关键词:
Digital soil mapping
Environmental factors
Frequent itemsets
Machine learning
Parent material
摘要:
【Objective】For the rational use of land resources, it is important to obtain accurate spatial distribution of soil types using digital soil mapping technologies.【Method】In this study, environmental factors were screened according to the soil parent material type based on field sampling points, and then three different mapping methods, random forest, SoLIM, and KNN, were used to map the zones according to the selected environmental factors, respectively. Each method was used individually to generate zoning maps, providing different reasoning for the spatial distribution of soil types. The zoning mapping results were obtained and combined to form a universal spatial distribution map of soil types, and then, we used the FP-Growth algorithm to effectively mine the internal correlation between environmental factors. By combining these associations with different mapping results obtained previously, the spatial distribution of soil types in the study area was deduced and used to obtain higher quality and precision inference results. 【Result】The mapping results revealed several key findings: (1) The independent mapping of soil type based on the parent material type of soil by three different mapping methods is more effective and accurate than the joint mapping of all parent materials, and the inference of spatial distribution of soil types is also more reasonable. (2) Among the three mapping methods adopted in this study, the method combining random forest and frequent itemset mapping had the highest accuracy of 70.73%. Moreover, the results obtained by this combined method are similar to the spatial distribution of soil types inferred by the other two combined methods. Through comparative analysis, we were able to determine the approximate spatial distribution of soil species in the study area. (3) After the three mapping methods were combined with frequent itemsets, we observed that all methods had different degrees of improvement in accuracy verification and Kappa coeff