spatial explorations of land use change and grain production in china

Agriculture, Ecosystems and Environment 82 (2000) 333–354

Spatial explorations of land use changeand grain production in China

Peter H. Verburga,∗, Youqi Chenb, Tom (A.) Veldkampaa Department of Environmental Sciences, Wageningen University, PO Box 37, 6700 AA Wageningen, Netherlands

b Institute of Natural Resources and Regional Planning, Chinese Academy of Agricultural Sciences,Baishiqiao Road 30, Beijing 100081, China

Abstract

Studies on land use change and food security in China have often neglected the regional variability of land use changeand food production conditions. This study explores the various components of agricultural production in China in a spa-tially explicit way. Included are changes in agricultural area, multiple cropping index, input use, technical efficiency andtechnological change. Different research methodologies are used to analyse these components of agricultural production. Themethodologies are all based on semi-empirical analyses of land use patterns in relation to biophysical and socio-economicalexplanatory factors. The results indicate that different processes and patterns of land use change are found in various parts ofthe country. Large inefficiencies in the use of agricultural inputs and relatively low input use are especially found in some ofthe rural, less endowed, western regions of China, which indicates that in these regions of China increases in grain yield arewell possible. The spatially explicit results might help to focus agricultural policies to the appropriate regions. © 2000 ElsevierScience B.V. All rights reserved.

Keywords:China; Grain production; Land use change; Yield; Technical efficiency; Spatial modelling

1. Introduction

Land use change is central to the interest of theglobal environmental change research community.Land use changes influence, and are determined by,climate change, loss of biodiversity, and the sustain-ability of human–environment interactions, such asfood production, water and natural resources andhuman health. Land use changes not only includechanges in land cover, but also the manner in whichthe land is manipulated and the intent underlying thatmanipulation (Turner II et al., 1995). Manipulation ofland refers to the specific way in which humans use

∗ Corresponding author. Tel.:+31-317-485208;fax: +31-317-482419.E-mail address:[email protected] (P.H. Verburg).

vegetation, soil, and water for the purpose in question,e.g., the use of fertilisers, pesticides, and irrigationfor mechanised cultivation.

Land use change in China is often seen in the lightof its impact on food security. With a rising demandfor agricultural products, as a consequence of popu-lation growth and changing consumption patterns, itis essential for China to increase its food production.In recent years much discussion is devoted to China’sability to maintain food self-sufficiency (Garnaut andMa, 1992; Brown, 1995; Rozelle and Rosegrant, 1997;Lin, 1998). China might become a major importer offood products as a consequence of its losses of agricul-tural land through urbanisation and degradation, andits limited possibilities to increase output per unit area.Brown (1995) takes the most extreme position, claim-ing that China’s grain production will fall in absolute

0167-8809/00/$ – see front matter © 2000 Elsevier Science B.V. All rights reserved.PII: S0167-8809(00)00236-X

334 P.H. Verburg et al. / Agriculture, Ecosystems and Environment 82 (2000) 333–354

terms while demand will rise, creating a shortfall ofmore than 200 million metric tons by 2030. Other au-thors, using more sophisticated analyses, have shownthat shortfalls in production are probably lower (Paarl-berg, 1997; Huang et al., 1999). Recent projections bythe United States Department of Agriculture have nowagain sharply reduced projected growth in China’sgrain import demand (USDA, 2000). The largest un-certainties in the projections of changes in China’sfood economy are found on the supply side of the foodbalance, largely as a consequence of differences inthe analysts’ perception of prospects for technologicalchange and other factors affecting growth of cultivatedland and productivity (Fan and Agcaoili-Sombilla,1997). Almost all of these assessments of the impactof land use changes on food supply in China are basedupon an analysis of the economy at the national level.Such aggregated assessments lack the analysis of re-gional variability in land use change and potentials toincrease production. The highly diverse natural andsocio-economic conditions in China cause land usechange to have differential impacts across the country.Spatially explicit assessments are needed to identifyareas that are likely to be subject to dramatic land usemodifications in the near future. Such information isespecially important for land use planners since, inorder to focus policy interventions, one needs not onlyto know the present rates and localities of land usechange, but also to anticipate where conversions aremost likely to occur next. Such predicative informa-tion is essential to support the timely policy response.

This paper studies potential near-future changes inChina’s agricultural production from a spatially ex-plicit perspective. The different components of landuse change influencing grain production, are studiedin order to identify potential ’hot-spots’ of change.Regional variability is related to its explanatory fac-tors, to identify the processes underlying the changesin land use.

2. Methodology and data

2.1. Overview

Changes in agricultural production result fromchanges in one or more of the components of theland use system. Changes in agricultural production

can either originate from changes in the sown areaor from changes in the production level per unit ofland (yield). For regions where it is possible to sowthe land more than once a year, changes in sown areacan be divided into changes in agricultural area andchanges in the multiple cropping index. Changes inyield can be subdivided into three components. First,the traditional source of growth stems from increasesin agricultural inputs, e.g., irrigation and fertilisers.The second source of growth comes from increases inthe efficiency of production. Increases in productionefficiency make more output available with the sameamount of inputs. Institutional innovations can be animportant source of efficiency growth, as these elimi-nate restraints in resource allocation. The third sourceof growth is technological change, which shifts theproduction function upward. So, similar to efficiencyincreases, more outputs become available out of thesame amount of inputs. New, improved varieties canbe an important source of technological progress.Fig. 1 shows the thus derived five sources of changein agricultural production. These different sourcesof change in agricultural production all have theirown drivers and constraints. Therefore, the differentsources of change in production have been analysedseparately by different methodologies, described inmore detail hereafter. Some of these methodologieshave been described elaborately in other papers, soonly the main characteristics are repeated here. Allthese methodologies have in common that they usedata on the spatial variability of production conditionsand (proximate) driving forces to analyse the dynam-ics of the land use system. Another similarity betweenthe methodologies is the use of statistical methodswhich are used to establish relationships between thepattern of land use and its supposed driving factors.The following paragraphs describe the data sets usedin the study and the different methodologies.

2.2. Data

A nation-wide, spatially explicit database has beenused containing biophysical, demographic and agri-cultural data. The data set is based on a large setof maps and statistical data gathered from varioussources. Demographic and agricultural data are basedupon statistics for 1986, 1991 and 1996 linked to ad-ministrative units (county-level). Soil, geomorphology

P.H. Verburg et al. / Agriculture, Ecosystems and Environment 82 (2000) 333–354 335

Fig. 1. Land use change components that determine changes in grain production.

and climatic characteristics are based on maps andinterpolated climate data respectively, land cover datawere unfortunately only available for 1986 and 1991whereas data on agricultural inputs and productionare also available for 1996. Therefore, the base-yearfor all calculations is 1991, whereas 1986 and 1996data are used to capture the temporal dynamics. Thedata and their various sources are summarised in Ap-pendix A of this paper and described in more detail inVerburg and Chen (2000). All data are converted intoa regular grid to match the representation of the dif-ferent data and facilitate the analysis. The basic gridsize, to which all data are converted, is 32× 32 km(∼1000 km2), which equals the average county sizein the eastern part of China. There has been consider-able discussion about the reliability of Chinese landuse statistics, especially with respect to the amount ofcultivated land (Crook, 1993). Official statistics havealways underestimated the area of cultivated land.However, the cultivated area in the agricultural surveywe have used, equals 133 million hectares, which cor-responds with the area generally assumed to be reli-able (Alexandratos, 1996; Smil, 1999). For grain yieldwe had to use official statistics. Underreporting ofarable land in official statistics has led to inflated esti-mates of grain yields and the appearance that China’syields are high by world standards. Statisticians haveadmitted that they have overstated grain yields to com-pensate for the underreported land area. They rely onsample survey cuttings to determine actual yields, andthen inflate them 20–30% (Crook and Colby, 1996).The same overestimation holds for fertiliser appli-

cation rates as these are also calculated with arableareas derived from official statistics. Unfortunately, itis not possible to correct for the unreliability of theChinese land use statistics. Any calculation presentedin this paper should therefore be seen in the light ofthis problem. However, we believe that it is still pos-sible to use the data to indicate ‘hot-spots’ of changeand make conclusions on the relative importance ofthe processes throughout the country.

