether-lands and explain the neighbourhood characteristics. This pattern may be reinforcedby planning policies in the last decades that aimed to protect open areas, that wereseen as characteristic for Dutch landscapes, by redirecting residential development tomore sheltered areas where it is not visible from a large distance.

3.2. Region-dependent variability in neighbourhood characteristics

Different biophysical conditions and a different settlement history can cause dif-ferences in the spatial pattern of land use in different parts of the study area. Thiscould lead to different characteristics of the neighbourhood composition. Thesedifferences are represented by the standard deviations listed in Table 2. However,these differences in neighbourhood composition are partly caused by differences inthe distribution of land use types over the country. Most forest/nature is found inthe central and eastern part of the Netherlands, whereas the western part of theNetherlands is more heavily urbanized. When regional enrichment factors are cal-culated to compare the neighbourhood characteristics this effect is compensated,because the enrichment factor is based on the regional frequency distribution of landuse types ( N k ) in Eq. (1). The variation in enrichment factor between regionstherefore represents differences in relative neighbourhood composition.

We studied regional variation in neighbourhood characteristics for two cases. Theeffect of differences in biophysical conditions was studied by dividing the Nether-lands into two regions based on the geomorphology. It was expected that thelandscape pattern, and therefore also the neighbourhood characteristics, are verydifferent for the high and low parts of the Netherlands. The country was divided intoa high region, dominated by sandy soils, and a low region with mostly clayey (uvialand marine) and peaty soils. The neighbourhood characteristics for these regionswere calculated independently and compared to characteristics for the country as awhole (Fig. 4). The general composition of the neighbourhood of grassland is similarin both landscapes: a large positive enrichment with grassland in the neighbourhoodand, mostly, negative enrichments for all other land use types. However, a moredetailed analysis of the results also shows large differences in neighbourhood char-acteristics between the two landscapes. In the low lying part of the Netherlands thelandscape is dominated by grassland areas in which villages and cities are developedat the somewhat higher locations (levees) along rivers and creeks. Residential areaand grassland are direct neighbours. Arable land is situated in separate patches in the

polders that have clayey soils suitable for arable land. In the southern and easternpart of the Netherlands the pattern of villages, grassland and arable land is moremixed. This results in a lower enrichment of the neighbourhood with grassland itself


8/9/2019 paperceus

13/24

and a somewhat higher contribution of arable land in the neighbourhood. Maizecultivation and grassland occur on the same elds, often as a cropping rotation.

The variability of neighbourhood characteristics was also addressed by calcu-lating the neighbourhood characteristics for the 12 provinces of the Netherlandsseparately. Fig. 5 gives the results for the different provinces for the enrichmentwith respectively grassland and industrial land in the neighbourhood of residentialarea. The shape of the curves for the different provinces is very similar, indicatingthat, in spite of large differences in settlement history, biophysical and socio-eco-nomic conditions, similar neighbourhood relations exist. The strength of the rela-tion differs by province, indicating differences in the distance over which therelation affects land use. A detailed analysis of the processes underlying these dif-ferences is out of the scope of this paper. One of the provinces included in the

analysis only has a limited industrial/commercial area, mainly concentrated in onecompound, causing a negative value for the relation with residential area at largerdistances.

Fig. 4. Neighbourhood characteristics (log F grass ;k ;d ) for grassland for the Netherlands as a whole (A), forthe higher region (B) and the lower region (C).


8/9/2019 paperceus

14/24

3.3. Scale dependence of the results

The inuence of data resolution on the neighbourhood characteristics was testedby conducting a similar neighbourhood characterization as presented in the pre-

ceding sections on a data set with another resolution. For this purpose we used theoriginal data set of land use in the Netherlands described in the methods section withthe original 25 25 m resolution. Since this is the same data set on which the ag-gregated 500 500 m data were based, results for both data sets can be compared.Fig. 6 presents the resulting neighbourhood characteristics for residential area in1989. The neighbourhood characteristics are different from the 500 500 m resolu-tion results (Fig. 3). At short distances, residential area is only positively related to

-0.5

-0.3

-0.1

0.1

0.3

0.5

0.7

0.9

1.1

0 5 10 15 20-0.5

-0.3

-0.1

0.1

0.3

0.5

0.7

0.9

1.1

0 5 10 15 20

enrichment of grasslandin neigbourhood ofresidential area

enrichment of industrialland in neigbourhood ofresidential area

l o g

( e n r i c

h m e n

t f a c t o r )


l o g

( e n r i c

h m e n

t f a c t o r )


A B

Fig. 5. Region-specic neighbourhood characteristics for residential area for all provinces of the Nether-lands based on the land use pattern in 1989; (A) enrichment with grassland and (B) enrichment with in-dustry/commercial.

