e c o l o g i c a l m o d e l l i n g 2 0 8 ( 2 0 0 7 ) 378–390

available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/ecolmodel

Global-scale modelling of future changes in

sown areas of major crops

Wenbin Wu a,b,∗ , Ryosuke Shibasaki a , Peng Yang b,c , Guoxin Tan d ,

Kan-ichiro Matsumura e , Kenji Sugimoto a
a Center for Spatial Information Science, University of Tokyo, Tokyo 153-8505, Japan
b Institute of Agricultural Resources and Regional Planning, Chinese Academy of Agricultural Sciences, Key Laboratory of Resources
Remote Sensing & Digital Agriculture, Ministry of Agriculture, Beijing 100081, China
c Center for Climate System Research, University of Tokyo, Chiba 277-8568, Japan
d Engineering Research Center for Education Information Technology, Hua Zhong Normal University, Hubei 430079, China
e Department of Applied Informatics, School of Policy Studies, Kwansei Gakuin University, Hyogo 669-1337, Japan

a r t i c l e i n f o a b s t r a c t

Article history: Land use and its dynamics have attracted much attention from researchers due to their
Received 30 November 2006 ecological and socio-economic implications. Many studies have used a modelling approach
Received in revised form 1 June 2007 to evaluate land use changes and their effects. Most of these models were designed for the
Accepted 12 June 2007 analysis of past, present and future cropland changes at different scales and few have been
Published on line 23 July 2007 designed for the study of dynamic changes in sown areas of crops within croplands. This
paper presents an integrated modelling approach to simulate dynamically the changes in
Keywords: sown areas for the world’s major crops at a global scale. This approach was based on three
Crop sown area core models. A crop choice decision model, the Multinomial Logit model, was used to track
Land use change and simulate the crop choice decisions made by individual farmers. A crop yield model, the
Crop choice GIS-based Environmental Policy Integrated Climate (EPIC) model, was utilized to estimate
Crop yield yields of different crop types under a given biophysical and management environment,
Crop price while a crop price model, the International Food Policy and Agricultural Simulation (IFPSIM)
Simulation model, was employed to assess the price of crops on the international market. Through
data exchange, the crop choice decision model was linked with the crop yield and crop price
models to allow the study of the dynamic feedback loop between changes in agricultural land
use and biophysical and socio-economic driving factors. Sensitivity analysis and empirical
validation for the model were conducted after the construction of the model. The model
validation indicated the reliability of the model for addressing the complexity of current
agricultural land use changes and its capacity for investigating long-term scenarios in the
future. Finally, the model was used to simulate future scenarios over a time frame of 30 years
with five-year increments, beginning from the year 2000. The simulation results provided
insights into potential global cropping patterns, variation in rates and trajectory of changes
in sown areas for major crops over the test period. These results can improve understanding
of projected land use changes and explain their causes, locations and consequences, and
provide support for land use planning and policy making.
© 2007 Elsevier B.V. All rights reserved.

∗
Corresponding author at: Center for Spatial Information Science, University of Tokyo, Tokyo 153-8505, Japan. Tel.: +81 3 54526417;
fax: +81 3 54526414.
E-mail addresses: [email protected] (W. Wu), [email protected] (R. Shibasaki).
0304-3800/$ – see front matter © 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.ecolmodel.2007.06.012
 e c o l o g i c a l m o d e l l i n g 2 0 8 ( 2 0 0 7 ) 378–390 379

