Analysis of the Driving Force of Land Use Change Based on Geographic Detection and Simulation of Future Land Use Scenarios

Abstract

Land use and land cover changes (LULCC) are the result of the combined action of many influencing factors such as nature, society, economy and politics. Taking Chongqing as an example, the driving factors of urban land expansion in Chongqing from 1999 to 2019 are analyzed using a geographic detection (GD) method. Based on this analysis, a land use scenario of Chongqing in 2029 is simulated by an Artificial Neural Network-Cellular Automata model. The results of the analysis of factors affecting land use change show that five factors have a significance >0.05: population, distance from central city, school density, GDP and the distance from railway, showing that these factors have a high impact on LULCC in Chongqing. In addition, the results of risk detection analysis show that areas with a population >50/km²; the areas with a distance <200 km from the city center; areas with a school density >5/km²; areas with a high GDP; and areas with a distance <25 km from the railway have a greater impact on urban land use change than other areas. The land use scenario in 2029 also is simulated based on the land use situation in 2019. The predicted results clearly reflect a land use change trend of increasing urban land and decreasing agricultural land in the region. These land use changes are especially related to the expansion of the population, economy, roads, and schools in the process of urbanization. This analysis also shows that the GD-ANN-CA model developed in this paper is well suited to urban land use simulation.

1. Introduction

Land use and land cover changes (LULCC) are the result of the combined influence of many factors such as nature, society, economics, and politics [1]. It is an indicator of regional economic vitality and reflects the mode of land use, the intensity of development, the degree of economic investment, and the direction of policy at a specific time and in a specific place [2]. An objective simulation of future land use can not only understand the change and development of future land use, but also test the feasibility of the current government’s social and economic policies.

Land is a basic element for human survival and development [3,4]. LULCC directly affect our environment by affecting climate, soil, vegetation, and biodiversity [5,6]. It is an important component of global climate and environmental change research. LULCC are the result of many complex factors acting together in time and space. Therefore, it is necessary to select simplified and abstract models based on a deep understanding of the driving forces behind LULCC, and use dynamic models to simulate the interaction between the various components of the system.

The use of remote sensing and GIS techniques [7,8] has become the main means of urban LULCC research now, and mathematical modeling has provided new theoretical and technical methods for the research of urban expansion. Model analysis based on mathematics has become a focus in the field because it effectively describes, simulates and predicts the dynamic characteristics of urban development and change. Cellular automata (CA) is the most current and most in-depth research into the simulation of urban growth, diffusion and land use evolution. Different researchers have adopted various frameworks and designed diverse models to simulate LULCC from different perspectives based on their fields. One example of these models is land use and land cover (LULC) prediction based on constrained condition [9,10,11]. The constraint conditions include farmers’ wishes and government policy constraints, etc. Other models, such as scenario-based future LULC prediction [12,13], the prediction of LULC changes based on geostatistical analysis model [14], the prediction of urban land expansion based on a hybrid model [15], and the combination of the logistic regression model and CA model. For example, the CA-Markov model can be applied to predict future LULCC [16,17].

The CA model is a dynamic model based on a grid with the distinct characteristics of integration of time and space [18,19]. It creates a bottom-up simulation of the spatio-temporal evolution process of complex systems.

Although the CA model solves the problem that the interaction of geographical phenomena in spatial proximity cannot be simulated, two limitations exist in the current LULCC simulation models: (1) Most of LULCC models train and estimate the conversion probabilities of each land use type independently, resulting in a separation between the different land use types. The interactions are not well explored in these models. (2) The calculation parameters of the CA model have modeling uncertainty [20,21]. Many kinds of evaluation factors are necessary for CA simulation. Too many factors may have multicollinearity and have not been objectively evaluated, resulting in low accuracy of evaluation results or simulation results [22,23]. Therefore, factor selection is the key to constructing an effective CA model [24,25,26,27].