-1.2

-0.8

-0.4

0

0.4

0.8

1.2

0 10 20 30 40 50 60 70

other agriculture (a)grassland (b)arable land (c)residential (d)industrial (e)forest/nature (f)recreation (g)

l o g

( e n r i c

h m e n

t f a c

t o r )


residential

a b

c

d

ge

f

Fig. 6. Neighbourhood characteristics (log enrichment factor) as a function of the neighbourhood size(cell size: 25 m) for residential area in 1989 using high resolution data.


8/9/2019 paperceus

15/24

other neighbouring residential land use. However, at a somewhat larger distance,recreational area and industrial area appear frequently in the neighbourhood of residential area. For a neighbourhood size of 20 units ( %500 m) the characteristics of

the high-resolution neighbourhood are comparable with the results for the smallestneighbourhood at the coarse resolution, corresponding with the same distance fromthe central cell. The maximal occurrence of recreational area in the neighbourhood isfound for approximately 20 units while the maximal occurrence for industrial/commercial area is found at 40 units. This latter observation is also found for thecoarse resolution data.

3.4. Neighbourhood characteristics of changes in land use between 1989 and 1996

The results presented above all apply to the neighbourhood of the land uses, asthey existed in 1989. However, for better understanding land use change, it is of interest to investigate the neighbourhood characteristics of the locations where landuse actually changed. The major land use changes in the Netherlands during theperiod 19891996 are the increase in industrial/commercial area, residential area andrecreational area. We have characterized the neighbourhoods of the new locations of these three land use types. Table 3 lists the existing land uses in 1989 at the locationsthat have been developed between 1989 and 1996. This table shows that a large partof the newly developed locations were formerly used for a land use type classied asother agriculture. Almost 4% of the recreational area in 1989 is converted into

residential area in 1996, while new recreational areas are developed on agriculturallands. The use of other agricultural land and grasslands for urban developments isrelated to the occurrence of this land use type in the western part of the Netherlandswhere many new residential areas are developed. The other agriculture land usetype includes horticultural land use, which is historically developed surrounding thecities, because of labour availability, transportation time and costs, but also becauseof the location of the cities on the somewhat higher, more suitable grounds, on thelevees of the rivers and behind the coastal dunes where the soil is especially suitable

Table 3Land use before conversion for locations converted into respectively industrial/commercial area, resi-dential area or recreational area between 1989 and 1996

Land usein 1989

Distribution (%) of converted land overland use types in 1989

% of the total area of a land use type in1989 converted between 1989 and 1996

Industrial/commercial

Residential Recreational Industrial/commercial

Residential Recreational

Otheragriculture

57.7 60.4 38.3 2.4 3.7 1.6

Grassland 21.7 15.0 34.6 0.3 0.3 0.4Arable land 7.0 7.0 9.3 0.2 0.3 0.3

Greenhouses 0.8 1.1 1.0 1.1 2.1 1.2Forest/nature 7.3 3.8 16.8 0.3 0.2 0.7Recreational 5.5 15.0 1.1 3.8


8/9/2019 paperceus

16/24

for the culture of ower-bulbs. Careful analysis of the data indicates that buildinglots are often, unjustly, classied as other agriculture. This is another reason for thehigh percentage of other agricultural soils that have been converted into residential

and/or industrial/commercial area.Fig. 7 presents the neighbourhood characteristics of the newly developed loca-

tions. The characteristics are determined by calculating the average characteristics of the neighbourhood in 1989 of all locations that have changed into a particular landuse type between 1989 and 1996. For all considered land use types we nd that thenew developments are close to already existing occurrences of the same land usetype. New residential area, however, is also located in the neighbourhood of existingrecreational and industrial areas, indicating the outward expansion of existing townsand cities. Urban growth is clearly much more important than new residential de-velopments far from existing residential areas. For residential areas the same ob-

