Modeling urban land use change by the integration of cellular automaton and Markov model

Ecological Modelling 222 (2011) 3761–3772

Contents lists available at SciVerse ScienceDirect

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

Modeling urban land use change by the integration of cellular automaton and Markov model DongJie Guan a,∗ , HaiFeng Li b , Takuro Inohae c , Weici Su d,e , Tadashi Nagaie c , Kazunori Hokao b,c a

Heihai Department, Chongqing Jiaotong University, No. 66 Xuefu Rd., Nan’an Dist., Chongqing 400074, China Faculty of Science and Engineering, Chongqing University, 1, Honjo-machi, Saga 840-8502, Japan Institute of Lowland and Marine Research, Saga University, 1, Honjo-machi, Saga 840-8502, Japan d Faculty of Geographic Science, Chongqing Normal University, No.12 Tianchenlu Road, Shapingba District, Chongqing 400047, China e Institute of Mountain Resources, Guizhou Academy of Science, Guiyang 550001, China b c

a r t i c l e

i n f o

Article history: Received 12 July 2011 Received in revised form 13 September 2011 Accepted 14 September 2011 Keywords: Land use change GIS Markov model Cellular Automata model Saga in Japan

a b s t r a c t Spatially land use models are indispensable for sustainable land use planning. This study demonstrates a combined Markov–Cellular Automata model to analyze temporal change and spatial distribution of land use stressed by natural and socioeconomic factors in Saga, Japan. Firstly, area change and spatial distribution of land use are calculated using GIS technology, and then the transition among different land use types is analyzed to obtain the transformation matrices during a period of 1976–2006. Meanwhile, an integration evaluation procedure with natural and socioeconomic data is used to generate the transition potential maps. Secondly, using the transition potential maps and transition matrices, a Markov–Cellular Automata model is established to simulate spatial distribution of land use in 2006. Finally, we use this Markov–Cellular Automata model to forecast the future land use changes during the period of 2015–2042. As a consequence, area change simulation predicts a continuing downward trend in agriculture land and forestland areas, as well as an upward trend in built-up areas; spatial distribution simulation indicates that built-up land will expand toward suburban regions, and land use of urban center is at the decline stage. Hence, if the current trends keep constant without holistic sustainable development measures, severe land use decline will ensue. The study is anticipated to help local authorities better understand and address a complex land use system, and develop the improved land use management strategies that can better balance urban expansion and ecological conservation. © 2011 Elsevier B.V. All rights reserved.

1. Introduction Land use change is one of the main research subjects of global environmental change and sustainable development. The intensity of land use change in response to world population growth and its consequences for the environment warrant in-depth studies of these transformations (Wu et al., 2006). Several international interdisciplinary research projects have been initiated during the past two decades for this purpose. These include the International Geosphere-Biosphere Project (IGBP) and International Human Dimensions Program (IHDP) (Messerli, 1997). Both of these projects indicated the need to construct an updated and accurate database concerning these changes, their meaning, their pace and the explanatory factors prompting their appearance (Mather, 1999). Meanwhile, IGBP and IHDP have launched a plan of “Land Use/Cover Change (LUCC)” in 1995, since then LUCC has been an advanced and hot subject in global environment change research

∗ Corresponding author. E-mail address: guandongjie [email protected] (D. Guan). 0304-3800/$ – see front matter © 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.ecolmodel.2011.09.009

(Meyer and Turner, 1996; Verburg et al., 1999; Dai et al., 2001; Geist and Lambin, 2001; Veldkamp and Lambin, 2001; Susanna and Chen, 2002; Honnay et al., 2003; Quan et al., 2006; Luciana et al., 2007; Ge et al., 2007; Fikir et al., 2009). In an international comprehensive review, studies on LUCC can be summarized as three core issues: dynamic analysis of process, driving forces, and global and regional models of LUCC (Henk and Latesteijn, 1995; Fischer and Sun, 2001; Pijanowski et al., 2002; Gautam et al., 2003; Kline, 2003; Aspinall, 2004; Patma et al., 2004; Erika et al., 2005; Shao et al., 2005; Guan et al., 2008). Land use models are core subject of LUCC. In recent years, the LUCC community has produced a large set of operational models that can be used to predict or explore possible land use change trajectories (Verburg et al., 2006). Models cannot only support the exploration of future land use changes under different scenario conditions, Scenario analysis with land use models can but also support land use planning and policy. Until now, all these models were divided into three classes: empirical and statistical models, such as Markov chains and Regression model, etc.; dynamic models, such as Cellular Automata (CA) model, Agent-based model and System dynamic model, etc.; integrated model, such as CLUE (Conversion of Land Use and its Effects)

