Determinants of land-use change patterns in the Netherlands · 2 Determinants of land-use change Land-use change can be described by the complex interaction of behavioural and structural

1 IntroductionThe spatial configuration of land use is an important determinant of many ecologicaland socioeconomical processes (Lambin et al, 2001). A better understanding of thedeterminants of the spatial configuration of land use is necessary to assess the impactof possible, future developments on environment, economy, and society at large.

Theoretical understandings of the processes that lead to the spatial pattern of landuse have been provided by researchers from different disciplines (Alonso, 1964;Christaller, 1933; Fujita et al, 1999; Von Thu« nen, 1966). Many empirical studies ofland-use-change patterns have focused on tropical deforestation (for example, Angelsenand Kaimowitz, 1999; Nelson et al, 2001; Walsh et al, 2001). Other researchers havefocused on urbanisation from the perspective of either economic geography (Arthur,1994; Fujita et al, 1999; Krugman, 1999) or from a more technology-driven perspectivethat focuses on the interaction of land uses through cellular automata models (Torrensand O'Sullivan, 2001; White and Engelen, 2000; Wu, 1999). In practice, many processesthat influence land-use change interact and lead to complex patterns, depending on the

Determinants of land-use change patterns in the Netherlands

Peter H VerburgDepartment of Environmental Sciences,Wageningen University, PO Box 37, 6700 AAWageningen,The Netherlands and Faculty of Geographical Sciences, Utrecht University, PO Box 80115,3508 TC Utrecht, The Netherlands; e-mail: [email protected]

Jan R Ritsema van EckFaculty of Geographical Sciences, Utrecht University, PO Box 80115, 3508 TC Utrecht, TheNetherlands and Netherlands Institute for Spatial Research, PO Box 30314, 2500 GH The Hague,The Netherlands; e-mail: [email protected]

Ton C M de NijsNational Institute of Public Health and the Environment, PO Box 1, 3720 BA Bilthoven,The Netherlands; e-mail: [email protected]

Martin J Dijst, Paul SchotFaculty of Geographical Sciences, Utrecht University, PO Box 80115, 3508 TC Utrecht,The Netherlands; e-mail: [email protected], [email protected] 30 January 2003; in revised form 3 July 2003

Environment and Planning B: Planning and Design 2004, volume 31, pages 125 ^ 150

Abstract. Land-use-change patterns are the result of the complex interaction between the human andthe physical environment. Case studies of the determinants of land-use change can help to analysewhich theory is appropriate in a particular region and stimulate the development of new theoreticunderstandings. In this paper an empirical method is presented to analyse the pattern of land-usechange that allows a wide range of factors, from different disciplines, to contribute to the explanationof land-use change. The method is applied to the Netherlands, based on an extensive database of land-use change and its potential determinants. Historic as well as recent land-use changes are studied.Historic land-use change is related mainly to the variation in the biophysical environment. Levels ofexplanation are low because of the inability to address the temporal variation in location factors. Forthe recent changes in land use high levels of explanation are obtained. The most important changesduring this period are expansions of residential, industrial/commercial, and recreational areas. Thelocation of these changes can be explained by a combination of accessibility measures, spatialpolicies, and neighbourhood interactions. On the basis of these results it is possible to define prioritytopics for in-depth analysis of land-use-change processes and suggest factors, relations, and processesthat need to be included in dynamic land-use-change models that support land-use-planning policies.

DOI:10.1068/b307

mailto:[email protected]





local cultural, socioeconomic, and biophysical context at different spatial scales (Lambinet al, 2001). For that reason, in this paper, we do not restrict ourselves to a singledisciplinary perspective or typical land-use conversion but will introduce a multidisci-plinary approach to analyse the impact of different determinants, derived from differenttheories explaining land-use allocation, for a range of different land uses. For thisanalysis we will use an extensive database from the Netherlands. Although urbanisationis the most important process of land-use change in the Netherlands we will also addressthe interaction of urban land uses with other land-use types and address recent as well ashistoric land-use-change processes that have shaped the land-use pattern. This analysismay contribute to our understanding of the interaction of land-use-change processes inthe Netherlands and provide an empirical foundation for the specification of land-use-change models. The next section discusses some theories of land-use change relevant tothe analysis of land use in the Netherlands. Based on these theories an empirical model isspecified (section 3) which is applied to the long-term historic changes in land use(section 4) and the more recent changes during the period 1989 ^ 96 (section 5). Thefindings are discussed and the applicability of the empirical method is evaluated in thelast section.

2 Determinants of land-use changeLand-use change can be described by the complex interaction of behavioural andstructural factors associated with the demand, technological capacity, social relationsaffecting demand and capacity, and the nature of the environment in question. It isdoubtful that a relatively simple explanation of why we transform the environment theway we do will be forthcoming (Turner et al, 1990). A theory of land-use change needsto conceptualise the relations among the driving forces of land-use change, theirmitigating processes and activities, and human behaviour and organisation. Althoughno single all-compassing theory of land-use change exists, different disciplinary theoriescan help us to analyse aspects of land-use change in concrete cases, their synthesis beingessential. In this paper we will use theoretical understandings from a range of disci-plines focused on the analysis of the spatial aspects of land-use change. Macroprocessesthat drive land-use conversion, the proximate causes and underlying driving factors ofland-use change, such as population growth, migration, and economic change are notanalysed as long as they are not directly related to the processes that determine thespatial pattern of land-use change.

2.1 Biophysical constraints and potentialsThe natural sciences, especially physical geography and agro-ecological disciplines,have contributed insights into the constraints and possibilities of the natural environ-ment for different land uses (Penning deVries et al, 1992). Each location has specific soilcharacteristics and climatic conditions that determine the possibilities for natural andagricultural vegetation. Some location characteristics constrain the growth of particularcrops so that they are not an option for that location. Other conditions cause lowerpotential yields or require additional inputs (fertiliser, land preparation, irrigation, etc)to make the location suitable for that particular land-use type. Agro-ecologists use land-evaluation techniques to determine the potential land uses for a location based on thebiophysical characteristics of that location (Fischer et al, 2002). Biophysical conditionsare also important for other land-use types, for example, the soil, geology, and drainageconditions determine the suitability of a location for residential construction. It takesconsiderable investment to reclaim swampy land for residential construction; therefore,a preference will exist to direct these developments to locations better suited forconstruction. In most land-use-change analyses the biophysical characteristics have

126 P H Verburg, J R Ritsema van Eck, T C M de Nijs, and coworkers

been incorporated as the potential gains that can be achieved by a particular land-usetype at a location.

2.2 Economic factorsA wide range of land-use-change studies are based on economic theory. Economistsmodel the relation between the location factors and land use assuming that, in equilib-rium, land is devoted to the use that generates the highest potential profitability[reflected in land rents (Chomitz and Gray, 1996; Irwin and Geoghegan, 2001; Munroeet al, 2001; Nelson et al, 2001)]. This assumption builds on the insights of Von Thu« nen(1966) and Ricardo (1817). In its original application only the location relative to themarket is important, all other landscape features are ignored. Land-use structure wasthe outcome of the competition for accessibility, leading to the well-known concen-tric circles of land use. In later years this theoretical model was extended by manyother factors that determine the rent as a location. Best known is Alonso's (1964)land-market theory and model that include a number of other economic and envi-ronmental characteristics that determine land rent, such as infrastructure and soilquality. In spite of developments in societies, accessibility and locational trade-offsare still seen as central elements in the creation of land-use structure in the 21stcentury (Clark, 2000). The main assumption of this approach: land-use decisionsare made by utility-maximising individuals, underlies all applications. Psychologicalresearch has suggested various modifications to this conception of human choice andhas indicated the limitations of using this theory in understanding land-use structure(Rabin, 1998).

2.3 Social factors, irreversibility, and uncertaintySocial theories focus on factors which influence locational choices of households.Important determinants of these choices are individuals' cultural values, norms, andpreferences (lifestyles), and their financial, temporal, and transport means (Pahl, 1975;van Kempen et al, 2000). In terms of spatial indicators, households focus on site andsituation characteristics of locations. Examples of site characteristics are housing prices,level of supply of consumer services, quality of landscapes, and social composition ofneighbourhoods (Faust et al, 1999; Geoghegan et al, 1997; 2001; Leggett and Bockstael,2000; Mertens et al, 2000). Furthermore, the role of historical events and arbitrarydecisions (uncertainty) is stressed to be important. A small, arbitrary, decision can leadto new development through the path dependency of some of the processes thatdetermine land-use allocation. The development of cities or commercial compoundsis an example: the initial decision to locate a company at a certain location can be ratherarbitrary. Economies of scale (Arthur, 1994) and other processes can reinforce such adevelopment and cause expansion of urban land uses at that location. Another causeof the reinforcement of certain land-use-change patterns relates to the irreversibility ofsome land-use changes, for example, tropical forest cannot be recreated and desertifiedland is very difficult to reclaim.