-0.3

-0.1

0.1

0.3

0.5

0.7

0.9

0 5 10 15 20

e

dg

a

b

fc

new industrial/commercial area

-0.3

-0.2

-0.1

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0 5 10 15 20

new recreation area

gd

e

af

b

c

-0.5

-0.3

-0.1

0.1

0.3

0.5

0.7

0 5 10 15 20

d

g e

a

bc

f

new residential area

other agriculture (a)

grassland (b)arable land (c)residential (d)industrial (e)forest/nature (f)recreation (g)

l o g ( e n r i c

h m e n

t f a c

t o r )


l o

g ( e n r i c

h m e n

t f a c

t o r )


l o g

( e n r i c

h m e n t

f a c

t o r )


A

B

C

Fig. 7. Neighbourhood characteristics (log enrichment factor) as a function of the neighbourhood size forlocations converted into industrial/commercial land (A), recreational area (B) and residential area (C)between 1989 and 1996.


8/9/2019 paperceus

17/24

servation holds as for the 1989 land use pattern: although a positively inuence of industrial/commercial compounds is found at all distances, the new residential areasare preferably not located in the direct neighbourhood of these industrial/commer-

cial compounds. The maximum enrichment of the neighbourhood with industrial/commercial area is located at a distance of 4 units (23 km from the new residentiallocation). The maximum enrichment of the neighbourhood with industrial/com-mercial area for new recreational areas is found at a smaller distance, indicating thatthe area between residential and industrial land uses is often used for recreationalpurposes. For all three presented land use types a large similarity between neigh-bourhood characteristics for the newly developed locations and the existing locationsin 1989 is found (Figs. 3 and 7).

3.5. Logistic regression: conversion probabilities based on neighbourhood character-istics

In order to assess the explanatory power of the enrichment factors, a logisticregression model was tted explaining the spatial distribution of newly developedresidential areas between 1989 and 1996 using the enrichment factors of the indi-vidual cells as explanatory variables. The goodness of t of this regression model wasevaluated by the ROC. Locations classied as water, residential or industrial/com-mercial areas in 1989 were excluded from the analysis. Based on the neighbourhoodcharacteristic of the newly developed residential areas (Fig. 7) and our under-standings of the processes leading to locational decisions we decided to include onlythe enrichment of the neighbourhood with residential, industrial/commercial andforest/nature as explanatory variables. In spite of the large neighbourhood effectobserved for recreational area, we did not include the enrichment with recreation asan explanatory factor because the development of recreational areas mostly followsafter new residential areas are established. We included the enrichment for thenearest neighbours ( d 1) and neighbours located at between 1500 and 2100 m fromthe location ( d 3) to represent the inuence of neighbouring land use types. Theneighbourhood relation is most pronounced for the immediate neighbours. For therelation between residential and industrial/commercial area the enrichment factorwas highest for locations between 1500 and 2100 m from the central cell ( d

3). In a

number of trials these neighbourhood sizes turned out to give the highest level of explanation. The enrichment factor for forest/nature in the larger neighbourhood(d 3) did not give a signicant contribution to the regression model and wastherefore excluded. The resulting coefficients of the regression equation are given inTable 4. The model has a good explanatory power as can be seen from the ROCvalue of 0.91. The ROC value can vary between 0.5 (completely random) and 1(perfect t). The value of 0.91 for the spatial distribution of new residential areaindicates that it is possible to predict new residential area locations reasonably wellbased on neighbourhood characteristics. Based on the regression model conversion

probabilities can be calculated for every location that can potentially be convertedinto residential area. Locations with high probabilities are shown in Fig. 8 and canbe visually compared to locations that were actually converted into urban area


8/9/2019 paperceus

18/24

during 19891996. This visualization conrms the high ROC value: most conver-sions did indeed take place at locations with a high conversion probability. However,the gure reveals that many other locations in the neighbourhood of the actualconversions have a high probability as well. This indicates that other factors inu-ence the allocation decisions as well, such as policy considerations, ecological values,soil suitability, tenure status etc.

