Estimating Street-level Temperature Using a Machine Learning Approach, a Case Study in Chicago

The accurate modeling of urban microclimate is a challenging task given the high surface heterogeneity in cities. Efforts have been made in various approaches to simulate street-level meteorology. The conventional methods include Computational Fluid Dynamic (CFD) models and urban canopy models. Both methods attempted to capture the dynamics of air flow and energy distribution in street canyon influenced by the diverse land use and the complex geometry of urban terrain. However, the demanding computational cost of CFD models makes it practically impossible to simulate an entire city. Urban canopy models are usually much coarser than CFD models in spatiotemporal scales. On the other hand, recent years have witnessed advancements in urban observations, represented by the hyper-resolution terrain digitalization by LiDAR, and street-level monitoring by affordable environmental sensors. These data provide a new perspective for urban microclimate modeling.

In this study, we proposed an approach to fuse the results from the urban resolving regional climate model with hyper-resolution urban data to estimate street-level air temperature. We implemented this approach in the City of Chicago as a case study. Specifically, a group of Gaussian Process Regression (GPR) models will be used to learn the relations between the 1-km output from the urbanized Weather Research and Forecast (uWRF) model and the corresponding urban features. Those urban features are derived from the Illinois Height Modernization LiDAR dataset and NLCD 2019 land use dataset; while the street-level air temperatures are derived from the Array of Things monitoring network in Chicago. Based on trained GPR models, we then dynamically downscaled the predicted air temperature from 1-km to 30-meter resolution and at the hourly temporal intervals. This approach will vastly improve the resolution of temperature predictions in cities, therefore can locate the vulnerable areas during heat waves precisely. This study also aims to gain insights for city officials and urban planners on the mitigation of urban heat.