Soil Texture and Prediction Dataset for Shaanxi, China

The dataset includes structured CSV files containing the environmental covariates and target variables required for predicting soil texture at different depths. The target variables — sand, silt, and clay — are expressed in percentage units. These CSV files serve as the primary input for the machine learning modeling process, encompassing all feature variables and corresponding values for each soil texture component. A supplementary Excel file is provided, which consolidates the performance evaluation results of the predictive models. The file is organized into three separate sheets, labeled sand, silt, and clay, each containing detailed statistical metrics and validation outcomes specific to the corresponding soil texture fraction. This structured format allows for clear and systematic comparison of model accuracy across different soil components. Spatial predictions of soil texture are delivered in raster format, with each file representing a specific depth interval. The naming convention follows a depth-code system: 05for 0–5 cm, 520for 5–20 cm, 2040for 20–40 cm, 4060for 40–60 cm, and 60100for 60–100 cm.