Spatial prediction of soil organic carbon content in coal mining subsidence area based on GIS and partition Kriging

Abstract: China's coal production accounts for 46.9% of total world coal production, and the annual subsidence area is about 200 km2, which greatly disturbs the soil carbon pool of farmland. As the soil organic carbon (SOC) pool of farmland is the most potential factor reducing carbon emissions from terrestrial ecosystems, China and other major coal mining countries in the world must do better in managing the SOC pool in the coal mining area, which is also an important way for low-carbon coal mining. A spatial prediction method with good prediction accuracy of SOC content in subsidence area of coal mining is the prerequisite of scientific management of SOC pool in coal mining subsidence area. In order to determine a prediction method with high accuracy, taking Jiuli mining area in Xuzhou as a case, this paper predicted SOC content in the study area by the ordinary Kriging and the partition Kriging method. The ordinary Kriging directly spatially interpolated the SOC content based on the regionalized variable theory. In the partition Kriging method, the study area was categorized into non-waterlogged subsidence area, seasonally waterlogged subsidence area and wetland according to the subsidence and water-logging condition in the study area. With the method of partition Kriging, the spatial prediction value can be gained by summing up the mean value of SOC content in each separated area and the predictive residuals through spatial interpolation based on the residual data of the SOC content at each sampling point. Lastly this paper compared the prediction accuracy between the 2 methods with the measured values to determine the better method. It is found that the SOC content obtained by ordinary Kriging ranged from 9.34 to 16.252 g/kg, while the result of the partition Kriging was from 9.333 to 18.058 g/kg. The estimated ranges of the 2 methods were approximately same. Through the comparison, the spatial distribution of the SOC content obtained by ordinary Kriging showed no regularity because of neglecting the smooth transition between regions, while the spatial gradient features of SOC could be enhanced by the partition Kriging. The result showed that the correlation coefficient between the predicted and observed values obtained by the partition Kriging method was 0.7564, which was far higher than that by the ordinary Kriging method (0.5086). The root-mean-square error (RMSE), the mean absolute error (MAE) and the mean error (ME) of 2 methods were 0.35 and 0.55, 1.8511 and 1.2878, 0.0202 and 0.018, respectively. The partition Kriging performed much better than ordinary Kriging as it revealed much lower values of ME, MAE, and RMSE and a higher R2, indicating that the former has higher prediction accuracy. It is concluded that the partition Kriging interpolation model is a more appropriate spatial prediction model for SOC content in coal mining subsidence area, which provides the scientific basis for low-carbon land reclamation in mining area and even utilization of land resources in the region.