Deep Learning-Based Super-Resolution of Remote Sensing Images for Enhanced Groundwater Quality Assessment and Environmental Monitoring in Urban Areas

This study presents a novel deep learning-based super-resolution framework for enhancing remote sensing imagery to assess groundwater quality and environmental conditions in Lahore, Pakistan. We developed a convolutional neural network architecture that upscales low-resolution satellite imagery to generate high-resolution (0.5 m) outputs, achieving a peak signal-to-noise ratio of 32.4 dB and structural similarity index of 0.91. The enhanced imagery enabled precise delineation of urban features and environmental parameters affecting groundwater quality. Using the super-resolved images alongside traditional water quality parameters (pH, hardness, TDS) analyzed through fuzzy analytic hierarchy process, we calculated the groundwater quality index (GWQI) for 33 areas across four years (2008–2020). Results showed most areas achieved “Better water” quality status by 2020, though two regions (Old City and Anarkali) were classified as “Poor water” quality. We observed a moderate negative correlation (r = -0.62, p < 0.001) between GWQI and static water level depth, with significant depth increases in areas such as Dholanwal (37.1 m), Ichhra (47.06 m), and Township (49.35 m) by 2020. The integration of super-resolution remote sensing with conventional water quality assessment demonstrates promising applications for urban environmental monitoring and groundwater resource management.