To assess the consequences of cultivation practices for

1. Introduction global food production and the health of the environment,
understanding and modelling future global changes in sown
Human land use activities – whether converting natural areas of various crops are critical. In this paper, a new
landscapes for human use or changing management prac- integrated modelling approach was developed to simulate
tices on human-dominated lands – have transformed a dynamic changes in sown areas for four major crops at a
large proportion of the planet’s land surface (Foley et al., global scale. This paper is organized into five sections. After a
2005). These land use practices have enormous consequences brief introduction about the study background in Section 1, the
for the environment as they alter the structure and func- methodology of this modelling approach is presented in Sec-
tioning of ecosystems, and they influence how ecosystems tion 2. The complete set of input datasets required for model
interact with the atmosphere, aquatic systems and sur- construction and validation is described in Section 3, then the
rounding land (Vitousek et al., 1997; Audsley et al., 2006). model sensitivity and validation is introduced in Section 4.
Agricultural land use changes are very important among A scenario simulation for sown area changes over a future
the globally significant processes in land use changes, due period of 30 years is presented in Section 5, followed by the
to their consequences for both environmental sustainabil- conclusions in Section 6.
ity and food security (Verburg et al., 2000; Rounsevell et al.,
2003; Heistermann et al., 2006). The extent of these changes
and the possible impacts on environment, landscapes and
rural livelihoods are largely unknown (Verburg et al., 2006). 2. Methodology
Therefore, decision-makers and scientists have identified the
need to explore potential changes in agricultural land use 2.1. Modelling framework
and their associated impacts at different regions and scales
(Fischer and Sun, 2001). Consequently, a number of studies The general hypothesis of the modelling approach presented
have been conducted to unravel the geographic distribution in this paper is that the sown area of particular crops is directly
of cropland and its dynamic changes over time, space and linked with human decisions on crop choice for farmland.
scale. Land users make decisions about crops for their land based on
These previous and ongoing studies can be grouped an understanding of ecological, socio-economic, technological
roughly into two categories. The studies in one category and political factors at local, regional, national and interna-
focused mainly on analyzing or monitoring the past or present tional levels (Duffy et al., 2001). Land use on a regional scale
cropland distribution and its temporal and spatial changes is the sum of the results of decision making at the farm level
with the aid of remote sensing (Liu et al., 2005; Xiao et al., (Rounsevell et al., 2003). Thus, through capturing the essen-
2006), statistical methods (Yang and Li, 2000) and a combi- tial features of individual human decision processes on crop
nation of these (Ramankutty and Foley, 1998). Their results choice, it is possible to track and estimate changes in the crop
showed the past or present cropland cover over the world on sown areas over time and space.
a continuous scale, but they were still limited in the extent This approach was based on three core models. The crop
to which they delineated the distribution of specific crop choice decision model is a Multinomial Logit model, which
types and cropping systems. Leff et al. (2004) synthesized was used to model the crop choice decisions among a variety
satellite-derived land cover data and agricultural census data of available alternatives by using an optimization approach.
to produce global data sets of the distribution of 18 major crops A crop yield model, the GIS-based Environmental Policy Inte-
across the world. The resulting data described the fraction of grated Climate (EPIC) model, was adopted to estimate the
a grid cell occupied by each of the 18 crops, but the results potential yields of different crop types under a given bio-
were representative of the early 1990s, with a noticeable short- physical and agricultural management environment (Tan and
age of time horizons and a limitation in its usefulness for Shibasaki, 2003). A crop price model, the International Food
the study of future scenarios. Studies in the other category Policy and Agricultural Simulation (IFPSIM) model, was uti-
attempted to develop numerous Land Use and Land Cover lized to evaluate the price of the test crops in the international
Change (LUCC) models to explore possible changes in crop- market (Ohga and Yanagishima, 1996). The relationships
land in future scenarios (the next 10–80 years) as a function of between these three models are shown schematically in Fig. 1.
driving forces (De Koning et al., 1999; Verburg et al., 1999, 2000, The crop choice decision model is the main mechanism deter-
2006; Fischer and Sun, 2001; Verburg and Veldkamp, 2001). All mining changes in crop sown areas and it is affected by the
of these studies provided important information about the crop yield and crop price models, which determine directly
scope and impact of future agricultural land use changes, and the crop utilities or profits, as well as by the potential multiple
they can be used to support analyses of the vulnerability of cropping systems, which provide different crop choice sets to
ecosystems and to provide an explanation of how agricultural decision-makers. These three models are seamlessly linked
land use might respond to a range of future environmen- through data flow and exchange between them. A geograph-
tal changes. However, these models were generally focussed ical information system (GIS), with the unique capabilities of
solely on the simulation of temporal and spatial changes data handling, spatial analysis and map input and output, was
in agricultural land use or croplands, and they were unable used for the preparation of the database required by the mod-
to address possible future changes in individual crops (such els, as well as being used as a platform for the presentation of
as rice, wheat and maize) occurring inside croplands in the the modelling results. The three models are described in more
future. detail later.
380 e c o l o g i c a l m o d e l l i n g 2 0 8 ( 2 0 0 7 ) 378–390

Fig. 1 – General structure of the modelling framework.

2.2. Crop choice decision model of those involved. In general, these driving factors include
biophysical factors (e.g. temperature, rainfall, soil physic-
Land users (i.e. farmers or households) make their decisions ochemical properties and topography) (Gobin et al., 2002),
on crop choice in the context of their own strategies or rules, demographic factors (e.g. rural population density), socio-
which affect the conversion of land from the cultivation of economic factors (e.g. farming income per capita, agricultural
one crop to another and the preservation of land in its cur- mechanization, road accessibility and international trade
rent state. But how do people make crop choice decisions and price) (Müller and Zeller, 2002) and technological factors (e.g.
how can we model these behaviours? In this model, farmers irrigation, fertilizers and pesticides). However, it was not pos-
were assumed to be generally autonomous and to maximize sible to include all of these driving factors in a LUCC model,
their long-term profits within the constraints of their situa- especially when modelling land use changes over large areas.
tion, taking account of uncertainty in prices and yields, which Thus, instead of using all of the underlying driving factors, the
causes farmers on otherwise identical farms to perceive differ- model generally used some proximate variables that represent
ent gross margins for the same crops. We introduced the term the underlying driving factors. Determining the proximate
‘utility’ to describe a mathematical function that expresses variables was often problematic and an issue of discussion,
the preferences of farmer’s crop choices (Evans et al., 2001). as there is no unifying theory describing the selection of
The utility of each possible crop is determined by not only the most appropriate variables. In this modelling approach,
the essential biophysical variables (level and variance of rain- some variables that were highly correlated to others were
fall, soil suitability, altitude and slope of land), but also by the excluded from the model analysis for the sake of simplifi-
socio-economic aspects (crop price, population density and cation and the elimination of computation redundancy. For
road accessibility) of a specific crop. Using these relative crop instance, crop yield itself is a measure of performance of
utilities, farmers seek to maximize their income by allocat- the crop plant, which is enhanced or reduced by biophysi-
ing their lands to those crop cultivation activities that they cal factors (e.g. temperature, rainfall, soil and topography) as
perceive will provide the greatest return or that will carry the well as by agricultural management practices such as irriga-
least risk. The allocation of land to specific crop types is then tion and fertilizing, therefore crop yield can be used to reflect
translated into the conversion of an area from one crop cover- the impacts and interactions of most biophysical and agri-
age to another. Clearly, the impetus for changes in sown crops cultural management variables. In the construction of the
will depend on the difference in their crop utilities or prof- model, four main variables, namely, crop yield, crop price,
its, where a change in crop utilities may drive changes in crop rural population density and road accessibility were selected
choice decisions, resulting in further changes in crop sown as the explanatory variables for the computation of crop
areas over time and space. utilities.
Following the above considerations, we explained changes A Logit model was employed to explore the causal relation-
in crop sown areas as a function of the driving factors deter- ships between predetermined variables and changes in sown
mining the crop utilities, which influence the crop choices area. This type of model estimates the direction and inten-
 e c o l o g i c a l m o d e l l i n g 2 0 8 ( 2 0 0 7 ) 378–390 381

sity of the explanatory variables on the categorical dependent systems under which they are to be grown. Using multiple
variable by predicting a probability outcome associated with cropping systems, different farmers or households across the
each category of the dependent variable (i.e. different crops). world may choose between the possible farming systems and
The probability that a crop, i, is chosen for cultivation can be make their decisions on crop choice.
stated as (Greene, 1997): Since there are no existing multiple cropping systems, a
number of potential multiple cropping systems and possible
eui
Pi = N (1) crop combinations were defined by matching temperature and
i=1
eui water requirements of individual crops and crop combina-
tions with the agro-ecological environment available for crop

