Spatial Distribution of Soil Organic Carbon Along a Toposequence Using Quantile Regression Forest in Lesotho

Abstract

Soil organic carbon (SOC) is a dynamic component of soil organic matter, playing numerous roles in soil health, quality, and fertility for agronomic and environmental purposes. However, SOC distribution is very variable both vertical and horizontal, therefore understanding its variability is critical in order to provide a more detailed soil carbon dynamics and enhance our capacity to manage and conserve SOC. The aim of this study was to map the spatial distribution of SOC along a toposequence and its distribution at 0–20 and 20–40 cm. This study was conducted across 6 subcatchments of Lesotho and their soil observations were combined and overlayed with environmental covariates to create a model. The Quantile Regression Forest (QRF) model was implemented to produce predicted and uncertainty SOC maps. The model performance was evaluated using 10-fold cross-validation. Results showed that SOC was significantly higher at downslope positions (valleys and toeslopes) than upslope positions (summits and shoulders) with an average of 3.93% at 0–20 cm and 2.02% at 20–40 cm. The results further demonstrated that SOC decreased with depth across all the subcatchments. SOC distribution was mostly influenced by valley depth, channel network base level, temperature, and NDVI. The model performance was best at 0–20 cm (R2 = 0.72, RMSE = 1.45%, and MAE = 1.11%) compared to 20–40 cm (R2 = 0.69, RMSE = 0.97, and MAE = 0.74) making it more suitable for SOC prediction. These findings highlight the importance of understanding SOC spatial distribution, providing insights that can inform soil management and conservation planning across different topographic locations.