2.3. Methodologies

2.3.1. Changes in agricultural areaChanges in agricultural area can result from the

reclamation of land, e.g., forest and grassland, intoagricultural land as well as from the conversion ofagricultural land into other land use types through,e.g., encroachment of cities or desertification. Whereasthe former conversion is constrained by availabilityof suitable land resources, the latter is the result ofcompetition between land use types. To explore thesechanges in agricultural area for China a dynamic, spa-tially explicit land use change model is used. Thismodel, the CLUE (Conversion of Land Use and itsEffects) modelling framework, has been described inmore detail in Verburg et al. (1999a) whereas the ap-plication of the model for China has been described inVerburg et al. (1999b). Therefore, in this section onlythe main characteristics of the modelling frameworkand the scenarios used in this paper are mentioned.

Fig. 2 gives an overview of the structure of themain components of the CLUE modelling framework.


Fig. 2. General structure of the CLUE modelling framework for the spatially explicit calculation of changes in the land use change pattern.

The demand module contains, at the national level,scenarios for changes in demand for different land usetypes. In this study these scenarios are entirely basedon previous studies published in literature. Most ofthese studies are based on the demand for agricul-tural products, taking into account population growth,changes in diet and import/export quantities. The pop-ulation module calculates changes in population andassociated demographic characteristics based uponprojections and historic growth rates. The central partof the model is the allocation module. This modulecalculates for all grid cells the changes in land useon a yearly basis. The allocation itself is based upona spatial analysis of the complex interaction betweenland use, socio-economic conditions and biophysicalconstraints. The Appendix A gives an overview of thedemographic, socio-economic, soil related, geomor-phologic and climatic variables evaluated in the anal-ysis for China. All these variables can, as is knownfrom literature (Turner II et al., 1993), influence thedistribution of land use. However, not in all situationsall of these variables will add a significant contribu-tion to the explanation of the land use distribution.Therefore, a stepwise regression procedure is used tostraightforwardly select variables with a significantcontribution.

The allocation module uses these spatial relations tocalculate changes in relative land cover, given either a

change in one of the determining factors, e.g., changesin population density or urbanisation, or a change incompetitive advantage upon a change in demand at thenational level. For every year the allocation modulecalculates changes, until the demand for the differentland use types is equalled. For every grid cell alloca-tion is constrained by the total land area in the gridcell, so that competition between the land use types isexplicitly taken into account.

The land use types included in the CLUE modelfor China are cultivated land, horticultural land, for-est, grassland, built-up land and unused land (mainlydesert). In addition, a nested model run is made tosimulate the relative occupation of agricultural landby different crops or groups of crops. We have mod-elled rice, wheat, corn, other grain crops, cash crops,vegetables and other crops separately.

The scenario used for the simulations in this paperis a baseline scenario of which the demands for thedifferent land use types and population growth as-sumptions are presented in Table 1. These demandsare based on an extrapolation of recent trends andestimates of land use change at the national levelfor the period 1989–2000 by Smil (1993). The pop-ulation growth rates, which stem from internationalprojections, are differentiated throughout the countrybased on observed growth rates between 1986 and1996. Also income was assumed to grow steadily


Table 1Baseline scenario for CLUE simulations

Demand for land cover types Area in 1991 (million hectares) Yearly change (1991–2010; 1000 ha)

Cultivated land 132 −574Horticultural land 5 182Forest 195 107Grassland 256 0Built-up land 25 221Water 34 0Unused 296 64

Population growth assumptions Value Source of projection

Total population 17% increase (1991–2010) US Census (1997)%Urban population 27% (1991) to 43% (2010) UN (1995)%Rural labour force of rural population Constant Shen and Spence (1996)%Agricultural labour force of rural labour force 78% (1991) to 65% (2010) Trend and author’s estimates

with growth rates proportional to the growth ratesobserved in the grid cells between 1986 and 1991.

The demand for the different crop types is basedon projections to 2007 by the United States Depart-ment of Agriculture (USDA, 1998) which providesyearly projections for all major crops. No projectionsare given for potato (included in the other grain cate-gory), some cash crops, vegetables and crops groupedin the category other crops, e.g., fodder and medici-nal crops. These crop projections are made based ontrends and literature (e.g., Lin and Colby, 1996).

China is expected to become a large vegetableexporter as China has a comparative advantage in

Fig. 3. Baseline scenario for changes in the relative share of different crop types used in the CLUE simulations between 1991 and 2010.

producing vegetables for export, primarily becauseof its abundant rural labour resources (Crook, 1996).As China invests in transportation and storage infras-tructure and as firms improve grading and packagingstandards, China is likely to become a fierce com-petitor in world vegetable markets (Guoqian, 1997;Lu, 1998). Domestic vegetable consumption is alsoexpected to increase due to rather high income elas-ticity for vegetables, especially in the rural areas(USDA, 1998; Han and Wahl, 1998). Therefore, wehave assumed a doubling of the area sown to veg-etables between 1991 and 2010. Fig. 3 summarisesthe scenario for the different crops types. Up to 1997


observed data have been used, which can be seen fromthe fluctuations in relative harvested area.

2.3.2. Cropping indexThe cropping index denotes the number of times a

year that a piece of land is sown to a crop. The possibil-ities for multiple cropping are constrained by climaticconditions and water availability. The temperature inwinter is in a large part of China too low to supportcrop growth. Other parts, with more favourable tem-peratures, but with a distinct dry season, lack the waterresources needed for crop growth unless irrigation isavailable. To understand the spatial variability in crop-ping index it is therefore essential to study both theactual and the potential cropping index. Agro-climaticand crop growth models can be used to calculate thevariation in potential cropping indices throughout thecountry (Cao et al., 1995). However, this requiresdetailed information on crop growth and hydrologi-cal conditions. Because at many places China’s crop-ping systems already operate at the maximum possi-ble cropping index, it is also possible to obtain themaximum cropping indices by an analysis of the ac-tual cropping index with a frontier function approach.Following the stochastic production function literature(Coelli et al., 1998) the following model was specified:

CIi = f (Xi, β) + vi − ui (1)

Fig. 4. Schematic representation of frontier function approach: (A) frontier function as compared to ordinary least squares (OLS) function;(B) observed multiple cropping index (CI) versus stochastic frontier value of the cropping index (adapted from Battese, 1992).

where CIi is the average cropping index in theith gridcell, Xi denotes the vector with climatic and hydro-logical factors determining the cropping index,vi is arandom error term which is assumed to be identicallyand independently distributed asN(0, σ 2

v ), andui arenon-negative truncations of theN(0, σ 2

u ) distribution(i.e., half-normal distribution). The frontier functionis represented byf(Xi , β), and is a measure of themaximum cropping index for any particular vectorXi . Both vi and ui cause the actual cropping indexto deviate from this frontier. The random variability,e.g., measurement errors or temporary constraints, isrepresented byvi . The non-negative error termui rep-resents deviations from the maximum potential crop-ping index attributable to inefficiencies. Inefficiencymeans in these circumstances a non-optimal use of theagro-climatic conditions. The basic structure of thestochastic frontier model is depicted in Fig. 4 for a hy-pothetical relation between CI andX. Fig. 4A indicateshow the frontier function might relate to an ordinaryleast squares function (OLS) while Fig. 4B focuses onthe condition of two grid-cells, represented by 1 and2. In grid-cell 1 a cropping index CI1 is found for theagro-climatic conditionsX1. The stochastic frontiercropping index, CI∗1, exceeds the value of the frontierfunction, f(X1, β), because the random error,v1, ispositive. The cropping index in grid-cell 2, CI2, underconditionsX2, has a corresponding stochastic frontier


cropping index, CI∗2, which is less than the value ofthe deterministic frontier function,f(X2,β), becausethe random error,v2, is negative. The efficiency inan individual grid-cell is defined in terms of the ratioof the observed cropping index to the correspond-ing stochastic frontier cropping index, conditional onagro-climatic conditions. Thus, the efficiency in thecontext of the stochastic frontier function is

TEi = CIiCI∗i

(2)

where TEi is the efficiency in grid celli.In this study the vectorX consists of the (long-term

average) yearly temperature (TMPAVG), the differ-ence in temperature between the warmest and coldestmonth (TMPRNG), the number of months that the av-erage monthly temperature is above 10◦C (TMP 10C),the number of months that more than 50 mm rain iscollected (PRC50M), the average percentage of sun-shine (SUNTOT) and the fraction of all cultivatedland in a certain grid cell that is irrigated (IRRI). Thelogarithm of the cropping index was used, because itresulted in a significantly better model fit. The frontierfunction is estimated for the three years that data wereavailable with Maximum Likelihood procedures usingFRONTIER4.1 software (Coelli, 1994). The actualchanges in cropping index between 1986 and 1996are used to evaluate the relevance of the calculatedinefficiencies for future changes in cropping index.