M growth. Potential multiple cropping systems and their possi-
ui = ai + bj xj (2) ble crop combinations were defined for a region only where
j=1 both temperature and moisture conditions in that region per-
mitted crop growth. The global climate dataset in the form
In (1), i denotes the crop type used for analysis (i = 1, 2, . . ., of 5 min latitude/longitude grids was taken from the Climate
N), Pi is the probability for crop type i, ui is the utility of crop Research Unit (CRU) at the University of East Anglia (New et
type i. In (2), ai is a constant for crop type i, j is the number al., 1999), and the criteria for defining multiple cropping sys-
of explanatory variables (j = 1, 2, . . ., M), xj is the explanatory tems were derived from the FAO Global Agro-Ecological Zones
variable and bj is the coefficient to be estimated for the variable 2000. Fig. 2 shows the estimated global multiple cropping sys-
xj (McFadden, 1973). tems for the four crops. In this figure, triple cropping systems
were not taken into account due to their low percentage area.
2.3. Multiple cropping systems The potential multiple cropping systems were classified and
mapped into seven zones: (1) zone of no cropping; (2) zone
Analyzing crop choice decision making requires knowledge of single cropping (spring wheat); (3) zone of single cropping
of what will possibly be chosen by land users. Therefore, (winter wheat); (4) zone of double rice cropping, or double
assumptions must be made about available crop options or maize/soybean cropping; (5) zone of double maize/soybean or
alternatives that land users consider in their decision mak- rice/winter wheat cropping; (6) zone of double cropping with
ing. According to the FAO statistical database (FAOSTAT), the one winter wheat; and (7) zone of limited double cropping
four crops of rice, maize, wheat and soybean make up nearly (only maize/winter wheat or soybean/wheat, otherwise single
80% of the global cereal croplands. Only these four major crops cropping).
were taken into account in this study, and they were called
the ‘choice set’ available to farmers in the process of decision
making. However, two points should be recognized about the 2.4. Crop yield model
crop choice options. One is that not all four crops can be culti-
vated everywhere in the world, since each crop has its specific Crop yield, which determines the direct output of farming
regions that are most suitable for planting or its own growth activities, is likely to have substantial implications for farmer’s
ecosystem. This means that the crop choice for farmers in crop choices. Changes in crop yield depend on different bio-
different regions of the world needs to be a subset of the total physical and socio-economic factors, of which climate change,
choice set. The other point to note is that harvesting two or increasing CO2 concentrations and technological develop-
more crops from a single plot in a single year is very common ment are the most important drivers (Ewert et al., 2005;
in many regions in the world (i.e. multiple cropping) (Frolking Rounsevell et al., 2006), thus making crop yield difficult to
et al., 1999). For this reason, farmers possibly choose two or assess.
more suitable crops for cultivation in a plot on their land in Process-based models are used increasingly to estimate
one year. Therefore, having selected the four crops to be con- crop productivity (Parry et al., 2004). Of these models, the
sidered, it is necessary to define the global multiple cropping EPIC model is used by many researchers due to its particu-

Fig. 2 – Potential multiple cropping systems for the globe.

382 e c o l o g i c a l m o d e l l i n g 2 0 8 ( 2 0 0 7 ) 378–390

model. Running the model simulation requires a variety of

input databases to be prepared and processed in advance with
the aid of GIS. When all the necessary data were available, crop
yields for each grid cell could be simulated for different crops
using the GIS-based EPIC. Simulated crop yields for different
years were introduced into the crop choice decision model
as a basic input variable for the computation of crop utili-
ties.

2.5. Crop price model

Crop price is another important determinant in human crop

choice decision making, since it affects the income or returns
from the crop types to which farmers may change. In this
Fig. 3 – Brief schematic presentation of the GIS-based EPIC study, crop price was assessed by a crop price model, the
model (Priya and Shibasaki, 2001). IFPSIM model. IFPSIM is a multi-commodity, multi-regional
and multi-period world agricultural trade and policy simu-
lation model developed and designed on the Ohga Model
lar features (Dumesnil, 1993). The EPIC model was initially Building System (OMBS) (Ohga and Yanagishima, 1996). It
developed by United States Department of Agriculture, Agri- is a partial equilibrium and interactive model, allowing for
cultural Research Service (USDA-ARS) in 1984 with the purpose the simultaneous determination of supply, demand, trade,
of understanding the relationship between soil erosion and stock levels and prices for 14 commodities for 31 regions
crop productivity. In EPIC, a general plant growth model with of the world. A complete description of the regions used
crop-specific parameters is used to simulate the growth of rice, in the model has been documented by Ohga and Gehlar
wheat, maize, sorghum and soybean, among others (Williams (1993).
et al., 1990). EPIC was originally a site-specific model, and Fig. 4 presents schematically the structure of the IFPSIM
uses a daily time increment to simulate weather, hydrology, model. Food demand in each region is divided into three kinds:
soil erosion by wind and water, nutrient cycling, tillage, crop demand for food for human consumption, for livestock and for
management and growth, and field scale costs and returns. the production of processed food, and it is described by indi-
It is thus not possible to use the original EPIC model directly vidual income, population and the consumer purchase price of
for large-area applications. However, by integrating EPIC with the crop in question. Food supply in one region is comprised of
GIS, the EPIC model gains the possibility of estimating crop the supply of crops and the supply of livestock products. The
yields from field level to small country or sub-regional scale supply of crops is described by crop yields, sown areas and
(Priya and Shibasaki, 2001; Yang et al., 2006). Subsequently, the producer price for each crop. The total food demand or
Tan and Shibasaki (2003) expanded this GIS-based EPIC model supply in the world is determined from the summation of the
to a global level and applied it to detecting crop yields and demand or supply in each region. In the international market,
predicting the effects of future global warming on the yields crop price is determined by the level at which world supply
of major crops at a global level. In this study, the GIS-based is equal to world demand, where all variables are simulta-
EPIC model (Version 8120) was used to simulate the potential neously determined, while world market clearing prices are
yields of different crops under given biophysical and man- derived by equating the sum of gross imports and the sum
agement conditions at a global level. Fig. 3 shows a brief of gross exports (Ohga and Yanagishima, 1996). One of the
schematic presentation of the structure of this crop yield important features of the IFPSIM model is that it can deal with

