High-resolution temperature downscaling for global cities based on satellite imagery, weather station data and NWP model data

The growth of urban areas in combination with an increased number of heatwaves worldwide caused by the anthropogenic climate change can make cities more vulnerable. Increasing number of buildings and sealed surfaces are changing the energy budget in urban areas towards higher longwave radiation fluxes due to the greater heat storage capacity.

Since WMO stations are typically located outside the city, where air temperatures normally are lower than in the city center, initial conditions of NWP models do not accurately represent the air temperatures in urban areas. Hence, NWP models tend to underestimate the air temperature in urban areas since NWP models cannot fully resolve the urban heat island effect. Without any post-processing the MAE is 1.7 K and the MBE is -1 K. This study focuses on an analysis of 17 different European cities in the year 2020. It quantifies the improvement of the statistical downscaling model over an NWP model by a) including dense air temperature measurements in the urban and rural areas, b) including satellite derived variables as model input and c) including both dense air temperature measurements and satellite derived variables.

Dense air temperature networks in cities help to better understand the micro-scale air temperature field in an urban environment. These air temperature data train a statistical high-resolution air temperature downscaling model for urban environments in 10 m horizontal resolution. Including official measurement stations, the statistical model can be transferred to other cities for an operational use to calculate micro-scale air temperatures on an hourly basis. The model is further forced by surface texture parameters from the high-resolution satellites Sentinel-2 and Landsat-8, as well as digital elevation models, and raw model output from meso-scale NWP models.

With a dense air temperature network (a), the urban heat island effect can be resolved, resulting in a reduced bias to almost 0 K. Including satellite derived variables as model input (b) the downscaling approach ensures to decrease the MAE by 0.4 K and to better represent the inner-city temperature variability. To better take dynamic processes into account, the downscaling approach can be extended with a dense measurement network (c) which also further reduces the MAE.

The statistical model approach enables to resolve high-resolution temperature fields in the past, making it possible to calculate high-resolution urban heat island maps. Furthermore, real-time temperature fields can help to significantly enhance the initial conditions for NWP models, thus improving forecast models in urban areas. A statistical downscaling of the numerical weather forecast can help decision-makers improving the heat wave management in cities.