Spatial downscaling of urban temperatures: evaluation of an approach using satellite and reanalysis data

Climate change is expected to exacerbate heat stress, particularly in urban areas where the urban heat island (UHI) effect tends to amplify warming compared to surrounding rural regions. Due to the heterogeneity of urban environments, heat stress can vary significantly within cities. Heat vulnerability maps, which combine data on heat sensitivity, heat exposure and adaptive capacity, are valuable tools for identifying areas that should be prioritized for heat stress mitigation measures. One important component of such heat vulnerability maps is data on the spatial distribution of heat. The present study explores the use of satellite data to generate high-resolution temperature maps, addressing two key challenges in the process.

The first challenge arises from the fact that satellites measure land surface temperature (LST) rather than air temperature (AT), whereas the latter is needed as input for most heat stress indicators. While linear models calibrated with weather station data are frequently used to estimate AT from LST, there are cities where the availability of weather stations is insufficient for calibrating models with multiple control variables. Additionally, the LST-AT relationship depends on the prevailing atmospheric conditions. The second challenge of using satellite data is that satellite images are usually not available on an hourly or daily basis due to factors such as satellite scheduling or excessive cloud cover.

To address the first challenge, we adopt a technique introduced by the ECOSTRESS mission, which leverages reanalysis data (GEOS-5) to estimate AT using LST, the normalized difference vegetation index (NDVI), and albedo. We apply this method to spatially downscaled LST data (100m) from the VIIRS instrument aboard the Suomi NPP satellite, AT reanalysis data from ERA5-Land (9km), as well as NDVI and albedo derived from Harmonized Landsat Sentinel (HLS) data (aggregated to 100m). Applying the method to individual satellite images enables day-specific adjustments for varying atmospheric conditions. To overcome the second challenge, we utilize high-resolution AT maps derived from LST images to calculate spatial patterns of air temperature distribution, which are then used to downscale ERA5-Land AT data for those times without satellite images available.

To evaluate the approach described, it is exemplary applied to various cities, whereby the downscaled temperature estimates are validated against (i) temperature estimates based on alternative methods than the ECOSTRESS technique to derive AT from LST, (ii) weather station data, and (iii) existing results from urban climate models.