Physics-Informed AI for Enhanced Climate Downscaling and Extreme Event Prediction in the Energy Sector

Climate change is a pressing reality, with increasing extreme weather events such as droughts, floods, and heatwaves causing significant damages and casualties [1]. To mitigate these impacts, climate neutrality has become a global priority. This requires a transition to renewable energy, necessitating accurate weather and climate models [2]. High-resolution Regional Climate Models (RCMs) offer detailed projections but are computationally expensive. Statistical downscaling techniques provide a more efficient alternative but have limitations, such as the inability to capture relevant climate change signals and underestimating extreme events. To address these issues, we propose a physics-informed artificial intelligence (AI) model bridging the gap between data-driven and model-driven approaches, by incorporating known physical principles and domain knowledge into the learning and prediction process [3,4].

In this research, we focus on developing a physics-informed AI model for efficient downscaling of climate and weather data, enabling high-resolution projections that enhance renewable energy predictions. More specifically, we aim for improving downscaling techniques, reducing uncertainties, and accurately representing extreme weather events. Key research questions include identifying suitable datasets for downscaling, evaluating errors, and improving multivariate downscaling from coarse (100 km for GCM, 10 km for RCM) to high resolutions (5 km to 1 km). Our developed method is compared against dynamical downscaling techniques across different temporal and spatial resolutions. This research aims to advance climate and weather predictions for impact sectors in need of very high spatial resolutions through providing an efficient and fast AI-based downscaling method, particularly for renewable energy applications, aiming at supporting decision-making and adaptation strategies in the face of climate change.

References

[1] IPCC, 2021 IPCC, 2021: Climate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change [Masson-Delmotte, V., P. Zhai, A. Pirani, S.L. Connors, C. Péan, S. Berger, N. Caud, Y. Chen, L. Goldfarb, M.I. Gomis, M. Huang, K. Leitzell, E. Lonnoy, J.B.R. Matthews, T.K. Maycock, T. Waterfield, O. Yelekçi, R. Yu, and B. Zhou (eds.)]. Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, In press, doi:10.1017/9781009157896.

[2] Schaeffer, Roberto, Alexandre Salem Szklo, André Frossard Pereira De Lucena, Bruno Soares Moreira Cesar Borba, Larissa Pinheiro Pupo Nogueira, Fernanda Pereira Fleming, Alberto Troccoli, Mike Harrison, and Mohammed Sadeck Boulahya. 2012. ‘Energy Sector Vulnerability to Climate Change: A Review’. Energy 38 (1): 1–12. https://doi.org/10.1016/j.energy.2011.11.056.

[3] Harder, Paula, Alex Hernandez-Garcia, Venkatesh Ramesh, Qidong Yang, Prasanna Sattigeri, Daniela Szwarcman, Campbell Watson, and David Rolnick. 2022. ‘Hard-Constrained Deep Learning for Climate Downscaling’. arXiv. https://doi.org/10.48550/ARXIV.2208.05424.

[4] Karniadakis, George Em, Ioannis G. Kevrekidis, Lu Lu, Paris Perdikaris, Sifan Wang, and Liu Yang. 2021. ‘Physics-Informed Machine Learning’. Nature Reviews Physics 3 (6): 422–40. https://doi.org/10.1038/s42254-021-00314-5.