2.4 Spatial interaction and neighbourhood characteristicsGeographers have contributed to the analysis of land-use pattern through the analysisof spatial interactions. Land use does not develop independently at each individuallocation; each development affects the conditions of neighbouring and distant loca-tions. Simple mechanisms for interaction between locations are provided by centralplace theory (Christaller, 1933) which describes the uniform pattern of towns and citiesin space as a function of the distance that consumers in the surrounding region travelto the nearest facilities. More complex spatial interactions are the subject of urbannetwork theories (Hohenberg and Lees, 1985). The location of land uses is dependent

Determinants of land-use change patterns in the Netherlands 127

on interregional and international networks of economic, cultural, and politicalrelations (Castells, 2000). Spatial interaction between the location of facilities, resi-dential areas, and industries has been given more attention in the work of Krugman(Fujita et al, 1999; Krugman, 1999). Spatial interactions are explained by a number offactors that either cause the concentration of urban functions (centripetal forces:economies of scale, localised knowledge spillovers, thick labour markets) and othersthat lead to a spatial spread of urban functions (centrifugal forces: congestion, landrents, factor immobility, etc). Direct interaction between neighbouring land uses is oftenassessed through cellular automata models that specifically address the push ^ pullbetween neighbouring land uses (Webster and Wu, 2001; Yeh and Li, 2001).

2.5 Spatial policiesPolicies at national or subnational level have a bearing on land use. In particular,policies that have a spatial manifestation (for example, the creation of conservationareas, land-tenure changes, or designated areas for subsidised developments) influencethe spatial pattern of land-use change. Spatial policies are specific for the study areaand time period studied. Therefore, any study of land-use-pattern change should bepreceded by an inventory of the relevant spatial policies.

A hypothesis that unifies the different disciplinary theories is that actors respond tocues both from the physical environment and from their sociocultural and spatialcontext and behave to increase both their economic and their sociocultural well-being;where well-being is dependent on perception, scale, and time. The empirical modelused in this paper is based on this hypothesis which allows the incorporation of a largenumber of factors important to land-use change derived from the different disciplinaryunderstandings of land-use change described above.

3 Methods and data3.1 MethodsHumans are expected to optimise well-being by allocating land-use conversions atlocations with the highest `preference' for the specific type of land-use conversionat that particular moment in time. `Preference' is an unobserved, dimensionless vari-able, defined by economic returns, market competition, sociocultural context, arbitrarypreferences, and policy regulations. Preferences can differ for the different actorsinvolved in land-use decisions. In this study, preference represents the outcome of theinteraction between the different actors and decisionmaking processes that result in aspatial land-use configuration. The preference of a location is empirically estimatedfrom a set of factors that are assumed to influence the preference based on the differentdisciplinary understandings described in the previous section. This definition of pref-erence allows us to include a wide range of factors that can also include factors thatexplain deviation from rational behaviour. Preference is calculated as follows:

Rik � AkGi � BkDi � .:: , (1)

where R is the preference to devote location i to land-use type k, Gi is one of thebiophysical characteristics of location i, and Ak is the relative impact of this biophysicalcharacteristic on the preference for land use k. Di represents the distance from location ito the market and Bk the relative importance of being located close to the market forland-use type k. Other factors are added depending on the land-use type considered andthe region addressed. The exact specification of the model should be based on athorough review of the processes important to the spatial allocation of land use inthe studied region.


The statistical model developed is a binomial logit model of two choices: convertlocation i into land-use type k or not. Assuming that the preference Rik is the underlyingresponse to this choice, we can specify the following regression:

Rik � bk xi � ui , (2)

where the xi are the exogenous (explanatory) variables of the preference (Rik ), bk arethe coefficients to be estimated, and the ui are random errors. Clearly, the preferencefor a location cannot be observed or measured directly, so Rik is actually unobserved.Therefore a dummy variable, y, is created:

y �1; if Rik > Rij for all uses j 6� k ,

0; for all others .

((3)

That is, location i is converted into land use k if the preference for land use k is higherthan for all other land-use types. Otherwise no conversion into land use k is observed.If it is assumed that the ui have a logistic distribution, then this is a binomial logit ofthe form:

P� y � 1�that is, converted�� exp� bx�1� exp� bx� , logitP � bx . (4)

In this study we have estimated the coefficients in equation (4) for land-use changesat two different time scales. First, we have formulated a long-term model to describethe land-use conversions that led to the land-use pattern in 1989 in the Netherlands.More recent developments have been studied in a model that describes the conversionsin the main land-use types between 1989 and 1996, a period in which the Netherlandsexperienced strong urbanisation. As a consequence of the regional differences inbiophysical and socioeconomic conditions between regions in the Netherlands it ispossible that different relations between land use and location characteristics areprevalent in different regions. To verify this we have conducted region-specific analysisfor a number of land-use types.

Instead of the binomial logit specification we could have used a multinomial logitmodel addressing the probability of conversion of the different land-use categoriesrelative to a reference category. However, in this study we are interested mainly inthe probability of a land-use change at a certain location relative to all other options.Therefore, a dichotomous specification of the regression is chosen. Furthermore, thisspecification allows different explanatory factors for different types of land-use change,which is essential for an appropriate specification of the model.

Stepwise logistic regression was used to estimate the coefficients of the definedmodel. The dependent variable is a binary presence or absence event, where1 � presence (occurrence of a land-use conversion) and 0 � other. Odds ratios areused to facilitate model interpretations. The odds ratio [exp ( b)] can be interpreted asthe change in the odds for the considered event (that is, land-use change) upon anincrease of one unit in the corresponding factor.

All analysis was made using pixels of 500� 500 m as unit of observation. Depend-ent and independent variables were therefore converted to a raster-based format. Spatialdependence can be expected in models in which relative location matters. Regressionsbased on dependent variables that exhibit spatial autocorrelation are influenced by thespatial structure (Anselin, 1988; Bell and Bockstael, 2000). In this study we haveminimised the influence of spatial autocorrelation on the regression results by usinga sample (10% of all observations) of randomly distributed observations instead of thefull dataset. The inclusion of neighbourhood characteristics in the model specificationaddresses the issue of autocorrelation in land-use patterns explicitly.


The goodness of fit of the logistic regression models is measured by the ROC[relative operating characteristic (Pontius and Schneider, 2001; Swets, 1986)]. TheROC is based on a curve relating the true-positive proportion and the false-positiveproportion for a range of cutoff values in classifying the probability. The ROC statisticmeasures the area beneath this curve and varies between 0.5 (completely random) and 1(perfect discrimination).

3.2 Data3.2.1 Land useIn this study we used the 1989 and 1996 maps of land use produced by the CentralBureau for Statistics (CBS Bodemstatistiek). Since 1989 these maps have been based onaerial photographs (scale 1:18 000), distinguishing thirty-three different land-use types.These maps were checked for inconsistencies by the National Institute of Public Healthand the Environment, rasterised to 25� 25 m (Raziei and Evers, 2001) and combinedwith the LGN database (Landelijk Grondgebruiksbestand Nederland), another seriesof high-resolution land-use maps produced by Wageningen University and ResearchCenter (Alterra Green World Research) based on the interpretation of Landsat TMremote sensing images (de Wit et al, 1999). The integration of both maps made itpossible to subdivide the agricultural land-use type (de Nijs et al, 2001). For this studythe dataset was aggregated to a 500� 500 m resolution. This aggregation was basedon the majority rule. Such a procedure would yield a bias, which could cause land-usetypes with a relative small coverage to disappear (He et al, 2002; Milne and Johnson,1993; Moody and Woodcock, 1994). Therefore, we constrained the aggregation byrequiring the total areas of each land-use type to correspond in both the high-resolutionand low-resolution maps.

Based on this procedure, consistent land-use maps at 500� 500 m were producedfor 1989 and 1996 and reclassified into ten land-use types relevant to the analysispresented in this paper. Table 1 gives a description of the land-use types.

3.2.2 Biophysical characteristicsBiophysical characteristics determine the suitability of a location for agricultural use,adding to the potential profit that can be gained by a certain type of land use at acertain location. For agricultural use, soil conditions are very important. Insteadof characterising the soil conditions by the soil-unit classification we have translatedthe soil units into properties relevant to agricultural and natural land uses, such as thealkalinity, texture, and organic matter content of both top soil and subsoil. Drainage is

Table 1. Land-use classification used in the analysis.

Land-use type Description

Grassland Grasslands, including seminatural grasslandsArable land All arable landsGreenhouses GreenhousesOther agriculture Agricultural land not belonging to greenhouses, grasslands, or

arable land, includes horticulture, orchards, etcResidential areas Residential areas and social ± cultural facilities, including houses,

roads within residential areas, schools, hospitals, churches, etcIndustrial/commercial Mining areas, industries, harbours, shopping malls, prisons, and

all service industriesForest/nature Forests and natural areas, including peat areas, swamps, heather, etcRecreational Parks, sports fields, kitchen garden complexes, camp sites, etcAirports AirportsWater Water


characterised by the groundwater depth and fluctuation throughout the year. Anotherbiophysical characteristics included in this study is elevation, directly related to agri-cultural suitability and historic settlement patterns when flooding was more frequent.

