Integrating river discharge and Sentinel-2 satellite imagery for enhanced turbidity mapping in arid region rivers: A machine learning approach

Abstract

This study pioneers a novel mapping application, addressing the state of water quality assessment along Iran's Karun River in order to advance this field. The current work utilizes a state of the art machine learning method that seamlessly combines the inversion of multi-temporal satellite imagery with river discharge data. Focusing on the crucial water resources of arid regions, four machine learning models were developed and compared: These are K-Nearest Neighbors, Random Forest, Gradient Boosting and Multi-Layer Perceptron. The research introduces two key models: (I) R_rs(Tur), which refers to remote sensing reflectance based retrieved turbidity maps, and the more advanced, (II) R_rs&Q(Tur), which denote incorporating both remote sensing and river discharge data. The results show that the R_rs & Q(Tur) model significantly outperforms its counterpart in Root Mean Square Error (≃15% Reduction) and Bias Errors (≃13%). By integrating hydrological and remote sensing data, this study demonstrates how to help environmental monitoring, especially in assessing water quality in arid regions through machine learning approaches.