Physics-Informed Graph Neural Networks for Air Pollution Forecasting in the Netherlands

Nikolas Assiotis, Rachel Hau, Valentijn Oldenburg, Rik Verbiest, Julian Koellermeier, Matthia Sabatelli, Juan Cardenas-Cartagena
Proceedings of the 2nd ECAI Workshop on "Machine Learning Meets Differential Equations: From Theory to Applications", PMLR 277:47-70, 2025.

Abstract

Accurate air pollution forecasting is critical for public health and environmental policy, particularly in densely populated regions like the Netherlands. This work introduces a physics-informed graph neural network (PI-GNN) framework for urban nitrogen dioxide (NO2) forecasting, which integrates domain-specific physical constraints into graph-based deep learning models. By combining spatial and temporal learning with physical knowledge, the proposed physics-informed graph convolutional network with gated recurrent units significantly outperforms purely data-driven recurrent and graph neural networks in terms of accuracy, generalizability, and environmental efficiency. Moreover, physics-informed models demonstrated progressively better relative performance over purely data-driven models in conditions with scarce data.

Cite this Paper

BibTeX
@InProceedings{pmlr-v277-assiotis25a, title = {Physics-Informed Graph Neural Networks for Air Pollution Forecasting in the Netherlands}, author = {Assiotis, Nikolas and Hau, Rachel and Oldenburg, Valentijn and Verbiest, Rik and Koellermeier, Julian and Sabatelli, Matthia and Cardenas-Cartagena, Juan}, booktitle = {Proceedings of the 2nd ECAI Workshop on "Machine Learning Meets Differential Equations: From Theory to Applications"}, pages = {47--70}, year = {2025}, editor = {Coelho, Cecı́lia and Zimmering, Bernd and Costa, M. Fernanda P. and Ferrás, Luı́s L. and Niggemann, Oliver}, volume = {277}, series = {Proceedings of Machine Learning Research}, month = {26 Oct}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v277/main/assets/assiotis25a/assiotis25a.pdf}, url = {https://proceedings.mlr.press/v277/assiotis25a.html}, abstract = {Accurate air pollution forecasting is critical for public health and environmental policy, particularly in densely populated regions like the Netherlands. This work introduces a physics-informed graph neural network (PI-GNN) framework for urban nitrogen dioxide (NO2) forecasting, which integrates domain-specific physical constraints into graph-based deep learning models. By combining spatial and temporal learning with physical knowledge, the proposed physics-informed graph convolutional network with gated recurrent units significantly outperforms purely data-driven recurrent and graph neural networks in terms of accuracy, generalizability, and environmental efficiency. Moreover, physics-informed models demonstrated progressively better relative performance over purely data-driven models in conditions with scarce data.} }
Endnote
%0 Conference Paper %T Physics-Informed Graph Neural Networks for Air Pollution Forecasting in the Netherlands %A Nikolas Assiotis %A Rachel Hau %A Valentijn Oldenburg %A Rik Verbiest %A Julian Koellermeier %A Matthia Sabatelli %A Juan Cardenas-Cartagena %B Proceedings of the 2nd ECAI Workshop on "Machine Learning Meets Differential Equations: From Theory to Applications" %C Proceedings of Machine Learning Research %D 2025 %E Cecı́lia Coelho %E Bernd Zimmering %E M. Fernanda P. Costa %E Luı́s L. Ferrás %E Oliver Niggemann %F pmlr-v277-assiotis25a %I PMLR %P 47--70 %U https://proceedings.mlr.press/v277/assiotis25a.html %V 277 %X Accurate air pollution forecasting is critical for public health and environmental policy, particularly in densely populated regions like the Netherlands. This work introduces a physics-informed graph neural network (PI-GNN) framework for urban nitrogen dioxide (NO2) forecasting, which integrates domain-specific physical constraints into graph-based deep learning models. By combining spatial and temporal learning with physical knowledge, the proposed physics-informed graph convolutional network with gated recurrent units significantly outperforms purely data-driven recurrent and graph neural networks in terms of accuracy, generalizability, and environmental efficiency. Moreover, physics-informed models demonstrated progressively better relative performance over purely data-driven models in conditions with scarce data.
APA
Assiotis, N., Hau, R., Oldenburg, V., Verbiest, R., Koellermeier, J., Sabatelli, M. & Cardenas-Cartagena, J.. (2025). Physics-Informed Graph Neural Networks for Air Pollution Forecasting in the Netherlands. Proceedings of the 2nd ECAI Workshop on "Machine Learning Meets Differential Equations: From Theory to Applications", in Proceedings of Machine Learning Research 277:47-70 Available from https://proceedings.mlr.press/v277/assiotis25a.html.

Related Material