A 2-m temperature deep learning emulator of the UrbClim model for European cities

The average air temperature in urban areas is usually elevated relative to the surrounding rural areas, commonly referred to as the Urban Heat Island (UHI) effect, leading to increased heat exposure and associated health risks. Under ongoing climate change and with a large and increasing share of the global population living in cities, the study of urban temperature is increasingly important as it provides a basis to study associated health risks and urban adaptation strategies.

Hectometre-scale heat risk information is crucial to implement effective measures to protect vulnerable groups in exposed neighbourhoods. Physics-based urban climate models provide this information, but their substantial computational costs limit their spatiotemporal coverage and practical usability. By leveraging deep learning, we provide rapid surrogates that deliver high-resolution temperature information at low computational cost.

We present EU-HEAT (European Urban High-resolution Emulator for Air Temperature), a machine-learning emulator of the UrbClim model to enable rapid inference of 2-m air temperature (t2m) fields at 100 m resolution across European cities. UrbClim is an urban-slab model that downscales t2m using large-scale meteorological forcing and surface representation data [1]. EU-HEAT v1 is a U-Net model trained on a dataset covering 100 European cities, generated with UrbClim [2]. Since we focus on emulating UrbClim, input features are selected to closely mirror the drivers of the physics-based model.

We will present a qualitative and quantitative validation of EU-HEAT v1 for representative hold-out cities. These results will be compared with scores obtained from the European Random Forest Urban Climate Emulator (Eu-RaFUCE), which was trained on the same UrbClim dataset [3]. 

 

References

[1] De Ridder, K., Lauwaet, D. and Maiheu, B. (2015), ‘UrbClim – A fast urban boundary layer climate model’, Urban Climate, Vol. 12, pp. 21–48, https://doi.org/10.1016/j.uclim.2015.01.001.

[2] Lauwaet, D., Berckmans, J., Hooyberghs, H., Wouters, H., Driesen, G. et al. (2024), ‘High resolution modelling of the urban heat island of 100 European cities’, Urban Climate, Vol. 54, p. 101850, https://doi.org/10.1016/j.uclim.2024.101850.

[3] Top, S., Blancke, J., Covaci, A., Caluwaerts, S., Hamdi, R. et al. (2025), ‘Emulation of a numerical urban model to create high-resolution near surface air temperature over European cities.’, ESS Open Archive, https://doi.org/10.22541/essoar.173671231.11835703/v1.