2.3.1. CLUE-S Model
This study used the CLUE-S model developed and improved by Verburg et al. [
7] from Wageningen University & Research in The Netherlands. The model was developed to simulate land use change using the intrinsic quantitative and empirical relationships between land type change and drivers. It allows for relatively accurate spatial dynamic simulations at small scale scales based on different scenarios and top-down spatial allocation of land use type changes. The CLUE-S model is divided into two different modules, the non-spatial demand module and the spatial allocation module [
18]. The non-spatial demand module, which needs to calculate the land use demand in the region according to the settings of different scenarios, usually relies on external models to accomplish this. In this study, Markov chains were used as an external prediction model to forecast the future land use quantity demand. The spatial allocation module is based on the theoretical foundation of where changes in a particular land class are most likely to occur. In this module, the probability of land use occurring at each location is calculated and the land use change is iterated year by year in conjunction with the land use demand. Thus, the spatial and temporal changes in future land use are simulated [
18]. The principle of the CLUE-S prediction model is [
19]:
where TPROP
i,u is the probability of occurrence of land type u on grid i; P
i,u is the probability of spatial distribution calculated by the regression model; ELAC
u is the elasticity coefficient of land use type u; ITER
u is the iterative variable of land use type u.
The following four components need to be considered as inputs to the CLUE-S model:
In this study, logistic regression was used to calculate the relationship between the distribution of land use types and the driving factors, linking the probability of land type occurrence with the location characteristics as a way to characterize the suitability of each area land type, which is often based on natural environmental, socioeconomic and other factors. Meanwhile, elevation, slope, slope direction, distance to rivers, and distance to lakes were selected as natural environmental factors, population density and nighttime lighting index as socio-economic factors, and distance to highway and distance to railway as transportation factors [
20]. The logistic regression model can be written as following:
where P
i is the probability of occurrence of the specified land use type in a grid cell; X
n is each driver, i.e., independent variable; β
0 is the constant term; β
n is the coefficient of each independent variable.
A binary logistic regression model stepwise regression was constructed in IBM SPSS Statistics ver.25 software (Armonk, NY, USA: IBM Corp.) to achieve this, and the Receiver Operating Characteristic (ROC) method was adopted to assess the goodness of fit as a way to estimate the regression results [
21]. Judging from the Area under the Curve (AUC) of the ROC, the value of AUC is between 0.5 and 1, and the closer to 1, the better the regression result.
Firstly, according to the land use remote sensing monitoring data from the Resource and Environment Science Data Center of the Chinese Academy of Sciences, the land use is divided into five categories: cultivated land, woodland, grassland, water body, and construction land. Because the amount of unused land is too small, this study classified the unused land into the category of construction land. The raster files of 5 land categories and 9 drivers were converted to ASCII files by the Raster to ASCII tool in ArcGIS. The single record file of each land category about 9 drivers was produced by using the convert software included in the CLUE-S model package, which was converted to an SPSS-applicable format and run SPSS for binary logistics. The results of the regression are shown in
Table 1. The selected drivers have good fits for the five land categories of cultivated land, woodland, grassland, water body and construction land, with the values of AUC of 0.772, 0.781, 0.897, 0.805, 0.875, respectively.
The Markov chain is a stochastic time series that uses the present state and movement of a variable to predict the future state of that variable and its movement [
22]. Due to its excellent “memorylessness”, it is widely used in the simulation of land use demand quantities. The basic formula is shown below [
23]:
where S
t+1 and S
t are the number of each land use type at moments t + 1 and t, respectively; P
i, j denotes the transfer probability matrix.
In this study, the Markov chain model was used to forecast the land use demand in 2030 based on the land use data in 2010 and 2020. The Markov model was also applied to obtain land use demand for 2023 based on data from 2013 and 2018, and 2027 based on data from 2013 and 2020. It was assumed that the speed of each category varies uniformly across the years 2020–2030, and the data for the remaining years were obtained using linear interpolation [
15,
24].
- 3.
Transfer matrix and transfer elasticity;
The transition matrix indicates whether one land type can be converted into another land type; 1 means it can be converted, 0 means cannot be converted. The settings of the transition matrix are shown in
Table 2.
The transfer elasticity was used to measure the reversibility of land use change, and its value is between 0 and 1; the closer to 0, the less likely the conversion will occur. According to the land use transfer matrix and transfer probability from 2010 to 2020, combined with expert experience, the transfer elasticity of cultivated land, woodland, grassland, water body and construction land under different scenarios was formulated (see
Table 3). In the ecological protection scenario, in order to effectively protect the ecological land, the transfer of woodland, grassland and water bodies is increased, and the transfer elasticity of these three types of use is increased by 0.1. The transfer elasticity of cultivated land increased to 0.8.
- 4.
Space policy and regional restrictions.
This part refers to the protected areas in the area where changes in land use are not allowed. According to the documents of the National Ecological Protection Plan of Jiangsu Province and the General Land Use Plan of Jiangsu Province (2006–2020), this study made ecological protection restrictions area and cultivated land protection restricted area document respectively.
2.3.3. Estimation of Carbon Emissions for Different Land Use Type
For the estimation of carbon emissions from different land types, there are two methods [
28]. The first is the direct estimation method based on the Intergovernmental Panel on Climate Change (IPCC) inventory [
29], where land use carbon emissions are calculated directly through carbon emission (carbon sequestration) factors. This method is applicable to areas with few human activities generating large amounts of carbon emissions, such as cropland, forest land, grassland, and water. The second one is to use the energy emission factor method in the IPCC report [
30], and this method is applicable to construction land with large amount of human activities generating carbon emissions [
31]. The carbon emission calculation formula for non-construction land is as follows:
where E
k is the total direct carbon emission from non-construction land;
i is the land use type;
ei is the carbon emission occurring on type
i land; T
i is the total area of type
i land; δ
i is the carbon emission or carbon absorption coefficient of type
i land.
The carbon emission of cultivated land consists of the difference between carbon emission from conducting agricultural activities and carbon uptake by photosynthesis of crops on cultivated land, and the carbon emission factor of cultivated land was determined to be 0.0422 kg/(m
2·a) according to the study of Sun He et al. [
32]. Woodland and grassland produce less carbon emissions from human activities and have better carbon sequestration capacity. The carbon emission coefficients of woodland and grassland were determined to be −0.0631 kg/(m
2·a) and −0.0021 kg/(m
2·a), according to Shi Hongxin et al. [
33]. General water bodies include wetlands and lakes, which has better ecological conditions and is mostly carbon sinks. The carbon emission factor of water body was determined to be −0.0253 kg/(m
2·a) according to Shi Hongxin et al. [
33].
Carbon emissions from construction sites are mainly composed of the sum of fossil energy consumption during production and construction and carbon emissions from population breathing. This study selected raw coal, washed coal, coke, natural gas, crude oil, gasoline, kerosene, diesel, fuel oil, and liquefied petroleum gas as the energy sources [
12]. The formula for calculating carbon emissions from construction land is shown below:
where
Et is the total carbon emissions from construction land;
Et1 is the carbon emissions from the consumption of fossil energy for production activities;
Et2 is the carbon emissions from population respiration;
Eti1 is the carbon emissions from the consumption of fossil energy of category I;
Ei1 is the total amount of fossil energy consumption of category
i;
i is the converted standard coal coefficient of fossil energy of category
i;
fi is the carbon emission coefficient of each type of energy.
According to the China Energy Statistical Yearbook and the IPCC Guidelines for National Greenhouse Gas Inventories, the
Table 4 shows converted standard coal coefficients and the carbon emission coefficients of the 10 fossil energy sources.