Fig. 4 – General structure of the IFPSIM model.

 e c o l o g i c a l m o d e l l i n g 2 0 8 ( 2 0 0 7 ) 378–390 383

Table 1 – Complete set of input data used in this study

Main variable name Time period Format Level of Data source
details

Crop choice decision model

Crop yield 1995–2035 Grid 6 min Simulation by crop yield model
Crop price 1995–2035 Text Region Simulation by crop price model
Population density 1998 Grid 30 s Landscan Global Population Database
(http://www.ornl.gov/sci/landscan/)
Road accessibility 1993 Grid 6 min ESRI DCW database (Digital Chart of the
World, http://www.maproom.psu.edu/
dcw/)
Global land cover dataset 1992–1993 Grid 1 km IGBP-DIScover global land cover database
(http://edcsns17.cr.usgs.gov/glcc/globe int.
html)

Crop yield model

Weather data
Monthly maximum/minimum 1995–2050 Grid 6 min IPCC-CGCM1 (http://www.cccma.ec.gc.ca/
temperature, and precipitation data data/cgcm1/cgcm1.shtml)

Soil data
Depth, percent sand, percent silt, bulk 1999 Grid 5 min Global Soil Data Products (http://mercury.
density, pH, percent organic carbon, ornl.gov/)
and percent calcium carbonate
Management data
Maximum annual irrigation volume 1995 Grid 30 min Global map of irrigated areas (http://www.
geo.uni-frankfurt.de/ipg/ag/dl/forschung/
global irrigation map/index.html)
Maximum annual fertilizer volume 1995 Text Country FAO statistical database (http://faostat.
fao.org/)
Crop price model
Population 1995–2035 Text Country World population prospects (http://esa.
un.org/unpp)
Economic growth rate 1990–2050 Text Region IPCC-SRES (http://www.grida.no/climate/
ipcc/emission/)

Model validation
FAO statistical database 1995–2004 Text country FAO statistical database (http://faostat.
fao.org/)
MODIS global land cover dataset 2001 Grid 1 km Boston University
(http://www-modis.bu.edu/landcover/)
Global distribution map of major crops 1992 Grid 5 min Center for Sustainability and the Global
Environment, University of Wisconsin–
Madison (http://www.sage.wisc.edu/
pages/datamodels.html)

changes in demand and supply both inside and outside one Some were used only for calibration and validation of the
region. This is especially important in relation to trends in model.
global trade. Thus, the crop price in one region estimated by Owing to a large degree of variation in data from sources
the IFPSIM model reflects not only the demand (and supply) of with different spatial and temporal resolutions, it was nec-
the internal market, but also the demand (and supply) of the essary to perform a procedure of data reprocessing and
external market. The simulation results for crop price were standardization. To do this, all spatial data were converted
saved as text format data and input into the crop choice deci- into GIS grid data with a cell size of 6 min by 6 min in a
sion model for dynamically updating the utility of the related standard GIS software environment (ESRI ArcGIS 9.1), while
crop. the socio-economic data were processed and stored as text
format data. Additionally, for all spatial data we excluded
from the model estimation some geographical regions of
3. Data preparation the world (mainly those covered by ocean or permanent
glaciers) in both the Northern and Southern polar regions.
A very large amount of input data, including spatial and The final test area covered the globe from longitude 180.0◦ W
socio-economic data, was required in this study (Table 1). Of to 180.0◦ E and from latitude 84.0◦ N to 56.5◦ S. The C pro-
these data, some were input directly into the crop choice gramming language was used to develop the model program,
decision model for the calculation of crop utilities while allowing the model to access directly the multiple input
some were used indirectly by the crop choice decision model data in GIS grid format and text format from numerous
by being input into the crop yield and crop price models. sources.
384 e c o l o g i c a l m o d e l l i n g 2 0 8 ( 2 0 0 7 ) 378–390

Table 2 – Sensitivity analysis of main input parameters

Parameters Rice (%) Maize (%) Wheat (%) Soybean (%)

Crop yield 6a 15a 5a 3a

Crop price 11a 27a 7a 8a

a
A two-sample paired t-test at the 0.05 level was used to test whether the prediction was different from the baseline data.

