PollutionNet: A Vision Transformer Framework for Climatological Assessment of NO$_2$ and SO$_2$ Using Satellite-Ground Data Fusion

Prasanjit Dey; Soumyabrata Dev; Bianca Schoen-Phelan

cs.CVMar 31, 2026

PollutionNet: A Vision Transformer Framework for Climatological Assessment of NO $_2$ and SO $_2$ Using Satellite-Ground Data Fusion

Authors: Prasanjit Dey, Soumyabrata Dev, Bianca Schoen-Phelan

Organizations: School of Computer Science, Technological University Dublin, Ireland. · ADAPT Research Ireland Centre, Ireland. · School of Computer Science, University College Dublin, Ireland.

Abstract

Accurate assessment of atmospheric nitrogen dioxide (NO $_2$ ) and sulfur dioxide (SO $_2$ ) is essential for understanding climate-air quality interactions, supporting environmental policy, and protecting public health. Traditional monitoring approaches face limitations: satellite observations provide broad spatial coverage but suffer from data gaps, while ground-based sensors offer high temporal resolution but limited spatial extent. To address these challenges, we propose PollutionNet, a Vision Transformer-based framework that integrates Sentinel-5P TROPOMI vertical column density (VCD) data with ground-level observations. By leveraging self-attention mechanisms, PollutionNet captures complex spatiotemporal dependencies that are often missed by conventional CNN and RNN models. Applied to Ireland (2020-2021), our case study demonstrates that PollutionNet achieves state-of-the-art performance (RMSE: 6.89 $μ$ g/m $^3$ for NO $_2$ , 4.49 $μ$ g/m $^3$ for SO $_2$ ), reducing prediction errors by up to 14% compared to baseline models. Beyond accuracy gains, PollutionNet provides a scalable and data-efficient tool for applied climatology, enabling robust pollution assessments in regions with sparse monitoring networks. These results highlight the potential of advanced machine learning approaches to enhance climate-related air quality research, inform environmental management, and support sustainable policy decisions.

Figures & tables

Figure 1: Study region for satellite and ground observations of NO 2 and SO 2 concentrations. Grid cells represent the spatial domains for each pollutant.

Figure 2: The Proposed frameworks of the PollutionNet for estimating and prediction of NO 2 , and SO 2 concentration.

Figure 3: End-to-end spatial-temporal fusion workflow showing: (1) input data acquisition from TROPOMI satellite and ground stations, (2) quality control and preprocessing, (3) core fusion algorithm execution, and (4) gap-filled output generation.

Figure 4: Technical implementation of the fusion algorithm showing: (a) neighborhood selection criteria based on spatial similarity thresholds, (b) weight allocation methodology, and (c) reconstruction results for complete and partial gap scenarios.

Figure 5: Architecture of the Vision Transformer (ViT) for NO 2 and SO 2 concentration prediction, showing: (1) patch embedding and positional encoding, (2) multi-head self-attention layers, and (3) MLP projection head.

Parameter	PollutionNet	CNN	LR	XGB	LGBM
Epochs	30	30	–	–	–
Learning Rate	0.01	0.01	–	0.1	0.1
Optimizer	Adam	Adam	–	–	–
Activation	GELU	ReLU	–	–	–
Batch Size	8	8	–	–	–
Patch Size	16	–	–	–	–

Table 1: Optimal hyperparameter configurations for all models

Panel of Figure 6 — Figure 6: Depicting the daily average of ground-level concentration vs. the predicted NO 2 concentration from various models: PollutionNet, CNN, linear regression, XGBoost, and LGBM.

Panel of Figure 7 — Figure 7: Depicting the daily average of ground-level concentration vs. the predicted SO 2 concentration from various models: PollutionNet, CNN, linear regression, XGBoost, and LGBM.

	Models	Proposed		CNN		Linear		XGB		LGBM
		NO 2	SO 2	NO 2	SO 2	NO 2	SO 2	NO 2	SO 2	NO 2	SO 2
RMSE	Fold1	7.86	4.91	8.38	6.33	8.95	6.83	8.98	6.84	8.97	6.23
	Fold2	6.51	4.34	6.97	5.54	7.55	6.24	7.50	6.24	7.74	6.82
	Fold3	7.09	4.71	7.59	5.13	8.22	6.68	8.11	6.72	8.10	6.70
	Fold4	6.45	4.92	8.00	5.41	7.62	6.79	7.59	6.83	7.59	6.81
	Fold5	6.56	3.57	7.30	3.86	7.63	5.44	7.60	5.42	7.59	5.41

Table 2: Performance comparison of PollutionNet with CNN, Linear Regression, XGBoost, and LGBM using RMSE, MAE, and Pearson’s Correlation Coefficient (R 2 ).

Figure 8: Daily time-series (March to April 2021) heatmap of PollutionNet for predicting NO 2 concentrations over the Ireland region.