Because of the assumed collinearity between the soil variables we have tested thecoefficients of determination (R 2 ) of the multivariate relationships between one ofthe variables against all the others. High correlations were found between the clayand loam contents of the soils and between the calcium contents of topsoils and subsoils.In the analysis the calcium content of the subsoil, the loam content of the topsoil, andthe clay content of the subsoil are omitted. Now all multivariate relationships are wellbelow the critical value of 0.80 (Menard, 1995).

3.2.3 Socioeconomic conditionsMany socioeconomic conditions do not relate to the location itself but to the regionalconditions instead, for example, the presence of employment opportunities in theneighbourhood, the level of facilities, etc. The influence of the socioeconomic condi-tions in the region can be best characterised by the access that a location has to thesefacilities. In many studies (for example, Chomitz and Gray, 1996), therefore, thedistance to the nearest location of employment or facility is calculated to representthese location factors. Often airline distance is used (for example, Pijanowski et al,2002; Thompson et al, 2002), without taking into account the layout of the infra-structures network and natural barriers (rivers, lakes, etc). This can cause unrealisticdistance measures. Therefore, we generated advanced accessibility measures usingFLOWMAP to produce a better representation of the accessibility faced by the agentsof land-use change. FLOWMAP is a software package for spatial analysis with empha-sis on interaction modelling, network analysis, and accessibility measures. It allowseasy data exchange with geographic information system packages where the results canbe further analysed and visualised using more sophisticated cartographic techniques(de Jong and Ritsema van Eck, 1996).

A common variable in economic models of land-use change is the distance betweenthe residential location and employment, as a proxy for the costs associated withtravel. We have included this proxy by calculating the average distance of a locationto the location of employment, identified by 100 000 and 500 000 jobs to distinguishbetween regional and national centres of employment, respectively. Besides economicmotives, travel times can be important for the decision to live or work at a certainlocation. Therefore, all accessibility measures are expressed in travel times as well as indistance measures. Another important factor for decisions on land-use allocation is theaccessibility of facilities, including schools, shops, and medical and social services. Wehave developed a proxy for this factor by calculating average distances and travel timesto the nearest 100 000 and 500 000 inhabitants. Furthermore, we have included acces-sibility of forest, nature areas, lakes, and the coast as a proxy for the ease of access torecreational facilities that is assumed to have a positive influence on the decision to livesomewhere. Because all accessibility measures are calculated by means of roads, usingthe car as the vehicle, we have also included the travel time and distance to the nearestrailway station as a factor indicating the ease of access to public transport for alocation.

Location factors that are important for the allocation decisions of industry andcommercial compounds are the ease of access to the main airport (Schiphol AmsterdamAirport) and Rotterdam harbour. Location next to a motorway is also importantfor companies, not only because of easy access (as measured by the accessibility tomotorway intersections), but also because of visibility.


132PH

Verburg,

JR

Ritsem

avan

Eck,T

CM

deNijs,

andcow

orkersN:/psfiles/epb3101w

/Table 2. Description of spatially explicit allocation factors used in this study.

Variable Description Source

ALTITUDE Mean elevation calculated from a high-resolution digital elevation model Rijkswaterstaat (http://www.minvenw.nl/rws/mdi/geoloket/)

GRWATER Groundwater dynamics characterised by depth and fluctuation throughthe year (grondwatertrappen); higher categories are better drained soils

1:50 000 soil map of the Netherlands (Steurand Heijink, 1991)

Soil characteristics Properties of soils based on the 1:50 000 soil map 1:50 000 soil map of the NetherlandsORGMAT TOP Organic matter content of the topsoila (Steur and Heijink, 1991) combined withORGMAT SUB Organic matter content of the subsoilb physical and chemical characteristicsCA TOP Calcium content of the topsoil of the different soil types (BodemkundigCA SUB Calcium content of the subsoil Informatie Systeem, Alterra, WageningenLOAM TOP Loam content of the topsoil University and Research Centre),LOAM SUB Loam content of the subsoilCLAY TOP Clay content of the topsoilCLAY SUB Clay content of the suboilpH TOP Acidity of the topsoilpH SUB Acidity of the subsoil

Socioeconomic variablesD STATION Road distance to nearest railway station National Road DatabaseT STATION Travel time to nearest railway station (Basisnetwerk Nederland) 1990;D AIRPORTt Road distance to Schiphol airport Adviesdienst Verkeer en VervoerT AIRPORT Travel time to Schiphol airport Rijkswaterstaat (AVV); employment andD HARBOUR Road distance to Rotterdam harbour population statistics: Central BureauT HARBOUR Travel time to Rotterdam harbour for StatisticsD WORK1 Average distance to nearest 100 000 jobsT WORK1 Average travel time to nearest 100 000 jobsD WORK2 Average distance to nearest 500 000 jobsT WORK2 Average travel time to nearest 500 000 jobsD HIGHWAY Distance to nearest motorway intersectionT HIGHWAY Travel time to nearest highway entranceD CITY Average distance to nearest 500 000 inhabitantsT CITY Average travel time to nearest 500 000 inhabitantsD TOWN Average distance to nearest 100 000 inhabitantsT TOWN Average travel time to nearest 500 000 inhabitants

http://www.minvenw.nl/rws/mdi/geoloket/

http://www.minvenw.nl/rws/mdi/geoloket/

Determ

inantsof

land-usechange

patternsin

theNetherlands

133Table 2 (continued).

Variable Description Source

D HISTCITY Historic travel time index for accessibility to nearest main city(>10 000 inhabitants in 1550)

Network structure derived from Ministerievan Verkeer en Waterstaat (2001)

D HISTTOWN Historic travel time index for accessibility to nearest main town(5 000 ± 10 000 inhabitants in 1550)

D RECREATION Distance to the nearest main forest/water area frequently used for recreation National Institute for Public Health andthe Environment

T RECREATION Travel time to the nearest main forest/water area frequently used for recreation National Institute for Public Health andthe Environment

D OPENWATER Distance to the coast or main river (no network) National Road Database [BASNET:NOISE Areas with cumulative noise exceeding 50dB(A) Basisnetwerk Nederland (1990)];AIRP NOISE1 Area with moderate noise [20 Ke (a Dutch unit of noise)] due to the national

Schiphol airportAdviesdienst Verkeer en VervoerRijkswaterstaat (AVV);

AIRP NOISE2 Area with high noise (35 Ke zone) due to the national Schiphol airportHIGHWAY Areas located within 750 m of a motorway

Neighbourhood Enrichment factors for neighbourhood at average distance of 500 m and 1800 m Calculations based on 1989 land-use mapinteraction characteristicsINFL RES1 Enrichment with residential land use at 500 mINFL RES3 Enrichment with residential land use at 1800 mINFL INDUST1 Enrichment with industrial/commercial land use at 500 mINFL INDUST3 Enrichment with industrial/commercial land use at 1800 mINFL REC1 Enrichment with recreational land use at 500 mINFL REC3 Enrichment with recreational land use at 1800 mINFL FOREST1 Enrichment with forest/natre at 500 mINFL FOREST3 Enrichment with forest/nature at 1800 m

PolicyGRCENTRE Designated municipalities to cater for a large part of new home construction to

avoid suburbanisation and overcrowding of the main cities of the NetherlandsGRHEART Delineation of the central open space in the western part of the Netherlands

(the Green Heart)a 0 ± 35 cmb 35 ± 125 cm

Another factor that is assumed to influence land-allocation decisions is noise. It isexpected that the proximity of highways and airports has a negative influence on thepreference to use a location for new residential land use because of noise. We haveincluded a map with all locations subject to high noise levels as calculated by theNational Institute for Public Health and the Environment. This map is supplementedwith a special map showing the noise and safety contours of air traffic to the nationalairport.

3.2.4 Neighbourhood characteristicsThe land-use situation in the direct neighbourhood of a location can be an importantdeterminant of land-use change. The neighbourhood characteristics can be quantifiedby an enrichment factor' (Verburg et al, 2004) that is defined by the occurrence ofa land-use type in the neighbourhood of a location relative to the occurrence of thisland-use type in the study area as a whole, according to

Fikd �nkdindi

�Nk

N, (5)

where Fikd characterises 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 identified by d ; nkdi is the number of cells of land-use type kin the neighbourhood d of cell i, ndi is the total number of cells in the neighbourhood,Nk is the number of cells with land-use type k in the whole raster, and N is the totalnumber of cells in the raster. So, if the neighbourhood of a certain grid cell contains50% grass whereas the proportion of grass in the country as a whole is 25% wecharacterise the neighbourhood by an enrichment factor of 2 for grassland. Whenthe proportion of a land-use type in the neighbourhood equals the national average, theneighbourhood is characterised by a factor of 1 for that land-use type. An underrepre-sentation of a certain land-use type in the neighbourhood results in an enrichmentfactor between 0 and 1.

