GeoPython 2020

A reusable computational workflow to assess urban heat islands in Python
2020-09-22, 09:45–10:15, Room 2

We show how remote sensing data, air temperature measurements and cadastral datasets can be used to simulate urban heat islands with the urban cooling model of the InVEST software platform.

The global increase of urban populations brings new challenges to urban planning and policy making. In the context of climate change, urban heat mitigation is a major priority for many cities that are exposed to rising temperatures and heat waves. To that end, it is central to understand the relationship between the spatial pattern of urban land cover and the intensity and distribution of urban heat islands.

While most research suggests a significant relationships between the spatial composition and configuration of urban tree canopy and the distribution of land surface temperature, many contradicting results regarding the magnitude and direction of such correlations have been reported. Overall, two major weaknesses might be noted in such approaches. On the one hand, land surface temperature cannot fully represent human exposure to heat, which is better represented by air temperature, e.g., at 2-m above the surface. On the other hand, and most importantly, even when accounting for spatial auto-correlation, statistical approaches are hard to relate to the physical mechanisms driving urban heat islands and therefore can hardly enlighten the understanding of the phenomenon.

The aim of this presentation is to propose a reusable computational workflow to model urban heat islands that addresses the shortcomings reviewed above. The approach is based on the urban cooling model of the InVEST software platform, which models the urban air temperature as a function of three main physical mechanisms, namely the shade, evapotranspiration and albedo. Four main inputs are required, which are a categorical land cover raster map, a biophysical table containing model information of each LULC class of the map, a high-resolution tree canopy map, and a reference evapotranspiration raster. The proposed computational workflow consists of three steps. Firstly, a supervised learning approach combines openly available satellite data and air temperature measurements from monitoring stations in order to downscale the latter and obtain a map of air temperature at an appropriate resolution. Such map will then serve two purposes, i.e., to compute the input reference evapotranspiration raster, and to calibrate the urban cooling model. Secondly, in order to better represent for the spatial heterogeneity of urban landscapes, the land cover raster map is coupled with the tree canopy map to refine the land cover classes depending on their proportion of tree cover. For instance, pixels of a sidewalk'' class could be reclassified to further distinguish pixels of asidewalk with high tree cover'' and ``sidewalk with low tree cover''. The reclassification will be accounted for in the biophysical table so that pixels with high tree cover have greater shading coefficients. Finally, the preprocessed data will be fed into the urban cooling model to achieve a distribution of air temperature that mimics the patterns observed empirically.

An example application to the study of urban heat islands in the Swiss urban agglomeration of Lausanne will be presented in order to show how the computational workflow might be applied in realistic settings. Additionally, the proposed approach aims at providing a reusable method with moderate data requirements that can be applied to very distinct urban areas, and might thus provide new insights into many of the contradicting results encountered in the literature.