2.3.3. Agricultural inputsFarming systems can be characterised by their in-

puts and outputs. Therefore, each grid cell is classifiedby the average farming intensity calculated from ananalysis of agricultural inputs and outputs. A disjointcluster analysis is used to summarise the differentinputs and outputs into groups of farming systems.In the cluster analysis grain yield (YGRAIN), chem-ical fertiliser input (FERT), irrigation (IRRI), labouravailability (LABOUR), mechanisation (MACH) andmanure application (MANURE) are used to charac-terise the farming system groups. To understand thespatial distribution of the distinguished farming sys-tems, we have studied the spatial differences of theindividual inputs and the farming system groups as awhole in relation to a number of environmental andsocio-economic factors. The relation between grainyield and agricultural inputs is studied by fitting a

production function with a Cobb–Douglas functionalform given by

ln(Yi) = β0 + β1 ln(FERTi ) + β2 ln(IRRIi )

+β3 ln(LABOURi ) + β4 ln(MANUREi )

+β5 ln(MACHi ) (3)

2.3.4. Production efficiencyThe notion of stochastic frontier functions and effi-

ciency, as described in Section 2.3.2, originates fromeconomic literature where frontier functions are usedto determine the efficiency with which a firm producesa certain output given the level of inputs (Farrell, 1957;Battese, 1992; Bravo-Ureta and Pinheiro, 1993). Thesame holds for agricultural firms, or groups of agri-cultural firms in a certain area. Output, in this studydefined as grain production per unit area (Yi), is afunction of agricultural inputs and the efficiency withwhich these inputs are used. Therefore, we can write

ln(Yi) = f (Xi, β) + vi − ui (4)

where f(Xi ,β) presents the frontier production func-tion andXi denotes the vector of agricultural inputssimilar to the Cobb–Douglas function (Section 2.3.3).vi is the random variability in production and is iden-tically and independently distributed asN(0, σ 2

v ).The non-negative error termui represents deviationsfrom the frontier output attributable to technical in-efficiency. Instead of giving these deviations a fixeddistribution, as we do in the function for the crop-ping index, we follow the specification of Batteseand Coelli (1995) where the inefficiency effects (µi)are expressed as an explicit function of a vector ofgrid-cell specific variables and a random error. There-fore, ui is assumed to be independently distributed astruncations at zero of theN(µi, σ

2u ) distribution. For

this study the inefficiency function is defined by

µi = δ0 + δ1 TMP AVG i + δ2 PRCTOTi

+δ3 SOILi + δ4 MEANELEVi + δ5 DISTCITYi

+δ6 ILLIT i + δ7%AGLFi + δ8 INCOMEi

+δ9 EROSIONi (5)

All variables in this function are assumed to bedeterminants of the inefficiency in grain produc-tion. Climatic conditions (TMPAVG; PRC TOT) areincluded because it is hypothesised that the large


differences in climate over the country will influencecrop growth and therefore the efficiency in grain pro-duction. Soil fertility (SOIL) is assumed to be impor-tant because in areas with a high natural soil fertilityless fertiliser is needed to obtain the same crop yield.Furthermore, the elevation (MEANELEV) is expectedto be negatively related to the efficiency as ruggedterrain asks for relatively more labour and hamperscultivation practices. A second set of variables relatesto the socio-economic conditions. Agriculture in areaswhich are relatively distant from cities (DISTCITY)is assumed to be hampered due to poor access to in-formation and appropriate inputs, causing inefficien-cies. In a similar way it is assumed that the illiteracylevel (ILLIT), representative for the received educa-tion, influences efficiency. The percentage of the totalpopulation that is part of the agricultural labour force(%AGLF) is supposed to be indicative for the oppor-tunities for off-farm labour, which is thought to berelated to the efficiency of grain production. Averageincome (INCOME) is supposed to influence efficiencyby the ability the farmer has to invest in his land (e.g.,terracing) and the possibilities to buy a balanced setof appropriate inputs. The extent and impact of wa-ter erosion (EROSION) is assumed to be negativelyrelated to efficiency as erosion can damage the crops,remove nutrients and have a negative effect on soilstructure.

The frontier function and efficiencies are againcalculated with the use of Maximum Likelihoodprocedures through FRONTIER4.1 software (Coelli,1994).

2.3.5. Technological progressTechnological progress can be defined as a shift of

the production function upwards. So, out of the sameamount of inputs more output becomes available. Thisshift can easily be defined by including a time trendvariable in the following production function:

ln(Yi) = βt t + f (Xi, β) (6)

wheret is a time trend variable. The production func-tion is calculated with data for the years 1986, 1991and 1996. Data for years in between, which wouldmake the estimates of the coefficient for the time trendvariable more robust, are not available.

3. Results and interpretations

3.1. Changes in agricultural area

The results of the simulation of changes in the agri-cultural area with the CLUE model are presented inFig. 5. Fig. 5A shows the spatial distribution of culti-vated land in 1991, the reference year, as derived fromthe agricultural survey. The distribution of changes incultivated area as simulated for the period 1991–2010are indicated in Fig. 5B. The results suggest thatalthough the total extent of the cultivated area de-creases, some regions showing an increase can still bedistinguished. A more extended analysis of a similarmodel run, as presented by Verburg et al. (1999b),indicates that in different areas different land use con-versions are responsible for the decrease in cultivatedarea. Degradation of arable land causes large lossesof cultivated land on the Ordos Plateau of Inner Mon-golia and on the Loess Plateau. These are areas thatare well known for their marginal agriculture and sus-ceptibility to degradation (Smil, 1993; Qinye et al.,1994; Liu and Lu, 1999). Other ‘hot-spots’ of land usechange can be found in the main agricultural regionsof east, central, south and southwest China as resultof the expansion of the built-up and horticultural area.In these regions large losses of agricultural land arefound around all major urban areas, especially thelarger area surrounding Shanghai.

Fig. 5C displays the relative importance of graincrops in the cultivated area for 1991. Grain crops area more important crop in the northern and westernparts of China than in the central, southern and easternparts. In the southern and eastern parts of China cashcrops make up an important part of the cropping sys-tem. Decreases in the share of the area devoted to graincrops are found throughout the whole country. Promi-nent areas are, however, the urban regions of easternand central China, where large increases in vegetablearea and other cash crops are expected as a result ofthe growing demand for these products by the urbanresidents.

3.2. Cropping index

Table 2 presents the results of the calculation ofthe frontier function for the cropping index as for1986, 1991 and 1996. Higher cropping indexes are


Fig. 5. (A) Cultivated land in 1991; (B) predicted changes in cultivated area between 1991 and 2010 by the CLUE model; (C) percentageof grain of harvested area in 1991; (D) changes in total grain sown area between 1991 and 2010.