types under different multiple cropping systems, since there

4. Sensitivity analysis and validation is only a finite amount of the Earth’s surface available for agri-
cultural land use. When the crop yield or crop price varies
4.1. Sensitivity analysis in space and time, one crop type will have either a physical
or an economic advantage over other crop types and, there-
Sensitivity analysis considers the robustness of a model’s fore, will be more likely to be selected for cultivation by a land
results to relatively small changes in the main input parame- user.
ters (Snowling and Kramer, 2001). In general, a highly sensitive
model is undesirable, given the uncertainty in model input
4.2. Model validation
(Li et al., 2005). In this study, a ‘one-at-a-time’ (OAT) method,
which is used widely by other researchers (Pohlert et al., 2007;
To validate a model empirically, one may employ either tem-
Cariboni et al., 2007), was used to examine the relative sensi-
porally or spatially independent data (Pontius et al., 2004).
tivity of the crop choice decision model to its important input
The first model validation approach used here was to com-
variables. We ran the model simulations twice for the year
pare model estimates with independently recorded historical
2001 to identify whether modifications to the two inputs, crop
data at a regional scale. The resulting similarity between pre-
yield and crop price, produced significant changes in the value
dicted and measured values at a regional level can improve
of outputs. The sensitivity was measured by the response of
our understanding of whether the simulated output is fol-
the model output to a fractional change in one single input
lowing the trend of a reported aggregate average. Due to the
parameter, while other parameters remained constant. To test
difficulty in collecting data on crop price, this kind of vali-
the model further, each of the major parameters (crop price
dation was limited to evaluating simulations for crop yield
and crop yield) was changed by a magnitude of 50%, and the
and crop sown area by using a time series of FAO statisti-
sensitivity index was calculated using the following equation
cal data for the period of 1995–2004. Figs. 5 and 6 show a
(Jorgensen, 1986):
comparative analysis between model estimates and FAO sta-
tistical data for crop yield and crop sown area, for four crops.
(dX/X) It can be seen that, although there were some places where
Sx = (3)
(dP/P) the model simulation more or less deviated from the reported
FAO values, in general the simulated and reported values were
where P is the value of the independent variable, dP is the similar to each other. The model estimates of crop yield and
value for a small change of P, X is the value of the dependent sown area for rice, maize and wheat had a higher correla-
variable and dX is the corresponding change in X in response tion with the corresponding FAO historical data, while those
to the change in P. for the soybean crop had a relatively higher deviation from
The sensitivity analysis results (Table 2) showed that the the aggregated FAO data. The main reason for these differ-
model is sensitive to both input parameters. In particular, the ences between the simulation and the observed data was
model appeared to be more sensitive to the economic factor possibly the uncertainty in estimations of crop yield made by
(crop price) than it did to the ecological factor (crop yield). This the GIS-based EPIC model and those of crop price made by
reflects the fact that changes in the crop sown areas are gen- the IFPSIM model. The results for soybean (Figs. 5d and 6d)
erally related more closely to fluctuations in crop price than to may describe this kind of relationship between the model
fluctuations in crop yield. This can probably be attributed to input and output. Most of the GIS-based EPIC crop param-
the assumption that farmers are autonomous and are capable eters established by the USDA were not modified and were
of perceiving changes in the social system. Thus, farmers are applied directly at a global level in this study, which may
more easily and intuitively able to regulate their crop choices have resulted in some underestimation or overestimation of
in response to the changes in crop price, which directly deter- crop yields that was then used by the crop choice decision
mine the profits or returns of crop cultivation on their land. model.
By contrast, farmers react gradually in response to changes in However, the non-site-specific nature of the above vali-
crop yield, since they may take a period of time to understand dation approach has limitations for the crop choice decision
new crop varieties or breeding measures that result in yield model, as a spatially explicit model could evaluate relatively
increases. easily the crop sown areas in the correct proportions but in the
The model also showed different sensitivities for different incorrect locations (Foody, 2002). Thus, another spatial com-
individual crops. Given the same size of change in both crop parison approach was needed to evaluate further the model’s
yield and crop price, the sown area of maize crop increased simulation of the spatial locations of crop sown areas. In doing
more than that of the other three crops. This interprets, to so, the MODIS 1 km global land cover dataset in 2001 and the
some extent, the competition between these different crop global distribution map of major crops in 1992 produced by
 e c o l o g i c a l m o d e l l i n g 2 0 8 ( 2 0 0 7 ) 378–390 385

Fig. 5 – Time series validation of the crop yield model (a: rice; b: maize; c: wheat; d: soybean).

Leff et al. (2004) were used as reference data for comparing One of the reasons for the difference between the model esti-
with model-simulated sown areas in 2001. Fig. 7 illustrates mates and the Leff database is that the Leff database was
the comparison between the model-predicted spatial distri- produced using the statistical data from the 1990s, and it
bution of the four test crops and the MODIS-derived cropland describes the fraction of a grid cell occupied by each of 18
distribution (Fig. 7a), and the comparison between the model crops. Uncertainties or bias in the reference data can dis-
estimation and the global distribution map of the same four tort the performance of the model validation in some way.
crops (Fig. 7b). It is obvious that in both cases the model From the evaluation results described above, it can be con-
estimates largely coincided with the reference data in the cluded that this integrated modelling approach appears to be
major agricultural regions of the globe, such as China, India, adequate for the purpose for which it was designed, and it
USA and Europe. The discrepancies between them occurred appears to be applicable for the analysis of long-term future
mainly in some regions of North America, Eastern Europe and scenarios.
Africa. From Fig. 7a and b, it is noticeable that the model
simulation was much closer to the global distribution map
of these four crops in all regions across the world than to 5. Results of future simulation
the MODIS-derived data. This was expected, since the spa-
tial distribution map of the four crops generated from the The following demonstrates the potential uses of this mod-
database of Leff et al. (2004) consisted only of these four crops, elling approach for assessing future changes in the sown
while the MODIS-derived data represented the distribution of areas of crops in a given scenario. The model application was
global cropland as a whole rather than the individual crops. designed to run over a period of approximately 30 years with a

Fig. 6 – Time series validation of the crop choice decision model (a: rice; b: maize; c: wheat; d: soybean).
386 e c o l o g i c a l m o d e l l i n g 2 0 8 ( 2 0 0 7 ) 378–390

Fig. 7 – Spatial comparison between the model estimate in 2001 and (a) MODIS dataset; and (b) global distribution map of
major crops.