In this paper, we present a method different from existing method: its advantage is that it can objectively evaluate the driving factors, and can carefully design the interaction and competition between different land use types. In the proposed model, firstly, we incorporated natural, environmental factors and socio-economic developments into the models. Secondly, it is necessary to choose which factors to use as evaluation factors based on mathematical methods. Thirdly, the Artificial Neural Network (ANN) is designed to address the complex local land use interactions and estimate the transition probabilities of different land use types. Taking Chongqing as a case study, as the fourth municipality under the direct jurisdiction of the central government of China, this paper uses the geographic detector technique to study and select the factors affecting land cover change, then uses the CA model and selects the artificial neural network as its conversion rule [28]. Following this method, a simulation model of land use change in Chongqing is developed. The results can be used to provide a scientific basis for the sustainable development and utilization of local land resources and the decision-making of government bodies.

2. Study Area

Chongqing is located in the southwest of China (Figure 1), in the transition zone between the Qinghai Tibetan Plateau and the plains in the middle and lower reaches of the Yangtze River. With a total area of 82,300 km², it is one of the largest cities in China, and it has a humid and warm climate. In 2019, Chongqing had a permanent resident population of 32 million. The land is used mainly as cultivated and forested land. As one of the municipalities directly under the jurisdiction of the central government of China, Chongqing has the characteristics of a big city, and the area includes countryside, mountains, and a large reservoir. It is dominated by hills and low mountains. The terrain fluctuates dramatically. The average altitude is 400 m, and the elevation difference within the jurisdiction is 1.4 km [29]. In recent years, the process of urbanization has been particularly rapid, making the study of the temporal and spatial distribution of land use in Chongqing of great practical value.

3. Material and Methods

3.1. LULC Data

The LULC data used in this study are mainly derived from the “ESA CCI Land Cover project” and “Year-Based Project for On-Board Autonomy-Vegetation (PROBA-V)”, and the resolution of these data is 300 m. Three LULCC data spanning 20 years (1999, 2009, 2019) were selected for research. The types of land use in the study area are urban areas, forested land, grassland, water bodies, cropland and bare areas. Based on an analysis of other studies on the factors driving land use change, 10 factors (

Table 1. List of predictors used in the transition potential modeling process.

Acronym	Variable	Description	Source	Time Interval
Land use
LULC	land use/land cover	Land cover maps at 300 m spatial resolution.	CRDP (The Land Cover CCI Climate Research Data Package. https://cds.climate.copernicus.eu (accessed on 13 November 2020))	1999; 2009; 2019
Environmental variables
DEM	Elevation	Elevation in m	GLSDEM (Global Land Survey Digital Elevation Model www.gscloud.cn/ (accessed on 13 November 2020))	(-)
SL	Slope	Slope in degrees	GLSDEM	(-)
TE	Temperature	Temperature Spatial Distribution	NMICC (National Meteorological Information left of China http://data.cma.cn/ (accessed on 15 November 2020))	2015
DRR	Distance from the river	Euclidean distance from river network in m	OSM (Open Street Map. https://www.openstreetmap.org (accessed on 15 November 2020).)	(-)
Socio-economic variables
DC	Distance from city left	Euclidean distance from city left in m	OSM	(-)
DRD	Distance from the road	Euclidean distance from road network in m	OSM	(-)
DRW	Distance from the railway	Euclidean distance from railway in m	OSM	(-)
GDP	Gross domestic product	Spatial distribution of GDP at a resolution of 1 km2. The units are numbered by 10,000 RMB per pixel.	RESD (Resource and Environment Science and Data left www.resdc.cn (accessed on 18 November 2020))	2005; 2015
POP	Population	Spatial distribution of population at a resolution of 3 arc. The units are numbered by people per pixel.	Worldpop (www.worldpop.org (accessed on 4 March 2021))	2005; 2015
SC	School	School Distribution Density	Baidu Map Open Platform (www.baidu.com (accessed on 4 March 2021))	(-)

) were finally selected [30]. These include physical variables such as surface temperature, DEM, and slope; socio-economic variables including gross domestic product (GDP) and population density; and human disturbance variables such as distance from the city, distance from the road, distance from the railway, and distance from the river. The coordinate system used for all data systems is GCS_WGS_1984, the projection is ALBERS, and the spatial resolution is 300 × 300 m.

