Scenario simulation and landscape pattern dynamic 1 changes of land use in mining area 2

6 In this study, we selected 11 townships with severe ground subsidence located in Weishan County as the study area. Based on the interpretation data of Landsat images, the Binary logistic regression model was used to explore the relationship between land use change and the related 7 driving factors at a scale of 60m × 60m. Using the CLUE-S model, combined with Markov model, the simulation of land use under three scenarios–namely, natural development scenario, ecological protection scenario and farmland protection scenario–were explored. Firstly, using land use map in 2005 as input data, we predicted the land use spatial distribution pattern in 2016. By comparing the actual land use map in 2016 with the simulated map of land use pattern in 2016, the prediction accuracy was evaluated based on the Kappa index. Then, after validation, the distribution of land use pattern in 2025 under the three scenarios was simulated. The results showed the following: (1) The driving factors had satisfactory explanatory power for land use changes. The Kappa index was 0.82, which indicated good simulation accuracy of the CLUE-S model. (2) Under the three scenarios, the area of other agricultural land and water body showed an increasing trend; while the area of farmland, urban and rural construction land, subsided land with water accumulation, and tidal wetland showed a decreasing trend, and the area of urban and rural construction land and tidal wetland decreased the fastest. (3) Under the ecological protection scenario, the farmland decreased faster than the other two scenarios, and most of the farmland was converted to ecological land such as garden land and water body. Under the farmland protection scenario, the area of tidal wetland decreased the fastest, followed by urban and rural construction land. We anticipate that our study results will provide useful information for decision-makers and planners to take appropriate land management measures in the mining area.


Introduction
In this study, mining area with severe ground subsidence problems was selected as the study area. We used BLRM to 33 analyze the driving forces of land use and landscape pattern evolution in mining area. On this basis, we selected the CLUE-S 34 model and Markov model to simulate and predict the spatial characteristics of land use patterns under different scenarios in 35 the future. We anticipate that our study results will provide theoretical foundation for the optimal allocation and sustainable 36 utilization of land resources in the mining area.

37
Study Area 38 Weishan county(34 • 27 N to 35 • 20 N, 116 • 34 E to 117 • 24 E), is located in the southern part of Jining City, Shandong Province. 39 The study area is 120 km long from north to south, 8-30 km wide from west to east. Nansi Lake, the largest freshwater lake 40 in northern China, is located within the study area. Weishan county comprises 3 sub-districts, 10 towns, 2 townships and 1 41 economic development zone (2014 administrative division), with a total area of 1779.8 km 2 . We selected 11 townships with a 42 total area of 1176.86 km 2 , because this study area has more mines, more severe land subsidence, and is spatially coherent. The 43 geographical location of the study area and the distribution of mining area are shown in Figure 1.

45
Data source 46 Considering factors such as amount of cloud and time intervals of image, four remote sensing images with a spatial resolution 47 agricultural land, urban and rural construction land, subsided seeper area, water area, and tidal wetland. In the process of image interpretation, we used support vector machine (SVM) classification, manual visual interpretation and decision tree 57 classification, hierarchical classification combined with many exponential models, to obtain the 2000, 2005, 2010, 2016 land 58 use type maps (Figure 2). The accuracy of the interpretation results was verified by confusion matrix and kappa coefficient. 59 The kappa coefficients of the four interpretation maps were 0.84, 0.85, 0.82 and 0.86, respectively. The accuracy could meet the 60 needs of further research. 61 The interpreted land use maps with a resolution of 30m × 30m exceed the upper limit of the CLUE-S model data, so the 62 land use maps were resampled to multiple resolution scales including 60 m, 90 m, 120 m, and 150 m to facilitate logistic 63 regression analysis of land use types and driving factors.

65
To interpret the relationship between the land use and its driving factors in the mining area, we not only need to identify the 66 driving factors that have greater explanatory power for land use change, but also need to quantitatively describe the relationship 67 between driving factors and land use types.

