Predicting spatial distribution of soil organic carbon and total nitrogen in a typical human impacted area

Abstract

The relationship between soil properties and environmental covariates is mostly nonlinear for human impacted areas. Herein, we proposed a nonlinear model for mapping soil organic carbon (SOC) and total nitrogen (TN) in a typical human impacted area (a small watershed of Poyang Lake, China), namely radial basis function neural network combined with agricultural land use (RBFNN_ALU). The results showed that the RBFNN_ALU performs better than the ordinary kriging (OK), the OK combined with agricultural land use (OK_ALU), the geographically weighted regression (GWR), the multiple linear regression (MLR), the MLR combined with agricultural land use (MLR_ALU) and the RBFNN. In addition, RBFNN_ALU provided a more detailed and accurate description of the spatial SOC and TN patterns. The results indicate that when predicting spatial distribution of SOC and TN for human impacted areas, non-linear models are critical for predicting the spatial distribution of soil properties.