The DEM data are derived from GLSDEM (www.gscloud.cn (accessed on 13 November 2020)). The SL data are generated based on DEM; The TE data are downloaded from the website and mapped by interpolation method. The DRR data are generated by ArcGIS software based on Euclidean distance and river network set. DC data are also generated by ArcGIS software based on Euclidean distance and urban central point data. DRD is mapped by ArcGIS software based road network data and Euclidean distance. DRW is mapped by ArcGIS based on railway network and Euclidean distance. GDP data are downloaded from RESD (www.resdc.cn (accessed on 18 November 2020)). Pop data are downloaded directly from Worldpop (www.worldpop.org (accessed on 4 March 2021)). The SC density map is generated by ArcGIS software based on point density function.

3.2. Methods

3.2.1. Research Method

The degree of influence of each factor on LULCC were analyzed by Geographic Detection (GD). The future LULCC of Chongqing were simulated by the Artificial Neural Network-Cellular Automata (ANN-CA).

• Geographic detector-based modeling

Geographic detector is a statistical tool used to measure a spatially stratified heterogeneity of factors. It is based on the assumption that if an independent variable has an important impact on a dependent variable, the spatial distribution of the independent variable and the dependent variable should be similar [31]. It can effectively determine if the relationship between multiple factors is independent or interactive, if they are enhancing or weakening one another, and if they have a linear or nonlinear relationship. (A geographic detector can be downloaded for free from http://www.geodetector.cn/ (accessed on 14 May 2021)) [32]. Geographic detectors include 4 calculation methods: risk detection, factor detection, ecological detection and interaction detection [33].

Factor detector: the factor detector $q$ statistic measures the degree of stratified heterogeneity of a variable $Y$; and the determinant power of an explanatory variable $X$ of $Y$. The extent to which factor $X$ determines the spatial differentiation of $Y$ measured by $q$ value, is shown in Figure 2 and Equation (1):

$$q=1-\frac{{\displaystyle \sum _{h=1}^L{\displaystyle \sum _{i=1}^{N_h}{\left(Y_{hi}-{\overline{Y}}_h\right)}^2}}}{{\displaystyle \sum _{i=1}^N{\left(Y_i-\overline{Y}\right)}^2}}=1-\frac{{\displaystyle \sum _{h=1}^L{N}_h}{\sigma }_h^2}{N{\sigma }^2}=1-\frac{SSW}{SST}$$

(1)

where the total sum of squares:

$$SST={\displaystyle \sum _i^N{\left(Y_i-\overline{Y}\right)}^2}=N{\sigma }^2$$

(2)

and the internal sum of squares:

$$SSW={\displaystyle \sum _{h=1}^L{\displaystyle \sum _i^{N_h}{\left(Y_{hi}-{\overline{Y}}_h\right)}^2}}={\displaystyle \sum _{h=1}^L{N}_h}{\sigma }_h^2$$

(3)

where: $h=1,\dots$, $L$ is the stratum of variable $Y$ or factor $X$; $N$ is the number of stratum in the whole area; $Y_i$ and $Y_{hi}$ denote the value of unity in the population and in stratum h, respectively; the stratum mean is represented by ${\overline{Y}}_h=\left(1/N_h\right){\displaystyle \sum _{i=1}^{N_h}{Y}_{hi}}$; the stratum variance is ${\sigma }_h^2=\left(1/N_h\right){\displaystyle \sum _i^{N_h}{\left(Y_{hi}-{\overline{Y}}_h\right)}^2}$.

