Maximum urban heat island intensity: how to account for multi scales effects and spatial interactions in a linear model?

Environmental issues such as human thermal discomfort or air quality are getting particularly important for policy makers and city planners.

However, meteorological variables within the Urban Boundary Layer show a very strong spatial and temporal heterogeneity, and therefore, numerical models aiming to properly predict this variability are often complex and computationally heavy. These characteristics can represent a major barrier to the needed dialogue with policy makers and city planners.

This contribution proposes a model to predict the spatial variability of the night-time Urban Heat Island Intensity (UHII) based on input data that can be simply understood by stakeholders.

Relatively simple methods, combining linear models and variables mainly derived from Local Climate Zones, can approximate the results of computationally heavier numerical models with an acceptable normalized residual mean square error (nrmse) and a good spatial representation.

In order to account for multiple scales of the phenomenon, we built aggregation variables according to wind direction and strength and variables relative to the shape of the urban sprawl. We improved the nrmse compared to [Garde et al 2020] who worked on the same data.

The spatial correlation of the residuals is also explored through spatial regression models.

The loss of precision is counter balanced by the gain in explainability and rapidity of use. Once the parameters of the model are estimated, one can explore the impact of a major urban renovation project with almost no delay as long as the geographical information is available.

To illustrate the method, we based our training and evaluation on numerical simulations performed with the SURFEX/MesoNH models for 42 French cities.