Lake total suspended matter retrieval by wind speed: A machine learning model trained by time-series satellite imagery

This study aims to develop an affordable and continuous method for monitoring water quality in arid, remote areas with high erosion rates. It presents a hypothesis that establishes a link between the concentration of Total Suspended Matter (TSM) and wind speed, emphasizing its ecological importance in lakes with dry conditions and high erosion rates. Building upon this hypothesis, the study introduces Wind2TSM-Net, a machine learning model that effectively bridges between different regression and neural network algorithms. This model connects on -site wind speed measurements with TSM data obtained through a physically remote sensing approach. It accurately predicts TSM concentration values, overcoming challenges such as cloud interference and reducing reliance on satellite imagery. The model was applied to Iran ' s Chah-Nimeh Reservoirs (CNRs) as a case study in an arid and remote area. The results revealed a significant correlation between TSM concentration and wind speed measurements, with impressive performance metrics (coefficient of determination (R 2 ) = 0.88, root mean square error (RMSE) = 1.97 g/m 3 , mean absolute error (MAE) =1.33). These findings highlight the effectiveness of Wind2TSM-Net in monitoring TSM values in remote and dry regions, particularly during extreme weather conditions when on -site measurements are impractical or cloud cover obstructs satellite observations.