The value range of $q$ is [0,1], and the larger the value, the more obvious the spatial differentiation of $Y$. In extreme cases, a $q$ value of 1 indicates that factor $X$ completely determines the spatial distribution of $Y$, whereas a $q$ value of 0 indicates that factor $X$ has no relationship with $Y$.

Risk detector: Compares the difference of average values between sub-regions; the bigger the difference, the greater the influence on urban expansion of the sub-region.

$$t_{{\overline{y}}_{k-1}-{\overline{y}}_{k-2}}=\frac{{\overline{Y}}_{h=1}-{\overline{Y}}_{h=2}}{{\left[\frac{\mathrm{Var}\left({\overline{Y}}_{h=1}\right)}{n_{h=1}}+\frac{\mathrm{Var}\left({\overline{Y}}_{h=2}\right)}{n_{h=2}}\right]}^{1/2}}$$

(4)

where: ${\overline{Y}}_h$ is the mean value of attributes in sub-region of $h$, $n_h$ is the number of samples in sub-region of $h$, and $\mathrm{Var}$ is the variance. The statistic $t$ approximately follows the Student’s t distribution, and the calculation method of the degree of freedom (DF) is as follows:

$$df=\frac{\frac{\mathrm{Var}\left({\overline{Y}}_{h=1}\right)}{n_{s=1}}+\frac{\mathrm{Var}\left({\overline{Y}}_{h=2}\right)}{n_{h=2}}}{\frac{1}{n_{h=1}-1}{\left[\frac{\mathrm{Var}\left({\overline{Y}}_{h=1}\right)}{n_{h=1}}\right]}^2+\frac{1}{n_{h=2}-1}{\left[\frac{\mathrm{Var}\left({\overline{Y}}_{h=2}\right)}{n_{h=2}}\right]}^2}$$

(5)

This study uses the GD method to analyze the driving factors of urban LULCC in 1999–2009 and 2009–2019 [34,35,36]. Comprehensively considering the land resources, social and economic development level, and environmental conditions in the study area, this study takes 10 indicators such as DEM, GDP, POP, DR, and School [37,38] as its research subject. Firstly, the factor detector is calculated to represent the intensity of the impact on urban LULCC. Secondly, the influence of each factor is classified, and the significance between the different levels of each factor is calculated based on the risk detector formula [39,40].

• Simulation of LULCC in Chongqing Based on ANN-CA

Markov prediction is a stochastic process. It is a prediction method to predict the change of each time in the future based on the current situation of the event. Markov chain prediction method is suitable for predicting the dynamic change of land use. In this paper, the transfer matrix is calculated based on the land use in 2009 and 2019. Then, the amount of land use types in 2029 is predicted based on Markov chain.

CA is a dynamic system that is spatially and temporally discrete. It can simulate the temporal and spatial evolution process of complex systems, and can effectively reflect micro patterns of land use. CA is composed of cells, cell space, neighborhood, time and rules. Transformation rules are the core of CA. The advantage of the ANN-CA model is that the cell transformation rules are extracted by the artificial neural network method, and there is no need to manually determine the model structure and transformation rules [41]. Compared with other models of CA, ANN-CA can be applied to more complex nonlinear systems, obtain higher simulation accuracy, and have less constrained independent variable requirements.

3.2.2. Workflow

In this paper, a GD-ANN-CA model is developed to analyze the driving force and simulate future land use. The model consists of three parts (Figure 3): spatial data processing, factor analysis based on GD, and model training and prediction based on ANN-CA.

Firstly, a grid map of each factor is generated using ArcGIS software. For example: DRD is mapped by ArcGIS software based road network data and Euclidean distance.

Secondly, the influence of 10 factors is calculated using a factor detector method [39], which can be used to compare the effects of different factors on land change. Additionally, then the factors with a value > 0.5 are selected as the factors of the ANN-CA model.

