Using Deep Learning to simulate the urban heat island over Paris

Half of the global population lives in cities and this percentage is expected to increase throughout the 21st century. Therefore, we must learn how climate and climate change affect cities to understand how to properly devise adaptation and mitigation policies. Spatial resolutions of physically-based models are often either too coarse or too expensive to run to obtain high-resolution and high-quality simulations of the urban environment. Deep Learning is a promising computationally efficient technology that has recently been successfully implemented in various applications in Earth Sciences. Here, we present an application of Convolutional Neural Networks (CNNs) to simulate 2-meter temperature (T2m) and Land Surface Temperature (LST) over the city of Paris for the 2004-2022 period. Subsequently, we analyse the CNNs quality in simulating the urban heat island (UHI) over Paris. DL presents substantial improvements over ERA5 (the benchmark of this study) in simulating LST, T2m, and both surface and air UHIs. This study supports the potential of DL as a technology to help improve the simulation of urban climate (namely, temperature extremes).   Acknowledgements: This work was funded by the Portuguese Fundação para a Ciência e a Tecnologia (FCT) I.P./MCTES through national funds (PIDDAC) – UIDB/50019/2020-IDL. Frederico Johannsen was supported by FCT with the doctoral grant UI/BD/151498/2021. Pedro M. M. Soares was supported by a grant through the project “ELABORAÇÃO DA ESTRATÉGIA REGIONAL DE ADAPTAÇÃO ÀS ALTERAÇÕES CLIMÁTICAS DO ALENTEJO – ERAACA”.