3762

D. Guan et al. / Ecological Modelling 222 (2011) 3761–3772

model, etc. Empirical and statistical models can complement dynamic simulation. Dynamic models appear to be better suited to predict land use changes in the future than empirical and statistical models. An integrated model that is based on multidisciplinary and combing elements of different modeling techniques will probably best serve the objective of improving and understanding land use change processes. A Markov–CA model incorporated with geographic information system (GIS) data is claimed to be a suitable approach to model the temporal and spatial change of land use (Myint and Wang, 2006; Courage et al., 2009). In the Markov–CA model, the Markov chain process controls temporal change among the land use types based on transition matrices (Lópeza et al., 2001); CA model controls spatial pattern change through local rules considering neighbourhood configuration and transition potential maps (White and Engelen, 1993; Clarke et al., 1994; Wu, 2002; Li and Yeh, 2004; Thomas and Laurence, 2006; Liu et al., 2007; He et al., 2008). GIS can be used to define initial conditions, to parameterize the Markov–CA model, to calculate transition matrixes, and to determine the neighbourhood rules (Batty et al., 1999; Weng, 2002; Aitkenhead and Aalders, 2009). Although the potential of the Markov–CA model has been recognized by some researchers (Li and Reynolds, 1997; Courage et al., 2009), few studies have combined natural and socioeconomic data to simulate land use changes. Recently, several reports have attempted to incorporate natural and socioeconomic data into land use simulations by using Markov–CA model. For example, Courage et al. (2009) combined physical and socioeconomic data into Markov–CA model to simulate future land use change. However, the model did not successfully forecast the location of the bareland class due to the shortage of spatial data. Yu (2009) forecasted future land use change based on Markov–CA model. However, authors used the relatively outdated data as a forecasting base and did not consider the yearly-changing impact intensity of socio-economic development, policy changes, and other factors on land use change, leading to inaccurate prediction result. Therefore, the integration of natural and socioeconomic data still remains a major research challenge because of the discrepancy among these different datasets. The objective of this study is to simulate future land use changes based on the Markov–CA model with natural and socioeconomic data in Saga, Japan. Firstly, transition matrix is computed from the land use maps (1976, 1987, 1997 and 2006) using the Markov model to forecast area change of land use. Secondly, an integration evaluation procedure is used to generate transition potential maps based on natural and socioeconomic indicators. Finally, transition matrix and transition potential map are implemented in the Markov–CA model to simulate spatial distribution of land use from 2015 to 2042. 2. Materials and methods 2.1. Study area Saga is the capital of Saga Prefecture, located on the island of Kyushu, Japan. After merger in 2005, the city became very long in the direction of north–south, as shown in Fig. 1. It borders the Ariake Sea to the south and Fukuoka Prefecture to the southeast and north. Its total area is 431.42 km2 , population is 238,934 till February 1, 2009, and population density is 554 per km2 . For a long time, the region has been chiefly aiming at urban expansion and ignoring reasonable adjustment of land use structure. So, potential disadvantages are threatening its sustainable development of urban land use. In this paper, land use types are divided into 6 classes (agricultural land, forestland, water, built-up land, road and other land) according to national classification standard of land use. Other land contains barren land and beaches.

2.2. Methods This study employs a coupled Markov–CA model that integrates GIS software to simulate land use changes and spatial distribution in the future. The detailed steps are shown in Fig. 2. Firstly, we obtained land use maps from 1976 to 2006 with GIS technology. Then, transition matrixes were established using Markov chain analysis. Secondly, seven assessment indicators are selected to compute transition potential maps of land use. Finally, we used the transition matrices and transition potential maps to simulate spatial distribution of land use on the basis of the transition rule of CA model. 2.2.1. Computation of transition matrix using Markov model Markov process is a special random moving from one state to another state at each time step. A first-order Markov model is the model of a system in which probability distribution over next state is assumed to only depend on current state, but not on previous ones (non-aftereffect) (Veldkamp and Lambin, 2001; Fischer and Sun, 2001; Pijanowski et al., 2002). This characteristic of Markov process is appropriate to be applied in change of land use structure, because dynamic change of land use also possesses the properties of Markov process under the following certain conditions: (1) within a certain region, different land use types may be transformed into each other; (2) mutual conversion process between land-use types includes many incidents which are difficult to be described precisely by some special function; (3) during the studied periods, average transfer state of land use structure is relatively stable and accordant with the requirements of Markov chain. At first, original transition probability matrices of land use type need to be defined prior to Markov process. Its mathematics expression is as follows:

   P11 P12 ...P1n     P P ...P2n  P = (Pij ) =  21 22   ...   Pn1 Pn2 ...Pnn 

In the above matrix, Pij is the transformation probability of the ith type land into the jth type land from prophase to telophase; n is the land use type of studied area. Pij should meet the following conditions 0 ≤ Pij ≤ 1(i, j = 1, 2, 3, ..., n) n 

Pij = 1(i, j = 1, 2, 3, ..., n)

i=1

According to non-aftereffect of Markov process and probability formulae of Bayes condition, forecast model of Markov is obtained: P(n) = P(n−1) Pij P(n) is state probability of any times; P(n−1) is preliminary state probability. Firstly, land use map used in the paper spans 30 years in which four year nodes of 1976, 1987, 1997 and 2006 were selected for the studied years. Area change and spatial distribution of land use types (agriculture land, forestland, built-up land, road, water and other land) at the four year nodes are shown in Figs. 3 and 4, respectively. Then, land use maps at different year nodes were spatially overlaid and operated with GIS technology. Finally, we applied the map algebra to calculate transition values of land use maps in the period between two year nodes. Successively, transition matrices

D. Guan et al. / Ecological Modelling 222 (2011) 3761–3772

3763

Fig. 1. Location of study area.

