INTEGRATING HIGH RESOLUTION SATELLITE IMAGERY FOR WASTEWATER QUALITY ASSESSMENT : DEVELOPMENT OF A REGRESSION MODEL AND A SPECTRAL INDEX FOR RESOURCE-LIMITED WASTEWATER TREATMENT PLANTS

Sammanfattning

The limited dissemination of knowledge and the lack of access to advanced laboratory instruments and monitoring systems, particularly in remote regions and developing countries, contribute to the continued challenges in achieving the SDGs. This degree project performed wastewater parameter estimation from high-resolution WorldView Legion 05 satellite imagery, combined with statistical analysis approaches such as Partial Least Squares Regression (PLSR). Both PLSR and remote sensing have been widely used in earlier studies, however, a gap remains in wastewater applications, especially for wastewater still undergoing treatment, due to the high complexity and pollution gradients involved. As an exploratory data analysis step, Principal Component Analysis (PCA) and spectral signatures were calculated and analyzed to improve understanding of wastewater behavior when interacting with electromagnetic radiation. Chemical Oxygen Demand (COD) and Total Suspended Solids (TSS) samples from a local laboratory for two Wastewater Treatment Plants (WWTPs) in Portugal were then used for model development and the proposal of a band-ratio index. The images showed distinct but consistent behavior across multispectral bands, with low reflectance values in Band 1 (Coastal Blue) and Band 2 (Blue), followed by a gradual increase toward Band 6 (Red Edge 1), and peak reflectance in Band 7 (Red Edge 2). Key challenges included limited dataset size, data availability, atmospheric and operational conditions during acquisition, and the novelty of the Legion 05 satellite (launched in 2025). This resulted in good predictive accuracies with LOOCV of R² = 0.72, Q² = 0.59 and RMSE = 18.03 for TSS (Improved Model 5), and excellent performance with R² = 0.98, Q² = 0.96 and RMSE = 22.26 for COD (Improved Model 6). A comparison between PLSR and index-based approaches shows that both methods effectively capture bulk water conditions and represent the expected behavior in different treatment stages. However, the index-based approach is more sensitive to spectral variability and non-water objects, while PLSR provides more stable and robust estimations, particularly in complex scenes. Overall, the results highlight the importance of considering operational conditions and contextual factors within wastewater treatment plants, as these strongly influence spectral response and model performance. The findings support the potential of multispectral remote sensing as a low-cost complementary tool for wastewater monitoring in complex and resource-limited environments, while also emphasizing the need for further data and methodological refinement to improve reliability and practical application.