Spatial Prediction Mapping and Driving Force Analysis of Cultivated Land Soil Bulk Density Based on Soil Workability Threshold Partitioning Strategy

Soil bulk density (BD), a key soil physical property, reflects the physical, chemical, and biological attributes of soil and indicates the stability of cultivated land ecosystems. However, the inherent spatial heterogeneity of soil presents challenges to stable and transferable regional BD prediction and mapping. In digital soil mapping (DSM), accurately capturing the spatial variability of BD remains challenging due to complex soil-environment interactions. Therefore, a WT-based partitioning strategy was developed to divide the study area into soil-condition-based regions, and BD in 2020 was predicted separately within each partition to improve mapping accuracy and identify region-specific drivers. A total of 362 soil samples were collected in 2020 from central Jilin Province, China, and combined with multi-source environmental covariates for BD spatial prediction. The performance of the WT-based partitioned prediction strategy was compared with that of the global prediction strategy. The optimal model was used for BD mapping, and SHAP analysis was applied to reveal the dominant factors controlling BD spatial variations. The results indicated that: 1) the WT-based partitioning strategy effectively captured soil-condition differences and improved the representation of BD spatial heterogeneity. Compared with the global prediction strategy, partitioned models achieved higher prediction accuracy, with increases in R² of up to 0.18, and generated more detailed BD spatial distribution maps. 2) The predicted BD showed clear spatial variation, with lower values mainly distributed in the northeastern part and higher values in the southwestern part of the study area. Area 1 generally exhibited lower BD values than Area 2, indicating distinct spatial distribution patterns between the two WT-based partitions. 3) SHAP analysis revealed distinct driving mechanisms across partitions: soil organic matter was the dominant factor controlling BD variation in the higher-WT area (Area 1), whereas geographic factors played a greater role in the lower-WT area (Area 2). These findings demonstrate that WT-based partitioning provides an effective framework for improving BD spatial prediction and understanding the environmental drivers of cultivated soil bulk density variation.