Thirdly, the parameters of the model are calculated by training ANN based on random samples. In each cycle, the conversion probability of land use corresponding to each grid unit is automatically calculated by the neurons of the output layer. Then the amount of future land use is calculated by Markov chain, which has been successfully employed by many studies.

Finally, the future land use scenario is simulated by the selected factors, conversion matrix and future pixel amount. The LULCC of Chongqing in 2019 were simulated by ANN-CA based on the LULCC of 2009 as the initial state. When the simulation results of 2019 were correctly modeled, then LULCC in 2029 are simulated based on the LULCC in 2019.

This paper uses the ANN-CA model of GeoSOS-FLUS (http://www.geosimulation.cn (accessed on 14 May 2021)) and ArcGIS 10.2 software to simulate LULCC in Chongqing City [42,43]. First, based on a factor detector value > 0.5, the factors with the greatest influence are selected as alternative factors. Second, using the ANN-CA model and the above factors, the land use scenario of Chongqing in 2019 is simulated based on the LULCC in 2009. The simulation results are compared with the observed values in 2019 to test the accuracy of the model [44,45,46]. Finally, the amount of land use types in 2029 is calculated based on a Markov chain, and then simulated 2029 land use scenarios.

4. Results and Discussion

4.1. Driving Force Analysis of LULCC

The degrees of influence of 10 factors such as GDP, TE, and DC on Chongqing LULCC in 1999–2009 and 2009–2019 were calculated by the factor detector formula based on GD. The result is shown in Figure 4.

Analyzed from Figure 4, it was found that the index of factor detector in 2009–2019 is POP(0.187) > DC(0.136) > SC(0.124) > GDP(0.087) > DRW(0.052) > DRD(0.046) > DEM(0.037) > SL(0.037) > TE(0.029) > DRR(0.011), which is in the similar sort order to 199\1–\2. Therefore, POP, DC, SC, GDP and DRW are the main factors affecting urban LULCC. POP has the greatest influence, with a maximum of 0.19 in 2009–2019. The distance from the central city (DC) has the second-highest impact. However, SC and GDP also play a major role in the development of urban land, about 0.09–0.12. The influence degree of DRW and DRD is about 0.05, which indicates that the influence of different traffic land is almost the same. Additionally, DEM has less impact on the growth of urban land, which is 0.04. The distribution of rivers (DRR), slope (SL)and temperature (TE) have the least impact on the LULCC of urban areas, about 0.01–0.03 only. The significance of the spatial distribution of population on urban land can also be seen because the government plans urban development based on the size of the population. The significance of DC shows that the urban area of Chongqing is developing rapidly, and a large amount of land around the city has been transformed into urban land. The SC factor’s impact reflects the aggregation effect of “School District Housing” [47]. The land around schools, especially around high-quality schools, is more easily transformed into residential land, roads, shopping malls and other urban lands. Especially, it should be noted that the development of GDP requires a large number of urban land, which has greatly accelerated the conversion of agricultural land and other land to urban land.

Comparing the LULCC between the two stages of 1999–2009 and 2009–2019, it shows that the 10 factors considered had less impact on the expansion of urban land in 1999–2009 than in 2009–2019, with a maximum of 0.08 and a minimum of 0.003 only. However, with the rapid development of city in 2009–2019, the degree of influence of the 10 factors on LULCC is quite different, with the highest POP at 0.19 and the lowest DRR at 0.01. POP, DC, and SC have a significant gap when compared with two stages results. By 2009–2019 index minus 1999–2009 index, the result is that POP is 0.10, DC: 0.09, SC: 0.07, GDP:0.02, DRW: 0.03, DRD: 0.03, DEM: 0.03, SL: 0.02, TE: 0.01, DRR: 0.01. One reason for these results is that with social and economic development, people tend to flow to big cities and suburbs near the center of urban areas. Other factors, such as DRD, DEM, SL, TE and DRR, had a lower impact on the expansion of urban land in 2009–2019. It demonstrates that economic and social factors are more significant in urban land expansion, and natural factors are less important, which is consistent with Chongqing being a famous “Mountain City” and “Fire City” in China. If natural factors were a significant barrier to urban expansion, there would be no available urban land in Chongqing [48].

