Non-contact hyperspectral monitoring of urban wastewater quality: Optimization of model calibration and performance

Monitoring pollution in urban drainage systems (UDS) is challenging due to their inaccessibility and harsh conditions. In this study, we investigate the use of visible and near-infrared hyperspectral imaging for non-contact monitoring of turbidity, dissolved organic carbon (DOC), and ammonium nitrogen (NH₄-N) in raw urban wastewater. We collected data from five separate locations that included a stormwater basin, foul sewers, and a combined sewer. We trained data-driven models based on the partial least squares method to predict pollutants from hyperspectral spectra using global, local, and hybrid calibration approaches. Local models are trained on data from a single site, global models on combined data from all sites but one, and hybrid models combine global models with a small amount of site-specific data to enhance performance. The efficacy of those methods was robustly evaluated using cross-validation and Mean Absolute Percentage Errors (MAPE) as performance metric. Our results show that local calibration methods with thirty training samples perform best, with cross-validated median MAPEs of 6.5% for turbidity, 14.1% for DOC, and 22.3% for NH4-N. Global models performed satisfactorily only for turbidity, with a median MAPE of 11.8%. Hybrid models with five local samples represent a potential compromise, utilizing limited local samples to enhance global models. We also find that model performance is best for foul sewers, followed by stormwater ponds, while it is worst for more variable combined sewers. We finally showed that model performance for turbidity can be explained by its correlation with reflected light intensity, while performance for DOC can mainly be explained by its correlation with turbidity. For NH4-N, correlations with turbidity play an important role, but other factors that could not be identified improved model performance. Overall, our findings are very promising for the development of precise, cost-effective, and low-maintenance UDS monitoring techniques.