干旱区生态环境知识资源中心

Digital soil mapping (DSM) can predict the spatial distribution of soil classes. In the present study, DSM was investigated to predicted soil classes in a flood plain. Taxonomic classes including soil great groups, subgroups, and families were modeled using Random forest (RF) technique and covariate sets for an area of similar to 60,000 ha in Sistan Region, eastern Iran. 108 soil profiles were excavated in the study area. Overall model accuracy and the Kappa statistic were used to evaluate the model using 10-fold cross-validation. Results showed fluvial activities are the main factors affected soil formation and development in the studied area. The channel networks, valley depth, convergence, NDSI, and catchment area were the most important covariates. The overall accuracy was 46, 44 and 46.4% for soil great groups, subgroups, and family levels, respectively. Predicted map at family level showed more details than legacy map but at the great group and subgroup levels were similar to legacy maps. Results showed channel networks were the most important covariate at all taxonomic levels in this deltaic area and RF showed high potential to map soil classes in the arid deltaic regions.