Multi-modal ensemble deep learning model for microclimate prediction using urban morphological maps

Buildings are the main energy consumers in cities, and accurately assessing their energy usage is crucial for improving their operational energy efficiency. However, conventional building energy simulation tools use a homogeneous Typical Meteorological Year (TMY) weather file, ignoring inter-city microclimate variation. This study aims to develop a spatial prediction model for microclimate conditions by considering the impacts of urban morphology. A multi-layer urban morphological mapping approach is developed to describe the urban morphologies graphically. The proposed multi-modal ensemble deep learning (DL) model predicts microclimate conditions using both microclimate data and morphological maps. The ensemble model outperforms individual multi-modal DL models trained on microclimate data from a single weather station. A validation experiment was conducted on a university campus using local weather stations. The proposed model reduces air temperature prediction errors by 78.1 %, 24.5 %, 19.4 %, 13.2 %, and 8.9 %, compared to TMY data, Kriging model, DL model, Factor-based DL model, and Map-based DL model, respectively. For relative humidity, the error reductions are 81.6 %, 32.6 %, 22.3 %, 29.8 %, and 10.9 %, respectively. The proposed method can be used to estimate microclimate conditions at the urban scale and provide more accurate weather data for urban building energy simulation.