5-year time increment, taking the year 2000 as baseline, and to seen clearly that all four crops showed a constant growth
analyse potential changes in sown areas for four major crops. in total sown areas, but there were noticeable differences in
For this simulation, some input data, such as road accessi- the changes in total sown areas for the different crops. Total
bility, and agricultural management data were assumed to be sown areas showed a considerable, stable increase for wheat
the same as they are currently since it is not possible to collect and rice, where sown areas were predicted to be about 209
these data for the future. and 380 million hectares in 2035 for rice and wheat, respec-
tively. Wheat showed the greatest rate of changes among the
5.1. Temporal changes in total global sown areas for four crops. Although total sown areas of maize and soybean
four major crops were predicted by the model to increase to 186 and 75 million
hectares, respectively, by 2035, changes in their sown areas
Fig. 8 presents the general trend of changes in total global generally fluctuated instead of increasing linearly, e.g. area of
sown areas for rice, maize, wheat and soybean crops as maize during the period of 2015–2025 and soybean area during
predicted by model for the period of 2005–2035. It can be the period of 2025–2030.

Fig. 8 – Temporal changes in total global sown areas for major crops during the period of 2005–2035.
 e c o l o g i c a l m o d e l l i n g 2 0 8 ( 2 0 0 7 ) 378–390 387

Fig. 9 – Sown area changes for major crops in different continental areas during the period of 2005–2035.

5.2. Spatial variation of sown area changes for four mainly in Asia, which accounts for about 90% of the total area
major crops of rice cultivation in the world. Just like wheat, maize is also
a geographically ubiquitous crop, and its cultivation area is
Fig. 9 illustrates the simulated crop sown areas and their pre- the third largest in the world. Soybean has the smallest sown
dicted changes for four crops during the period 2005–2035 area of the four tested crops, and its cultivation is distributed
in six continental areas (Asia, North America, Latin Amer- mainly in North America, Latin America and Asia.
ica, Africa, Europe and Oceania) of the world. The possible Fig. 10 shows the simulated global geospatial distribution of
global cropping patterns for major crops for this period can sown areas for four crops in 2005 (Fig. 10a) and 2035 (Fig. 10b).
be interpreted from Fig. 9. Wheat is the most abundant and The characteristics of the predicted changes in the distribu-
widespread crop in the world and the most intensive wheat tion of the major crops in different regions can be interpreted
cultivation occurs in three regions: Europe, North America and from Figs. 9 and 10. Excluding the soybean crop in North Amer-
Asia. Rice is the second largest crop in the world and is found ica and the maize crop in Asia, the sown areas of rice, maize,

Fig. 10 – Global distribution of sown areas for major crops in 2005 (a) and 2035 (b).
388 e c o l o g i c a l m o d e l l i n g 2 0 8 ( 2 0 0 7 ) 378–390

wheat and soybean were predicted to increase at different ing LUCC issues, but are aimed at climate change, biodiversity
rates from 2005 to 2035 in each of the continental areas. Specif- loss and so on (Lambin and Geist, 2006). Second, the objects
ically, rice, wheat and soybean crops in all regions showed a of analysis of this model differed from other LUCC mod-
significant trend for a constant increase in sown areas over els in that this model specifically simulates the dynamic
time. In particular, rice in Asia and Latin America and wheat changes in sown areas of crops occurring in agricultural
in all regions except Oceania, showed a substantial increase lands while other models are focused mainly on the simu-
during the later period of the model simulation. The changes lation of conversions between croplands and other types of
in sown areas for maize were uneven across the world. Asian land use like forests, grassland and urban areas. The third
areas declined, while the other regions showed a tendency to and final feature is that LUCC over time and space results
increase the sown area of maize, but with significant fluctua- from the complex interactions of a coupled human–natural
tions over time. system. The structure of this coupled model provides impor-
tant insights into the relationships between different factors
contributing to land use changes, and enables us to under-
6. Discussion and conclusions stand the process of land use changes as a system rather
than as an isolated set of independent variables (Müller et
This paper describes a new modelling approach for simulating al., 2004). In this regard, the integrated model can provide
dynamically changes in the sown areas of four major crops on a better explanation of systems of land use changes than
a global scale. The basic hypothesis was that decisions on crop could be provided by either of these research approaches
choice made by farmers mediate the impacts of biophysical applied individually (Evans et al., 2001; Castella and Verburg,
and socio-economic aspects on changes in agricultural land 2007).
use. This basic hypothesis was considered when developing The model on which future changes in crop sown areas
the model, which attempts to establish a dynamic interface were simulated also contained some uncertainties. It used
of a human–natural environment relationship in an inte- a simplified modelling approach and it was based on a
grated modelling framework. Within the overall model, a Logit few assumptions about the driving factors behind land use
model was developed to track the decision-making processes changes. Some other causes, e.g. policy change (Van Meijl
of farmers by using a crop utility function. The crop utility was et al., 2006), technological development (Verburg et al., 2006)
explained by the four independent variables: crop yield, rural and social preferences (Serneels and Lambin, 2001), which
population density, road accessibility and crop price. Of these, may also have a great influence on changes in crop sown
crop yield, which represents biogeophysical factors, was sim- areas, were not taken into account in this study. Second,
ulated from the GIS-based EPIC model and crop price, which when models are used to construct future scenarios, their
represents the socio-economic variables, was estimated by the input parameters are changed according to a set of rules
IFPSIM model. Through data exchange, the crop choice deci- that is designed to explore or depict these future scenarios.
sion model was linked with crop yield and crop price models, The inherent uncertainty of these parameter values can bring
and the dynamic feedback loop between agricultural land use about some bias in the outputs from model simulations. Fur-
changes and biophysical and socio-economic driving factors thermore, even when the causal relationship between the
was studied. parameters and land use changes was well constructed, future
A sensitivity analysis and an empirical validation using changes in land use may not necessarily be described by
data from historical observation and from other studies indi- the relationships derived from past and present observa-
cated that the integrated model is reliable for addressing tions, as land use activities are so dynamic and mechanistic
the complicated dynamic changes in agricultural land use at understanding of land use change is insufficient (Rounsevell
present and that it has the capacity to be used for inves- et al., 2006). Therefore, the model results are neither real
tigating long-term scenarios and applications in the future. predictions nor facts (Veldkamp and Lambin, 2001; Verburg
Taking the year 2000 as a benchmark, the model was applied et al., 2006), and they depend to a certain extent on the
to a simulation of future scenarios until 2035 in time incre- specified scenario. However, it can be helpful to explore
ments of 5 years. The simulation results showed the potential what might happen given certain assumptions about societal
global cropping patterns of rice, maize, wheat and soybean development and environmental changes (Rounsevell et al.,
in the future, as well as the temporal and spatial variation 2005).
in rates and direction of changes in the sown areas of these
four crops. The model outcomes can help us to understand
and explain the causes, locations and consequences of land Acknowledgements
use changes (Verburg and Veldkamp, 2005), and can provide
significant support for land use planning and policy making The research described in this paper was supported and
(Veldkamp and Lambin, 2001). financed by the Japan Society for the Promotion of Science
The bottom-up, process-oriented approach of analyzing (GASR-A-073400000002), and by the Ministry of Finance of
the dynamics of LUCC presented in this paper has three main China through Non-profit National Research Institute (IARRP-
features. First, it is characterized by the fact that it models 2007-025). All persons and institutes who kindly made their
on a global scale. Currently, numerous different models exist data available for this analysis are acknowledged. We would
that apply at a local, national or continental scale. Only a also like to thank the anonymous reviewers whose valuable
few global models of LUCC have been developed and those comments greatly helped us to prepare an improved and
global model analyses are not typically aimed at investigat- clearer version of this paper.
 e c o l o g i c a l m o d e l l i n g 2 0 8 ( 2 0 0 7 ) 378–390 389

