An approach for broad‐scale predictive soil properties mapping in low‐relief areas based on responses to solar radiation

Abstract

AbstractIn low‐relief areas, easily observed landscape features such as terrain and vegetation often do not spatially co‐vary with soil conditions to the level that they can be effectively used in predictive soil mapping. This paper proposes an approach to predicting soil spatial variation over such areas at a coarse resolution by constructing environmental covariates from the dynamic feedbacks of land surface in response to solar radiation. These feedbacks are captured by the Moderate Resolution Imaging Spectroradiometer (MODIS) land surface temperature observations as a time series of four temperatures per day (1:30 AM, 10:30 AM, 1:30 PM, 10:30 PM) converted into five covariates representing the features of the time series. The approach is demonstrated by mapping sand, silt, soil organic matter (SOM), and pH over a 12,000‐km2 low‐relief area of Jiangsu Province, China, using a random forest model trained on 144 soil observations. The model was built and evaluated for 13 single days. For the non‐winter days with relatively high solar radiation, the covariates were strongly correlated with the soil properties, so that their use in predictive mapping gives good results for soil texture and useful results for SOM and pH. We also present plausible physical interpretations for the action of the covariates based on thermal properties of soils. We conclude that, despite the coarse resolution of the MODIS‐derived covariates, the proposed approach is a feasible solution for predictive soil mapping over large areas of low relief and should be investigated at finer resolutions.