4. Discussion and conclusions

This paper has introduced a simple method for exploring and quantifying theneighbourhood characteristics of land use. The results can be used to verify

Table 4Coefficients of a logistic regression model explaining the spatial distribution of new residential area by theenrichment of the neighbourhood ( n 166 ; 079)

Variable Estimated coefficient Odds ratio

Constant ) 5.537* F residential ;i;1 0.364* 1.438 F industrial ;i;1 0.023* 1.023 F forest =nature ;i;1 ) 0.201* 0.818 F residential ;i;3 0.090* 1.094 F industrial ;i;3 0.028* 1.029

) 2 Log likelihood 9478v2 2594*ROC 0.91

*Signicant at p < 0:01.

Fig. 8. Locations with high probability for new conversion into residential area in 1989 based on logisticregression using neighbourhood characteristics as independent variable (left) and actual locations withnew development of residential area between 1989 and 1996 (right).


8/9/2019 paperceus

19/24

hypotheses concerning the interaction of land use types and suggest unaccountedneighbourhood relations between land use types. Neighbourhood effects alone are,however, not the only location factor relevant for describing the spatial pattern of

land use. Other factors, such as accessibility, environmental suitability, spatial poli-cies etc. do also inuence the pattern of land use. The neighbourhood characteristicspresented in this paper are, therefore, not suitable for explaining the large scalepatterns in land use. These large-scale patterns are determined by other factors. Inthe Netherlands, remaining nature areas are very much related to the location of poor, sandy soils that were unsuitable for agricultural use. Many cities are located inthe western part of the Netherlands, because of the historic advantage of these lo-cations for industrialisation and urbanisation. At a more detailed level, neighbour-hood interactions explain spatial differences in land use pattern and can indicatewhich locations are suitable for future conversions based on the neighbouring landuse types. In the case study for the Netherlands predictability of land use changebased on neighbourhood characteristics alone turned out to be relatively high,stressing the importance of including neighbourhood interactions within studies of land use change. A more detailed analysis of the different factors underlying theallocation is presented by Verburg, Ritsema van Eck, de Nijs, Dijst, and Schot(2003).

The presented method for quantifying the neighbourhood characteristics of landuse is of special interest for land use change modellers. The method can be used toassist the denition of transition rules related to neighbourhood interactions in thesemodels. These neighbourhood interactions can be incorporated in land use changemodels in different ways. Models based on transition probabilities, such as Markovchain models (Li & Reynolds, 1997; Thornton & Jones, 1998), statistical models(Chomitz & Gray, 1996; Schneider & Pontius, 2001; Serneels & Lambin, 2001;Veldkamp & Fresco, 1997) or models based on other probabilistic methods, such asthe weights of evidence approach (Almeida et al., 2003) can incorporate the en-richment factor as an explanatory variable. Many models that explicitly addressneighbourhood interactions are based on cellular automata (CA). The enrichmentfactor can assist the modeller in formulating the transition rules for cellular auto-mata. It is, however, not possible to directly translate the neighbourhood charac-teristics into transition rules for cellular automata. Not all neighbourhood relationsidentied with the enrichment factor have a causal explanation, and some of theobserved neighbourhood effects are the indirect result of other interactions. Fur-thermore, CA models work with smaller time-steps (usually 1 year) for calculatingland use transitions than is usually possible for the empirical analysis of neigh-bourhood characteristics. These short time-steps result in the emergence of complexpatterns that cannot be deduced analytically back into transition rules (Torrens &OSullivan, 2001). Therefore, the results of this analysis can only be used to derivea set of CA transition rules directly when land use data with a high temporal re-solution, similar to the time-steps of the CA model, are available. The increasing

availability of remote sensing images makes this a realistic option. When land usedata are not available at a high temporal resolution, the results of land use patternexplorations still have value for CA modellers. The empirical results identify the


8/9/2019 paperceus

20/24

most important neighbourhood relations and their spatial extent for a particularcase-study. After interpreting the empirical results a selection of interactions can bemade that is used as a starting point for dening the transition rules in the CA

model. Such a selection should be based on a careful analysis from the researcher of the theoretical considerations that give rise to spatial interactions between land usetypes and the empirical results obtained by the method presented in this paper.Calibration methods for CA models (Clarke et al., 1996; Messina & Walsh, 2001;Straatman et al., in press) can now be used to modify the transition rules such thatmodel results are closer to reality. The empirical analysis can (most likely) narrowdown the solution space for such a calibration considerably. This procedure issummarized in Fig. 9. The use of empirically derived relations provides a betterstarting point for calibration than transition rules based on expert judgement alone.This approach will be tested in the near future.