of land use types in each period were obtained by using Markov model shown in Table 1, and then land use change process was further analyzed. 2.2.2. Computation of transition potential maps Typical natural and socioeconomic indicators, including slope, elevation, distance to the nearest road and distance to the nearest river, population density, GDP per capita, and land price, are selected to compute transition potential maps of land use. Using GIS technology, spatial distribution of each indicator is obtained as shown in Fig. 5. Then, we used Analytic Hierarchy Process (AHP) to determine indicator weight of land use

transition potential. The AHP provides a comprehensive and rational framework for structuring a decision problem, representing and quantifying its elements, correlating those elements to overall goals, and evaluating alternative solutions. The AHP is most useful to deal with complex problems, especially those with high stakes, involving human perceptions and judgments, whose resolutions have long-term repercussions (Hafeez et al., 2002; Júlíus, 2003; Xiong et al., 2007; Li et al., 2007). It has unique advantages when important elements are difficult to quantify and compare, or where communications among the working team members are impeded by their different specializations, terminologies, or perspectives.

Table 1 Transition probability matrix of land use change in different periods of 1976–2006 in Saga. Land use type

Year

Agriculture land

Forestland

Built-up land

Road

Water

Other land

Agriculture land

1976–1987 1987–1997 1997–2006

0.9354 0.9549 0.9350

0.0138 0.0000 0.0200

0.0304 0.0308 0.0242

0.0044 0.0026 0.0009

0.0043 0.0001 0.0023

0.0117 0.0116 0.0176

Forestland

1976–1987 1987–1997 1997–2006

0.0049 0.0063 0.0059

0.9900 0.9867 0.9897

0.0005 0.0017 0.0024

0.0003 0.0002 0.0000

0.0000 0.0001 0.0002

0.0043 0.0050 0.0018

Built-up land

1976–1987 1987–1997 1997–2006

0.1124 0.0000 0.0054

0.0014 0.0000 0.0070

0.8660 0.9992 0.9858

0.0046 0.0005 0.0009

0.0058 0.0000 0.0002

0.0098 0.0003 0.0007

Road

1976–1987 1987–1997 1997–2006

0.0000 0.0000 0.0000

0.0000 0.0000 0.0000

0.0000 0.0385 0.0000

1.0000 0.9615 1.0000

0.0000 0.0000 0.0000

0.0000 0.0000 0.0000

Water

1976–1987 1987–1997 1997–2006

0.0370 0.0214 0.0026

0.0006 0.0000 0.0026

0.0148 0.0214 0.0013

0.0017 0.0017 0.0000

0.9413 0.9549 0.9922

0.0046 0.0006 0.0013

Other land

1976–1987 1987–1997 1997–2006

0.0916 0.0785 0.0101

0.1388 0.0000 0.1180

0.1089 0.1189 0.0373

0.0066 0.0023 0.0008

0.0133 0.0000 0.0016

0.6408 0.8003 0.8322

3764

D. Guan et al. / Ecological Modelling 222 (2011) 3761–3772

Fig. 2. Research flow chart.

15%

12%

40%

9%

30% Built-up land Water Agriculture land

6%

Road Other land Forestland

20%

3%

10%

0%

0% 1976

1987

1997

advice (Liang et al., 2004). A detailed analytic process is written below. (1) Expert judgment Relative importance of indicators C1 , C2 . . ., C7 is analyzed by Delphi method, also so-called ‘Expert Judgment’. Expert judgment is summarized in Table 1. (2) Calculate the product of every row Mi Mi =

50%

Agriculture land and Forest land

Built-up land, Road, Water and Other land

In detail, we invited 23 experts in research fields of land use, urban planning, and ecological science to give the relative importance of each factor. Each of invited 23 experts gave a matrix of relative importance of assessment indicators. Then, we employed a reliability-determining method in group decision proposed by Liang et al. to determine a consensus matrix of all experts, where interaction between direct and indirect judgments need be established to evaluate the consistency and reliability of experts’

m 

Cij (i = 1, 2, . . . , m)

j=1

Calculation results: M1 = 2.25, M2 = 0.1317, M3 = 0.0176, M4 = 0.0176, M5 = 288, M6 = 16.856, M7 = 2.25. (3) Get the cubic root of Mi 1/m

ˇi = Mi

Calculation results: ˇ1 = 1.1228, ˇ2 = 0.7486, ˇ3 = 0.5614, ˇ4 = 0.5614, ˇ5 = 2.2456, ˇ6 = 1.4971, ˇ7 = 1.1228. (4) Get the weight of C1 , C2 , C3 , C4 and C5

2006

Year Fig. 3. Area change of land use from 1976 to 2006 in Saga.

wi =

ˇ

mi

ˇ i=1 i

(i = 1, 2 . . . m)

D. Guan et al. / Ecological Modelling 222 (2011) 3761–3772

3765

Fig. 4. Space distribution of land use type from 1976 to 2006 in Saga.

In the equation, CI = (max − n)/(n − 1) = 0.0013, RI is average random consistence index. When m = 7, RI = 1.32, CR (random consistence index) = 0.000985 ≤ 0.10, thus weight of assessment indicators is thought to be acceptable.

According to the above calculation, weight of assessment indicators is gotten (Table 2). (5) Calculate the maximum eigenvalue 1  (Bw)i = 7.0072 mwi m

max =

i=1

(6) Use CR = CI/RI to carry out consistent test, when CR ≤ 0.10, it means that the consistence of this matrix is acceptable.