This neighbourhood characteristic results for each grid cell i in a series of enrich-ment factors for the different land-use types (k ). The procedure is repeated for differentneighbourhoods located at different distances (d ) from the grid cell to study theinfluence of distance on the interaction between land-use types. In this study we haveused square rings at different distances (d ) from the central cell as the neighbourhood.

3.2.5 Spatial policiesPolicies that have spatial manifestations are included as a spatially explicit layerdesignating the specified areas. During the period for which land-use changes werestudied in detail (1989 ^ 96) few new specific spatial policies were implemented. Thisperiod is characterised by a disengagement of the central government from manypolicies associated with the Dutch welfare state. This led to weak government controlover housing. Emphasis was given to promoting compact urbanisation by developingsites within and directly adjacent to cities (Dieleman et al, 1999). During the 1970s and1980s so-called growth centres and growth towns were designated to cater for a largepart of the new home construction to avoid suburbanisation and overcrowding of themain cities of the Netherlands and to protect the Green Heart as an open space inthe Randstad. This policy of controlled dispersal was dubbed concentrated deconcentra-tion' (Faludi and Van der Valk, 1990) and included nineteen municipalities, two of themnew towns on land reclaimed from the sea. It is assumed that this policy still affectsresidential construction after 1989. Therefore, we have included the growth-centremunicipalities as an independent variable.


Another important item of Dutch spatial planning doctrine has always been theGreen Heart policy (Dieleman et al, 1999). Spatial policies have been relatively successfulin keeping the Green Heart as a central open space surrounded by urban development.We included the delineation of the green heart as an explanatory factor in the analysis.

A summary of all data included in the analysis is given in table 2.

4 Long-term land-use changesIn this section an analysis is made of the long-term changes in land use that haveresulted in the land-use pattern found now. In the analysis we have used the landscapepattern of 1989 as the dependent variable. This pattern is the result of many historicdevelopments that have converted the Dutch landscape from an uninhabited naturallandscape with large waterlogged areas into a heavily urbanised landscape.

4.1 Model definitionMany factors that are commonly used to explain land-use change patterns are endog-enous to the processes studied at a long time scale. Measures indicating currentaccessibility can be used when studying land-use change for one or several decades asa factor explaining the allocation of new residential or commercial locations. At longertime scales, accessibility and urban expansion are closely linked, with complex cause êffect relations. The same type of reasoning makes other socioeconomic location factors,such as population pressure and labour availability, unsuitable explanatory factors forthe land-use pattern. Instead, we assume that the historic development of the land-usepattern is closely related to the biogeophysical conditions of the land. Therefore weincluded soil properties and elevation as explanatory factors.We have left out drainagecharacteristics because in almost the whole of the Netherlands the groundwaterdynamics are artificially managed given the requirements for the specific land use.Such management weakens the connection between land use and the constraints setby the abiotic template (Bu« rgi and Turner, 2002). Therefore, groundwater data cannotbe considered as independent variables in the analysis. Soil properties, especiallyorganic matter content, can also be influenced by land use and management (Sonneveldet al, 2002). For this part of the study, in which we have derived the soil conditions fromthe 1:50 000 soil map that is not directly linked to management units, we assume thatthis effect is negligible.

At the time scale considered we could include only two variables representing thesocioeconomic opportunities for urban development. These factors are the distance toopen water (coast and main rivers) that is seen as a proxy for market access and trade,which has, historically, always been an important factor for the economic developmentof the Netherlands. The second variable included is a rough map providing an index ofthe historic accessibility (valid for around 1600) to the important towns. It is assumedthat the historic settlement places have triggered further development of the urban landuses because of the advances of larger economies (Arthur, 1994; Krugman, 1998).

The tested logit models for residential and industrial or commercial land are

logitP � a� b1SOIL� b2ALTITUDE� b3D HISTTOWN� b4D OPENWATER ,

where SOIL includes organic matter, loam content, clay content, pH, and calcium contentlisted in table 2. For grassland and forest we have included only the biophysical factors:

logitP � a� b1SOIL� b2ALTITUDE ,

and for arable land we also include the distance to historic towns because it is expectedthat food production would preferably be located near to the town:

logitP � a� b1SOIL� b2ALTITUDE� b3D HISTTOWN .


4.2 ResultsTable 3 gives the estimated coefficients for the logit models describing the land-usepattern for the main land-use types in 1989. The presented exp(b) values (odds ratio)indicate the change in odds of the dependent variable (land use) upon a change in theindependent variable. Values between 0 and 1 mean that the probability will decreaseupon an increase in the value of the independent variable.Values above 1 indicate anincrease in probability upon an increase in the independent variable. The ROC valuesindicate that the spatial pattern of forest/nature can be reasonably explained bythe independent variables. Model fit for arable land and grassland is less goodand the independent variables explain only a small fraction of the spatial variabilityof residential and industrial/commercial land.

The model fit is illustrated in figure 1 with a probability map of forest/nature basedon the logit model in table 3. A number of soil variables and the elevation have asignificant contribution to the explanation of the spatial pattern of forest/nature. Ingeneral, forest/nature is found on the somewhat higher locations with soil propertiesthat are clearly different from those of arable land, indicating that forest/natureremains in areas unsuitable for agriculture. The association with low pH of the topsoilrelates to the podzolisation process that is dominant in most poor sandy soils in theNetherlands. Soil characterstics are important determinants of the distribution ofarable land: the main arable areas are located on marine sediments in the northernand southwestern part of the Netherlands and within the polder areas (positive con-tributions of loam and calcium content). An area with intensive potato cultivation isfound in the northern part of the country on reclaimed peat bogs. After reclamation,the soils have a high organic matter content in the topsoil and low subsoil organicmatter contents. The positive contribution of topsoil organic matter and negativecontribution of subsoil organic matter makes it possible to distinguish this area fromthe lowland peat areas in the western part of the Netherlands that are unsuitable forarable land.

Table 3. Binary logit estimates for land-use patterns in 1989 [exp(b ) values].

Forest/ Arable Grassland Residential Industrial/nature land land commercial land

Constant 0.14 0.10 0.62 0.70 0.06ALTITUDE 1.03 * 0.98 * 1.01SOIL

ORGMAT TOP 0.99 1.02 0.996 * *ORGMAT SUB 1.04 0.98 * * *CA TOP 1.10 1.29 0.72 * *LOAM SUB 0.93 1.02 1.01 1.01 *CLAY TOP * 0.97 1.01 0.95 *pH TOP 0.64 * 1.17 * *pH SUB 2.04 0.90 0.86 0.77 *

D HISSTTOWN ± 1.00004 ± 0.9999 0.99998D HISTCITY ± ± ± * 0.99999D OPENWATER ± ± ± * 0.99

ROC statistica 0.82 0.73 0.73 0.67 0.65

* not significant; excluded from model specification, all other exp(b ) values significant at 0.05level.± not included in model specification.a ROCÐrelative operating characteristic statistic.


We are less successful in explaining the spatial distribution of residential areas. Fewof the selected variables have a significant contribution to the model. The positiveassociation with loam content indicates the historic preference to use solid soils, mostlythe levees of rivers, instead of the swampy areas in between the rivers. The distance tohistorical towns is an important factor; however, the low level of explanation indicatesthat many more processes have been important in the spatial allocation of newresidential areas that are not included in the present model. The distance to historiccities and open water does not add to the explanation of residential land distributionbut has an influence on industrial/commercial land allocation, indicating the impor-tance of access to national and international markets for this land-use type. Possibly,the location of residential and industrial/commercial concentrations can be betterexplained by self-reinforcement of industrial and residential locations through econo-mies of scale and the emergence of new centres (Fujita et al, 1999; Krugman, 1999).Arthur (1994) showed that, as soon as self-reinforcements or economies of scale areimportant, chance events become important determinants of the system of cities. There-fore, the observed pattern of residential and industrial locations cannot be explained bydeterminism alone without reference to chance events, coincidences, and circumstancesin the past. Chance events can never be adequately included in an empirical analysis ofland use. Reinforcement of patterns and path dependence can be studied at shorter timescales. In the next section we present an analysis of land-use change between 1989 and1996 that includes proxies (that is, neighbourhood characteristics) for processes thatlead to path dependence in the model specification.

Another reason for the low explanatory power of the regression model for urbanland use is the time dependency of factors important to land-use-change decisions.Historically, the distance between urban and agricultural areas might have been animportant consideration for land-use decisions. Modern transport facilities and storagecapacity have made this an unimportant factor. Furthermore, land-use policies as wellas socioeconomic conditions have seen large changes during the long period over whichthe land-use pattern has developed. Therefore, it is difficult to explain the observed

(a) (b)

Figure 1. (a) Location of forest/nature areas in 1989 (black) and (b) probability of the occurrenceof forest/nature areas based on the logit model (dark colour: high probability).


land-use changes that have occurred over such a long period of time without explicitlyaddressing the changes in land-use-change processes over time.

