Integration of Google Earth Engine, Sentinel-2 images, and machine learning for temporal mapping of total dissolved solids in river systems

Abstract

One of the important indicators of water quality (WQ) in inland water systems is total dissolved solids (TDS). Collecting and maintaining in situ TDS data with high spatial and temporal resolution is time and money-consuming. This study highlights an advanced approach integrating Google Earth Engine (GEE), remote sensing techniques using Sentinel-2 imagery, and machine learning algorithms to map TDS in a spatially explicit manner. We extracted relevant spectral indices and used them to train machine learning models, specifically Random Forest (RF) and Support Vector Machines (SVM), to classify TDS levels across the stretch of the Little Miami River (LMR). We analyzed TDS for August, September, October, and November, and over three years, from 2020 to 2023. Results showed RF to be more effective than SVM in mapping TDS levels, with overall accuracies and Kappa coefficients up to 0.88 and 0.85, respectively, for November 2021. Further, TDS levels remained a concern, particularly in the midstream LMR sections. Temporal rainfall variations corresponded with elevated TDS levels. Areas with higher percentages of natural cover (forests and wetlands) showed greater resilience to TDS fluctuations (r = -0.632) compared with developed or barren lands (r = 0.298). Our findings provide spatial insight into the current state of TDS as well as the success of management steps taken to manage and prevent eutrophic problems in the LMR.