This study uses a linear combination method to calculate transition potential map of land use. Because the areas of water and road are small, transition potentials to water and road are not computed. We compute the transition potential maps of land use, using natural and socioeconomic data and weights derived from the AHP procedure (Fig. 6).

Table 2 Relative importance and weight of assessment indicators from expert judgment results. B: Y to X

X C1

Y

C1 C2 C3 C4 C5 C6 C7

Slope Elevation Distance to the nearest river Distance to the nearest road Population density GDP per capita Land price

1 2/3 1/2 1/2 2 4/3 1

Weight C2 3/2 1 3/4 3/4 3 2 3/2

C3

C4

2 4/3 1 1 4 8/3 2

2 4/3 1 1 4 8/3 2

Note: Numbers 1, 2, 3, and 4, represent the nominal scales of preferred importance of indicators.

C5

C6 1/2 1/3 1/4 1/4 1 2/3 1/2

C7 3/4 1/2 3/8 3/8 3/2 1 3/4

1 2/3 1/2 1/2 2 4/3 1

0.143 0.095 0.071 0.071 0.286 0.190 0.143

3766

D. Guan et al. / Ecological Modelling 222 (2011) 3761–3772

Fig. 5. Spatial distribution of assessment index for transition potential map.

2.2.3. CA model In this study, cellular lattice represents each land use cell, and each lattice has 8 neighboring cells; cellular state represents land use type of cell; time step is 9 years; transition rule uses a 3 × 3 neighbourhood and follows the special rules depending on the actual objective. CA’s transition rules use a 3 × 3 neighbourhood to judge land use type in the future. State of each cell is affected by the states of its 8 neighboring cells in the filter. The 8 neighbors have six cell states: agriculture land, forestland, built-up land, other land, water, and road. Correspondingly, future land use of a cell is decided by the land use type which is the most one among the land use types in its 8 neighboring cells; if several land use types have the same number of cells, the future cell state is decided analogically according to land use types of 14 cells around the 8 cells. Because different land use types have different priorities, for instance, urban areas can not become crops, water areas are not expected to change in the future (Thomas and Laurence, 2006), as well as in accordance with land use policy and urban planning guideline in Saga, we define the following priority sequence rules of land use transition: if the land use type is built-up land, its spatial allocation to various land use types follows a priority sequence of forestland → agriculture → road → other land → water; if the land use type is agriculture, its spatial allocation to various land use types follows a priority sequence of built-up land → forestland → other land → water → road; if the land use type is forestland, its spatial allocation to various land use types follows a priority sequence of agriculture → built-up land → other land → water → road; if the land use type is other land, spatial allocation of various land

use types follows a priority sequence of forestland → built-up land → agriculture → water → road. Besides CA transition rule and priority sequence rule, land use transition also follows the other two rules: (1) maximum transition probability rule: a land use type is successively allocated into the cells as a descending sequence of transition probability of this land use type in all the cells; and (2) hysteresis rule: if a cell is allocated with a land use type, the cell will be not changed to other land types within the simulation period (Li and He, 2008). 2.2.4. Combination process of Markov model and CA model According to the rule of CA model, future land use of the cell will be determined by a most land use type quantitatively among the 8 neighboring land use types. Three datasets, (1) land use base map in 2006, (2) transition probability matrix from 1997 to 2006, and (3) transition potential maps in 2006, are integrated using CA neighbourhood to simulate land use map in the future. Shown in Fig. 7 is a detailed combination process. Firstly, transition potential maps in 2006 are input to identify transition potential probability of each land use type, subsequently, transition potential map of built-up land is first extracted by using selection function of GIS. Secondly, according to transition probability matrix of built-up land, we can determine how many transition cells of built-up from 2006 to 2015 are converted to all other types of land use. Also, we can determine which built-up land cell will be converted with the above result, maximum transition probability rule, and transition potential map. Thirdly, on a basis of the amount and position of transition cells determined above, CA transition rule is utilized to spatially allocate the built-up from 2006 to next year. Fourthly, if one land

D. Guan et al. / Ecological Modelling 222 (2011) 3761–3772

3767

Fig. 6. Transition potential maps of land use type in 2006.

use type (e.g. built-up land) which is undergoing the spatial allocation reaches to the total amount, spatial allocation of next land use type begins; otherwise, transition cell number from built-up to all land use type should be slightly modified until the number of converted cells is equal to the value obtained from the above transition probability matrix. Finally, following the same procedure, spatial distribution maps of all four land use types are merged with water and road maps in 2006 to obtain an overall land use map in next year. 3. Results 3.1. Model validation For model validation, we compare simulated land use area with actual one, and also analyze the simulated land use maps with the actual ones in 2006 with GIS software. The testing results of area change are shown in Fig. 8, in which six land use types have low relative errors lower than 5%. The best agreement is shown in the forestland type, where the actual area is 12.3 km2 , while the corresponding simulated area is 11.6 km2 . So, the developed Markov model is verified to effectively

predict area change of land use in the future. Visual analysis results seen from Fig. 9 indicate that agriculture, forestland, and built-up land in the simulated land use map are relatively close to the corresponding classes in the actual land use map. Meanwhile, the Markov–Cellular Automata’s overall simulation success is 95.13% in 2006. However, the overlay method cannot provide information about the morphology of the urban spatial structures. Considering that landscape indicators are capable of describing the characteristics of spatial patterns and provide useful insights about urban morphology (Liu et al., 2008), for the model validation in terms of spatial pattern, we further calculate the simulated and actual landscape indicators (fractal dimension and shape index) to examine the differences between the simulated patterns and the actual ones on the landscape metrics as seen in Fig. 10. The small differences on these metrics show a good conformity between the simulated and actual patterns in term of landscape structure, which indicates that similarity of the spatial distribution of land uses in both maps is in consistency. Therefore, the Markov–CA model can be used to forecast the spatial distribution of land use in the future. Therefore, the Markov–CA model can be used to forecast the spatial distribution of land use in the future.