Region-specific regression models for arable land were estimated for three differentregions in the Netherlands (figure 2; table 4). For the northern and western regions theROC value was higher than the ROC value for the country as a whole and differentparameter estimates were found for the different regions. The organic matter contentof the topsoil has a positive relation with the occurrence of arable land only inthe northern region (the previously mentioned reclaimed peat bogs). Especially in thewestern region the spatial pattern of arable land could be described particularly wellbecause arable land is, in this area, found only on marine sediments. In the southernregion maize cultivation is the most important crop of the arable land-use type. Maizecultivation is less dependent on the soil suitability, leading to a lower level of explanationfor the distribution of arable land in the southern region.

40 0 40 km

(a) (b)

NTravel timeto town

Distance torailway station

Designatedgrowth centres Elevation

Loam contentof the topsoil

Calcium contentof the subsoil

Figure 2. (a) Map of the Netherlands with provincial boundaries and regional subdivisions and(b) examples of six variables used in this study.

Table 4. Binary logit estimates for the spatial pattern of arable land in 1989 [exp(b ) values].

Western Northern Southernregion region region

Constant 0.02 0.07 0.11ALTITUDE 0.78 0.99 *SOIL

ORGMAT TOP 0.97 1.02 0.97ORGMAT SUB 0.96 0.97 0.99CA TOP 1.40 1.36 1.19Loam SUB 1.02 1.02 1.01CLAY TOP 0.98 0.98 *pH TOP * * 0.95pH SUB * 0.92 1.14

D HISTTOWN 1.00005 1.00006 1.00002

ROC statistica 0.89 0.76 0.63

* not significant; excluded from model specification, all other exp(b ) values significant at 0.05level.± not included in model specification.a ROCÐrelative operating characteristic statistic.


The results of the regional analysis suggest that spatial disaggregation leads tomore specific relations between the location characteristics and land use. Thematicdisaggregation of the land-use classes will certainly improve the results as well, butwas not considered within the scope of this study.

5 Recent land-use changes: 1989 ^ 96In this section we analyse the determinants of the changes in land-use patternbetween 1989 and 1996. In the first part of this section we define and estimatea model including only location characteristics. In the second part we expand thismodel by including neighbourhood characteristics.

5.1 Location characteristics5.1.1 Model definitionModels are defined to explain the major land-use changes between 1989 and 1996.These changes include new residential developments (0.75% of the total area of theNetherlands), new areas devoted to industrial/commercial uses (0.49%) and newrecreation areas (0.44%). Because different factors are assumed to determine thelocation of these land-use changes a different model was estimated for each of thesethree conversion types. The model for new residential developments is:

logitP � a� b1GRWATER� b2ALTITUDE� b3ACCESSIBILITY� b4NOISE

�b5POLICY .

ACCESSIBILITY includes accessibility measures that represent the distance and/or traveltime to employment opportunities, urban facilities, infrastructure, and recreation. Weconsidered the following variables: D STATION, T STATION, D WORK1, T WORK1,D WORK2, T WORK2, D HIGHWAY, T HIGHWAY, D CITY, T CITY, D TOWN, T TOWN,D RECREATION, and T RECREATION (table 2). POLICY variables include the growthcentre policy (GRCENTRE) and the Green Heart policy (GRHEART). Noise was expectedto influence negatively the decisions to plan new residential areas. We includedcumulative noise (NOISE) and noise and safety contours for the national airport(AIRP NOISE).

The estimated model for new industrial developments is:

logitP � a� b1GRWATER� b2ALTITUDE� b3ACCESSIBILITY� b4POLICY

� b5HIGHWAY ,

where accessibility includes a somewhat different set of variables including accessibilityto the airport and Rotterdam harbour as potential factors determining the location ofnew industrial/commercial compounds. The full list of variables that represent theaccessibility is: D STATION, T STATION, D WORK1, T WORK1, D WORK2, T WORK2,D HIGHWAY, T HIGHWAY, D CITY, T CITY, D TOWN, T TOWN, D AIRPORT,T AIRPORT, D HARBOUR, and T HARBOUR. In addition to the variables that describethe accessibility to infrastructure, we include a variable (HIGHWAY) that indicates thezone of 750 m along the main motorways to represent the preference to locate newindustrial/commercial compounds as much as possible near to the main infrastructure.For new recreational areas:

logitP � a� b1ACCESSIBILITY� b2POLICY ,

where ACCESSIBILITY includes the distance or travel time to residential areas, existingrecreational areas, and infrastructure (D HIGHWAY, T HIGHWAY, D CITY, T CITY,


D TOWN, T TOWN, D RECREATION, and T RECREATION). Biogeophysical variableswere supposed not to influence the allocation of new recreational areas.

Locations with land use classified as residential, industrial/commercial, airport, ornature/forest in 1989 were considered not to be potential locations for new residentialor industrial/commercial development and were therefore excluded from the analysis.For the analysis of new recreational areas the locations that were designated forrecreational use in 1989 were also excluded from the analysis.

5.1.2 ResultsFor each of the three types of land-use change the characteristics of locations chosenfor development and the locations not chosen are calculated (table 5). Existing urbanlocations and forest/nature areas are excluded from the analysis. Accessibility isimportant for all three land-use changes studied: for all accessibility measures differ-ences were observed between the accessibility of developed and undeveloped locations.The closer the distance and shorter the travel time to city, town, job opportunities, androad and rail infrastructure, the greater the development. It should be noted that allaccessibility measures used in this study are positively correlated, but still exhibit cleardifferences in spatial variability and pattern. Although it was expected that newdevelopments would avoid noisy locations, this holds only true for the immediateneighbourhood of the national airport (AIRP NOISE2: 35 Ke zone), in other partsnew developments take place at locations that have higher levels of noise than others.This is a direct consequence of concentrations of population and the infrastructureneeded to serve this population. Almost 14% of all new residential areas were con-structed within the growth centres that were part of the 1980s housing policy. Thisillustrates that there is still in the period considered in this study a tendency toconcentrate housing in these growth centres. The growth centres were also importantfor the allocation of new commercial/industrial compounds, partly because of theattractive industrial location of some of the growth centres and the incentives providedby the growth-centre municipalities (Faludi and Van der Valk, 1994). For commercial/industrial land the influence of the proximity of a highway was a main determiningfactor. Not only was the distance or travel time to a motorway intersection important,but also the location in the immediate neighbourhood of the motorway: 40% of all newindustrial/commercial areas were within 750 m of a motorway but the figure was only15% for locations not developed into industrial/commercial land.

The results are summarised in logit models that were estimated independently forthe three land-use changes considered in this study. Table 6 (over) gives the estimatedcoefficients and odds ratios for these models. Through multicollinearity only a smallnumber of variables were actually included. Because of the supposed lack of causalityin the relation between residential land use and NOISE we excluded this variable fromthe model. Variables that are significant in the model for new residential areas relateto the biophysical conditions (dryer locations, lower reclamation costs), accessibility,and policy. For new industrial/commercial developments only the accessibility andlocation near highways is of significant importance and for new recreational areasthe location close to concentrations of population and the growth-centre policy. Prob-ability maps for new residential areas and new industrial/commercial areas are shownin figure 3 (over). Many more locations are indicated to have a high probability forconversion than the actually converted locations. A considerable part of these locationsis, however, already occupied by residential or industrial land uses and is therefore nota candidate for conversion.

The logistic regressions mentioned above are based on an analysis for the country ofthe Netherlands as a whole. Regionally specific conditions are expected to influence the


Determ

inantsof

land-usechange

patternsin

theNetherlands

141Table 5. Average characteristics of locations with and without new residential, industrial/commercial, and recreational developments, 1989 ^ 96.

Variable Residential Industrial/commercial Recreational

without with without with without with(N � 104 402) (N � 1009) (N � 104 752) (N � 659) (N � 101 279) (N � 593)