The method is particularly suitable for CA models that include multiple land usetypes and a exible denition of the neighbourhood, such as the models of Whiteand Engelen (1997, 2000). However, the enrichment factor can also be used insituations where only one land use transition is considered (e.g., the conversion of non-urban to urban) or where a xed size of the neighbourhood is assumed.

Theories explaining

land use structure

Empiricalquantification of

neighbourhoodcharacteristics

Selection ofrelevant interactions

Translation intoCA transition rules

CA modelcalibration

CA transition rules

Fig. 9. Procedure for dening transition rules within a cellular automata (CA) land use change modelbased on an empirical characterization of neighbourhood characteristics.


8/9/2019 paperceus

21/24

Although some of the results from our study for the Netherlands suggest thatstraightforward distance-decay functions for the interaction between land uses suf-ce, this does not mean that it is easy to determine the neighbourhood interactions

based on expert knowledge alone. The results have also indicated that interactionsbetween land uses differ by region, scale and time-period. The introduced methodallows an exploration of the implications of these differences for land use modelling.Most land use change models that include neighbourhood interactions (e.g. Soares-Filho, Cerqueira, & Pennachin, 2002; Ward et al., 2000; Wu, 1999; Yeh & Li, 2002)use uniform transition rules for the whole study area. Our analysis has shown that itis possible that neighbourhood interactions among land use types differ in differentparts of a study area, due to different environmental or social-cultural conditions. If so, it would be advisable to use regional specic transition rules instead of uniformrules for the whole study area. The empirical analysis presented in this paper pro-vides a good means to explore regional differences in neighbourhood interactionsand can help to decide upon the need for region specic transition rules.

Another aspect of transition rules in land use models is the temporal stability of these rules. Changes in society, policy and land use pressure might lead to differentinteractions among land use types. The method presented could be used to comparethe neighbourhoods in different periods to analyse temporal stability. However,high-resolution land use data are most often difficult to obtain for different periods.If available, changes in data gathering and classication system make it difficult toanalyse time series. In the case study we were able to analyse the land use pattern in1989 as well as the changes between 1989 and 1996. The land use pattern in 1989reects land use as it has changed since historic times. This pattern reects how theland use structure has evolved from the interactions between land use and the social-economic and biophysical environment and among land use types itself.

One could argue that changes in the processes underlying the neighbourhoodinteractions would have caused differences between the neighbourhood character-istics of the land use pattern in 1989 and the neighbourhood characteristics of thechanges between 1989 and 1996. The neighbourhood characteristics for all studiedland use types do, however, indicate a large similarity between the two time periodsstudies. Subtle differences do reveal changes in the processes underlying the neigh-bourhood relations, e.g. the increasing distance over which residential and industrialareas are related due to the policy to avoid disturbance of residential areas by in-dustrial activities. Similar observations of temporal stability of the functional rela-tionships were found for studies in which the relation between land use and theunderlying biophysical and demographic conditions were studied for differentperiods (Hoshino, 1996; Veldkamp & Fresco, 1997).

Several remarks must be made concerning the use of the introduced methodology.The resulting neighbourhood characteristics are dependent on the resolution of theinput data. Maps of different spatial resolution exhibit different spatial variabilityand structure. The resulting neighbourhood characteristics are therefore scale

dependent. Land use types that tend to be neighbours at a certain resolution canbe distant at another resolution. When the derived neighbourhood characteristicsare used to assist the parameterisation of land use models with neighbourhood


8/9/2019 paperceus

22/24

interactions (e.g. CA models) care must be taken that the neighbourhood charac-terization is made at the same scale as the transition rules are used for.