Table 2Stochastic frontier functions for cropping index in China for 1986 and 1991a

Coefficient 1986 Coefficient 1991 Coefficient 1996

Frontier functionINTERCEPT 0.168 (7.31)∗ 0.367 (15.8)∗ 0.611 (22.8)∗TMP AVG 0.0438 (39.9)∗ 0.0416 (37.4)∗ 0.0369 (25.9)∗TMP RNG 0.00442 (10.5)∗ 0.00253 (5.86)∗ −0.000240 (−0.462)∗TMP 10C −0.0462 (−20.6)∗ −0.0410 (−18.0)∗ −0.0306 (−10.6)∗PRC50M 0.0347 (28.9)∗ 0.0342 (28.1)∗ 0.0279 (18.6)∗SUN TOT −0.00505 (−16.7)∗ −0.00719 (−23.5)∗ −0.00941 (−24.7)∗IRRI 0.00180 (23.6)∗ 0.00126 (15.1)∗ 0.00141 (13.8)∗

Statistical parametersSigma-squared (σ 2

s = σ 2u + σ 2

v ) 0.0375 (26.6)∗ 0.0349 (25.8)∗ 0.0587 (33.4)∗Gamma (γ = σ 2

u /σ 2s ) 0.790 (40.7)∗ 0.723 (30.8)∗ 0.803 (65.3)∗

Log likelihood function 2733 2683 1768LR test of the one-sided error 171.5∗∗ 122.5∗∗ 539.0∗∗

a t-Ratios between the parentheses.∗ Significant at the 0.01 level.∗∗ Significant at the 0.01 level (mixed chi-square distribution).


found at higher temperatures, longer rainy periods anda larger proportion of the cultivated area that is irri-gated. The number of months with temperatures above10◦C and the range in temperature modify this rela-tionship. The frontier functions for the different yearsare very much similar, indicating a stable relation be-tween the agro-climatic variables and the cropping in-dex. The Likelihood Ratio test statistic is calculatedfor testing the absence of inefficiency effects from thefrontier (Coelli et al., 1998). For all three years thevalue is highly significant, hence the null hypothesis ofno inefficiency effects is rejected. The gamma statis-tic is indicative for the partitioning of the deviationsfrom the frontier. A zero-value forγ indicates that thedeviations from the frontier are entirely due to noise,while a value of 1 would indicate that all deviationsare due to inefficiency. The values found in this study(0.72–0.80) indicate that still a considerable propor-tion of the deviations from the frontier in this studycan be attributed to noise.

Fig. 6 presents the results in a spatially explicit way.Fig. 6A and B indicates the actual and frontier crop-ping index in 1991, respectively. The observed patterncorresponds fairly well with the climatic variabilitythroughout the country whereas local variations aremainly due to differences in irrigation. The differencebetween the actual cropping index and the stochasticfrontier cropping index is the inefficiency, which is in-dicated in Fig. 6C. This figure shows that most poten-tial for higher cropping indices is found in the centraland southern regions of China. However, in these ar-eas the cultivated areas are generally small, hence theincrease in sown area upon an increase in croppingindex is small. When the inefficiencies in cropping in-dex are multiplied by the cultivated area the potentialincrease that can be attained in sown area is indicated(Fig. 6D). As a consequence of its extended cultivatedarea, the North China plain area is identified as themain area where considerable increases in sown areaare possible, in spite of the relatively small potentialincrease in cropping index. The total sown area thatcould be gained by increasing the cropping index to itsstochastic frontier value is about 25 million hectares,which would mean an increase of about 12% in sownarea.

The feasibility of increasing grain productionthrough an increase in cropping index can be deter-mined by an analysis of the reasons underlying the

deviations from the frontier cropping index. Correla-tion analysis between the estimated efficiencies andthe available socio-economic and biophysical vari-ables did not result in strong conclusions. The bestrelations were found with the mean elevation (−0.10,significant at 0.01 level) and the available labour perunit area of cultivated land (0.14, significant at 0.01level), meaning that mountainous areas might ham-per high cropping intensities and that labour shortagemight cause less intensive use of the land. Anotherreason for deviations between actual and frontiercropping index might be found in the choice of crops.Some crops, e.g., sugar cane, have a longer growingseason, inhibiting multiple cropping. However forindividual farmers such crops can often make muchhigher profits. This hypothesis is confirmed by asignificant, negative correlation (−0.07) between therelative share of sugar in the cropping system and theefficiency. Because of large demands for high qualityrice farmers increasingly cultivate varieties that havea much longer growing period, also decreasing themultiple cropping index. It was expected that ineffi-ciencies would also be higher near urban centres, withlarge possibilities for off-farm labour. However, noevidence was found in the data. It can be hypothesisedthat regulations by county governments that specifyminimum cropping indices overrule this effect.

The potential for increases in cropping index canbe seen as an indicator for near future increases of thecropping index. In Fig. 6E and F the observed changesin cropping index between 1986 and 1996 are com-pared with the potential for increase in cropping in-dex determined for 1986. If evaluated for individualgrid-cells, the correlation is relatively low (0.34, sig-nificant at 0.01 level). However, the general pattern isvery similar. When evaluated at the aggregated levelof the seven main geographical regions of China, acorrelation between the observed changes in croppingindex and the potential for change of 0.83 is found.This indicates that it is probable that also for longertime periods most increases in cropping index will befound in southern China.

3.3. Agricultural inputs

Four different farming system groups are distin-guished based on the disjoint cluster analysis. Thecharacteristics of the different groups are presented


Fig. 6. (A and B) Cropping index in 1991 and frontier cropping index in 1991; (C and D) inefficiencies in 1991 expressed in croppingindex (C) and multiplied by cultivated area, indicative for potential expansion of sown area (D); (E and F) real changes in cropping indexbetween 1986 and 1996 (E) and difference between actual cropping index and frontier cropping index (inefficiencies) in 1986 (F).


Table 3Mean value of agricultural parameters for farming system groups in China as determined by a cluster analysisa

Farmingsystems group

Numberof cells

Grain yield(kg/ha)

Fertiliserapplication(kg/ha)

Irrigated(%)

Labour (persons/hectare sown area)

Machinecultivated (%)

Manureapplication(ton/ha)

Group 1 61 (1%) 7205 (495) 292 (74) 58 (35) 2.43 (2.47) 63 (21) 8.97 (4)Group 2 1485 (32%) 5089 (565) 218 (84) 65 (23) 3.73 (2.02) 49 (28) 9.59 (6)Group 3 2385 (51%) 3270 (611) 139 (67) 39 (25) 2.58 (1.77) 38 (30) 8.64 (5)Group 4 712 (15%) 1467 (512) 70 (80) 15 (13) 1.14 (1.00) 34 (26) 6.31 (5)

Mean 4643 (100%) 3627 (1404) 156 (91) 44 (28) 2.74 (2.07) 41 (29) 8.59 (5)

a Standard deviations between the parentheses.

in Table 3. The groups indicate the intensity of thefarming systems. Farming system groups 1 and 2 arecharacterised by high inputs and high yields whilefarming system groups 3 and 4 are characterised bylow inputs and low yields. Only labour availabilitydoes not obey the general pattern. Fig. 7 shows thespatial distribution of the different farming systemgroups. From this map it can be seen that the farmingsystem groups have a clear distribution throughout thecountry. The high input farming systems of group 1and 2 are mainly located in the central and eastern part

Fig. 7. Farming system groups as distinguished by cluster analysis; group 1 represents the most intensive land use systems with respectto inputs and outputs whereas group 4 represents the most extensive systems (see Table 3).

of China, whereas the low input farming systems aremainly found in the western region, far from the maincities. Although farming conditions in this region aregenerally less favourable for very intensive farming,as a consequence of higher altitudes and lower pre-cipitation, socio-economic conditions are the maindeterminants of the lower land use intensities. Thiscan be seen from Table 4 which presents correlationcoefficients between the quantity of agricultural in-puts and a number of biophysical and socio-economicconditions. The use of chemical fertiliser, irrigation


Table 4Pearson correlation coefficients between agricultural inputs and a number of biophysical and demographic factors

Chemical fertiliserapplication

Irrigatedarea (%)

Machine culti-vated area (%)

Manureapplication

Agricultural labourforce per sown area

Agricultural population density 0.48∗ 0.39∗ 0.17∗ 0.01 0.47∗Average temperature 0.38∗ 0.46∗ −0.39∗ 0.21∗ 0.70∗Mean elevation −0.34∗ −0.25∗ −0.27∗ 0.22∗ −0.21∗Suitability rice 0.25∗ 0.12∗ 0.36∗ −0.19∗ −0.05∗Percentage of illiterate population−0.28∗ −0.27∗ −0.44∗ 0.11∗ 0.01Distance to city −0.23∗ −0.05∗ −0.33∗ 0.07∗ 0.15∗

∗ Significant at 0.01 level.

and agricultural machinery tends to decrease with thedistance to a city and increasing percentages of illiter-ate population. In areas with higher population densi-ties generally more inputs are used, probably a conse-quence of strong competition for land resources. Thecorrelation of fertiliser and irrigation with the averagetemperature is a result of the location of intensivelymanaged, irrigated rice cropping systems in the south-ern part of China. Because rice cultivation is not asmechanised as the cultivation of other grains, we finda negative correlation coefficient with temperaturefor mechanisation and a positive one for the labouruse intensity. The connection between managementintensity and soil quality is expressed by the positivecorrelation between soil suitability for rice cultivationand input quantities. The distribution of manure appli-cation does not obey the general pattern found. Inputsof manure are mainly determined by the availability ofmanure, and thus the distribution of livestock (Verburgand van Keulen, 1999). Increasing labour intensitiesin agriculture with the distance to city are indicativefor the decreasing opportunities for off-farm labour.

