基于多时相遥感影像和随机森林算法的土壤制图

获取准确的土壤-环境关系是数字土壤制图的关键,目前遥感影像已作为环境因子应用于土壤-环境知识的建立过程,但单幅遥感影像所包含的光谱信息差异难以将不同土壤类型区分开来。因此本文提出了一种基于多时相遥感影像的土壤制图方法：选取红安县滠水河流域为研究区,以母质类型图、等高线数据和多时相哨兵二号遥感影像为基础,提取与土壤形成有关的环境因子,通过随机森林算法获取土壤-环境关系,预测研究区各土壤类型的空间分布并成图,利用野外实地分层采样点验证推理图的精度。结果表明：推理土壤图总体分类精度高达86%,与原始土壤图对比,各土壤类型的空间分布具有一定相似性,展现了更为详细的空间细节信息,该研究成果可为更新土壤图工作提供新方法。

Soil Mapping Based on Multi-temporal Remote Sensing Images and Random Forest Algorithm

Extracting accurate soil-environment relationship is the key to digital soil mapping. Nowadays, remote sensing images have been used as indicators of environmental factors in the process of obtaining soil-environment knowledge. However, the spectral differences in mono-temporal image are difficult to be used to distinguish soil types. In this study, we proposed a soil mapping method based on multi-temporal remote sensing images. The Sheshui River Basin in Huajiahe Town, Hongan County, Huanggang City of Hubei Province was selected as the study area, and the parent-material-type map, the multi-temporal sentinel-2 remote sensing images, and contour data were used to extract environmental factors related to soil properties. Soil environment relationships were obtained to infer the spatial distribution of soil types using the random forest algorithm. The field sampling points in the study area were used for validation, and the confusion matrix and Kappa coefficient of inferenced soil map were calculated to evaluate the map accuracy. The results demonstrated that the overall classification accuracy of the inferred soil map is as high as 86%. The soil type map obtained by inference was similar to the traditional soil map in the spatial distribution, but it could display more detailed information than the traditional soil map. This research can provide an effective alternative for updating the traditional soil map.