Mapping Soil Salinity Prediction Enhanced by Time‐Series Remote Sensing Features

Gaining timely knowledge of soil salinity's spatiotemporal dynamics is crucial for accurate mapping. Through temporal span comparisons, this study systematically extracts and utilizes time‐series features (TSFs) derived from Sentinel‐2 and Landsat 8 NDVI and SAVI data (2019–2023) to enhance prediction accuracy. Three datasets were constructed: (i) traditional geographic factors combined with single‐date remote sensing data, (ii) one incorporating Landsat 8‐based TSFs (L_TSFs), and (iii) one incorporating Sentinel‐2‐based TSFs (S_TSFs). These datasets were used with Random Forest regression (RFR) and Support Vector Regression (SVR) models to predict soil salinity. Model evaluation and mapping comparisons demonstrated that incorporating TSFs substantially improved performance relative to single‐date data, with growing‐season TSFs contributing most strongly. Among all models, the RF + S_TSFs combination achieved the highest accuracy ( R 2 = 0.83, RMSE = 0.84 g kg −1 ). These findings highlight the value of time‐series information for digital soil salinity mapping and provide a scalable, data‐rich framework to support precision agriculture and sustainable land management.