The shape of the neighbourhood can also inuence the results obtained. In this

paper we have used square Moore neighbourhoods because of computational ad-vantages. However, especially for larger neighbourhoods, the square shape causesdifferences in the distance between the neighbourhood and the central cell that hasno theoretical validity. It would, therefore, be better to use a more circular neigh-bourhood, such as the circular neighbourhoods used by White and Engelen (2000).Neighbourhoods based on the activity range of the agents of land use change, e.g.the network-based neighbourhoods used in graph-cellular automata (OSullivan,2001), are even more advanced options. Asymmetrical neighbourhoods may berelevant for other interactions, e.g., for interactions between heavy industry andresidential land use the neighbourhood shape can be based on the prevalent winddirection. The next version of the software to calculate the neighbourhood charac-teristics will enable a exible denition of the shape of the neighbourhood.

The simple and straightforward method for analysing neighbourhood interactionsin land use pattern introduced in this paper makes it worthwhile to include the ex-ploration of neighbourhood interactions in assessments of land use change. Theempirical characterization can contribute to the identication of spatial relationsbetween land uses underlying the spatial allocation of land use change. Future re-search should test the approach in a wider range of case-studies and evaluate the useof this method in the denition of transition rules in land use change models.

References

Almeida, C. M., Batty, M., Monteiro, A. M. V., C^

amara, G., Soares-Filho, B. S., Cerqueira, G. C., &Pennachin, C. L. (2003). Stochastic cellular automata modelling of urban land use dynamics: empiricaldevelopment and estimation. Computers, Environment and Urban Systems, 27 (5), 481509.

Anselin, L. (1988). Spatial Econometrics: Methods and Models . Boston: Kluwer.Bonham-Carter, G. F. (1994). Geographic Information Systems for Geoscientiests: Modelling With GIS .

Oxford: Pergamon.Burrough, P. A., & McDonell, R. A. (1998). Principles of Geographical Information Systems . New York:

Oxford University Press.Candau, J. (2000). Calibrating a cellular automaton model of urban growth in a timely manner. In B. O.

Parks, K. M. Clarke, & M. P. Crane (Eds.), Proceedings of the 4th International Conference onIntegrating Geographic Information Systems and Environmental Modeling: Problems, Prospects, and Needs for Research , September 28, University of Colorado, Boulder.

Chomitz, K. M., & Gray, D. A. (1996). Roads, land use, and deforestation: a spatial model applied toBelize. The World Bank Economic Review, 10 , 487512.

Christaller, W. (1933). Central Places of Southern Germany (Edition 1966). London: Prentice Hall.Clarke, K. C., Hoppen, S., & Gaydos, L. J. (1996). Methods and Techniques for Rigorous Calibration of a

Cellular Automaton Model of Urban Growth. In Third International Conference/Workshop Integrating GIS and Environmental Modeling , Santa Fe, NM, January 2125, 1996.

de Nijs, T., de Niet, R., de Hollander, G., Filius, F., & Groen, J. (2001). De LeefOmgevingsVerkenner:Kaartbeelden van 2030. Een verkenning van de inzet bij beleidsondersteuning (in dutch) . RIVM rapport408505 004, National Institute of Public Health and the Environment. Bilthoven.


8/9/2019 paperceus

23/24

de Wit, A. J. W., van der Heijden, Th. G. C., Thunnissen, H. A. M., (1999). Vervaardiging ennauwkeurigheid van het LGN3-grondgebruiksbestand (in dutch) . Report 633. DLO-Staring Centrum.Wageningen.

Engelen, G., White, R., Uljee, I., & Drazan, P. (1995). Using cellular automata for integrated modelling of socio-environmental system. Environmental Monitoring and Assessment, 34 , 203214.

Fujita, M., Krugman, P., & Mori, T. (1999). On te evolution of hierarchical urban systems. EuropeanEconomic Review, 43 , 209251.

He, H. S., Ventura, S. J., & Mladenoff, D. J. (2002). Effects of spatial aggregation approaches on classiedsatellite imagery. International Journal of Geographical Information Science, 16 , 93109.

Hoshino, S. (1996). Statistical analysis of land-use change and driving forces in the Kansai District, Japan .WP-96-120. 1996. Laxenburg, IIASA. IIASA working papers.

