Heat stress neighbourhood scale model combining urban simulations and deep neural network

Several indices have been proposed in the domain of urban climate and biometeorology for the assessment of heat stress or thermal comfort. In this study, the UTCI is selected as the primary index, which incorporates factors such as air temperature, humidity, wind speed and mean radiant temperature (Tmrt), where urban morphology plays a pivotal role. This paper presents the methodology and preliminary results of a heat stress neighbourhood scale model, accounting for the urban effect. To this end, SOLWEIG microscale urban simulations have been done for various LCZs (Local Climate Zones) dispersed across diverse latitudes in Europe, with hourly input data from ERA5-Land.The UTCI index is derived from the Tmrt outcomes of the microscale modelling. An artificial intelligence model is then trained using a deep neural network to estimate the number of hours of strong heat stress (UTCI > 32) using daily climate data, LCZ, day of the year, sun elevation, and sunset and sunrise time. Finally, a 30-year period heat stress database is calculated for different regions of Europe with a resolution of 100 meters and the average number of hours per year of strong heat stress is calculated for each region. Preliminary results show good agreement between the neural network and SOLWEIG results. This approach increases the agility and efficiency of the simulation, once the artificial intelligence model is trained, thereby facilitating its utilisation by urban and territorial planners, and decision-makers.