2.3.2. Present Situation Analysis
(1) Reclassification of Land Use
The original land use maps include seven types of land: cultivated land, woodland, grassland, shrubland, river, urban land, and wetland. Similar land types were combined to conduct an in-depth study on the land types with important characteristics in land use change in Qishan County. Among them, the grassland area is small, which is not a typical land type in Qishan County. Since these areas are mainly scattered sparse forest grassland and shrub grassland, we combined grassland into forest for analysis. The reclassification included four land types, i.e., cultivated land, forest, water, and urban land, as a foundation for subsequent analyses on the area change of land use and transitional type. Refer to
Table 2 for details of the reclassification system of land use:
(2) Dynamics of Land Use
The dynamics of land use can reflect its temporal and spatial variation, and an individual dynamic stands for the quantity change of a specific land use type of the research subject within a certain period [
18]. A positive value represents an increase in area, whereas a negative value represents a decrease in area. The greater its absolute value is, the greater the net change in land. The formula of the individual dynamic index is:
where K is the dynamic of a specific land use type during the study period,
and
represent the areas of certain land use types at the beginning and the end of the study, respectively. T is the study interval set as ten years, and K is the annual change rate of the land use type researched.
Compared with the individual dynamics, the integrated dynamic index focuses on the processes rather than the results of the change, which solves the problem that the final area of land types barely changes when the transfer in and transfer out of local classes are quite similar because it calculates the absolute value of the conversion area between land types [
11]. The formula of the integrated dynamic index is:
where
is the area of land use type “
a” at the beginning of the study,
is the absolute value of the area of a land use type (except for “
”) transferred from the land use type “
”.
T is the study interval defined as 10 years in this study, and
LC is the annual change rate (absolute value) of land use change of the research subject during the interval.
(3) Transition Matrix of Land Use
The land use matrix is crucial for analyzing the land type change in a region, as it describes the transition among various land use types at the beginning and the end of a certain period of time [
19,
20,
21]. This is the basis for the subsequent analyses on land use change and the establishment of LUCC models.
2.3.3. Model Establishment
(1) CA-Markov Model
The CA-Markov Model is used to predict the land-use transition quantity in the future according to the history of transition probability and follows the spatial distribution principle that land with high suitability will be transferred preferentially in land use prediction. It is a combination of the spatial operation of the cellular automata (CA) model and the quantity prediction of the Markov chain (Markov) model. The CA model is advantageous in generating the potential transition maps by taking the spatial structure and neighborhood state into consideration, and the Markov model can provide an LUCC transitional area matrix based on the time variation [
11].
CA-Markov is used to determine the trend of each state at each future moment according to the initial probability of different states of a system and the transition probability between the states [
22]. The simulation process of CA-Markov based on Terrset software is described as follows:
a. Transition Analysis with Markov Chain. The matrices of land use transitions, including the probability matrix and area matrix from 2000 to 2010 and 2010 to 2020, were obtained after 2000 and 2010 and were set as the reference years, respectively. The land use data were input into the Markov model, and the interval and prediction period were defined as ten years;
b. Creation of a multi-criteria evaluation model. The five driving forces, i.e., altitude, slope, road distance, residential area distance, and water distance, were selected, and upon determination of their weight, a multi-criteria evaluation model (MCE, Multi-Criteria Evaluation) was created [
11,
12];
c. Establishment of a suitability atlas. The suitability atlas is a key file for predicting land use distribution and the basis for the cellular evolution of the CA model. In this paper, the suitability images of each land use type were generated upon running the created MCE model and finally synthesized into a raster group file to establish the final suitability atlas; and
d. Establishment of the model. The land use classification map was selected as the starting point of prediction and inserted into the CA-Markov model. Then, the model was run to predict the land use status after the input of the historical transition matrix and the suitability atlas of land use types.
(2) LCM model
Similar to the CA-Markov model, the Land Change Modeler (LCM) predicts the change in the number of land types based on the Markov transition matrix, but its spatial distribution is based on the transition probability. Here, multi-Layer perceptron (MLP) was used to generate a transition probability atlas. The MLP model is a kind of artificial neural network model with empirical learning and decision-making driven by algorithms [
23]. It can randomly generate and verify the operational relationship between the driving force and transition probability. The same number of training samples and verification samples are selected in each iteration, with their root mean square error and prediction accuracy updated simultaneously. When the number of iterations or prediction accuracy reaches the threshold value, the training is immediately stopped, and the transition probability image is the output [
24,
25].
Similar to the suitability atlas, the transition probability atlas is the key data for predicting the spatial distribution of land use change. In this paper, a transition probability atlas was created using the MLP module based on Terrset software, and then a predicted map of land use classification could be generated based on the transition matrix and transition probability atlas.
(3) Model accuracy verification
The accuracy of the LUCC model was measured by the Kappa coefficient, which can be calculated using the CROSSTAB tool in GIS Analysis under IDRISI. Under the same operational rules, if the Kappa coefficient calculated as per the actual annual land use data and the simulated predicted data is less than 0.4, the simulation is unsuccessful; if it is between 0.4 and 0.75, the simulation result is average; and if it is greater than 0.75, the simulation result is good and consistent with the actual situation. The equation used in this process is [
26].
where
is the error precision;
is the predicted area of type
land; and
is the actual area of type
i area.
2.3.4. Multi-Scenario Prediction
In this paper, the land use change in Qishan County in 2030 was predicted using the LUCC model with a better simulation effect. However, given that future land planning interventions, incentives, and restrictions may also change the process of land development, three specific scenarios were set in the in-depth analysis based on its historical land use changes: “natural state” with the status quo unchanged, “economic development state” with the accelerated expansion of urban land, and “water conservation state”. The specific scenarios are as follows:
Scenario 1: Natural state. The present evolutionary trend will continue in Qishan County without any human intervention. The historical transition area matrix from 2000 to 2020 is used for this state.
Scenario 2: With a policy prioritizing economic development, urban land would further expand continuously. In this scenario, based on the historical matrix from 2000 to 2020, the grids of other types of land converted to urban land increased to 110%.
Scenario 3: Water conservation state. Qishan County is short in water resources and has serious soil erosion. In recent years, local governments have strengthened the comprehensive treatment of water pollution and implemented water resource protection policies. In this state, the conversion of water to other types of land is prohibited, and the area of water converted to other types of land is reset to zero.