Table 5 presents the Cobb–Douglas productionfunction as was derived for grain yield in 1991. Exceptfor LABOUR, all estimates in the production func-tion have the expected positive sign. The very small,but negative coefficient for agricultural labour forceis somewhat surprising. However, low elasticitiesfor labour are also found in other labour-rich Asiancountries (Huang and Rozelle, 1995). Besides, Bhat-tacharyya and Parker (1999) and Rawski and Mead(1998) have argued that Chinese statistics massivelyoverestimate the number of farm workers. This mightwell explain the negative value of the coefficient forlabour in the production function. Chemical fertiliser

is the most important input factor, followed by manureand irrigation. The elasticity for agricultural mechani-sation is, however, small. These results are consistentwith other studies (Yao and Liu, 1998; Fan, 1997) thatalso found low elasticities for machinery, probably aresult of abundance of cheap agricultural labour.

From the production function it is clear that chem-ical fertiliser is the most important input in Chineseagriculture. Inputs of chemical fertiliser have rapidlyrisen during the recent past. However, increases in fer-tiliser use have not made an equal pace throughout thecountry.

Table 6 presents the changes in fertiliser applicationduring the period 1986–1996 for the different farmingsystem groups. From this table it can be clearly seenthat the largest (absolute) increases in fertiliser useare found in areas that already have a high fertiliseruse. So, in spite of the doubling of fertiliser use inareas classified as farming system groups 3 and 4, the

Table 5Production function for grain yield in 1991a

Variable Parameter Average function

Production functionIntercept β0 5.450 (132)∗FERT β1 0.390 (44.0)∗IRRI β2 0.133 (18.7)∗LABOUR β3 −0.023 (−2.91)∗MANURE β4 0.132 (13.4)∗MACH β5 0.012 (2.56)∗∗F-statistic model 1430∗Adj. R2 0.61

a Values in parentheses are thet-ratios of the estimates.∗ Significant at the 0.01 level.∗∗ Significant at the 0.05 level.


Table 6Input use for the farming system groups in China as determined by a cluster analysisa

Fertiliser application(kg/ha)

Change in fertiliserapplication 1986–1996(kg/ha)

Change in fertiliserapplication 1986–1996(%)

1986 1991 1996

Group 1 232 (60) 292 (74) 378 (108) 146 63Group 2 166 (60) 218 (84) 290 (139) 124 75Group 3 101 (50) 139 (67) 205 (139) 104 104Group 4 44 (32) 70 (80) 99 (70) 56 127

a Standard deviations between the parentheses.

differences in fertiliser use between the groups haveincreased. In the areas of farming system groups 3 and4 both grain transportation and grain storage facilitiesare very backward and limited, the rural populationnot only has to produce food locally, but it also has todevote most of the arable land to food production forsurvival (Lin and Wen, 1995). So, under the presenteconomic and infrastructural conditions in these areas,individual households are not able to increase theiragricultural inputs (Lin and Li, 1995). Another reasonfor lower increases in fertiliser use in the areas offarming system groups 3 and 4 can be the allocationpolicy of subsidised fertiliser. The largest categoryof allocation is the ‘procurement-linked fertiliser’, itsuniform price nation-wide being set by the centralgovernment. The central leadership allocates this fer-tiliser, which, as its name suggests, is directly linkedto the quantity of state crop procurement. As a result,most of the subsidised urea flows into prosperousareas, where the state purchases most of its grain, cot-ton and oilseed crops. In contrast, farmers in poorerareas receive little or no subsidised urea because theyhave few surplus crops (Ye and Rozelle, 1994). Morerecently these subsidies have largely been dropped.

Given the functional form of the production func-tion and the already high levels of fertiliser use, di-minishing returns upon further increases of chemicalfertiliser input in the areas of farming system groups1 and 2 can be expected. Furthermore, excessive useof agricultural chemicals has already caused severedamages to the natural environment, e.g., groundwaterpollution and deterioration of soil fertility (Jin et al.,1999; Smil, 1993). The low inputs and yields in largeparts of western China suggest that there is still avast potential for raising grain output by using more

land-augmenting inputs such as fertilisers and irriga-tion in these medium and low yield regions.

3.4. Production efficiency

The parameters that are calculated for the fron-tier production and inefficiency function are given inTable 7. The values of the frontier production func-tion are very similar to those of the average productionfunction (Table 5), indicating that the frontier func-tion consists of a near-neutral upward shift of the av-erage function. The diagnostic statistics indicate thata relatively large part of the variation is explained byinefficiency effects whereas the likelihood-ratio statis-tic is highly significant, rejecting the hypothesis of noinefficiency effect in grain production in China.

All coefficients in the inefficiency function are sig-nificant. The relation with precipitation is as expected:the more precipitation, the smaller the inefficiency.Although higher temperatures might enhance cropgrowth they are associated with higher inefficiencies.Also the coefficient for soil fertility does not obey ourexpectations, grid-cells with large areas of fertile soilshave generally low efficiencies. So, natural soil fer-tility does not seem to enhance production efficiency.This can probably be attributed to the relatively smallcontribution of natural soil fertility to total nutrientsupply in intensive farming systems. Elevation hasthe expected positive contribution in the inefficiencyfunction. Illiteracy and distance to city, proxies forthe access to information and means of production,both hamper efficiency. This result corresponds withfindings by Huang and Kalirajan (1997) for a numberof China’s provinces. The share of agricultural labourforce of the total population has a positive influence


Table 7Stochastic frontier production function for grain production inChinaa

Variable Parameter Stochastic frontier

Production functionIntercept β0 6.282 (159)∗FERT β1 0.309 (38.7)∗IRRI β2 0.092 (15.0)∗LABOUR β3 −0.071 (−8.95)∗MANURE β4 0.144 (15.4)∗MACH β5 0.027 (6.98)∗

Inefficiency functionIntercept δ0 0.442 (5.97)∗TMP AVG δ1 0.0291 (7.51)∗PRCTOT δ2 −0.000839 (−12.8)∗SOIL δ3 0.00245 (6.67)∗MEANELEV δ4 0.000134 (6.04)∗DISTCITY δ5 0.00197 (4.33)∗ILLIT δ6 0.422 (3.33)∗%AGLF δ7 −0.00519 (−2.95)∗INCOME δ8 −0.000379 (−24.2)∗EROSION δ9 0.000615 (2.22)∗∗

Diagnostic statisticsGamma (γ = σ 2

u /σ 2s ) γ 0.88

Sigma-square(σ 2

s = σ 2u + σ 2

v )σ 2

s 0.16

Log likelihood −178.2Likelihood-ratio statistic 1827∗∗∗

a Values in parentheses are thet-ratios of the estimates.∗ Significant at the 0.01 level.∗∗ Significant at the 0.05 level.∗∗∗ Significant at 0.01 level according to Table 1 of Kodde and

Palm (1986).

on production efficiency. This suggests that the pos-sibilities for non-farm labour decrease the technicalefficiency of grain production. Income is an importantvariable in the inefficiency function, having a positiveeffect on the efficiency of input use, probably throughthe opportunities to buy higher quality, more balancedinputs and invest in the land (e.g., terracing). In cor-respondence with other studies (Yao and Liu, 1998;Lambert and Parker, 1998), we found that erosion isalso an important source of inefficiency in agricul-tural production. Apart from the variables included inthe inefficiency function there might be some otherdeterminants of technical efficiency, which were notincluded in the empirical investigation due to dataavailability. For instance, low technical efficienciesmay also be related to the availability of the appro-priate fertilisers. Jin et al. (1999) indicate that unbal-anced fertiliser application is very common in China.

