Spatial Estimation of Ground-Level Temperature for Climate-Sensitive Urban Mobility using Image-to-Image Deep Neural Networks

Global warming is reflected by the increase in air temperature at ground level, among other factors. This increase in temperatures is more pressing in urban environments, due to the phenomenon known as Urban Heat Island (UHI). This phenomenon consists of temperatures in urban environments being higher than those in rural areas, which can be due, among other factors, to urban morphology and activities (traffic, air conditioning). UHI poses a risk to people and affects habits of urban life, such as mobility. This is why estimating air tem-peratures at 2 meters above ground level with a street spatial resolution can help urban planners make better decisions to achieve less thermally stressed urban areas. This paper presents the results of a preliminary study aimed to explore the use of image-to-image deep neural networks to estimate the pedestrian level air temperature in urban areas. Specifically, we propose a U-Net architecture fed with meteorological variables to produce, at its output, a estimation of the spatial distribution of the target variable. Results over data belonging to 4 major European cities show that with a suitable methodology implementation and databases, Deep Learning can be very convenient and efficient when estimating the pedestrian level air temperature, highlighting its potential for climate change adaptation of urban mobility.