Spatiotemporal Dynamics and Driving Mechanisms of Water Erosion in the Shiyang River Basin, Northwest China (2001–2020)

This dataset was developed to investigate the spatiotemporal patterns and driving factors of water erosion in the Shiyang River Basin from 2001 to 2020. Water erosion intensity in this region is influenced not only by natural factors such as precipitation and topography, but also significantly by land use change and vegetation dynamics. Moreover, these driving factors may interact to amplify erosion risks. The dataset includes the following components:
(1)Simulated Water Erosion Data:
Annual raster layers of potential soil erosion, transport capacity, and net water erosion rate, generated using an integrated RUSLE and TLSD model. These data capture both the potential for soil detachment and the actual sediment transport across the basin landscape. In addition, multi-year spatial trend data and inter-annual change maps are provided to illustrate erosion dynamics over time.
(2)Erosion Classification and Transition Data:
Net water erosion intensity was categorized into seven classes. The spatial variation in erosion severity and the area transition matrix across different years were compiled to assess temporal changes and spatial dynamics in erosion patterns.
(3)Driving Factor Analysis (Random Forest):
Output from the Random Forest model quantifies the relative importance of various natural and anthropogenic drivers of erosion. Summary tables and maps illustrate erosion responses under different environmental gradients, including topography, LUCC, and NDVI.
(4)Supplementary Data:
Raster datasets for all RUSLE input factors (rainfall erosivity, soil erodibility, slope and slope length, cover management, and support practice), NDVI, topsoil texture, and LUCC data are included. The dataset also provides statistics on net soil erosion rate, sediment deposition, and NDVI values across different slope classes and under various land use change scenarios.
Data Collection and Processing:
Climate and topographic data were obtained from national and international geospatial databases. Slope and slope length were derived from DEM data, and rainfall data were used to calculate rainfall erosivity. NDVI and land use/land cover (LUCC) data were extracted from MODIS and Landsat imagery and reclassified based on regional characteristics.
Soil erosion was simulated using a calibrated RUSLE-TLSD model. Model inputs include rainfall erosivity, soil erodibility, topographic factors, vegetation cover, and land management practices. Sediment yield observations were used for model validation.
Random forest analysis was conducted in R using the “randomForest” and “rfPermute” packages. The model was trained with 500 trees to evaluate the importance of environmental variables in driving net water erosion.
All raster datasets were resampled to a uniform 90 m resolution for spatial analysis.