Especially potash fertiliser is underapplicated, dimin-ishing the effectiveness of nitrogen and phosphateuptake.

The average efficiency of grain production in Chinaas calculated with the derived frontier productionfunction is 0.74 (on a scale 0–1). The spatial distri-bution of the efficiency is displayed in Fig. 8. Lowefficiencies are especially found in the northwesternpart of the agricultural area and in the southwesternprovince Yunnan. High efficiencies are found in thenortheastern part of China and in the central region.The heavily urbanised strip along the southern coastalso has relatively low efficiencies in grain produc-tion. In the same Fig. 8 the most important variablesin the inefficiency function are denoted for the dif-ferent regions with low efficiencies, indicating thatinefficiencies have different causes in different partsof the country.

3.5. Technological progress

The coefficient for the time trend in the produc-tion function fitted for the data from 1986, 1991 and1996 indicates a 1.2% yearly change in productionlevel. All other coefficients in the production func-tion are similar to those presented in Table 5. Huangand Rozelle (1995) derived for a similar productionfunction a technical change of 2.9% yearly, basedupon provincial data between 1975 and 1990. Agri-cultural research is an important determinant of therate of technological change. Unfortunately, China’sagricultural research system itself is negatively af-fected by budget cutbacks and other measures inrecent years, which might further decrease the rateof technological change, and hence grain production(Lin, 1998). More detailed analyses of technologicalchange in Chinese agriculture are presented by Stone(1988), Huang et al. (1995) and Huang and Rozelle(1996).

4. Discussion

The methodologies used in this paper all explorechanges, or possibilities for change, in land use basedon an analysis of the regional variability of land usein China. Only the analysis of the changes in agricul-tural area is based on dynamic modelling, all other


Fig. 8. Technical efficiency in grain production 1991 and indication of areas and variables with high contribution to the inefficiency ingrain production.

assessments only explore the options for increasesin grain production. However, also the modelling re-sults should not be interpreted as forecasts of futureevents. Rather, they indicate possible patterns of landuse change, given the underlying assumptions of thescenario’s.

The methodologies used in this paper are specificfor the scale of analysis. As the basic unit of analy-sis, the individual grid-cells, measure approximately1000 km2, most research methods based on causal,deterministic understanding of processes of land usechange are inappropriate. Single units of observa-tion contain large numbers of different actors of landuse change with numerous interactions in a diversebiophysical environment. Simple aggregation of theprocesses known at the level of individual actors willgenerate large errors due to scale dependencies andsimple aggregation errors as result of non-linear sys-tem responses (Rastetter et al., 1992; Gibson et al.,2000). The methodologies used in this paper are

appropriate for the scale of analysis as the relationsbetween land use patterns and its explanatory factorsare quantified in an empirical way with data collectedat the same aggregation level as the analysis andthe presented results. The drawback of the empiricalquantification of the relations is the lack of causality,which forces us to interpret the results with caution.

Changes in agricultural area and grain-sown area inparticular, occur throughout the entire agricultural areaof China. Hot-spots of change are found in the Ordosand Loess plateau regions where degradation is themain land use change process, and around the grow-ing cities in eastern China. It can be argued to whatextent this process can be stopped. All decreases inagricultural area need to be compensated for by moreintensive cultivation on the remaining agricultural areato keep up food production. The model results indi-cate for some areas large decreases in the share ofgrain crops. Cash crops and vegetables take over alarge share in the cropping systems, especially in the


urban surroundings in southern and eastern China. Inthese areas the production of vegetables and cash cropsmake more intensive use of abundant labour per unitof scarce arable land. However, at the same time thisleads to increased grain imports.

The patterns of agricultural input use and of theefficiency of grain production correspond to a largedegree. Generally, low inputs are associated withlow efficiencies. Our analysis made clear that partof the lower efficiencies and the less intensive grainproduction can be explained by the less favourableenvironmental conditions in the western part ofChina. Agricultural research, aimed at new varietiesof higher adaptability, and better resistance and en-durance could help to overcome these constraints (Lin,1998). The main problem underlying grain produc-tion intensity and efficiency are the large differencesbetween well endowed areas around major cities andthe rural areas associated with income and illiteracy.Bridging the gap between urban and rural develop-ment is, therefore, essential to increase productivityin the less endowed regions. Policies to bridge thegap could include intensifying the construction ofinfrastructure including storage, communication, andtransport facilities and improving marketing condi-tions. Our results have also shown that increasinginvestment in rural education might, in the long-term,enhance agricultural production. The low income ofpeasants is correlated with their low education. Soit is an important measure to narrow the differencesin human capital and hence income gaps. This willcreate conditions for peasants to enhance their qualityof life.

These results suggest that China has, at least incertain areas, the potential to increase production andcompensate losses in agricultural area, by increasingproduction per unit area. However, China’s land usealready has a negative impact on its natural resourcesthrough land degradation, pollution and decreasingland qualities. A further intensification might threatenthe long-term sustainability of agricultural produc-tion (Smil, 1993). The transition towards intensive,but sustainable land use systems is therefore moreimportant for food security than a further intensifi-cation alone. Thus, more emphasis should be paidto production systems that not only strive for a highproduction but to maintaining environmental qualityas well.

5. Conclusion

This study has proven that a spatially explicit anal-ysis of land use change can reveal information thatis not accounted for in aggregated assessments thatis provided by economic analyses (e.g., Garnaut andMa, 1992; Paarlberg, 1997; Weersink and Rozelle,1997). Aggregated analysis, e.g., at the national level,cannot adequately shed insight in the production situ-ation because agricultural systems are, in nearly everycountry, very diverse and variable. Spatially explicitmethods focus on diversity of situations rather than onaverage situations. These deviations from the averagesituation, and the reasons underlying the deviations,provide insights into possibilities and constraints forincreasing agricultural production.

Although the spatial resolution of this study ismuch more detailed than those of most nation-wideassessments, the scale of analysis is still very coarse.At more detailed scales other forms of variability andother options and constraints for increasing agricul-tural production will be found. Similar types of analy-sis should therefore be used to analyse the variabilityof land use at more detailed scales, e.g., for indi-vidual provinces and counties located in ‘hot-spots’of land use change identified in this study. At thesemore detailed scales, studies on variability can alsolink up with socio-economic studies of the processesand actual motives of people for certain agricul-tural strategies. In this sense, the presented study isonly a first step towards a full, multi-scale, analy-sis of land use change and agricultural production;a more complete insight into the land use situationof China can only be attained by linking up thisresearch with other studies, using a set of comple-mentary research methodologies over a wide rangeof scales.

Acknowledgements

The research described in this paper is financed bythe Netherlands Organisation for Scientific Research(NWO/NOP-II). All persons and institutes who kindlymade there data available for this analysis are acknowl-edged. The authors thank Johan Bouma for commentson earlier versions of this paper.


Appendix A. Description of variables in data base

Variable name Description Source

Land use types Percentage of the total area used forCULT91 Cultivated lands INRRP/CLUE, 1998ORCH91 Horticultural lands INRRP/CLUE, 1998FOREST91 Forestry lands INRRP/CLUE, 1998GRASS91 Grasslands INRRP/CLUE, 1998URBAN91 Lands for settlement and industry (built-up land) INRRP/CLUE, 1998UNUSED91 Unused lands: deserts, glaciers, saline lands etc. INRRP/CLUE, 1998

DemographyPTOT91 Total population density (persons/km2) INRRP/CLUE, 1998PAG91 Agricultural population density (persons/km2) INRRP/CLUE, 1998PURB91 Urban/non-agricultural population density (persons/km2) INRRP/CLUE, 1998PRURLF91 Rural labour force density (persons/km2) INRRP/CLUE, 1998PAGLF91 Agricultural labour force density (persons/km2) INRRP/CLUE, 1998PAGPER91 Percentage of population belonging to agricultural population INRRP/CLUE, 1998PRLPER91 Percentage of population belonging to rural labor force INRRP/CLUE, 1998%AGLF91 Percentage of population belong-

ing toagricultural labour force

INRRP/CLUE, 1998

PAGRUR91 Percentage of rural labor force be-longing toagricultural labor force

INRRP/CLUE, 1998

Socio-economicsILLIT Fraction of population that is illiterate (1990) Skinner et al. (1997)INCOM91 Net income per capita (RMB/person) INRRP/CLUE, 1998DISTCITY Average distance to city (km) Tobler et al. (1995)

