https://doi.org/10.1140/epjp/s13360-026-07563-2
Regular Article
SentinelAirNet: a deep learning-based framework for identifying air pollution sources in urban environments using sentinel-5P satellite imagery
1
Department of Computer Science and Engineering, B.M.S. College of Engineering, Bull Temple Road, 560019, Bengaluru, Karnataka, India
2
Department of CSE, BMS Institute of Technology and Management, Bengaluru, Karnataka, India
3
Department of Information Technology, Chaitanya Bharathi Institute of Technology (CBIT), 500075, Gandipet, Hyderabad, India
4
Dept of ECE, Aditya University, 533437, Surampalem, Andhra Pradesh, India
5
Department of CSE (AI&ML), Vaagdevi College of Engineering, Warangal, Telangana, India
a
This email address is being protected from spambots. You need JavaScript enabled to view it.
Received:
2
May
2025
Accepted:
12
March
2026
Published online:
2
April
2026
Abstract
Heavy air pollution is a significant environmental and human health problem in large population centres, where emissions from mobile, point, and residential sources create hazardous atmospheric conditions. Existing air quality monitoring systems, mostly ground-based sensors, fail due to insufficient spatial coverage and real-time adaptability. Innovations in satellite remote sensing and deep learning have enabled widespread, data-driven worldwide pollution monitoring, whilst most existing methodologies suffer from multiple shortcomings. These include insufficient integration of spatial and temporal features, the non-use of multi-pollutant data, and the non-interpretability of model predictions. In addition, the models are developed primarily for either classification or regression but not both, limiting their overall utility in joint pollution scenario analysis. To address these difficulties, this work introduces SentinelAirNet, a novel deep learning-based framework for identifying image-inferred urban pollution sources and estimating their associated concentration levels from Sentinel-5P (TROPOMI) satellite imagery. For this purpose, a hybrid CNN-LSTM architecture is embedded into the framework to learn spatial features and temporal dynamics of different pollutants (i.e. NO2, CO, SO2, and CH4). The model accuracy is improved with a well-structured preprocessing pipeline, multi-channel data fusion, and interpretation of prediction outputs (using Grad-CAM). SentinelAirNet outperforms baseline models with 93.6% accuracy and 93.4% F1-score for classification and a low RMSE of 0.1363 and R2 score of 0.912 for pollutant estimation, as shown in the experimental results. It provides visual heatmaps, source attribution, and spatial overlays to reveal insights for actions. SentinelAirNet is a scalable, interpretable, and high-performance solution for real-time monitoring of urban air quality and supporting environmental decision-making.
Copyright comment Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
© The Author(s), under exclusive licence to Società Italiana di Fisica and Springer-Verlag GmbH Germany, part of Springer Nature 2026
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
