Digital soil mapping with adaptive consideration of the applicability of environmental covariates over large areas

https://doi.org/10.1016/j.jag.2022.102986Get rights and content
Under a Creative Commons license
open access

Highlights

  • Adaptively considering covariate applicability can improve soil prediction.

  • Terrain conditions well indicate spatial variation of soil–environment relations.

  • Prediction uncertainty considers both similarity and covariate applicability.

  • Parameters for covariate applicability are insensitivity in this large study area.

Abstract

The effective use of environmental covariates in characterizing soil–environment relationships is key to successful digital soil mapping. The typical way to use environmental covariates in digital soil mapping is by selecting diverse environmental covariates considering the overall geographical characteristics of the study area and considering these covariates to have consistent applicability across the whole area. However, this practice ignores the fact that the applicability of each environmental covariate in characterizing soil–environment relationships varies over complex environmental conditions, especially in large areas. This study proposed a method to adaptively consider covariate applicability in large-area digital soil mapping using soil–environment relationships. The applicability of each covariate at each location was quantified from the terrain conditions using the newly designed fuzzy functions in the study. Then the covariate applicability was regarded as the importance weight and integrated into an existing representative method, iPSM (individual predictive soil mapping). The integration was separately performed at the similarity calculation and soil estimation stages of iPSM to generate two new methods: iPSM weighting on the applicability of all covariates (iPSM_WCovar_all), and iPSM weighting on the applicability of the limiting covariate (i.e., the covariate with the minimum similarity between two locations that constrains the overall similarity) (iPSM_WCovar_limit). Experiments were carried in Anhui Province, China. The two new methods were used to predict the soil organic matter content of topsoil and outperformed the original iPSM and random forest kriging methods. The root mean square error of the iPSM_WCovar_all, iPSM_WCovar_limit, iPSM and random forest kriging methods were 8.14, 8.00, 8.88 and 9.65 g/kg, respectively, while the mean absolute error of those methods were 6.48, 6.31, 6.61 and 6.82 g/kg. Both proposed methods outperformed the iPSM method and the other commonly used method, i.e., random forest kriging. Moreover, the performance was stable under different parameter settings. Experimental results indicate that the idea of adaptively considering covariate applicability in digital soil mapping is feasible and effective.

Keywords

Digital soil mapping
Soil–environment relationship
Large area
Individual predictive soil mapping (iPSM)
Covariates applicability
Uncertainty

Data availability

Data will be made available on request.

Cited by (0)