Soil related variables Fraction of the total land area withGOODDRAI Well drained soils FAO, 1995MODDRAIN Moderately drained soils FAO, 1995BADDRAIN Badly drained soils FAO, 1995SHALLOW Shallow soils FAO, 1995DEEP Deep soils FAO, 1995S1IRRPAD Soils very suitable for irrigated rice FAO, 1995S2IRRPAD Soils moderately suitable for irrigated rice FAO, 1995NSIRRPAD Soils not suitable for irrigated rice FAO, 1995S1MAIZER Soils very suitable for rainfed maize FAO, 1995S2MAIZER Soils moderately suitable for rainfed maize FAO, 1995NSMAIZER Soils not suitable for rainfed maize FAO, 1995SMAXHIGH Soils that have high moisture storage capacity FAO, 1995SMAXLOW Soils that have low moisture storage capacity FAO, 1995FERT1 Poor soil fertility CAS, 1978/1996FERT2 Moderate soil fertility CAS, 1978/1996FERT3 High soil fertility CAS, 1978/1996TEXT1 Coarse soil texture CAS, 1978/1996TEXT2 Medium soil texture CAS, 1978/1996TEXT3 Fine soil texture CAS, 1978/1996


Variable name Description Source

GeomorphologyMEANELEV Mean elevation (m.a.s.l.) USGS, 1996RANGEELE Range in elevation (m) USGS, 1996SLOPE Slope (◦) USGS, 1996PHYSL (Fraction of the area with) level land FAO, 1994PHYSS (Fraction of the area with) sloping land FAO, 1994PHYST (Fraction of the area with) steep sloping land FAO, 1994PHYSC (Fraction of the area with) complex valley land forms FAO, 1994GEOMOR1 (Fraction of the area with) mountains CAS, 1994/1996GEOMOR2 (Fraction of the area with) loess CAS, 1994/1996GEOMOR3 (Fraction of the area with) eolian land forms CAS, 1994/1996GEOMOR4 (Fraction of the area with) tableland CAS, 1994/1996GEOMOR5 (Fraction of the area with) plain land CAS, 1994/1996DISTRIVER Average distance from major river (km) CAS, 1994/1996EROSION Index representing the extent and impact

of humaninduced water erosion

Based on OldemanandVan Lynden (1997)

ClimateTMP MIN Temperature in coldest month (oC) Cramer (experimental data)TMP MAX Temperature in warmest month (oC)TMP AVG Average temperature (oC)TMP RNG Difference between warmest and coldest month (oC)TMP 10C Number of months with temperature above 10◦ (months)PRC TOT Total yearly precipitation (mm)PRCRNG Difference between wettest and driest month (mm)PRC50M Number of months with precipitation

above 50 mm (months)SUN TOT Average percentage of sunshine (%)

Agricultural production Available for1986, 1991, 1996INDEX Multiple cropping index INRRP/CLUE, 1998GRAINa Percentage of total sown area sown to grain crops INRRP/CLUE, 1998CASHC Percentage of total sown area sown to cash crops INRRP/CLUE, 1998VEGE Percentage of total sown area sown to vegetables INRRP/CLUE, 1998OTHERS Percentage of total sown area sown to other crops INRRP/CLUE, 1998YGRAIN Yield of grain crops (kg/ha) INRRP/CLUE, 1998FERT Application of chemical fertiliser (kg/ha) INRRP/CLUE, 1998IRRI Percentage of the land that is irrigated INRRP/CLUE, 1998LABOUR Labour force density on arable land (persons/ha sown area) INRRP/CLUE, 1998MANURE Application rate of manurial fertiliser (ton/ha) Calculatedb

MACH Percentage of cultivated land cultivated by machine INRRP/CLUE, 1998

a In Chinese statistics grain includes rice, wheat, corn, sorghum, millet, other miscellaneous grains, tubers(potatoes), and soybeans.

b Since there are no statistics on the application of manurial fertilisers we have calculated the data based onlivestock numbers and manure production (see: Verburg and Thijssen, 2000).


Data sourcesCAS (Chinese Academy of Sciences) 1980/1996.Map of River System of China. Edited by Carto-graphic Publishing House of PR China; published(1989) by Cartographic Publishing House, Beijing.Digital version (1996) by the State Key Labora-tory of Resources and Environmental Informationsystem (LREIS). Chinese Academy of Sciences,Beijing, China.

CAS (Chinese Academy of Sciences) 1994/1996.Geomorphological Map of china. Edited by theGeographical Institute of the Scientific Academy ofChina; published (1994) by Science Press, Beijing.Digital version (1996) by the State Key Labora-tory of Resources and Environmental InformationSystem (LREIS). Chinese Academy of Sciences,Beijing, China.

CAS (Chinese Academy of Sciences). 1978/1996.Soil Map of China. Edited by the Nanjing Instituteof Soil Science, Chinese Academy of Sciences; pub-lished (1978) by the Map publishing house of PRChina, Beijing. Digital version (1996) by the StateKey Laboratory of Resources and Environment In-formation System (LREIS), Chinese Academy ofSciences, Beijing, China.

FAO (Food and Agriculture Organization of theUnited Nations), 1995. Digital soil map of theWorld and derived soil properties version 3.5. FAO,Rome, Italy.

FAO (Food and Agriculture Organization of theUnited Nations), 1994. A draft physiographicmap of Asia (excluding the former Soviet Union).Compiled by G.W.J. van Lynden. FAO, Rome,Italy.

INRRP/CLUE (Institute for Natural Resources andRegional Planning and CLUE-group Wageningen),1998. Statistical database of China for land cover,agriculture and population, collected by the Insti-tute for Natural Resources and Regional Planningof the Chinese Academy of Agricultural Sciencesand edited by You Qi Chen, P.H. Verburg andA.R. Bergsma. Chinese Academy of AgriculturalSciences and Wageningen Agricultural University,Wageningen, The Netherlands.

Oldeman, L.R. and Van Lynden, G.W.J., 1997. As-sessment of the status of human-induced soil degra-dation in China. In: Uithol, P.W.J., Groot, J.J.R.(Eds.), Proceedings Workshop Wageningen–China,

Wageningen, May 13, 1997. Report 84. AB-DLO,Wageningen.

Skinner, G.W. et al., 1997. China county-level dataon population (census) and Agriculture, Keyed to1:1M GIS Map. CIESIN, USA.

Tobler, W., Deichmann, U, Gottsegen, J., Maloy, K.,1995. The Global Demography Project. TechnicalReport TR-95-6. National Center for GeographicInformation and Analysis. Department of Geogra-phy. University of California, USA.

USGS (United States Department of Agriculture),1996. GTOPO30: Global Digital Elevation Model(DEM) with a horizontal grid spacing of 30 arcseconds. Electronic database.

References

Alexandratos, N., 1996. China’s projected cereals deficits in aworld context. Agric. Econ. 15, 1–16.

Battese, G.E., 1992. Frontier production functions and technicalefficiency: a survey of empirical applications in agriculturaleconomics. Agric. Econ. 7, 185–208.

Battese, G.E., Coelli, T.J., 1995. A model for technical inefficiencyeffects in a stochastic frontier production function for paneldata. Empirical Econ. 20, 325–332.

Bhattacharyya, A., Parker, E., 1999. Labor productivity andmigration in Chinese agriculture: a stochastic frontier approach.China Econ. Rev. 10, 59–74.

Bravo-Ureta, B.E., Pinheiro, A.E., 1993. Efficiency analysis ofdeveloping country agriculture: a review of the frontier functionliterature. Agric. Res. Eco. Rev. 22, 88–101.

Brown, L.R., 1995. Who Will Feed China? Wake-up Call for aSmall Planet. Norton, New York.

Cao, M., Ma, S., Han, C., 1995. Potential productivity and humancarrying capacity of an agro-ecosystem: an analysis of foodproduction potential of China. Agric. Syst. 47, 387–414.

Coelli, T., 1994. A computer program for stochastic frontierproduction and cost function estimation. A Guide to Frontier4.1. CEPA Working Paper 96/07. Centre for Efficiency andProductivity Analysis, University of New England, Armidale,Australia.