3768

D. Guan et al. / Ecological Modelling 222 (2011) 3761–3772

Transition potential maps in 2006 Input

Determine transition grid number from built-up land to all land use types

Identify transition potential probability of each land use type

Extract transition potential map of built-up land

Spatially transmitting as the descending transformation probability Transition probability matrix of built-up land No Spatial allocation from built-up land to all land use types

CA transition rule

If the demand for total amount is achieved? Yes

Determine transition grid number from agriculture land to all land use types

Spatially transmitting as the descending transformation probability

Extract transition potential map of agriculture land

Transition probability matrix of agriculture land No Spatial allocation from agriculture land to all land use types

CA transition rule

Determine transition grid number from forestland land to all land use types

Spatially transmitting as the descending transformation probability

If the demand for total amount is achieved? Yes Extract transition potential map of forestland

Transition probability matrix of forestland No Spatial allocation from forestland to all land use types

CA transition rule

If the demand for total amount is achieved? Yes

Determine transition grid number from other land to all land use types

Spatially transmitting as the descending transformation probability

Extract transition potential map of other land

Transition probability matrix of other land No Spatial allocation from other land to all land use types

CA transition rule

If the demand for total amount is achieved? Yes

Land use map in next year

Spatial allocation map of water in 2006 Output Spatial allocation map of road in 2006

Merge spatial allocation map of four land use types

Fig. 7. Combination process flowchart of transition potential map, Markov model and CA model.

3.2. Analysis of transition matrix Analysis of land use area changes in Fig. 3 indicates that agriculture, forestland and built-up land were the dominant types of changed land use in the studied area. From 1976 to 2006, forestland areas decreased from 43.33% to 43.03%, while in 1987 they slightly increased to 43.93%. However, agriculture area decreased significantly from 40.67% to 36.51% during 1976–2006. During this period, built-up area increased from 8.24% to 11.80%. Other land area reduced by 3.05% from 1976 to 1987, and then dramatically

increased by 3.38% from 1987 to 2006. The area of water and road was increased a little. Spatial distribution of land use types in Saga City at the four year nodes were also obtained in virtue of GIS spatial technology. From Fig. 4, we can see that spatial pattern of land use in the four periods was mainly characterized as the changes of patches distribution. Firstly, agricultural land, built-up land, and unused land patch showed continual fragmentation and dispersion, indicating that the human’s exploration extent and utilizing intensity became larger and larger. Secondly, patch numbers and patch areas

D. Guan et al. / Ecological Modelling 222 (2011) 3761–3772

3769

20

1.55

16

Actual shape index

Simulated shape index

Actual fractal dimension

Simulated fractal dimension

1.45 12

8 1.35 4

0

1.25 Agriculture land Built-up land Forestland

Fig. 8. Simulated change versus actual change of land use type in 2006.

of build-up land kept increasing and showed the diffused distribution patterns from urban center to suburban region. This is because urban construction of Saga City accelerated urban development and 1ed to rapid expansion of urban land use. Thirdly, urban center appeared more patches of other land type from 1987 to 2006, suggesting that land use extent of Saga City had been situated in a declining stage. The transition probability matrices of land use for the 1976–1987, 1987–1997 and 1997–2006 periods, calculated on the basis of the Markov model, are shown in Table 1. The diagonal of the transition probability represents the self-replacement probabilities, that is the probability of a land use type remaining the same (shown in bold in Table 1), whereas the off-diagonal values indicate the probability of a change occurring from one land use type to another. While there is a 11-year time lag for the 1976–1987 matrix, 10-year time lag for the 1987–1997 matrix and 9 year time lag for the 1997–2006 matrix. From 1976–1987 matrix, it is evident that the converted areas of agriculture land, built-up land, and other land were large; on another hand, built-up land was increased by 5.2 km2 from agriculture land and 1.64 km2 from other land, respectively. 1987–1997 matrix shows the different area change of land use types from 1976 to 1987. Although agriculture land and other land kept marked changes, converted area of forestland also

Other land

Road

Water

Fig. 10. Comparison of simulated and actual spatial pattern changes of land use types in 2006.

began to increase. Built-up land was not converted into agriculture land any more and a small fraction to road and other land. The increased built-up land was also mainly derived from agriculture land and other land. From 1997–2006 matrix, the converted areas of agriculture land, forestland, and other land were large. The increased built-up area was mainly derived from agriculture land and forestland. The total area of agriculture land decreased from 1997 to 2006, which was mainly transferred into built-up land and other land. Meanwhile, the conversion tendencies for agriculture land, forestland and built-up land into other land were enhanced. As a result, the total region transformation tendency of land use types from 1976 to 2006 appeared an unbalanced tendency of unidirectional-transformation. The built-up land was continuously increased, and the agriculture land was continuously decreased. The increasing source of built-up land was changed from the previous agriculture land and other land, to the present forest land and agriculture land. 3.3. Analysis of transition potential Seen from Table 2, ‘slope’, ‘population density’, ‘GDP per capita’, and ‘land price’ have more significant effects on the transition

Fig. 9. Actual map and simulated map of land use type in 2006.