Jenerette, G. D., & Wu, J. (2001). Analysis and simulation of land-use change in the central Arizona Phoenix region, USA. Landscape Ecology, 16 , 611626.

Kasperson, J. X., & Kasperson, R. E. (2001). International Workshop on Vulnerability and Global Environmental Change . 1719 May 2001, Stockholm Environment Institute (SEI), Stockholm, Sweden.A Workshop Summary prepared on behalf of workshop participants. Report 2001-01. Stockholm,Stockholm Environment Institute. SEI Risk and Vulnerability Programme.

Kasperson, J. X., Kasperson, R. E., & Turner, B. L., II (1995). Regions at Risk: Comparisons of Threatened Environments . Tokyo: United Nations University.

Krugman, P. (1999). The role of geography in development. International Regional Science Review, 22 ,142161.

Lambin, E. F., Turner, B. L., II, Geist, H. J., Agbola, S. B., Angelsen, A., Bruce, J. W., Coomes, O.,Dirzo, R., Fischer, G., Folke, C., George, P. S., Homewood, K., Imbernon, J., Leemans, R., Li, X. B.,Moran, E. F., Mortimore, M., Ramakrishnan, P. S., Richards, J. F., Skanes, H., Stone, G. D., Svedin,U., Veldkamp, A., Vogel, C., & Xu, J. C. (2001). The causes of land-use and land-cover change:moving beyond the myths. Global Environmental Change, 4 , 261269.

Li, H., & Reynolds, J. F. (1997). Modeling effects of spatial pattern, drought, and grazing on rates of

rangeland degradation: a combined Markov and cellular automaton approach. In D. A. Quattrochi, &M. F. Goodchild (Eds.), Scale in Remote Sensing and GIS (pp. 211230). New York: Lewis Publishers.Li, X., & Yeh, A. G. (2001). Calibration of cellular automata by using neural networks for the simulation

of complex urban systems. Environment and Planning A, 33 , 14451462.Li, X., & Yeh, A. G. (2002). Neural-network-based cellular automata for simulating multiple land use

changes using GIS. International Journal of Geographical Information Science, 16 , 323343.Messina, J. P., & Walsh, S. J. (2001). 2.5D morphogenesis: modeling landuse and landcover dynamics in

the Ecuadorian Amazon. Plant Ecology, 156 , 7588.Milne, B. T., & Johnson, A. R. (1993). Renormalization relations for scale transformation in ecology.

Lectures on Mathematics in the Life Sciences, 23 , 109128.Moody, A., & Woodcock, C. E. (1994). Scale-dependent errors in the estimation of land-cover

proportions: implications for global land-cover data sets. Photogrammetric Engineering and Remote

Sensing, 60 , 585594.OSullivan, D. (2001). Graph-cellular automata:a generalised discrete urban and regional model.

Environment and Planning B, 28 , 687705.Pontius, R. G., & Schneider, L. C. (2001). Land-use change model validation by an ROC method for the

Ipswich watershed, Massachusetts, USA. Agriculture, Ecosystems and Environment, 85 , 239248.Raziei, A., & Evers, W. (2001). Afstemming ruimtelijke gegevens voor de L(eef)O(mgevings)V(erkenner).

en R(uimte)S(canner) (in dutch) . Intern CIM-report M009/01. National Institute of Public Healthand the Environment. Bilthoven.

Schneider, L. C., & Pontius, R. G. (2001). Modeling land-use change in the Ipswich Watershed,Massachusetts, USA. Agriculture, Ecosystems and Environment, 85 , 8394.

Serneels, S., & Lambin, E. F. (2001). Proximate causes of land use change in Narok district Kenya: aspatial statistical model. Agriculture, Ecosystems and Environment, 85 , 6581.

Silva, E. A., & Clarke, K. C. (2002). Calibration of the SLEUTH urban growth model for Lisbon andPorto Portugal. Computers, Environment and Urban Systems, 26 , 525552.


8/9/2019 paperceus

24/24

Smith, D. M. (1975). Patterns in Human Geography . Newton Abbot: David and Charles.Soares-Filho, B. S., Cerqueira, G. C., & Pennachin, C. L. (2002). DINAMICAa stochastic cellular

automata model designed to simulate the landscape dynamics in an Amazonian colonization frontier.Ecological Modelling, 154 , 217235.