The classification of each factor is shown in

Table 2. Level table of variables.

Variable	Level 1	Level 2	Level 3	Level 4	Level 5
GDP (RMB/km2)	<50 million	50–150 million	150–300 million	>300 million	(-)
DC (km)	<20	20–100	100–200	200–300	>300
DEM (m)	<50	50–250	250–500	500–1000	>1000
TE (°C)	<25	25–33	>33	(-)	(-)
POP (/km2)	<50	50–150	150–250	>250	(-)
DRW (km)	<10	10–25	25–45	45–70	>70
DRR (km)	<1	1–3	3–5	5–10	>10
DRD (km)	<1	1–3	3–5	5–10	>10
SL (°)	<5	5–15	15–25	>25	(-)
SC(/km2)	<1	1–5	5–10	>10	(-)

. Additionally, the influence degree of each level of the 10 factors on urban LULCC is calculated by risk detector in 1999–2009 and 2009–2019 (Figure 5).

GDP plays a greater role in the growth of city land, because economic development requires a lot of land. In the level 1–4 region, GDP development promotes the transformation of non-urban land into urban land. Compared with level 1–2 and level 2–4, this index is at least 0.17 in level 2–4, which is much larger than the level 1–2. According to the vertical analysis of this index, in 2009–2019, the economy developed faster than before, and the demand for urban land increased by 10% compared with the previous stage. Additionally, this index indirectly reflects the relationship between GDP and real estate.

DC: Distance from the city center affects the expansion of urban land. In 1999–2019, the increased urban land area is primarily located within 200 km from the city center. Especially within level 1, the index is 0.28 in 2009–2019 and 0.11 in 1999–2009, indicating that the urban land growth is the largest in the area <200 km. These areas are mainly located in the suburbs and satellite cities of Chongqing. The rapid urban development occupies a lot of non urban land such as cultivated land and forest land. On the whole, the farther away an area is from the urban center, the lower the increase in urban land.

DEM: the elevation of urban land in Chongqing is mainly <500 m. The impact of elevation on urban land in Chongqing is relatively low, with a maximum of 0.11. The area with an elevation of less than 500 m is the main area for new urban development, and in areas with an elevation of greater than 500 m the amount of new urban land decreases sharply with the increase in elevation. From the red line and blue line (Figure 5c DEM), it can be seen that the newly added urban land in Chongqing is gradually developing from 250 m to 500 m. This is because Chongqing is a “mountain city” and there is less land suitable for urban development.

TE: Temperature has little effect on the distribution of urban land in Chongqing, with a maximum of 0.01 only. Chongqing is known as a “fire city” in China. The temperature distribution map of Chongqing was also interpolated based on data from monitoring stations. It is found that due to the “Heat Island Effect” in Chongqing and its surrounding counties, the expansion of urban land seems to be proportional to TE.

POP: It can be seen from POP (Figure 5e POP) that the change in urban land use is positively correlated with population density. It is found from the figure that this average index is relatively high. Population growth leads to the reduction of per capita cultivated land area, which leads to the transformation of the mode of production of the people who live in formerly agricultural areas. The transfer of agricultural population to non-agricultural population accelerates the speed of urbanization. Comparing the indexes of 1999–2009 and 2009–2019, in the level 2–4 area, the value based on the risk detector calculation increased to 0.26. This is because after Chongqing was identified as the fourth municipality directly under the control of the central government in China in 1997, a large number of people moved to Chongqing, which intensified the speed of urbanization. From the distribution trend of the two curves in Figure 5, it can be seen that the urbanization rate in densely populated areas is much higher than that in areas with low population density.