D. Guan et al. / Ecological Modelling 222 (2011) 3761–3772

20%

50%

16%

40%

12%

30%

8%

Built-up land

Road

Water

Other land

Agriculture land

Forestland

4%

20%

10%

0%

Agriculture land and Forestland

Built-up, Road, Water and Other land

3770

0% 2015

2024

2033

2042

into account, key problems in the land use development can be identified for immediate policy interventions. Seen from Fig. 5, the south region is urban center of Saga City, where population density, GDP per capita, and land price are high, and distance to river and road is small; slope and elevation are high in the north region. Transition potential of each land use type is obtained by combining spatial distribution, standard and weight of assessment index. Fig. 6 represents the transition potential from one land use type to all other land use types. For example, transition potential of agricultural land indicates the transition potential of agricultural land to all other land use types (forest, built-up, water, road, and other land). Also seen from Fig. 6, transition potential of agriculture land and forestland are high; transition potential of built-up land and other land are low.

Fig. 11. Area change of land use from 2015 to 2042 in Saga.

3.4. Analysis of simulation results potential maps. Thus, they determine the quantity and location of the simulated future changes of land use types, particularly built-up land. The other factors of ‘elevation’, ‘distance to the nearest road’, and ‘distance to the nearest river’ are listed with smaller weights to influence the spatial allocation of urban central land, such as extension of agriculture land and other land. Because the simulated land use map takes significant natural and socioeconomic factors

Because the accuracy of forecast cannot be guaranteed if different potentials are used, thus, we only used the transition probability matrix of the latest 1997–2006 period to forecast land use change in the future 30 years. Firstly, 2006 year is set as starting year; transition probability matrix of 1997–2006 periods is used to forecast 2015 year land use change; then, 2015 year is set as starting year; transition probability matrix of 1997–2006 periods is used

Fig. 12. Simulated map of land use type from 2015 to 2042 in Saga.

D. Guan et al. / Ecological Modelling 222 (2011) 3761–3772

to forecast 2024 year land use change; thirdly, 2024 year is set as starting year; transition probability matrix of 1997–2006 periods is used to forecast 2033 year land use change; finally, 2033 year is set as starting year; transition probability matrix of 1997–2006 periods is used to forecast 2042 year land use change. Based on the successful simulation of area change and spatial distribution in 2006, we forecast the area change of future land use and land use maps from 2015 to 2042 as shown in Figs. 11 and 12, by using land use base map in 2006, transition probability matrix of 1997–2006 period, and transition potential maps in 2006. Seen from Fig. 11, area change results show that agriculture land areas would decrease from 36% to 28% in the study area, while built-up land would increase slightly from 12% to 16%. Other land areas would also increase from 3.3% to 3.7%. Conversely, forestland areas would slightly decrease from 44% to 41%. Seen from Fig. 12, spatial distribution results indicate that all land use types would exhibit the concentrated spatial distribution patterns; urban built-up land would expand to suburban, because agriculture land in the suburban areas would rapidly convert into built-up land. Meanwhile, the transformation tendencies for forestland and built-up land into other land would be enhanced. In the view of total study area, we can see very strong transformation tendencies of all land types into built-up land. Generally, urban built-up area is based on the scale of urban population. The population of Saga City reached to a peak in 1995, and later goes to a downward tendency. According to Population Projections 001-2050 for Japan announced by National Institute of Population and Social Security Research, the population of Saga City will decrease by 17.2% from 246,674 in 1995 to 204,338 in 2035. The decreasing tendency in population is a phenomenon seen in almost all local cities in Japan. However, in the simulation results of this paper, the rate of built-up area will increase from 12% in 2006 to 16% in 2042. This result is consistent with the present status that some low-unused lands such as vacant building, empty stores and parking, etc. appear in the urban region. Meanwhile, with extension of urban planning region and coming of motorization society, some commercial lands and public lands in the urban center are moved into the suburbs. Population in the urban center also appears a decreasing tendency because of commerce transformation. These activities will lead to rapid increase of built-up land in suburban areas, although population of Saga City will decrease in the future. Thus, it is necessary to recover downtown for retaining the function of urban center and improving environment in Saga.

4. Discussion and conclusions In this study, using land use maps (1976, 1987, 1997 and 2006 year), and natural and socioeconomic data, we combine Markov–CA model with GIS technology to successfully simulate the land use changes in Saga, Japan. Our model is validated with the actual data of 2006 and shows satisfied reliability. Based on the validation, the Markov–CA model is used to simulate the future land use changes up to 2042. The area change results show the decrease in agriculture land and forestland, and the increase in built-up land. From the aspect of spatial pattern change of land use from 2015 to 2042, urban built-up land would expand to suburban, because a certain proportion of agriculture land would be transformed into urban land, reflecting that agriculture land would be used for urban land in the future; whereas, other land at the urban center would increase, suggesting that Saga City is at a declining stage of urban development. As a consequence, if the sustainable development policies with the cooperation of resident are not constituted to modify the current trends of land use, land use condition will seriously deteriorate to threat the urban sustainability. In summary, our study presents an important contribution to land use modeling, which logically integrates natural and

