Predicting the spatial distribution patterns of albic horizon thickness and burial depth in Northeast China

The albic horizon is a bleached diagnostic layer formed through combined processes of iron-manganese leaching and clay illuviation. Its thickness and burial depth collectively determine the critical role in impeding water movement and restricting root extension. However, understandings of the spatial variation of the albic horizon remain limited. Traditional understandings primarily rely on extensive field surveys combined with geostatistical methods. To expand such understandings on a broader geographical area, we developed a digital soil mapping (DSM) framework to predict the thickness and burial depth of the albic horizon. We integrated 111 soil profile points containing albic horizons with a rich set of environmental covariates to construct reliable prediction models. Following the feature selection process of recursive feature elimination (RFE), a quantile regression forest (QRF) model was employed for spatial prediction, cross-validation, and uncertainty estimation. The results from 50 repetitions of 10-fold cross-validation demonstrated robust model performance, with R² of 0.38 and 0.29 for thickness and burial depth, respectively, and RMSE% accounting for 34% and 38% of their mean values. The prediction interval coverage percentage (PICP) indicated that approximately 86.9% and 90.6% of the validation samples for thickness and burial depth, respectively, fell within the predefined 90% prediction interval (PI), affirming the reliability of uncertainty estimation. The relative variable importance indicated that climate factors were the dominant determinants in predicting both albic horizon thickness and burial depth, highlighting the necessity of incorporating climate data in the spatial modeling of albic soils. The prediction maps indicated a general decreasing trend in both the thickness and burial depth from northeast to south across the study area. Large prediction uncertainty mainly occurred in areas where soil survey points were lacking, highlighting the need for targeted supplementary surveys. Our findings offer valuable references for other similar large-scale mapping of soil layer thickness (or depth) in plain agricultural regions.