Abstract: In coastal plain regions, soil salinity serves as one of the abiotic stressors limiting crop growth. The content of soil salinity is a critical determinant for crop cultivation. The variety of crops grown can indirectly indicate the extent of soil salinization. Therefore, this paper proposes an integrated approach for the inversion of soil salinity, incorporating crop type information. Based on Sentinel-2 MSI imagery in a typical coastal saline soil area in the Yellow River Delta. Firstly, crop type information was extracted using random forest classification and coded based on the OneHot method. Then, by integrating crop type information, environmental covariate data, and ground-measured salinity data, the adaptive boosting decision tree (AB-DT) model is applied for soil salinity estimation. Finally, the accuracy of salinity estimation is compared with other machine learning methods, including support vector machines, random forests, K-nearest neighbors, and decision trees. The results indicated that ①Incorporating crop type information enhances the accuracy of soil salinity estimation models. Among all models, the AB-DT model with fused crop type variables achieves the highest modeling set R² of 0.86 and validation set R²of 0.61.②The inclusion of crop type information enable to correct misclassifications of salinity levels and yield sharper boundaries in soil salinity estimation results. In conclusion, the incorporation of crop type information improves the accuracy of soil salinity estimation, providing a more reliable basis for agricultural management and decision-making.

Key words: soil salinization, multi-spectral remote sensing estimation, machine learning