A Comprehensive Assessment of Spatiotemporal Dynamics of Urban NO2 using Sentinel-5P with Evaluation of the Predictive Capability of Sentinel-2 for Pollution Estimation

Urban NO₂ pollution poses major challenges to public health and environmental sustainability, particularly in rapidly expanding cities like Kanpur. Despite advances in remote sensing, limited attention has been given to the integration of high-resolution multispectral satellite data with advanced machine learning models for fine-scale NO₂ monitoring. This study addresses this gap by integrating Sentinel-2 spectral reflectance and Sentinel-5P NO₂ datasets through a Random Forest regression model to examine the spatial and temporal variability of NO₂ concentrations in Kanpur during 2024. The key objectives include identifying dominant temporal cycles, assessing spectral drivers, and evaluating model performance. Temporal analysis revealed distinct seasonal variation, with peak NO₂ in November and lowest levels in August, influenced by meteorological and anthropogenic factors. The 7th-degree polynomial model effectively captured these seasonal trends with best tradeoff between accuracy and complexity, while frequency-domain analysis detected dominant semi-annual and annual cycles. The Random Forest model achieved optimal performance with 50 trees with R 2 = 0.5442 and error of 7.2%). Spectral correlations indicated strong negative associations between NO₂ and Red Edge/NIR bands, underscoring vegetation’s mitigating role. In summary, this study demonstrates the effectiveness of integrating multi-sensor satellite data and machine learning for urban air quality estimation and offers a scalable, low-cost approach for data-scarce regions, supporting evidence-based pollution control and health planning.