Spatiotemporal changes of Mollic Horizon thickness based residual random forest in typical province of northeastern China

The thickness of the Mollic Horizon, a key indicator of soil quality, and its rate of decline are critical for ensuring food security. Current assessments of Mollic Horizon thickness primarily rely on statistical reports and soil erosion studies, which inadequately capture regional spatial patterns. Predicting spatial changes in Mollic Horizon thickness is essential for understanding large-scale trends. However, existing models, while accurate at small watershed scales, often yield substantial errors at regional scales. To address these limitations, this study applied a rapid investigation method (RIM), resulting in a dataset of 357 Mollic Horizon thickness measurements from 2022. A Converting Soil Horizon Depth into Mollic Horizon Thickness (CSMH) method was used to reconstruct the historical thickness dataset from 1746 soil profiles investigated in 1982. To enhance prediction accuracy, the study developed the Residual Random Forest (RRF) and Residual Quantile Regression Forest (RQRF – an uncertainty-aware extension modeling conditional quantiles of residuals) models, generating high-precision spatial distribution maps of Mollic Horizon thickness for two periods. As a benchmark of traditional soil profile survey (TSP) data, the improved RIM methods achieved an accuracy of 95.32 %, and CSMH achieved 86.17 % on Mollic Horizon thickness using surveyed profile data in 2023, demonstrating their data suitability from alternatives methods. The RRF model exhibited high spatial explanatory power (R² > 0.79) and a low error (RMSE < 10.01 cm), supporting accurate regional-scale mapping. Over the past 40 years, Mollic Horizon thickness of cultivated land in Heilongjiang Province decreased by 0.27 cm/year, based on both investigation site statistics and continuous spatial analysis. Significant thinning was observed in the Songnen and Sanjiang Plains due to long-term cultivation, whereas thickening occurred in low-lying cultivated lands near hilly and mountainous areas. These findings provide robust data and methodological advances for Mollic Horizon protection in Northeast China.