High-resolution and three-dimensional mapping of soil texture of China

• Published in: Geoderma Vol. 361; p. 114061
• Main Authors: Liu, Feng, Zhang, Gan-Lin, Song, Xiaodong, Li, Decheng, Zhao, Yuguo, Yang, Jinling, Wu, Huayong, Yang, Fei
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.03.2020
• Subjects: China; clay fraction; climate; Digital soil mapping; ecology; Environmental factors; heat; hydrology; landscapes; Large extent; Machine learning; prediction; regression analysis; sand; silt; soil formation; soil profiles; soil quality; soil texture; Uncertainty; weathering; wind; China; R2; MAT; CCC; Uncertainty; Digital soil mapping; Large extent; TWI; Environmental factors; RMSE; NDWI; LST; DEM; NDVI; MODIS; SD; CV; RI; Machine learning; ME; MAP; ETM
• ISSN: 0016-7061; 1872-6259
• DOI: 10.1016/j.geoderma.2019.114061

•High resolution soil texture maps were predicted at a national extent.•The maps were much more accurate and detailed than the existing maps.•Climate and terrain are major controllers for spatial patterns of soil texture.•Physical and chemical weathering and wind erosion largely shape clay pattern.•Erosion, deposition, transportation and sorting largely shape silt pattern. The lack of detailed three-dimensional soil texture information largely restricts many applications in agriculture, hydrology, climate, ecology and environment. This study predicted 90 m resolution spatial variations of sand, silt and clay contents at a national extent across China and at multiple depths 0–5, 5–15, 15–30, 30–60, 60–100 and 100–200 cm. We used 4579 soil profiles collected from a national soil series inventory conducted recently and currently available environmental covariates. The covariates characterized environmental factors including climate, parent materials, terrain, vegetation and soil conditions. We constructed random forest models and employed a parallel computing strategy for the predictions of soil texture fractions based on its relationship with the environmental factors. Quantile regression forest was used to estimate the uncertainty of the predictions. Results showed that the predicted maps were much more accurate and detailed than the conventional linkage maps and the SoilGrids250m product, and could well represent spatial variation of soil texture across China. The relative accuracy improvement was around 245–370% relative to the linkage maps and 83–112% relative to the SoilGrids250m product with regard to the R2, and it was around 24–26% and 14–19% respectively with regard to the RMSE. The wide range between 5% lower and 95% upper prediction limits may suggest that there was a substantial room to improve current predictions. Besides, we found that climate and terrain factors are major controllers for spatial patterns of soil texture in China. The heat and water-driven physical and chemical weathering and wind-driven erosion processes primarily shape the pattern of clay content. The terrain, wind and water-driven deposition, erosion and transportation sorting processes of soil particles primarily shape the pattern of silt. The findings provide clues for modeling future soil evolution and for national soil security management under the background of global and regional environmental changes.