68
Considering the accessibility, usability of the data and the actual conditions in the study area, seven driving factors were including the nearest distance between each grid pixel and the main roads, the major rivers, the residential area, the major

79
BLRM is often used for regression analysis of explanatory binary variables. The presence and absence of a certain type of land 80 use in a specific area is set as 1 and 0, respectively, which is characteristic for binary variable. Therefore, we used BLRM to 81 4/13 calculate the probability (P) of various land use types in a specific spatial location, and its mathematical expression is: Where P 1−P is the 'odds ratio' of an event, abbreviated as Ω, which represents the odds that an outcome will occur given a 83 particular condition compared to the odds of the outcome occurring in the absence of that condition; β 0 is a constant; β 1 is the 84 correlation coefficient of an explaining variable and an explained variable. Making mathematical transformation of the above 85 formula, we get: Ω = ( P 1−P ) = e β 0 +β 1 X .

86
Regression analysis using BLRM, we divided the study area into many grid cells. Taking each land use type as the explained 87 variable, and the driving factor causing land use change as the explanatory variable, we calculated the odds ratio of each land 88 use type in a specific spatial location, and analyzed the relationship between each land use type and the driving factors. The 89 calculating formula is: Making mathematical transformation of the above formula, we get: Where: P i is the probability of a certain land use type i in a grid cell, X 1,i ∼ X n,i are the driving factors of land use type i, β 0 92 is the constant, β n ∼ β n are the correlation coefficients of each driving factor and land use type i.

93
The relationship between each land use type and the driving factors was obtained using BLRM 11, 29 . The β coefficients 94 (listed in Table 1), derived from the logistic regression equation, were used as input parameters for the CLUE-S model. Table 1 95 shows that the distance to residential area was the main driving factor for the change of urban and rural construction land, and 96 there was obvious negative correlation between them, which suggested the probability of construction land occurrence was 97 relatively less in areas far away from the residential area. There was a significant negative correlation between subsided seeper 98 area and the distance from mines, main rivers, and roads, suggesting that the probability of subsidence water area occurrence 99 increased around mines, rivers and main roads. The distance to river was a negative explanatory variable for other agricultural  The receiver operating characteristic (ROC) was used to evaluate the accuracy of regression analysis results. As shown in grassland, and aquaculture land, belong to other agricultural land, which have regulatory effects on the local ecosystem, so their 133 conversion to other land use types should be restricted as well.

134
(3) Farmland protection scenario 135 According to the guidelines of "the general land use planning in Weishan County (2006-2020)", we should maximize the 136 potential use of current construction land, implement intensive and economical utilization of construction land, and use less or 137 not use farmland to economical construction, so as to ensure the dynamic balance of total farmland amount and the regional 138 food supply security, Therefore, in the farmland protection scenario, the conversion from farmland to other land use types 139 should be restricted. The projected land use demands for 2025 under the three different scenarios are shown in Table 2.
147 The BLRM was established and validated to explore the relationship between driving factors and land use types. Using the  to be converted to water area. Tidal wetland was mostly predicted to be converted to other agricultural land or water area.

177
Specifically, the large areas of tidal wetland, located in the east bank of Zhaoyang Lake and the north bank of Weishan Lake, 178 were projected to be converted to water area, and a large area of tidal wetland in the north of Liuzhuang Town was projected to 179 be converted to other agricultural land.

180
(2) Ecological protection scenario 181 In the ecological protection scenario, the change of land use types was similar to those in the natural development scenario. Xiazhen Street and the middle of Zhaomiao Township was projected to be converted to other land. In this scenario, the reduction 184 of construction land was faster than that in the other two scenarios, with -7.01% changing rate and total area reduction of 185 3434.58 hm 2 . The tidal wetland was mostly to be converted to water body. In addition, the reduction of farmland was also 186 faster in this scenario as compared with the other two scenarios, with an estimated changing rate of -2.91% and a total area 187 reduction of 4172.49 hm 2 . The farmland was mainly converted to more ecological land types such as garden land and water 188 area, due to the implementation of "Grain for Green Project " and "Grain for water Project". The subsided land with water 189 accumulation also had a faster conversion rate in this scenario and was mostly to be converted to water area.