Predictive models of non-optically active coastal water quality parameters by remote sensing imagery

The coastal waters of the Gaza Strip are significantly impacted by wastewater discharge and pollution from commercial and domestic activities. Timely monitoring and accurate assessment of water quality are crucial for detecting contamination. This study established correlations between spectral reflectance and non-optically active pollutants, specifically Dissolved Oxygen (DO), Total Phosphorus (TP), and Molybdenum (Mo). These parameters were selected for their relevance to wastewater pollution and compatibility with the employed measurement techniques. Non-optically active constituents, which do not exhibit distinct spectral signatures, may still be detectable under specific conditions. The study focused on testing the feasibility of retrieving concentrations of non-optically active components in seawater using Sentinel-2 MSI imagery. Sentinel-2 MSI was chosen for its high revisit frequency and broad wavelength range, making it suitable for assessing water quality. Empirical multiple regression models revealed varying performance among the pollutants. The TP model showed poor correlation and high uncertainty. In contrast, the predictive models for DO and Mo yielded more promising results. The DO model exhibited strong performance, achieving a significant coefficient of determination (R²) value of 0.73, with a low prediction error: a Root Mean Square Error (RMSE) of 0.21 % and a Mean Absolute Percentage Error (MAPE) of 6.6 %. The Mo model demonstrated moderate accuracy, with an R² value of 0.51, an RMSE of 0.35 %, and a higher MAPE of 171 %. The results indicated that DO and Mo concentrations correlated with specific Sentinel-2 bands. This study confirms that remote sensing can effectively retrieve concentrations of non-optically active pollutants, supporting rapid seawater quality assessments, particularly in regions where field surveys are challenging. Highlights Empirical models predict dissolved oxygen and molybdenum concentrations Study area affected by wastewater discharge and coastal pollution Sentinel-2 satellite data used for non-optically active elements prediction DO model shows high accuracy for hypoxic conditions detection Molybdenum model shows moderate accuracy, indicating prediction challenges