Improved mesoscopic meteorological modeling of the urban climate for building physics applications

Abstract

Meteorological mesoscale models with different urban parametrization are used to predict the local urban climate at 250 m resolution. The authors propose a hybrid machine learning approach to improve the mesoscale prediction accuracy using measured air temperature data from a sensor network and remove simulation bias. The simulation of the urban climate of Zurich during a hot summer is used as case study showing the improvements of the simulation accuracy. Based on the hybrid model results, a cumulative heat exposure index is proposed to map local hotspots in the city and assess the difference of cooling loads between rural and urban environments. Furthermore, intra-urban microclimatic differences of a typical mid-latitude city are explored to highlight the benefits of detailed simulations for building physics purposes.