DRW: Railway construction shortens the time necessary to travel between cities. The urban land growth area is mainly concentrated in level 1–2 (Figure 5f DRW), that is, the newly added urban land is mainly distributed within 25 km on both sides of the railway. The further an area is from the railway, the slower the urban development.

DRR: Rivers play an important role in human survival and production. The main rivers in Chongqing are the Jialing, Wujiang, Fujiang, etc. Since most cities are distributed along the river, the urban LULCC are centered on the river and spread to the surrounding areas. It is found that the biggest impact of rivers on urban land use change is that level 1–2 reached 0.09, and then fell sharply at level 2–4 with an index of 0.03 from Figure 5 in 2009–2019. It also shows that within 3 km from the river is a high-value area for new urban land. For areas greater than 3 km, the amount of newly added urban land tends to decrease as the distance from the river increases. The red line index in Level 1–2 is much higher than the blue line index, which shows that there has been more intense human activity in the area near the river in recent years.

DRD: The development of strong road networks promotes the expansion of urban land. As one moves further away from the road, urban land density decreases. It is found from the figure that the DRD impact is almost all in level 1–2, the highest is 0.10, and the impact degree of other levels is almost 0. In the past 20 years, the newly added urban land in Chongqing is primarily distributed within 3 km on either side of all types of roads. Beyond 3 km, there is less new urban land. A comparison between the lines representing 1999–2009 and 2009–2019 shows that the new urban land is increasingly concentrated within 3 km of the road.

SL: SL has little effect on the increase in urban land in Chongqing. The increase in urban land is mainly distributed in level 1–2 (0–15°). The higher the slope value, the more restricted the urban development. However, within the area of urban development, the new urban land tends to develop in high slope areas.

SC: Based on an analysis of Figure 5, the maximum index of school demand for urban land is 0.27, and its influence increases with the increase in school density. Additionally, the new urban land in Chongqing is almost proportional to the density of schools, which is because people in China tend to prefer “School District Houses.” The denser the schools, the more concentrated the urban land, because residential land and commercial land tend to be centered on schools. Comparing the two lines in this figure, the impact of schools on urban land in 2009–2019 is much greater than that of the previous 10 years.

4.2. Simulation of Future LULCC Based on ANN-CA

4.2.1. Model Calibration and Performance

In the ANN-CA model, the main function of ANN is to provide conversion rules, which calculate the probability of each cell from one land use type to another [49], and then this rule is applied to simulate the future land use change. The ANN-CA model consists of two relatively independent modules [50]: training and simulation. For the purposes of this study, the CA model conversion rules were trained based on 2009 land use data and spatial influencing factor data, and then these same conversion rules were applied to simulate land use in 2019. The accuracy of ANN-CA model in simulating LULCC is verified by comparing the simulated data to the observed data in 2019 [51].

The evolution of land use is affected by many factors, such as nature, society, economics, technology, etc. This makes it very difficult to accurately simulate LULCC [52]. The accuracy of the model simulation is judged by the total simulation accuracy, which is found by comparing the ratio of the number of cells correctly simulated to the total number of cells in the study area. The calculation results (

Table 3. Prediction accuracy.

Land Use Type	Commission Error	Omission Error	Producer’s Accuracy
cropland	0.05	0.05	0.95
forests	0.07	0.06	0.94
grassland	0.43	0.13	0.87
water bodies	0.01	0.02	0.98
urban areas	0.31	0.25	0.75
bare areas	0.16	0.32	0.68

) show that the Kappa index of the simulation is 88.26% with 1% random sampling, and the total accuracy is 90.15%. The simulation accuracy of cropland land and forested land is higher, at more than 94%, while the simulation accuracy of urban land is only 75%. This is mainly due to the relatively small area of urban land. The accuracy is within a reasonable range in all results.

Comparing the simulation results with the observed results (Figure 6), it is found that the commission error and omission error of urban land [53] are mainly distributed in the suburbs of Chongqing. This error is likely because the suburbs are the main area of new urban land. They are not only affected by natural, geographical, and economic factors, but also by government policies. The commission error and omission error of grassland and bare areas are high, mainly due to the small area of grassland and bare areas in Chongqing.