Coelli, T., Prasado-Rao, D.S., Battese, G.E., 1998. An introductionto efficiency and productivity analysis. Kluwer AcademicPublishers, Boston.

Crook, F.W., 1993. Underreporting of China’s cultivated landarea: implications for world agricultural trade. In: InternationalAgriculture and Trade Report, China. Doc. RS-93-4, ERSUSDA, Washington, DC.

Crook, F.W., 1996. The development of China’s vegetable markets.In: International Agriculture and Trade Report, China. Doc.RS-96-2, ERS USDA, Washington, DC.

Crook, F.W., Colby, W.H., 1996. The Future of China’s GrainMarket. Agricultural Information Bulletin No. 730. ERS-USDA,Washington, DC.


Fan, S., Agcaoili-Sombilla, M., 1997. Why projections on China’sfuture food supply and demand differ? Aust. J. Agric. Res.Econ. 41 (2), 169–190.

Fan, S., 1997. Production and productivity growth in Chineseagriculture: new measurement and evidence. Food Policy 22 (3),213–228.

Farrell, M.J., 1957. The measurement of productive efficiency. J.Roy. Statist. Soc. A 120, 253–258.

Garnaut, R., Ma, G., 1992. Grain in China. Department of ForeignAffairs and Trade, Canberra.

Gibson, C., Ostrom, E., Ahn, T.-K., 2000. The concept of scaleand the human dimensions of global change: a survey. Ecol.Econ. 32, 217–239.

Guoqian, C., 1997. Market prospects for upland crops in China.Working Paper Series No. 24. CGPRT Centre, Bogor.

Han, T., Wahl, T.I., 1998. China’s rural household demand forfruit and vegetables. J. Agric. Appl. Econ. 30 (1), 141–150.

Huang, Y.P., Kalirajan, K.P., 1997. Potential of China’s grainproduction: evidence from the household data. Agric. Econ. 17,191–199.

Huang, J., Rozelle, S., 1995. Environmental stress and grain yieldsin china. Am. J. Agric. Econ. 77, 853–864.

Huang, J., Rosegrant, M., Rozelle, S., 1995. Public investment,technological change, and agricultural growth in China. PaperPresented in the Final Conference on Medium- and Long-termProjections of World Rice Supply and Demand. Sponsoredby the International Food Policy Research Institute and theInternational Rice Research Institute, Beijing, April 23–26.

Huang, J., Rozelle, S., 1996. Technological change: rediscoveringthe engine of productivity growth in China’s agriculturaleconomy. J. Dev. Eco. 49, 337–369.

Huang, J., Rozelle, S., Rosegrant, M.W., 1999. China’s food eco-nomy to the twenty-first century: supply, demand, and trade.Eco. Dev. Cultural Change 47, 737–766.

Jin, J.Y., Lin, B., Zhang, W.L., 1999. Improving nutrient mana-gement for sustainable development of agriculture in China. In:Smaling, E.M.A., Oenema, O., Fresco, L.O. (Eds.), NutrientDisequilibria in Agroecosystems, Concepts and Case Studies.CAB International, Wallingford, pp. 157–174.

Kodde, D.A., Palm, F.C., 1986. Wald criteria for jointly testingequality and inequality restrictions. Econometrica 54 (5), 1243–1248.

Lambert, D.K., Parker, E., 1998. Productivity in Chinese provincialagriculture. J. Agric. Econ. 49 (3), 378–392.

Lin, J.Y., 1998. How did china feed itself in the past? Howwill china feed itself in the future? In: Second DistinguishedEconomist Lecture. CIMMYT, Mexico, DF.

Lin, W., Colby, H., 1996. China brings volatility to the worldsugar market. In: China, Situation and Outlook Series. ReportWRS-96-2. ERS-USDA, Washington, DC.

Lin, J.Y., Li, Z., 1995. Current issues in China’s rural areas. OxfordRev. Eco. Policy 11 (4), 85–96.

Lin, J.Y., Wen, G.J., 1995. China’s regional grain self-sufficiencypolicy and its effects on land productivity. J. Comp. Eco. 21,187–206.

Liu, J., Lu, Q. (Eds.), 1999. Land use and sustainable developmentin Loess Plateau, Northwest China. In: Proceedings of

the Annual International Workshop. Sustainable AgricultureWorking Group (SAWG), China Council for International Coo-peration on Environment and Development (CCICED). ChinaEnvironmental Science Press, Beijing.

Lu, F., 1998. Grain versus food: a hidden issue in China’s foodpolicy debate. World Dev. 26 (9), 1641–1652.

Paarlberg, R.L., 1997. Feeding China: a confident view. FoodPolicy 22 (3), 269–279.

Qinye, Y., Yili, Z., Guodong, L., 1994. Critical zones and theirsituation in China. Chinese J. Arid Land Res. 7 (2), 173–177.

Rastetter, E.B., King, A.W., Cosby, B.J., Hornberger, G.M.,O’Neill, R.V., Hoebbie, J.E., 1992. Aggregating fine-scaleecological knowledge to model coarser-scale attributes ofecosystems. Ecol. Appl. 2 (1), 55–70.

Rawski, T.G., Mead, R.W., 1998. On the trail of China’s Phantomfarmers. World Dev. 26 (5), 767–781.

Rozelle, S., Rosegrant, M.W., 1997. China’s past, present,and future food economy: can China continue to meet thechallenges? Food Policy 22 (3), 191–200.

Shen, J., Spence, N.A., 1996. Modelling urban-rural populationgrowth in China. Environ. Plann. A 28, 1417–1444.

Smil, V., 1993. China’s environmental crisis: an inquiry into thelimits of national development. M.E. Sharpe, New York.

Smil, V., 1999. China’s agricultural land. China Quart. 158, 414–429.

Stone, B., 1988. Developments in agricultural technology. ChinaQuart. 116, 767–822.

Turner II, B.L., Moss, R.H., Skole, D., 1993. Relating land useand global land-cover change: a proposal for an IGBP HDPcore project. IGBP Report No. 24/HDP Report No. 5.

Turner II, B.L., Skole, D., Sanderson, S., Fischer, G., Fresco,L., Leemans, R., 1995. Land-Use and Land-cover changescience/research plan. IGBP Report No. 35; HDP Report No. 7.

United Nations, 1995. World urbanization prospects: the 1994revision. UN Department for Economic and Social Informationand Policy Analysis, Population Division, New York.

US Bureau of the Census, 1997. International database.http://www.census.gov.

US Department of Agriculture (USDA), 1998. InternationalAgricultural Baseline Projections to 2007: Market and Trade.Agricultural Economic Report No. 767. Economics Division,Economic Research Service, US. Department of Agriculture,Washington.

US Department of Agriculture (USDA), 2000. USDA AgriculturalBaseline Projections to 2009. Staff Report WAOB-2000-1.Interagency Agricultural Projections Committee, Washington.

Verburg, P.H., Chen, Y.Q., 2000. Multi-scale characterization ofland-use patterns in China. Ecosystems 3, 369–385.

Verburg, P.H., de Koning, G.H.J., Kok, K., Veldkamp, A., Bouma,J., 1999a. A spatial explicit allocation procedure for modellingthe pattern of land use change based upon actual land use.Ecol. Modelling 116, 45–61.

Verburg, P.H., Veldkamp, A., Fresco, L.O., 1999b. Simulation ofchanges in the spatial pattern of land use in China. Appl. Geogr.19, 211–233.

Verburg, P.H., van Keulen, H., 1999. Exploring changes in thespatial distribution of livestock in China. Agric. Syst. 62, 51–67.


Verburg, P.H., Thijssen, G., 2000. Intensity and efficiency ofgrain production in China: a spatial perspective, submitted forpublication.

Weersink, A., Rozelle, S., 1997. Marketing reforms, market deve-lopment and agricultural production in China. Agric. Econ. 17,95–114.

Yao, Sh., Liu, Z., 1998. Determinants of grain production andtechnical efficiency in China. J. Agric. Econ. 49 (2), 191–207.

Ye, Q.L., Rozelle, S., 1994. Fertiliser demand in China’s reformingeconomy. Can. J. Agric. Econ. 42, 191–207.

spatial explorations of land use change and grain production in china

Documents