references Müller, D., Zeller, M., 2002. Land use dynamics in the central
highlands of Vietnam: a spatial model combining village
survey data with satellite imagery interpretation. Agric. Econ.
27, 333–354.
Audsley, E., Pearn, K.R., Simota, C., Cojocaru, G., Koutsidou, E.,
Müller, B., Birr-Pedersen, K., Schou, J.S., 2004. Combined
Rounsevell, M.D.A., Trnka, M., Alexandrow, V., 2006. What can
ecological and economic modelling in agricultural land use
scenario modeling tell us about future European scale
scenarios. Ecol. Model. 174, 5–18.
agricultural land use, and what not? Environ. Sci. Policy 9,
New, M.G., Hulme, M., Jones, P.D., 1999. Representing 20th century
148–162.
space–time climate variability. I. Development of a 1961–1990
Cariboni, J., Gatelli, D., Liska, R., Saltelli, A., 2007. The role of
mean monthly terrestrial climatology. J. Climate 12, 829–856.
sensitivity analysis in ecological modelling. Ecol. Model. 203,
Ohga, K., Gehlar, C., 1993. The International Food Policy
167–182.
Simulation (IFPSIM) Model: A Documentation and
Castella, J.C., Verburg, P.H., 2007. Combination of process-oriented
Application, Washington D.C., USA.
and pattern-oriented models of land-use change in mountain
Ohga, K., Yanagishima, K., 1996. JIRCAS working report No. 1:
area of Vietnam. Ecol. Model. 202, 410–420.
international food and agricultural policy simulation model.
De Koning, G.H.J., Verburg, P.H., Veldkamp, A., Fresco, L.O., 1999.
Japan International Research Center for Agricultural Sciences
Multi-scale modelling of land use change dynamics in
(JIRCAS) Ministry of Agriculture, Forestry and Fisheris.
Ecuador. Agric. Syst. 61, 77–93.
pp. 121.
Duffy, S.B., Corson, M.S., Grant, W.E., 2001. Simulating land-use
Parry, M., Rosenzweig, C., Iglesias, A., Liveemore, M., Fischer, G.,
decisions in the La Amistad Biosphere Reserve buffer zone in
2004. Effects of climate change on global food production
Costa Rica and Panama. Ecol. Model. 140, 9–29.
under SRES emissions and socio-economic scenarios. Glob.
Dumesnil, D. (Ed.), 1993. EPIC User’s Guide-Draft. USDA-ARS
Environ. Change 14, 53–67.
Grassland Soil and Water Research Laboratory, Temple, TX.
Pohlert, T., Huisman, J.A., Breuer, L., Frede, H.G., 2007. Integration
Evans, T.P., Manire, A., de Castro, F., Brondizio, E., McCracken, S.,
of a detailed biogeochemical model into SWAT for improved
2001. A dynamic model of household decision-making and
nitrogen prediction––Model development, sensitivity, and
parcel level land cover change in the eastern Amazon. Ecol.
GLUE analysis. Ecol. Model. 203, 215–228.
Model. 143, 95–113.
Pontius Jr., R.G., Huffaker, D., Denman, K., 2004. Useful
Ewert, F., Rounsevell, M.D.A., Reginster, I., Metzger, M.J., Leemans,
techniques of validation for spatially explicit land-change
R., 2005. Future scenarios of European agricultural land use. I.
models. Ecol. Model. 179, 445–461.
Estimating changes in crop productivity. Agric. Ecosyst.
Priya, S., Shibasaki, R., 2001. National spatial crop yield
Environ. 107, 101–116.
simulation using GIS-based crop production model. Ecol.
Fischer, G., Sun, L., 2001. Model based analysis of future land-use
Model. 135, 113–129.
development in China. Agric. Ecosyst. Environ. 85, 163–176.
Ramankutty, N., Foley, J.A., 1998. Characterizing patterns of global
Foley, J.A., Defries, R., Asner, G.P., et al., 2005. Global
land use: an analysis of global croplands data. Glob.
consequences of land use. Science 309, 570–574.
Biogeochem. Cycles 12, 667–685.
Foody, G.M., 2002. Status of land cover classification accuracy
Rounsevell, M.D.A., Annetts, J.E., Audsley, E., Mayr, T., Reginster,
assessment. Rem. Sens. Environ. 80, 185–201.
I., 2003. Modelling the spatial distribution of agricultural land
Frolking, S., Xiao, X., Zhuang, Y., Salas, W., Li, C., 1999.
use at the regional scale. Agric. Ecosyst. Environ. 95, 465–479.
Agricultural land-use in China: a comparison of area
Rounsevell, M.D.A., Ewert, F., Reginster, I., Leemans, R., Carter,
estimates from ground-based census and satellite-borne
T.R., 2005. Future scenarios of European agricultural land use.
remote sensing. Glob. Ecol. Biogeo. 8, 407–416.
II. Projecting changes in cropland and grassland. Agric.
Gobin, A., Campling, P., Feyen, J., 2002. Logistic modelling to
Ecosyst. Environ. 107, 117–135.
derive agricultural land use determinants: a case study from
Rounsevell, M.D.A., Reginster, I., Araujo, M.B., Carter, T.R.,
southeastern Nigeria. Agric. Ecosyst. Environ. 89, 213–228.
Dendoncker, N., Ewert, F., House, J.I., Kankaanpaa, S.,
Greene, W.H., 1997. Econometric Analysis. Prentice-Hall,
Leemans, R., Metzger, M.J., Schmit, C., Smith, P., Tuck, G., 2006.
Englewood Cliffs, NJ.
A coherent set of future land use change scenarios for Europe.
Heistermann, M., Muller, C., Ronneberge, K., 2006. Land in sight:
Agric. Ecosyst. Environ. 114, 57–68.
achievements, deficits and potentials of continental to global
Serneels, S., Lambin, E.F., 2001. Proximate causes of land use
scale land-use modeling. Agric. Ecosyst. Environ. 114,
change in Narok district Kenya: a spatial statistical model.
141–158.
Agric. Ecosyst. Environ. 85, 65–81.
Jorgensen, S.E., 1986. Fundamentals of Ecological Modeling.
Snowling, S.D., Kramer, J.R., 2001. Evaluating modeling
Elsevier, Amsterdam.
uncertainty for model selection. Ecol. Model. 138, 17–30.
Lambin, E.F., Geist, H. (Eds.), 2006. Land-Use and Land-Cover
Tan, G., Shibasaki, R., 2003. Global estimation of crop productivity
Change: Local Processes and Global Impacts. Springer,
and the impacts of global warming by GIS and EPIC
Germany.
integration. Ecol. Model. 168, 357–370.
Leff, B., Ramankutty, N., Foley, J.A., 2004. Geographic distribution
Van Meijl, H., van Rheenen, T., Tabeau, A., Eickhout, B., 2006. The
of major crops across the world. Glob. Biogeochem. Cycles 18,
impact of different policy environments on agricultural land
GB1009, doi:10.1029/2003GB002108.
use in Europe. Agric. Ecosyst. Environ. 114, 21–38.
Li, A., Linderman, M., Qi, J., 2005. Exploring complexity in a
Veldkamp, A., Lambin, E.F., 2001. Predicting land-use change.
human-–environment system: an agent-based spatial model
Agric. Ecosyst. Environ. 85, 1–6.
for multidisciplinary and multiscale integration. Ann. Assoc.
Verburg, P.H., Veldkamp, A., 2001. The role of spatially explicit
Am. Geogr. 95, 54–79.
models in land-use change research: a case study for cropping
Liu, J., Liu, M., Tian, H., Zhuang, D., Zhang, Z., Zhang, W., 2005.
patterns in China. Agric. Ecosyst. Environ. 85, 177–190.
Spatial and temporal patterns of China’s cropland during
Verburg, P.H., Veldkamp, A., 2005. Spatial modeling to explore
1990–2000: an analysis based on Landsat TM data. Rem. Sens.
land use dynamics. Int. J. Geogr. Inform. Sci. 19, 99–102,
Environ. 98, 442–456.
Editorial.
McFadden, D., 1973. Conditional logit analysis of qualitative
Verburg, P.H., De Koning, G.H.J., Kok, K., Veldkamp, A., Bouma, J.,
choice behaviour. In: Frontiers in Econometrics. Academic
1999. A spatial explicit allocation procedure for modelling the
Press, New York.
pattern of land use change based upon actual land use. Ecol.
390 e c o l o g i c a l m o d e l l i n g 2 0 8 ( 2 0 0 7 ) 378–390

