Machine Learning Model for Prediction of Indicative Water Parameters on the Danube River Based on Satellite Data

This study introduces a novel machine-learning approach using Sentinel-2 satellite images for water quality assessment in the Danube River. Utilizing a deep neural network to build a model that integrates multispectral satellite data with in-situ measurements, the research provides a comprehensive analysis with augmented data. It demonstrates high predictive accuracy for significant water quality indicators for the Danube River. The R2-result exceeds 0.98 for water temperature, electrical conductivity, and dissolved oxygen, while slightly less precision is achieved for chemical oxygen demand. Our method represents a scalable, efficient improvement of traditional assessment techniques, emphasizing the synergy of remote sensing and machine learning. It significantly advances monitoring water quality parameters in hydrology stations on river flows near urban environments with the possibility of implementing it in other locations of interest. In addition, our approach that uses filtering masks is adaptable to different in-land water surface satellite-based precise detections.