[Objective] To address the inefficiencies of traditional approaches for monitoring the soil moisture content （SMC） and soil organic matter content （SOMC） in saline-alkaline farmlands， an estimation method that integrates hyperspectral data with interpretable machine learning was investigated. The goal was to establish a theoretical foundation for the rapid acquisition of soil information and quality assessments of the Hetao Plain， China. [Methods] Ground-based hyperspectral reflectance data and field-measured SMC and SOMC were used as the primary data sources. Spectral data were processed using a fractional-order differential （FOD） transformation， and various spectral indices were constructed. The models were developed using partial least squares regression （PLSR）， support vector machines （SVM） and random forest （RF）. To enhance interpretability， the Shapley additive explanations （SHAP） method was employed to evaluate the relative contribution of each variable to model predictions. [Results] ① The spectral indices derived from the 1.25-order differential transformation showed the highest correlation with the SMC and SOMC. In particular， the generalized difference index （GDI） and optimal spectral index （OSI） exhibited the strongest correlations， with coefficients of 0.505 4 and 0.682 5， respectively. ② The RF model significantly outperformed the PLSR and SVM models in estimating both the SMC and SOMC. For the validation datasets， the RF models achieved R² values of 0.734 and 0.870， root mean square errors of 3.28 and 1.53， and recognition-primed decisions of 2.07 and 2.43 for SMC and SOMC， respectively. ③ The SHAP analysis indicated that the normalized plane domain index （NPDI） and ratio index （RI） were the most influential variables for estimating the SMC and SOMC， respectively. The combined contributions of the NPDI， OSI and difference index （DI） to SMC modeling reached 68.58%， whereas RI， GDI and NPDI collectively contributed 61.86% to SOMC modeling. [Conclusion] The integration of FOD and spectral indices enhanced the utility of hyperspectral data. The RF model demonstrated superior accuracy and robustness in estimating soil properties， whereas the SHAP analysis effectively elucidated the contribution of individual variables. Spectral indices （such as NPDI， RI， OSI and DI） played significant roles in modeling SMC and SOMC in saline-alkaline farmland.