Towards a satellite based long-term monitoring and prediction of urban heat stress

Climate change is causing temperatures around the globe to rise, leading to an increase in hot days and nights and a rise in the frequency and intensity of heat waves. High temperatures represent one of the key factors causing heat stress, which is affecting human health. Examples include the mega heat wave in August 2003 that took more than 70,000 lives across Europe or the unprecedented heat in 2018 with more than 100,000 heat-related deaths in the EU. Spatial exposure to heat stress varies, with urban areas usually heating up much stronger than their rural surroundings (urban heat island effect). With progressing global warming, the importance of spatially monitoring and predicting urban heat stress over large areas to prevent heat-associated morbidity and mortality is thus rising. For this purpose, thermal data obtained by satellites may represent a valuable alternative or addition to in-situ observations and climate modelling, which both show several advantages, but also shortcomings in terms of spatial coverage (in-situ), data needs (climate modelling) and costs.

One of the main challenges for the use of satellite thermal data for urban heat stress monitoring is to convert the obtained land surface temperature (LST) into air temperature (AT), since satellites only provide information on the former whereas the latter is needed as input for most heat stress indicators. In our presentation, we will address this challenge with emphasis on urban regions. Two main strands of methods are used in the literature for converting LST into AT. First, data driven methods that use the empirical relationship between in-situ observations of AT from weather stations and LST data from satellites. These methods usually work well for individual stations, but the transferability of the relationship to different locations is typically limited. In addition, especially for methods based on machine learning techniques, a significant amount of data is needed for model calibration. The second type of methods comprises physical models. They typically make use of the energy balance to estimate AT from LST. These models show the advantage of better transferability but need additional input data besides LST. Moreover, physical methods are often designed for applications in rural areas, which may differ from the situation in cities. We apply different methods to derive AT from LST in urban regions and discuss their suitability to monitor and predict urban heat stress based on satellite thermal data.