avg SD avg SD avg SD avg SD avg SD avg SD

D WORK1 20.14 7.33 14.29 6.45 20.11 7.34 15.18 6.79 20.27 7.23 15.99 8.15D WORK2 41.90 15.60 32.45 13.47 41.86 15.60 32.90 13.98 42.16 15.46 33.23 16.63T WORK1 24.26 11.80 17.69 6.76 24.24 11.79 18.37 8.23 24.38 11.59 19.75 9.06T WORK2 42.25 17.37 32.98 11.76 42.22 17.36 33.40 12.63 42.46 17.17 34.67 15.59D TOWN 12.70 4.51 8.38 3.78 12.68 4.52 8.98 4.21 12.79 4.44 9.84 4.85D CITY 24.26 8.51 18.12 7.10 24.23 8.51 18.97 7.43 24.41 8.41 19.74 9.40T TOWN 16.58 8.88 11.62 3.95 16.56 8.87 12.10 5.20 16.65 8.64 13.30 5.40T CITY 27.93 13.10 21.13 7.44 27.90 13.09 21.83 8.60 28.05 12.91 23.32 10.41D HIGHWAY 11.81 8.48 7.10 6.67 11.80 8.48 7.61 7.75 11.91 8.46 8.40 7.55T HIGHWAY 15.26 13.65 9.36 7.87 15.24 13.64 9.69 9.17 15.36 13.52 11.00 9.68D STATION 10.98 6.99 6.68 5.95 10.96 6.99 7.58 6.56 11.06 6.96 8.49 6.91T STATION 16.75 12.91 11.02 7.87 16.72 12.90 12.11 9.03 16.82 12.78 13.42 9.39GRCENTRE (%) 2.95 13.78 3.00 11.53 2.87 9.78GRHEART (%) 5.79 2.68 5.77 3.79 5.85 3.04GRWATER1 (%) 0.68 0.69 0.67 1.06 0.66 1.01GRWATER2 (%) 11.63 10.01 11.61 11.53 11.50 11.9GRWATER3 (%) 40.77 34.79 40.73 37.33 41.08 32.3GRWATER4 (%) 33.44 31.02 33.40 35.05 33.69 32.0GRWATER5 (%) 13.49 23.49 13.57 65.59 1 5.02 61.94 13.07 22.6AIRP NOISE1 (%) 0.96 2.57AIRP NOISE2 (%) 0.31 8.67 0.29 7.72D RECREATION 16.25 9.50 13.91 8.13T RECREATION 20.36 17.59NOISE (%) 17.60 44.20D HARBOUR 128.60 45.64 101.19 41.88T HARBOUR 96.72 48.34 74.78 51.46D AIRPORT 121.98 37.46 103.80 37.60T AIRPORT 98.24 82.33HIGHWAY (%) 15.33 40.36

relations between location factors and land-use change. Region-specific logit models fornew residential and industrial/commercial area in the different regions are given intables 7 and 8 (over). There are differences in the variables that prove most significantin explaining the pattern of land-use change in the different regions. Groundwater levelis a determinant of the location of new urban areas when the country is analysed as awhole, in the individual regions groundwater level does not add significantly to theexplanation of the land-use pattern. In the individual regions different accessibilitymeasures are best able to explain the new residential locations; in spite of these differ-ences it is clear that new residential areas are preferably located with easy access totowns, cities, and infrastructure. The growth-centre and Green Heart policies have asignificant influence only on the land-use pattern in the western region. It is in thisregion that most growth centres are located. Similar observations hold for the regionalvariability in location factors determining the allocation of new industrial/commercialareas. In all regions new industrial/commercial areas are preferably located in theimmediate neighbourhood of motorways with easy access to consumers and labourforces. The growth centres in the western and northern region attracted industrial/commercial compounds. The variability explained by the regression models was in allregions similar to the level of explanation for the country as a whole.

5.2 Location and neighbourhood characteristicsWe analysed the neighbourhood interactions between land uses through the quantifi-cation of possible interactions between the land-use types for locations with observedchanges during 1989 and 1996. The enrichment factor was calculated for the neighbour-hood of all locations with new residential, industrial/commercial, and recreational areasbased on the neighbouring land uses in 1989. Figure 4 (over) presents the logarithm ofthe enrichment factor as a function of the distance from the location of land-use change.Positive values indicate an enrichment of the neighbourhood of the location with theland-use type while negative values indicate that relatively little of the studied land-usetype is found in the neighbourhood. For all land-use changes, distance-decay functionsare found to indicate a decreasing deviation of the neighbourhood from the average land-use composition. The interaction between residential area and industrial/commercialarea does not follow a straightforward distance-decay function. Although at all distancesa positive interaction between residential and industrial uses is found, the highest valuesare found at distances between 1 and 2 km (d � 2 or 3) from the location itself. Thisindicates a tendency to locate industrial/commercial areas in the neighbourhood of

Table 6. Binary logit estimates of land-use changes between 1989 and 1996 (N �15750; thresholdsfor stepwise regression on 0.02/0.03).

New residential New industrial/ New recreationalarea commercial area area

coefficient exp(b ) coefficient exp(b ) coefficient exp(b )

Constant ÿ1.34 ÿ3.02 ÿ2.58HIGHWAY ± ± 1.05** 2.84 ± ±D TOWN ÿ2:3� 10ÿ4** 0.9998 ÿ2:00� 10ÿ4** 0.9998 ÿ2:40� 10ÿ4** 0.9998D STATION ÿ9:8� 10ÿ5** 0.9999 ± ± ± ±GRCENTRE 0.69** 1.98 ± ± 0.82* 2.27GRHEART ÿ1.45** 0.23 ± ± ± ±

ROCa 0.81 0.80 0.75

* significant at 0.05 level; ** significant at 0.01 level.a ROCÐrelative operating characteristic statistic.


Determ

inantsof

land-usechange

patternsin

theNetherlands

143Residential area

Industrial/commercial area

Observed newarea between 1989and 1996

Probability ofnew area (withoutneighbourhood)

Probability ofnew area (withneighbourhood)

Observed newarea between1989 and 1996

Probability ofnew area (withoutneighbourhood)

Probability ofnew area (withneighbourhood

Figure 3. Observed changes inresidential and industrial/com-mercial land use between 1989and 1996 and probabilities basedon the logistic regression modelwithout and with neighbourhoodfactors.

residential areas, but preferably in different compounds. The positive autocorrelationfor residential and industrial uses indicates that urban growth (reinforcement of existingcities) is more important than new developments far from existing residential areas. Thenegative association of new residential and industrial developments with grassland,arable land, and forest in the immediate neighbourhood confirms this observation.New recreation areas are positively associated with forest/nature in the neighbourhood,indicating that part of the new recreational facilities is created near to forest/nature areasthat also have a recreational function. Other recreational areas are located in theimmediate neighbourhood of residential areas providing facilities, such as sports fields,to the inhabitants.

Based on the results of this exploration we selected, for each of the studied land-usechanges, a number of interactions which were assumed to have a causal effect on the land-use allocation. These factors were included in the logistic regression models of table 6 thatdescribe the location of land-use change based on location characteristics.We included the

Table 7. Residential area: estimates for the coefficient of a logit model of changes between 1989and 1996 in different regions.

Whole country Western region Northern region Southern region

Constant ÿ1.34 ÿ1.39 ÿ0.62 0.57D TOWN ÿ2:3� 10ÿ4** ÿ2:7� 10ÿ4** ± ÿ4:4� 10ÿ4

T TOWN ± ± ÿ0.24** ±T WORK1 ± ± ± ±D WORK2 ± ± ± ÿ3:0� 10ÿ5**D STATION ÿ9:8� 10ÿ5** ÿ6:9� 10ÿ5** ± ±T STATION ± ± ÿ0.09** ±T HIGHWAY ± ± ± ±GRCENTRE 0.69** 0.70** ± ±GRHEART ÿ1.45** ÿ1.19** ± ±

ROCa 0.81 0.82 0.81 0.81

** significant at 0.01 level.a ROCÐrelative operating characteristic statistic.

Table 8. Industrial/commercial area: estimates for the coefficient of a logit model of changesbetween 1989 and 1996 in different regions.

Whole country Western region Northern region Southern region

Constant ÿ3.02 ÿ4.08 ÿ4.29 ÿ1.107ALTITUDE ± 0.09** ± ±HIGHWAY 1.05** 1.12** 0.81** 0.513**D TOWN ÿ2:3� 10ÿ4** ± ÿ3:5� 10ÿ4** ±T TOWN ± ÿ0.10** ± ÿ0.233**T WORK1 ± ± 3:0� 10ÿ4** ±D WORK2 ± ÿ8:0� 10ÿ5** ± ±T CITY ± 0.11** ÿ0.126** ±D STATION ± ± ÿ7:5� 10ÿ5** ±D AIRP ± ± ± ÿ5:8� 10ÿ6**GRCENTRE ± 0.89** 3.317** ±

ROCa 0.80 0.75 0.82 0.75

** significant at 0.01 level.a ROCÐrelative operating characteristic statistic.


log(enrichmentfactor)

log(enrichmentfactor)

0.9

0.7

0.5

0.3

0.1

ÿ0.1

ÿ0.3

0.7

0.5

0.3

0.1

ÿ0.1

ÿ0.3

ÿ0.5

0.7

0.5

0.3

0.1

ÿ0.1

ÿ0.30 5 10 15 20

0 5 10 15 20

Neighbourhood size (d )

Neighbourhood size (d )

(a) (b)

(c)

ed

a

b

c f

g

gd

e

f

a

b

c

d

e

a

b

f

c

g

Other agriculture (a)

Grassland (b)

Arable land (c)

Residential (d)

Industrial (e)

Forest/nature (f )

Recreation (g)

Figure 4. Neighbourhood characteristics (log enrichment factor) as a function of the neighbour-hood size (d: number of grid cells of 500 m each) for locations converted into (a) industrial/commercial land, (b) recreational area, and (c) residential area between 1989 and 1996.

Table 9. Estimation results of logit models including neighbourhood characteristics. Only thedirection and significance of coefficients for the added neighbourhood variables are presented.