4.2.2. LULCC Predictions for 2029

(1) Determine the total area of pixels of each land type in 2029

Based on the land observation data in 2009 and 2019, the number of pixels of each land type in 2019 and 2029 is predicted using a Markov chain (

Table 4. Area of each land use type (/km²).

Land Use Type	Observed 2019	Simulated 2019	Simulated 2029
cropland	49,556	49,548	48,992
forests	30,642	30,544	30,435
grassland	35	37	26
water bodies	813	792	786
urban areas	1182	1286	1949
bare areas	99	119	138

). The prediction accuracy is calculated based on the observed and simulated values in 2019 [54,55]. The simulation results show that the simulation value of cropland is 49,548 km² in 2019, which is 8 km² different from the observed value, and the simulation effect is good. However, the difference between urban areas simulation and observation value is 104 km². The simulation error for urban land is 8.76%. By comparing and analyzing the simulated and observed LULCC map in 2019, the average prediction accuracy of the total amount of each land type in 2019 is 93.32%, which is based on the Markov chain calculation.

(2) Land use scenario of Chongqing in 2029

The land use scenario of Chongqing in 2029 is simulated based on an ANN-CA model (Figure 7). In 2029, the total number of pixels in Chongqing is still 914,739, with an area of 82,326 km². The land type scenario maintains the previous pattern: forest land is mainly distributed in the north and east of the region. Cropland land is mainly distributed in the southern and western regions, although it is disconnected and interrupted by forest due to the influence of terrain. As the land type with the largest change, the distribution of urban land obviously shows a trend of diffusion from the center outwards. However, due to the limitations of the western and eastern mountainous areas of Chongqing, urban land is mainly developing to the north. Based on an analysis of the simulation results, the urban land area in 2029 increased by 766 km² compared with 2019. The main sources of new urban land are 564 km² of cropland and 206 km² of forest. The grassland and water areas are used less for urban expansion. Comparing the land use pattern between 2019 and 2029, it is found that most land use changes occur in areas with relatively flat terrain, mainly through the transformation of farmland, forest land, and grassland into urban land. The expansion of urban land has a great impact on the regional land use pattern.

5. Conclusions

(1) The driving factors affecting LULCC were analyzed using a geographic detector model. Five factors, including POP, DC, DRW, SC and GDP, were found to have a high impact on LULCC in Chongqing. The results show that the spatial areas with a population >50/km², areas within 200 km from the city center, areas with a school density >5/km², areas with high GDP and areas less than 25 km from the railway have a greater impact on the LULCC of urban areas. Therefore, the government should focus on the development of these areas in urban land planning.

(2) The ANN-CA simulation model is defined by the use of cellular automata and artificial neural networks. The advantages of this model are that it emphasizes the spatial interaction of geographic elements and provides a visual analysis of the driving forces of LULCC [56]. ANN-CA modeling provides a new method for the study of LULCC. Based on the simulation values and actual observed values in 2019, the total precision of the land use simulation prediction used in this study is 90% [57,58]. This shows that ANN-CA model can be accurately applied to the study of land use change simulation. Finally, the LULCC of Chongqing in 2029 is simulated using this model. Based on predictions, in the next 10 years, if no restrictions are imposed, the cultivated land in Chongqing will be reduced by 1%, and the increase in urban land will reach a surprising 65% of that in 2019. These results show that the government ought to strictly restrict the development of urban land, so that land use change can be planned to meet the rational land use needs of social and economic development to the greatest extent, and farmland can be protected.

(3) In summary, the proposed model is suitable for analyzing the driving factors of land change and predicting the future land use scenario. The GD-ANN-CA model can not only objectively select evaluation factors, but also be applied to more complex nonlinear systems. This model is especially suitable for land use analysis, ecological evaluation, etc.