Estimating soil salinity stock in cropland by remote sensing to save agricultural desalination irrigation quotas

Abstract

Improving land-water use efficiency in arid regions hinges on managing salt stress, which constrains both crop yields and irrigation productivity. While off-season desalination irrigation can alleviate salt stress, water allocation remains largely empirical because baseline soil salinity stock data are lacking, leading to substantial freshwater waste. We therefore advocate developing a rapid, cost-effective, and accurate method to estimate soil salinity stock and enable science-based quota irrigation. This study focused on cotton fields in the Alar Reclamation of Xinjiang, China, and integrated Sentinel‑1 SAR, Sentinel‑2 bare‑soil spectra, and multiple environmental covariates. Predictors were selected using recursive feature elimination, Geodetector, and Shapley, after which the random forest model was constructed to quantify the soil salinity density (SSD), thereby developing the desalination irrigation quotas. Results show that the SSD values range from 1.12 to 10.41 kg m −2 , with lower values in the central river corridor and higher values in peripheral patches. According to SSD estimates, the total soil salinity stock is 6.16 −6.27 Tg, corresponding to a simulated total irrigation volume of 4.19 × 10 8 m 3 to 4.30 × 10 8 m 3 . Theoretically, compared with empirical irrigation, 11.34–13.61% of freshwater may be saved. Among the covariates, planting years, human impact index, and soil sand content are dominant. The remote-sensing‑based estimation framework for SSD developed in this study provides technical support for precision irrigation in arid regions.