Integrating machine learning algorithms and remote sensing for High-Resolution mapping of soil crusting in agricultural lands

Soil crusting is a significant form of land degradation that adversely affects soil quality and its functional properties, particularly in agricultural lands. Accurate identification of degraded areas and optimization of soil management practices in croplands are essential to mitigate the negative impacts of soil crusting. In this study, 520 soil samples were collected from the top 10 cm of soil in a portion of agricultural lands in Shaanxi Province, China, and the Soil Crusting Index (SCI) was subsequently calculated. Two machine learning algorithms, Random Forest (RF) and Multiple Linear Regression (MLR), were evaluated using 26 indices derived from a digital elevation model along with 13 remotely sensed datasets. Results indicated that SCI values in the study area ranged from 0.27 to 2.69, with highest susceptibility in the northeastern parts and lowest in the southern parts. The RF model outperformed the MLR model, showing a higher coefficient of determination (R² = 0.79 vs. 0.65), lower root mean square error (RMSE = 0.161 vs. 0.242), and reduced bias (0.031 vs. 0.130). Variable importance analysis within the RF framework identified the Clay Index (CI) as the most influential predictor of SCI, while precipitation and runoff- and vegetation-related factors in agricultural lands also had substantial effects. Other examined indices contributed less significantly to model performance. Based on the results of this study, it is recommended that future research employ hybrid modeling approaches combining RF with other algorithms to identify the best model for predicting soil crusting and to better understand the dynamics of soil crust formation in croplands within the study area. This approach can aid in the development of more effective management strategies to mitigate soil crusting.