New residential New industrial/ New recreationalarea commercial area area

Residentialclose (INFL RES1) �*** ÿ �***distant (INFL RES3) ÿ �* �

Industrial/commercialclose (INFL INDUST1) � �***distant (INFL INDUST3) ÿ �

Recreationalclose (INFL REC1) �***distant (INFL REC3) �***

Forest/natureclose (INFL FOREST1) �**distant (INFL FOREST3) �**

ROC (without neighbourhood 0.81 0.80 0.75interaction)

ROC (with neighbourhood interaction) 0.90 0.91 0.85

* significant at 0.1 level, ** significant at 0.05 level, *** significant at 0.01 level.


influence of the immediate neighbourhood (d � 1; 500 ^ 700 m) and the neighbourhoodover which the interaction between industrial/commercial and residential areas is maximal(d � 3; 1500 ^ 2100 m). The influence of industrial and residential locations over largerdistances is captured by the accessibility measures. For new residential land and for newindustrial/commercial land we included the neighbourhood enrichments with residentialand industrial/commercial areas and for new recreational areas we included the enrich-ment with residential, recreational, and forest/nature areas. Table 9 summarises the resultsof adding these neighbourhood characteristics to the regression models. Not all neigh-bourhood factors have a significant contribution to the model. However, in all cases thevariability explained by the model increased considerably as indicated by the ROCvalues. Figure 3 shows that including the neighbourhood characteristics also leads tobetter probability maps for the location of land-use change.

6 Discussion and conclusionThe logit models for the location of land-use change estimated in this paper indicatewhich processes are important to land-use change in the Netherlands. Mostly, the resultscorrespond with our general understanding of the processes of land-use change in theNetherlands. The results show that the historic land-use pattern in the Netherlands can,in considerable part, be explained by the conditions of soil and landform. In particular,the location of agricultural land use and areas left to forest or nature can be explainedby the suitability of the soil for agricultural purposes. Recent land-use conversions,studied for the period 1989 ^ 96, are no longer related to the biophysical properties of alocation. Accessibility, spatial policies, and neighbourhood interactions are more impor-tant determinants of current land-use changes. These factors are, to some extent, relatedto the historic developments in land-use pattern. Recent developments in land use,therefore, follow a process of self-organisation in which centripetal forces lead to thegrowth of existing residential centres.

Residential land-use patterns as observed in 1989 are not explained very well bythe spatial variation in location characteristics. We may well have missed some of theimportant determinants of the suitability of the different locations for residential landuse. Furthermore, it is clear that location characteristics do not fully explain land-usepatterns: geography can engage in a process of self-organisation in which locationswith seemingly identical potential end up playing very different roles (Krugman, 1999;Venables, 1999). However, the most important reason for the low level of explanation ismost likely that a static analysis over a long time span is not appropriate to representthe temporal dynamics leading to land-use change. The method introduced, therefore,is much more appropriate for analysing the determinants of short-term changes in land-use pattern. At this time scale the effect of self-organisation can be analysed byincluding variables that are related directly to the process of self-organisation (neigh-bourhood interactions, accessibility). In our analysis these variables turned out to besignificant contributors to the explanation of the geographical pattern of land-usechange with relatively high levels of explanation of the spatial variability. This resultindicates that land-use-change analysis based on neighbourhood characteristics andaccessibility change is relevant and an important topic for further research. In modelapplications cellular automata might be suitable tools to simulate these land-use-changeprocesses.

The results of the analysis in this paper indicate that policies have an importantinfluence on land-use patterns in the Netherlands. In many studies they are not givenexplicit attention because they are difficult to include in a quantitative assessment.Our study indicates that omitting policy variables might cause an incomplete assess-ment. In the Netherlands the growth centres have catered for a large part of the new


residential developments. In the period of study, 1989 ^ 96, the official growth-centrepolicy ended and most attention was given to reinforcement of the larger cities, usingopen spaces within the city or on the edge of the cities as preferential places for newresidential and industrial use (Faludi and Van der Valk, 1990; 1994). This policy is visiblein the results of the analysis through the importance of parameters that represent thedistance to the cities. However, it is impossible to determine, based on the analysismade in this paper, if these relations are a direct effect of policy intervention or aconsequence of other socioeconomic processes that cause a concentration of residentialand industrial/commercial areas.

The analysis for the Netherlands can be summarised by the observation that multiplefactors, derived from different theoretical frameworks, are needed to explain the patternof land-use change. One single disciplinary theory would have been insufficient toaddress the complexity of the system adequately.

The methods used in this paper enable a simple straightforward analysis of changesin land-use pattern in relation to the geographic variability in location and neighbour-hood characteristics. The results help us to understand and reveal the processes thatunderlie the land-use-change decisions. However, empirical relations do not straight-forwardly translate into processes: different processes can lead to the same spatialpattern and the same process can lead to different spatial patterns, for example, byamplification of chance events. Empirical analysis can suggest mechanisms for andquantify the relation between land use and proximate factors. In-depth research, basedon the results of the empirical analysis, is necessary to reveal causality. Examples ofin-depth studies on specific determinants of land-use change for the Netherlands areprovided by Dieleman et al (2002) and van der Vlist et al (2002). In-depth researchshould also explicitly address the issue of spatial scale. It is uncorrect to assume thatresults from an analysis based on area-level aggregate values apply at the individuallevel. Incorrect cross-level inferences, the so-called ècological fallacy' (Robinson, 1950;Tranmer and Steel, 1998), are a common phenomenon in studies of complex systems.

The method applied in this paper can be of prime importance to modellers develop-ing decision-support systems for land-use planning and policy. Many land-use modelsuse a single set of factors, assumed to be of importance in the specific case study, toforecast the location of future land-use conversions. Mostly, the selection of factorsand processes assumed to be important is made without case-specific analysis andjustification (Torrens and O'Sullivan, 2001). This can cause a high level of uncertaintyin land-use projections because of the lack of causality in the underlying relations onwhich the extrapolations are based. When models are to be actively used in impactanalysis of future land-use changes, planning, and policymaking, theoretic understand-ing and empirical testing should underlie the specification of the simulation model. Forspecific case studies the method presented can be used to identify which processes andfactors are important for the case study and time-period studies. Such information isa suitable start to specifying variables and identifying functional relations in a simula-tion model of land-use change. In this sense the method presented can contribute toimprove models of land-use change and decision support.

ReferencesAlonsoW, 1964 Location and Land Use (Harvard University Press, Cambridge, MA)Angelsen A, Kaimowitz D, 1999, ` Rethinking the causes of deforestation: lessons from economic

models'' TheWorld Bank Observer 14 73 ^ 98Anselin L, 1988 Spatial Econometrics: Methods and Models (Kluwer, Boston, MA)Arthur W B, 1994, ` Urban systems and historical path dependence'', in Increasing Returns and

Path Dependence in the Economy Ed. W B Arthur (University of Michigan Press, Ann Arbor,MI) pp 99 ^ 110


Bell K P, Bockstael N E, 2000, Àpplying the generalized-moments estimation approach to spatialproblems involving microlevel data'' The Review of Economics and Statistics 82 72 ^ 82

Bu« rgi M, Turner M G, 2002, ` Factors and processes shaping land cover and land cover changesalong theWisconsin River'' Ecosystems 5 184 ^ 201

Castells M, 2000 The Rise of the Network Society (Blackwell, Oxford)Chomitz K M, Gray D A, 1996, ` Roads, land use, and deforestation: a spatial model applied to

Belize'' TheWorld Bank Economic Review 10 487 ^ 512Christaller W, 1933 Die zentralen Orte in Su« ddeutschland ; published in 1966 as Central Places of

Southern Germany translated by C W Baskin (Prentice-Hall, London)ClarkWAV, 2000, ` Monocentric to polycentric: new urban forms and old paradigms'', in

ACompanion to the City Eds G Bridge, S Watson (Blackwell, Oxford) pp 141 ^ 154de Jong T, Ritsema van Eck J R, 1996, ` Location profile based measures as an improvement on

accessibility modelling in GIS'' Computers, Environment and Urban Systems 20 181 ^ 190de Nijs T, de Niet R, de Hollander G, Filius F, Groen J, 2001, ` De LeefOmgevings Verkenner:

Kaartbeelden van 2030. Een verkenning van de inzet bij beleidsondersteuning [The EnvironmentExplorer: land use maps for 2030. An exploration of the possibilities for policy support], RIVMreport 408505 004, National Institute of Public Health and the Environment, Bilthoven

deWit A J W, van der HeijdenTh G C,Thunnissen H AM,1999, `Vervaardiging en nauwkeurigheidvan het LGN3-grondgebruiksbestand [Development and accuracy of the LGN3 land usedatabase], report 633, DLO-Staring Centrum,Wageningen

Dieleman F M, Dijst M J, Spit T, 1999, ` Planning the compact city: the Randstad Hollandexperience'' European Planning Studies 7 605 ^ 620

Dieleman F M, Dijst M, Burghouwt G, 2002, ` Urban form and travel behaviour: micro-levelhousehold attributes and residential context'' Urban Studies 39 507 ^ 527

