Optimal time-window for assessing soil salinity via Sentinel-2 multitemporal synthetic data in the arid agricultural regions of China

作者：Ju Xiong, Xiangyu Ge, Jianli Ding, Jinjie Wang, Zipeng Zhang, Chuanmei Zhu, Lijing Han, Jingzhe Wang · 发表于：Ecological Indicators · 年份：2025 · DOI：10.1016/j.ecolind.2025.113642 · 被引用次数：22 · 研究领域：Soil Geostatistics and Mapping、Geochemistry and Geologic Mapping、Soil and Land Suitability Analysis

Soil salinity is a critical issue affecting agricultural productivity in arid regions. Remote sensing is an effective tool for assessing and monitoring soil salinity to enable precision soil care. However, obtaining bare-soil information from agriculturally active regions remains challenging. Therefore, this study aimd to identify the optimal temporal window for assessing soil salinity. We developed three different time-synthesis strategies based on Sentinel-2 time-series images (1-month synthetic, 2-month synthetic, and seasonal synthetic image) through median and mean syntheses. We constructed estimation models (including random forest (RF) and gradient tree boosting (GTB)) using band and spectral indices information from synthetic images in the Google Earth Engine (GEE) platform. Additionally, we compared the results of different modeling strategies and assessed the uncertainty in soil salinity mapping. The results showed the optimal time-window for assessing soil salinization was the images synthesized in summer (June-August) (R 2 : 0.41–0.45), which was approximately 36.51% higher than that during the bare soil period (March-April). Assessment models constructed from summer synthetic imagery had a low uncertainty in soil salinity mapping. The median-based synthesis approach was the most effective, compared to the mean-based synthesis approach with an R 2 of 0.45 (RF validation mean). The six spectral indices including EVI, GYEX, TBI, GARI, NDSI, and NDVI proved more impo...