ABSTRACT The objective of our study was to develop a model to improve the risk prediction of soil salinization, one of the most severe forms of soil degradation, under different influencing factors. Soil salinization and nonsalinization points were identified in Northwestern China using Sentinel‐2 remote sensing images and were randomly divided into two groups of training (70%) and testing (30%) datasets. Information value (IV) was used to conduct correlation analyses between the influencing factors and the occurrence of soil salinization, and a high‐precision IV–RF (random forest) coupled model was established. The direct and indirect effects of different factors on soil salinization were evaluated using a structural equation model (SEM). The results showed that soil salinization mainly occurs on slopes of 0°–6.6°, precipitation < 240.21 mm, evaporation capacity of 1945.26–2298.64 mm, normalized difference vegetation index (NDVI) of 0.43–0.61, and on farmland among the land use types, with standardized influence size being −0.31, −0.74, 0.41, −0.17 and 0.07, respectively. IV–RF model (area under curve AUC = 0.95, accuracy = 0.77, recall = 0.99, F ‐score = 0.78) was superior to decision tree (DT), IV–DT and RF models. The IV–RF model calculation results showed that 18.84% of the whole region will experience very high risk of salinization in the future, and these high‐risk areas are mainly located in Xinjiang, Gansu, and Ningxia provinces. These results indicated that the IV–RF model can be widely used in the prediction of soil salinization, and it might provide a very good service for formulating strategies to prevent and manage land degradation from macroperspectives.
Gao et al. (Sat,) studied this question.