Evaluating the importance of street trees and their parameters for urban canopy model performance: Model updates and machine learning

In this study, the effects of street trees on the performance of an urban canopy model (UCM) and how the UCM sensitively responds to tree-related parameters compared with urban thermal parameters are examined. For this, a single-layer UCM is extended to represent street trees within urban canyons and multi-objective parameter optimizations and a global sensitivity analysis are conducted with the aid of machine learning (ML) technique. Including street trees markedly improves the simulation performances of net radiation, sensible heat flux, and latent heat flux at both dense and vegetated urban sites, substantially reducing their systematic errors. Moreover, including street trees reduces the trade-off between the simulation performances of net radiation and sensible heat flux. At the dense urban site, the simulation performances of net radiation and sensible heat flux are the most sensitive to roof albedo while that of latent heat flux is the most sensitive to tree fraction. At the vegetated urban site, the tree-related parameters are the most influential for the simulation performances of all the three energy fluxes. This study emphasizes the importance of street trees in better simulating urban surface energy balance and presents an ML-based framework for efficient parameter optimization and sensitivity analysis.