Statistical-dynamical downscaling of the urban heat island in Hamburg, Germany

Abstract

Regional climate models provide climate projections on a horizontal resolution in the order of 10 km. This is too coarse to sufficiently simulate urban climate related phenomena such as the urban heat island (UHI). Therefore, regional climate projections need to be downscaled. A statistical-dynamical method for the UHI was developed and applied to provide urban climate results at a high resolution with little computational costs. For the downscaling, weather situations relevant for the UHI are determined. This is done by combining objective weather pattern classification based on a k-means cluster analysis of ERA-40 reanalysis data and a regression-based statistical model of the observed UHI of Hamburg. The resulting days for each weather pattern are simulated with the mesoscale meteorological model METRAS at 1 km horizontal resolution. To obtain the average UHI for a climate period, the mesoscale model results are statistically recombined weighted by the frequency of the corresponding weather patterns. This is done for present-day climate (1971-2000) using reanalysis data to yield the current climate UHI. For the future climate periods 2036-2065 and 2070-2099 the results of regional climate projections are employed. Results are presented for Hamburg (Germany). The present day UHI pattern is well reproduced compared to temperature data based on floristic mapping data. The magnitude of the early night-time UHI is underestimated when compared to observed minimum temperature differences. The future UHI pattern does only slightly change towards the end of the 21st century based on A1B scenario results of the RCMs REMO and CLM. However, for CLM the number of days with high UHI intensities significantly increases mainly due to a decrease in near-surface relative humidity.