Faludi A,Van der Valk A J, 1990 De Groeikernen als Hoekstenen van de Nederlandse RuimtelijkePlanningsdoctrin [The growth poles as cornerstones of Dutch spatial planning doctrine](Van Gorcum, Assen)

Faludi A,Van der Valk A J, 1994 Rule and Order: Dutch Planning Doctrine in theTwentieth Century(Kluwer Academic, Dordrecht)

Faust K, Entwistel B, Rindfuss R R,Walsh S J, Sawangdee Y, 1999, ` Spatial arrangements ofsocial and economic networks among villages in Nang Rong District, Thailand'' SocialNetworks 21 311 ^ 337

Fischer G,Velthuizen H, Shah M, Nachtergaele F, 2002 Global Agro-ecological Assessment forAgriculture in the 21st Century: Methodology and Results (International Institute for AppliedSystems Analysis, Laxenburg, Austria)

Fujita M, Krugman P, Mori T, 1999, ` On the evolution of hierarchical urban systems'' EuropeanEconomic Review 43 209 ^ 251

Geoghegan J,Villar S C, Klepeis P, Mendoza P M, Ogneva-Himmelberger Y, Chowdhury R R,Turner II B L,Vance C, 2001, ` Modeling tropical deforestation in the Southern YucatanPeninsular Region: comparing survey and satellite data''Agriculture, Ecosystems andEnvironment 85 25 ^ 46

Geoghegan J,Wainger LA, Bockstael N E,1997, ` Spatial landscape indices in a hedonic framework:an ecological economics analysis using GIS'' Ecological Economics 23 251 ^ 264

He H S,Ventura S J, Mladenoff D J, 2002, ` Effects of spatial aggregation approaches on classifiedsatellite imagery'' International Journal of Geographical Information Science 16 93 ^ 109

Hohenberg P M, Lees L H, 1985 The Making of Urban Europe 1000 ^ 1950 (Harvard UniversityPress, Cambridge, MA)

Irwin E, Geoghegan J, 2001, ` Theory, data, methods: developing spatially-explicit economicmodels of land use change''Agriculture, Ecosystems and Environment 85 7 ^ 24

Krugman P, 1998, ` Space: the final frontier'' The Journal of Economic Perspectives 12 161 ^ 174Krugman P, 1999, ` The role of geography in development'' International Regional Science

Review 22 142 ^ 161Lambin E F, Turner II B L, 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: movingbeyond the myths'' Global Environmental Change 4 261 ^ 269

Leggett C G, Bockstael N E, 2000, ` Evidence of the effects of water quality on residential landprices'' Journal of Environmental Economics and Management 39 121 ^ 144

Menard S, 1995 Applied Logistic Regression Analysis (Sage, London)


Mertens B, SunderlinW D, Ndoye O, Lambin E F, 2000, ` Impact of macroeconomic change ondeforestation in south Cameroon: integration of household survey and remotely-sensed data''World Development 28 983 ^ 999

Milne B T, Johnson A R, 1993, ` Renormalization relations for scale transformation in ecology''Lectures on Mathematics in the Life Sciences 23 109 ^ 128

Ministerie vanVerkeer enWaterstaat, 2001Atlas Interfererende Netwerken. Historische ontwikkelingvan verkeer- enwaternetwerken [Atlas of interconnected networks: historical development of thetransport and water network] report 110643/CE1/0j0/000089, Uitgevoerd door Arcadis enCarto Studio Amsterdam, Ministerie van Verkeer enWaterstaat Arnhem

Moody A,Woodcock C E, 1994, ` Scale-dependent errors in the estimation of land-coverproportions: implications for global land-cover data sets'' Photogrammetric Engineering andRemote Sensing 60 585 ^ 594

Munroe D, Southworth J, Tucker C M, 2001, ` The dynamics of land-cover change in westernHonduras: spatial autocorrelation and temporal variation'', paper prepared for the 2001American Agricultural Economics Association Annual Meetings, http://www.uncc.edu/dkmunroe/publications.htm

Nelson G C, Harris V, Stone S W, 2001, ` Deforestation, land use, and property rights: empiricalevidence from Darien, Panama'' Land Economics 77 187 ^ 205

Pahl R E, 1975Whose City and Further Essays on Urban Society (Penguin Books, Harmondsworth,Middx)

Penning de Vries F, Teng P, Metsalaar K, 1992 Systems Approaches for Agricultural Development(Kluwer Academic, London)

Pijanowski B C, Brown D G, Shellito B A, Manik G A, 2002, ` Using neural networks and GISto forecast land use changes: a land transformation model'' Computers, Environment andUrban Systems 26 553 ^ 575

Pontius R G, Schneider L C, 2001, ` Land-use change model validation by an ROC method forthe Ipswich watershed, Massachusetts, USA''Agriculture, Ecosystems and Environment 85239 ^ 248

Rabin M, 1998, ` Psychology and economics'' Journal of Economic Literature 36 11 ^ 46Raziei A, EversW, 2001, Àfstemming ruimtelijke gegevens voor de L(eef)O(mgevings)V(erkenner)

en R(uimte)S(canner)'' [Tuning of spatial data for the Environment Explorer and the LandUse Scanner], Interim CIM-rapport M009/01, National Institute of Public Health and theEnvironment, Bilthoven

Ricardo D, 1817 Principles of Political Economy and Taxation reprinted in 1971 (Penguin Books,Harmondsworth, Middx)

RobinsonW,1950, ` Ecological correlations and the behaviour of individuals''American SociologicalReview 15 351 ^ 357

Sonneveld M P W, Bouma J,Veldkamp A, 2002, ` Refining soil survey information for a Dutchsoil series using land use history'' Soil Use and Management 18 157 ^ 163

Steur G G L, HeijinkW, 1991, ` Bodemkaart van Nederland schaal 1: 50 000. Algemene begrippenen indeligen'' [Soil map of the Netherlands scale 1:50 000. Definitions and classification],DLO Staring CentrumWageningen

Swets J A, 1986, ` Measuring the accuracy of diagnostic systems'' Science 240 1285 ^ 1293Thompson D M, Serneels S, Lambin E F, 2002, ` Land use strategies in the Mara ecosystem: a

spatial analysis linking socio-economic datawith landscape variables'', inLinking People, Place,and Policy: A GIScience Approach Eds S J Walsh, K A Crews-Meyer (Kluwer Academic,Boston, MA) pp 39 ^ 68

Torrens P M, O'Sullivan D, 2001, ` Cellular automata and urban simulation: where do we go fromhere?'' Environment and Planning B: Planning and Design 28 163 ^ 168

Tranmer M, Steel D G, 1998, ` Using census data to investigate the causes of the ecological fallacy''Environment and Planning A 30 817 ^ 831

Turner II B L, 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; definitional and spatial-scaleissues in their human dimensions'' Global Environmental Change 1 14 ^ 22

van der Vlist A J, Gorter C, Nijkamp P, Rietveld P, 2002, ` Residential mobility and local housing-market differences'' Environment and Planning A 34 1147 ^ 1164

van Kempen R, Hooimeijer P, Atzema O, Dijst M,Verhoeff R, 2000 Sociale en CultureleOntwikkelingen en Ruimtelijke Configuraties [Social and cultural development in relation tospatial pattern] (Van Gorcum, Assen)


http://www.uncc.edu/dkmunroe/publications.htm

http://www.uncc.edu/dkmunroe/publications.htm

Venables A J, 1999, ` But why does geography matter and which geography matters?'' InternationalRegional Science Review 22 238 ^ 241

Verburg P H, Ritsema van Eck J, Nijs T de,Visser H, Jong K de, 2004, À method to analyseneighborhood characteristics of land use patterns'' Computers, Environment and UrbanSystems in press

Von Thu« nen J H, 1966 Der isolierte Staat in Beziehung der Landwirtschaft und Nationalo« konomie[Isolated state], Ed. P Hall, translated by C M Wartenberg (Pergamon, Oxford)

Walsh S J, Crawford T W, Crews-Meyer K A,WelshW F, 2001, À multi scale analysis of landuse, land cover change and NDVI variation in Nang Rong District, Northeast Thailand''Agriculture, Ecosystems and Environment 85 47 ^ 64

Webster C J,Wu F, 2001, ` Coase, spatial pricing and self-organising cities'' Urban Studies 382037 ^ 2054

White R, Engelen G, 2000, ` High-resolution integrated modelling of the spatial dynamics ofurban and regional systems'' Computers, Environment and Urban Systems 24 383 ^ 400

Wu F, 1999, ` GIS-based simulation as an exploratory analysis for space-time processes'' Journalof Geographical Systems 1 199 ^ 218

Yeh A G, Li X, 2001, À constrained CA model for the simulation and planning of sustainableurban forms by using GIS'' Environment and Planning B: Planning and Design 28 733 ^ 753

ß 2004 a Pion publication printed in Great Britain


Documents

Determinants of land-use change patterns in the Netherlands · 2 Determinants of land-use change Land-use change can be described by the complex interaction of behavioural and structural