3771

socioeconomic data into a spatially explicit Markov–CA model with GIS technology to successfully simulate and forecast the temporal and spatial changes land use. Facing to declining problems of urban land in Saga, the simulated future land use maps can serve as an early warning system for understanding the future effects of land use changes. The acquired information is beneficial to other communal areas in Asia experiencing similar land use changes. The simulation results can also be considered as a strategic guide to urban land use planning, and help local authorities better understand a complex land use system and develop the improved land use management that can better balance urban expansion and ecological environment conservation. We also address that to further improve our Markov–CA model, e.g. using Markov model to reveal the transition rule of CA in the case of land use as proposed by Liu et al. (2008), and to demonstrate how such models can help solve practical planning issues are required. For instance, besides the prediction purpose, Markov–CA model may interactively simulate different evolution scenarios and further provides some solutions to the present problems as suggested recently (Inés et al., 2010). Acknowledgment This research is supported by Fund of Institute of Lowland and Marine Research in 2010, National “Twelfth-Five” Science and Technology Support Program (No. 2011BAC02B02) in China, and Innovative Team Project of Chongqing Municipal Education Commission in China (No. 201012). References Aitkenhead, M.J., Aalders, I.H., 2009. Predicting land cover using GIS, Bayesian and evolutionary algorithm methods. J. Environ. Manage. 90, 236–250. Aspinall, R., 2004. Modelling land use change with generalized linear models—a multi-model analysis of change between 1860 and 2000 in Gallatin Valley, Montana. J. Environ. Manage. 72, 91–103. Batty, M., Xie, Y., Sun, Z.L., 1999. Modeling urban dynamics through GIS based cellular automata. Comput. Environ. Urban. Syst. 23, 205–233. Courage, K., Masamu, A., Bongo, A., Munyaradzi, M., 2009. Rural sustainability under threat in Zimbabwe—simulation of future land use/cover changes in the Bindura district based on the Markov–cellular automata model. Appl. Geogr. 29, 435–447. Clarke, K.C., Brass, J.A., Riggan, P.J., 1994. A cellular-automaton model of wildfire propagation and extinction. Photogramm. Eng. Remote Sens. 60, 1355–1367. Dai, F.C., Lee, C.F., Zhang, X.H., 2001. Gis-based geo-environmental evaluation for urban land-use planning: a case study. Eng. Geol. 61, 257–271. Erika, L., Eric, F., Lambin, A.C., Janetos, R.D., Frederic, A., Navin, R., Robert, J.S., 2005. A synthesis of information on rapid land-cover change for the period 1981–2000. Bioscience 55, 115–124. Fikir, A., Nurhussen, T., Jan, N., 2009. The impacts of watershed management on land use and land cover dynamics in Eastern Tigray (Ethiopia). Resour. Conserv. Recycl. 53, 192–198. Fischer, G., Sun, L.X., 2001. Model based analysis of future land-use development in China. Agric. Ecosyst. Environ. 85, 163–176. Gautam, A.P., Webb, E.L., Shivakoti, G.P., 2003. Land use dynamics and landscape change pattern in a mountain watershed in Nepa1. Agric. Ecosyst. Environ. 99, 83–96. Geist, H.J., Lambin, E.F., 2001. What drives tropical deforestation. LUCC Report Series 4, l–2. Ge, J., Qi, J., Lofgren, B.M., Moore, N., Torbick, N., Olson, J.M., 2007. Impacts of land use/cover classification accuracy on regional climate simulations. J. Geophys. Res. 112 (D05), 107, doi:10.1029/2006JD007404. Guan, D.J., Gao, W.J., Watari, K., Fukahori, H., 2008. Land use change of Kitakyushu based on landscape ecology and Markov model. J. Geogr. Sci. 18, 455–468. Hafeez, K., Zhang, Y.B., Malak, N., 2002. Determining key capabilities of a firm using analytic hierarchy process. Int. J. Prod. Econ. 76, 39–51. He, C.Y., Okada, N., Zhang, Q.F., Shi, P.J., Li, J.G., 2008. Modelling dynamic urban expansion processes incorporating a potential model with cellular automata. Landscape Urban Plann. 86, 79–91. Henk, C., Latesteijn, V., 1995. Assessment of future options for land use in the European Community. Ecol. Eng. 4, 211–222. Honnay, O., Piessens, K., VanLanduyt, W., Hermy, M., Gulinck, H., 2003. Satellite based land use and landscape complexity indices as predictors for regional plant species diversity. Landscape Urban Plann. 63, 241–250. Inés, S., Andrés, M.G., David, M., Rafael, C., 2010. Cellular automata models for the simulation of real-world urban processes: a review and analysis. Landscape Urban Plann. 96, 108–122.