Model. 116, 45–61. Xiao, X., Boles, S., Frolking, S., Li, C., Babu, J.Y., Salas, W., Moore III,
Verburg, P.H., Chen, Y.Q., Veldkamp, A., 2000. Spatial explorations B., 2006. Mapping paddy rice agriculture in South and
of land-use change and grain production in China. Agric. Southeast Asia using multi-temporal MODIS images. Rem.
Ecosyst. Environ. 82, 333–354. Sens. Environ. 100, 95–113.
Verburg, P.H., Veldkamp, A., Rounsevell, M.D.A., 2006. Yang, H., Li, X., 2000. Cultivated land and food supply in China.
Scenario-based studies of future land use in Europe. Agric. Land use policy 17, 73–88.
Ecosyst. Environ. 114, 1–6. Yang, P., Zhou, Q., Chen, Z., Zha, Y., Wu, W., Shibasaki, R., 2006.
Vitousek, P.M., Mooney, H.A., Lubchenco, J., Melillo, J.M., 1997. Estimation of regional crop yield by assimilating
Human domination of earth’s ecosystems. Science 277, multi-temporal TM images into crop growth model. In:
494–499. Proceedings of IEEE International Geoscience and Remote
Williams, J.R., et al., 1990. EPIC-Erosion/Productivity Impact Sensing Symposium 2006 (IGARSS’06), Colorado, USA, pp.
Calculator. United State Department of Agriculture, 2259–2262.
Agricultural Research Service. Technical Bulletin Number
1768. Springfield, VA, USA.