Including soil spatial neighbor information improves model performance in predictive soil mapping

Digital soil mapping (DSM) is revolutionizing the understanding and management of soil resources by providing high-resolution spatial and temporal soil information essential for tackling environmental challenges. While integrating environmental covariates has significantly improved mapping accuracy, the potential of neighboring soil sample data remains underutilized. This study introduces soil spatial neighbor information (SSNI) as a novel approach to enhance the predictive performance of spatial models. Using two open-access datasets—LUCAS Soil and Meuse—our results demonstrate that incorporating SSNI improves the accuracy of random forest models for mapping soil organic carbon density (reducing %RMSE by 3.1%), cadmium (3.6%), copper (5.9%), lead (11.5%), and zinc (7.4%). Compared to methods utilizing buffer distances or oblique geographic coordinates, SSNI consistently outperformed for both datasets. These findings highlight the potential of SSNI to enhance digital soil maps by effectively capturing neighboring soil information. Adopting SSNI could advance soil management practices and offers promising opportunities for broader applications in future research across related disciplines.