3772

D. Guan et al. / Ecological Modelling 222 (2011) 3761–3772

Júlíus, S., 2003. Environmental quality indexing of large industrial development alternatives using AHP. Environ. Impact Asses. 23, 283–303. Kline, J.D., 2003. Characterizing land use change in multidisciplinary landscape-level analyses. Agr. Resour. Econ. Rev. 32, l03–l15. Li, H., Reynolds, J.F., 1997. Modeling effects of spatial pattern, drought, and grazing on rates of rangeland degradation: a combined Markov and cellular automaton approach. In: Quattrochi, D.A., Goodchild, M.F. (Eds.), Scale in Remote Sensing and GIS. Lewis Publishers, Boca Raton, FL. Li, X., Yeh, A.G.O., 2004. Data mining of cellular automata’s transition rules. Int. J. Geogr. Inf. Sci. 18, 723–744. Li, Y.C., He, C.Y., 2008. Scenario simulation and forecast of land use/cover in northern China. Chin. Sci. Bull. 53, 1401–1412. Li, Z.W., Zeng, G.M., Zhang, H., et al., 2007. The integrated eco-environment assessment of the red soil hilly region based on GIS—a case study in Changsha City, China. Ecol. Model. 202, 540–546. Liang, L., Xiong, L., Wang, G.H., 2004. A new method of determining the reliability of decision-makers in group decision. Syst. Eng. 22, 91–93. Liu, X.P., Li, X., Shi, X., Wu, S.K., Liu, T., 2008. Simulating complex urban development using kernel-based non-linear cellular automata. Ecol. Model. 211, 169–181. Liu, X.P., Li, X., Yeh, A.G.O., 2007. Discovery of transition rules for geographical cellular automata by using ant colony optimization. Sci. China Ser. D: Earth Sci. 37, 824–834. Luciana, P.B., Edward, A.E., Henry, L.G., 2007. Land use dynamics and land˜ Campeche, Mexico. Landscape Urban Plann. 82, scape history in LaMontana, 198–207. Lópeza, E., Boccoa, G., Mendozaa, M., Duhau, E., 2001. Predicting land-cover and land-use change in the urban fringe A case in Morelia City, Mexico. Landscape Urban Plann. 55, 271–285. Mather, A.S., 1999. Land use and cover change. Land Use Policy 16, 143. Messerli, B., 1997. Geography in a rapidly changing world. IGU Bull. 47, 65–75. Meyer, W.B., Turner, B.L., 1996. Land-use/land-cover change: challenges for geographers. Geo Journal 39, 237–240. Myint, S.W., Wang, L., 2006. Multicriteria decision approach for land use land cover change using Markov chain analysis and a cellular automata approach. Can. J. Remote Sens. 32, 390–404. Patma, V., Sukaesinee, S., Viriya, L., Somjai, S., Vidhaya, T.G., Vichai, S., 2004. From forest to farmfields: changes in land use in undulating terrain of northeast Thailand at different scales during the past century. J. Southeast Asian St. 41, 444–472.

Pijanowski, B.C., Brown, D.G., Shellito, B.A., Manik, G.A., 2002. Using neural networks and GIS to forecast land use changes: a land transformation model. Comput. Environ. Urban Syst. 26, 553–575. Quan, B.N., Chen, J.F., Qiu, H.L., Romkens, M.J.M., Yang, X.Q., Jiang, S.F., Li, B.C., 2006. Spatial–temporal pattern and driving forces of land use changes in Xiamen. Pedoshere 16, 477–488. Shao, J.G., Ni, J.P., Wei, C.F., Xie, D.T., 2005. Land use change and its corresponding ecological responses: a review. J. Geogr. Sci. 15, 305–328. Susanna, T.Y., Chen, W.L., 2002. Modeling the relationship between land use and surface water quality. J. Environ. Manage. 66, 377–393. Thomas, H., Laurence, H.M., 2006. Modelling and projecting land-use and land-cover changes with a cellular automaton in considering landscape trajectories: an improvement for simulation of plausible future states. EARSeL eProceedings 5, 63–76. Veldkamp, A., Lambin, E.F., 2001. Predicting land-use change. Agr. Ecosyst. Environ. 85, 1–6. Verburg, P.H., de Koning, G.H.J., Kok, K., Veldkamp, A., Bouma, J., 1999. A spatial explicit allocation procedure for modelling the pattern of land use change based upon actual land use. Ecol. Model. 116, 45–61. Verburg, P.H., Overmars, K.P., Huigen, M.G.A., de Groot, W.T., Veldkamp, A., 2006. Analysis of the effects of land use change on protected areas in the Philippines. Appl. Geogr. 26, 153–173. Weng, Q.H., 2002. Land use change analysis in the Zhujiang Delta of China using satellite remote sensing, GIS and stochastic modeling. J. Environ. Manage. 64, 273–284. White, R., Engelen, G., 1993. Cellular automata and fractal urban form: a cellular modelling approach to the evolution of urban land-use patterns. Environ. Plann. A 25, 1175–1199. Wu, F., 2002. Calibration of stochastic cellular automata: the application to rural–urban land conversions. Int. J. Geogr. Inf. Sci. 16, 795–818. Wu, Q., Li, H.Q., Wang, R.S., Paulussen, J., He, Y., Wang, M., Wang, B.H., Wang, Z., 2006. Monitoring and predicting land use change in Beijing using remote sensing and GIS. Landscape Urban Plann 78, 322–333. Xiong, Y., Guang, M.Z., Gui, Q.C., Lin, T., Ke, L.W., Dao, Y.H., 2007. Combining AHP with GIS in synthetic evaluation of eco-environment quality—a case study of Hunan Province, China. Ecol. Model. 209, 97–109. Yu, F., 2009. Study on forecast of land use change based on Markov–CA. Land Resour. Inf. 4, 38–46.