A lake salinity dataset produced via microwave and optical imageries

Abstract

Lake salinity is an important parameter to characterize physical and biogeochemical processes and a fundamental indicator to evaluate lake water quality. However, its estimation in inland waters has been challenging because passive microwave salinity satellites lack sufficient spatial resolution, and optical satellites cannot directly measure it. To address it, we constructed a framework for estimating lake salinity by combining Synthetic Aperture Radar (SAR) and Multi-Spectral Instrument (MSI) data. It can be summarized in step 1: construct a salinity mechanism model based on SAR data using the Elfouhaily spectrum, dielectric constant, and small perturbation method (SPM) models; step 2: develop four machine learning (ML) salinity algorithms using quasi-synchronous salinity and MSI with SAR imagery; and step 3: build an ensemble model to estimate salinity by coupling the mechanism and ML models via a generalized additive model. The proposed integrated algorithm (N= 84, RMSE = 0.60 ppt, and MAPE = 2.3 %) outperformed single-satellite microwave mechanistic or ML models across all eleven lakes in the Inner Mongolia Xinjiang Lake zone. On this basis, we reconstructed the lake salinity dataset for 2016–2024 and conducted independent validation (N= 65, R2= 0.97, and RMSE = 0.89 ppt) and pixel-level histogram validation confirmed dataset quality, with no significant systematic bias across lake types. The reconstruction revealed a spatial pattern of smooth transition from the nearshore to the center and trends with significant increases in Lake Daihai and Lake Dalinor. The dataset and its development framework will facilitate exploration of salinity status and trends in inland lakes, providing scientific evidence and methodological support for salinization prevention and global lake salinity budget research. The dataset (10 m spatial resolution, TIF format) is publicly available via Zenodo (https://doi.org/10.5281/zenodo.18371515, Deng et al., 2026) and includes annual/seasonal salinity rasters and statistical files.