High-resolution air temperature forecast for urban heat wave management

Urban areas are becoming more vulnerable with their growth and an increasing number of heatwaves caused by anthropogenic climate change. Differences in the surface structure, such as between green spaces and sealed surfaces, cause differences in surface energy budget due to the different heat storage capacity, resulting in a high inner-city air temperature variability.

Initial conditions of NWP models do not accurately represent air temperatures in urban areas, since official (WMO) ground weather stations are typically located in city outskirts where air temperatures are lower than in the city center. Hence, the urban heat island effect is not fully resolved in NWP models which results in an underestimation of urban air temperatures. Since the urban air temperature variability occurs on a micro-scale, state-of-the art NWP models, with resolution in the order of a few km, are not able to resolve the air temperature field.

This study focuses on the development of the meteoblue city-climate model (mCCM), a dynamic statistical downscaling model which was developed to resolve the urban heat island effect and the small-scale air temperature variability in urban areas.

The mCCM is based on a high-resolution model domain with 10 m horizontal resolution and trained on air temperature data from measurement networks. These networks are operated by the municipality or based on crowdsourcing and help understanding the air temperature variability and dynamics in an urban environment. To resolve the differences in the surface energy budget, surface texture parameters are derived from the polar-orbiting, high-resolution satellites Sentinel-2 and Landsat 8. Further improvement is achieved with the parameterization of the coefficients of the statistical model, which depend on the prevailing meteorological conditions. The trained model is then fully driven by meso-scale NWP models and thus independent of any air temperature measurements, which allows a direct forecast of high-resolution air temperatures in urban environments.

Through statistical downscaling, the mCCM can resolve the urban heat island effect and the inner-city air temperature variability, reducing the bias to almost 0 K. Implementing the parameterization, the model considers dynamic processes which makes the model sensitive to different meteorological conditions and further reduces the hourly MAE by at least 25 % compared to NWP models.

The mCCM allows more reliable and location-specific air temperature forecasts in cities for the upcoming six days for areas where vulnerable groups live, such as hospitals, kindergartens, or schools. The approach helps decision-makers to improve the heat wave management in their cities and allows data-driven decisions.