Geographical detector-based stratified regression kriging strategy for mapping soil organic carbon with high spatial heterogeneity

Abstract

Spatial variability of soil organic carbon (SOC) content is crucial in precision fertilization and climate change prediction. Stratified ordinary kriging (SOK), which considers the sharp spatial variability of SOC due to natural factors and human activities, is a feasible way to map SOC with high spatial heterogeneity. However, conventional stratified methods are arbitrary and may therefore result in unnecessary and ineffective stratification. In addition, SOK performs poorly when soil properties of some strata present a local trend. In this study, we propose a geographical detector-based stratified regression kriging (GD–SRK) strategy, which aims to guide the stratification, gain strata-dependent knowledges, and improve the performance of stratified kriging. The study area locates in Jianghan Plain, China. The soil and land use types are used as categorical variables, while land use percentages are used as continuous variables. Ordinary kriging (OK), regression kriging (RK), and geographical detector-based stratified ordinary kriging are used as references of GD–SRK. Results indicated that the spatial heterogeneity of SOC was attributed to the differences in soil types and land use types and suggested that the study area should be stratified into four. The spillover effect of land use on the SOC content depended on the strata. Specifically, the SOC in strata II (paddy fields ∩ Fluvisols) and IV (irrigated lands ∩ Fluvisols) was unaffected by land use percentages and was estimated via OK. The SOC in strata I (paddy fields ∩ Anthrosols and Luvisols) and III (irrigated lands ∩ Anthrosols and Luvisols) was negatively correlated with the percentage of irrigated lands. Thus, their estimation of SOC via RK was possible. Using these knowledges, the GD–SRK outperformed the other methods by obtaining highest prediction accuracy and R2. We conclude that GD–SRK strategy effectively guides the stratification and maps SOC with high spatial heterogeneity.