Straatman, B., Engelen, G., & White, R. (in press). Towards and Automatic Calibration Procedure forConstrained Cellular Automata. Computers, Environment and Urban Systems . doi:10.1016/S0198-9715(02)00068-6 .

Sui, D. Z., & Zeng, H. (2001). Modeling the dynamics of landscape structure in Asias emerging desakotaregions: a case study in Shenzhen. Landscape and Urban Planning, 53 , 3752.

Swets, J. A. (1986). Measuring the accuracy of diagnostic systems. Science (June), 12851293.Thornton, P. K., & Jones, P. G. (1998). A conceptual approach to dynamic agricultural land-use

modelling. Agricultural Systems, 57 , 505521.Tomlin, C. D. (1990). Geographic Information Systems and Cartographic Modeling . Englewood Cliffs, NJ:

Prentice Hall.Torrens, P. M., & OSullivan, D. (2001). Editorial. Cellular automata and urban simulation: where do we

go from here? Environment and Planning B, 28 , 163168.Turner II, B.L., Ross, R.H., & Skole, D.L. (1993). Relating land use and global land cover change . IGDP

report no. 24; HDP report no. 5.Turner, B. L., II, Kasperson, R. E., Meyer, W. B., Dow, K. M., Golding, D., Kasperson, J. X., Mitchel, R.

C., & Ratick, S. J. (1990). Two types of global environmental change; denitional and spatial-scaleissues in their human dimensions. Global Environmental Change, 1 , 1422.

Tyson, P., Steffen, W., Mitra, A. P., Fu, C., & Lebel, L. (2001). The earth system: regional-global linkages.Regional Environmental Change, 2 , 128140.

Veldkamp, A., & Fresco, L. O. (1997). Reconstructing land use drivers and their spatial scale dependencefor Costa Rica. Agricultural Systems, 55 , 1943.

Verburg, P. H., Ritsema van Eck, J. R., de Nijs, T., Dijst, M. J. & Schot, P. P. (2003). Determinants of land use change patterns in the Netherlands, Environment and Planning B , in press.

von Neumann, J. (1966). Theory of Self-Reproducing Automata . Illinois: University of Ilinois Press.Ward, D. P., Murray, A. T., & Phinn, S. R. (2000). A stochastically constrained cellular model of urbangrowth. Computers, Environment and Urban Systems, 24 , 539558.

White, R., & Engelen, G. (2000). High-resolution integrated modelling of the spatial dynamics of urbanand regional systems. Computers, Environment and Urban Systems, 24 , 383400.

White, R., Engelen, G., & Uijee, I. (1997). The use of constrained cellular automata for high-resolutionmodelling of urban land-use dynamics. Environment and Planning B, 24 , 323343.

Wu, F. (1998). Simulating urban encroachment on rural land with fuzzy-logic-controlled cellularautomata in a geographical information system. Journal of Environmental Management, 53 , 293308.

Wu, F. (1999). GIS-based simulation as an exploratory analysis for spacetime processes. Journal of Geographical Systems, 1 , 199218.

Yang, Y., & Billings, S. A. (2000a). Extracting boolean rules from CA patterns. IEEE Transactions on

systems, man, and cyberneticsPart A: systems and humans, 30 , 573581.Yang, Y., & Billings, S. A. (2000b). Neighborhood detection and rule selection from cellular automata

pattern. IEEE Transactions on systems, man, and cyberneticsPart A: systems and humans, 30 , 840 847.

Yeh, A. G., & Li, X. (2002). A cellular automata model to simulate development density for urbanplanning. Environment and Planning B, 29 , 431450.

http://dx.doi.org/10.1016/S0198-9715(02)00068-6http://dx.doi.org/10.1016/S0198-9715(02)00068-6http://dx.doi.org/10.1016/S0198-9715(02)00068-6http://dx.doi.org/10.1016/S0198-9715(02)00068-6http://dx.doi.org/10.1016/S0198-9715(02)00068-6

paperceus

Documents

Transcript of paperceus