Predication of the Post Mining Land Use Based on Random Forest and DBSCAN

Abstract

AbstractMine reclamation is one of the most important stages of the mining activities in line with the basic principles of sustainable development. In this study, different post-mining land uses are evaluated in the Hongliulin mining area, which is located in Shen mu country of China. 145 soil samples were collected in the May,2021 by using the soil auger, and the sampling depths were 0-20 cm. The sampling points contains 45 to be reclaimed samples and 100 existing classification land use types. 14 environmental factors including soil organic matter (SOM), total nitrogen (TN) and other soil nutrients and terrain factors were extracted and calculated based on laboratory test and digital elevation map. Meanwhile, random forest classier was utilized to determine the post-mining land use based on GINI index and 14 environmental factors by using 100 existing classification land use types. 82 of 100 samples were utilized to build the model and the other 18 samples were utilized to validate the accuracy of the classification model. Density-Based Spatial Clustering of Applications with Noise (DBSCAN) techniques were utilized to determine the specific land use types in cropland, shrub and grassland. the random forest classier showed a great prediction capability, with only 1 miss-classified sample in the validation data-set, the accuracy of the classification model was 95%. Among the 45 samples to be reclaimed, 15 samples were classified into cropland, 23 samples were classified into shrub and grassland, 5 samples were classified into arbor land, 2 samples were classified into solar station. By using DBSCAN, the 58 samples classified into cropland in terms of the post-mining were separated into cluster 1 and 2, cluster 1 comprises 24 samples while cluster 2 comprises 34 samples. Among the 61 samples classified into shrubland and grassland, Cluster -1 comprises 24 samples and Cluster 0 comprise 22 samples. The results could provide technological support for land reclamation determination. The content of TN of C1 is 5 times more than C2 and 4 times more than C3. Also, the K valve of C1 column is maximum and over 0.4, which means the soil particle is relatively smaller and the soil texture of it is sandy loam. The TWI of C1 is minimum, in consideration of a positive valve of profile curvature and a negative valve of plan curvature. In terms of the 45 to be reclaimed samples, 15 samples were classified into C1, 23 samples were classified into C2, 5 samples were classified into C3, 2 samples were classified into C4. The valve of K and content of soil nutrients of the samples classified to be C1 column(C1-C) is maximum. Simultaneously, the slope steepness(°) is below 5° and is perceived as gentle surface. Cluster 1 comprises 24 samples and the average content of TN, AP and SOM is 0.566g/kg, 11.93mg/kg and 19.975g/kg respectively, while Cluster 2 comprises 34 samples and the average content of TN, AP and SOM is 0.304g/kg, 3.12mg/kg and 8.36g/kg respectively. The result may contribute to the land use planning and idle land utilization strategy.