Panel of Figure 9 — Figure 9: Illustrating the intercomparison of the PollutionNet framework with CNN, linear regression, XGBoost, and LGBM models for NO 2 and SO 2 concentrations. The first row represents the comparison of NO 2 , and the last row represents the comparison of SO 2 . The color gradient, ranging from deep blue to light blue, represents the density of data points. Additionally, the outer line plot denotes the density area.

Panel of Figure 10 — Figure 10: Visualization of RMSE scores for various percentages of dataset sizes for (a) NO 2 , and (b) SO 2 .

Literature	Region	Period	Satellite Data	Res.	Model	RMSE ( $\mu g/m^{3}$ )
Li et al. (2019)	China	2014–2015	SO 2 from OMI	0.25°	RF-STK	SO 2 : 10.36
Zhang et al. (2020)	E. China	2014	NO 2 , CH 2 O from OMI	0.25°	GWR	–
Wang et al. (2021)	China	2018–2020	S5P-TROPOMI (O 3 , NO 2 )	0.05°–0.07°	LightGBM	NO 2 : 8.44, O 3 : 17.7
Wei et al. (2023)	China	2013–2020	NO 2 from OMI	0.25°	Decision Tree	NO 2 : 11.5
Xu et al. (2023)	BTH, China	2014–2019	NO 2 from OMI	0.25°	–	Avg NO 2 : 13.3
Zhu et al. (2023)	China, EU, USA	2012–2021	TROPOMI O 3	0.1°	Deep Forest	O 3 : 19.6

Table 3: Comparative analysis of PollutionNet with recent studies showing region, period, satellite data, resolution, model, and RMSE (in

\mu g/m^{3}

Air Quality Prediction

Air quality prediction estimates future pollutant levels and maps local exposure to support health warnings, planning, and pollution control.

See all 42 papers

Tracking Urban Atmospheric Pollutants using Sentinel-5P Satellite Data
May 21, 2026Alice Gomez-Cantos, Henry O. Velesaca
Facultad de Ciencias Naturales y Matemáticas, Escuela Superior Politécnica del Litoral, ESPOL, Campus Gustavo Galindo, Km. 30.5 Vía Perimetral, Guayaquil, 090902, Ecuador · Software Engineering Department, Research Center for Information and Communication Technologies (CITIC-UGR), University of Granada, 18071, Granada, Spain
TinyNina: A Resource-Efficient Edge-AI Framework for Sustainable Air Quality Monitoring via Intra-Image Satellite Super-Resolution
Apr 6, 2026Prasanjit Dey, Zachary Yahn, Bianca Schoen-Phelan +1Autoregressive Image Super-Resolution Air Quality Prediction
ADAPT Research Centre, Dublin, Ireland · School of Computer Science, Technological University Dublin, Ireland · School of Computer Science, University College Dublin, Ireland +1
AirMPA: A Meteorology-to-Pollution Adapter for Global Air Quality Forecasting
Oct 19, 2026Yiheng Wang, Kai Zheng, Yuetan Lin +5Air Quality Prediction Spatiotemporal Forecasting
Fudan University · Shanghai Academy of AI for Science · University of Shanghai for Science and Technology

May 21, 2026J/K move · Enter open · S save

Tracking Urban Atmospheric Pollutants using Sentinel-5P Satellite Data

Alice Gomez-Cantos, Henry O. Velesaca

Facultad de Ciencias Naturales y Matemáticas, Escuela Superior Politécnica del Litoral, ESPOL, Campus Gustavo Galindo, Km. 30.5 Vía Perimetral, Guayaquil, 090902, Ecuador · Software Engineering Department, Research Center for Information and Communication Technologies (CITIC-UGR), University of Granada, 18071, Granada, Spain

Open paper AskPDF arXiv

Urban nitrogen dioxide (

NO_2

) is a key indicator of combustion-related air pollution and exhibits strong spatial and temporal variability in cities. This study presents a satellite-based framework for tracking urban

NO_2

pollution using tropospheric column observations from Sentinel-5P/TROPOMI over Guayas Province, Ecuador. Rather than estimating surface concentrations, the methodology emphasizes robust distributional metrics, including the median and upper-tail percentiles (

P_{90}

P_{95}

, and

P_{99}

), to characterize background conditions and localized pollution extremes at the canton scale. Multi-year satellite observations are aggregated annually and analyzed using unsupervised K-means clustering to identify characteristic pollution regimes without predefined thresholds. Results show that highly urbanized cantons consistently exhibit elevated extreme

NO_2

values and greater variability, while less urbanized areas display lower and more homogeneous patterns. The proposed approach provides an interpretable and scalable tool for urban air-quality assessment in data-scarce regions using satellite observations alone. The implementation is publicly available on GitHub https://hvelesaca.github.io/sentinel-5P-clustering/.

Abstract

Air Quality Prediction

Explore similar work

Tracking Urban Atmospheric Pollutants using Sentinel-5P Satellite Data

TinyNina: A Resource-Efficient Edge-AI Framework for Sustainable Air Quality Monitoring via Intra-Image Satellite Super-Resolution

AirMPA: A Meteorology-to-Pollution Adapter for Global Air Quality Forecasting