Modeling Urban Trees and Neighborhoods Archetypes in Chennai to inform Heat Resilience

Abstract

Urban trees play an important role in heat resilience. However, urban trees are difficult to map bottom up, especially in the Global South. Doing a tree census is expensive and time consuming and there are limited data products at a fine enough scale such that individual trees or clusters of trees can be detected. Data fusion and machine learning using SkySat satellite imagery could help detect individual trees at the intra-urban scale, creating a cheap and scalable approach for tree canopy mapping. Intra-urban heat (air, surface, and mean radiant temperature) is also difficult to model or measure at fine scale. Having these models could help inform neighborhoods on heat mitigation strategies. Combining tree canopy maps and Google building polygons with ENVI-met could be used to model neighborhoods and simulate the impact of tree canopy to inform heat resilience. This thesis develops and evaluates a machine learning framework for urban land cover mapping that can detect individual trees and distinguishes between trees, grass, and non-vegetation. Performing data fusion, this thesis will define five neighborhood archetypes based on real-world neighborhoods. Using a 24-hour microclimate simulation in ENVI-met, this thesis explores the role of trees and white painted roofs and roads in urban heat resilience. This will be done as a case study in Chennai, India. The results of this thesis show that SkySat and machine learning can detect and map individual trees in urban areas with an overall accuracy over 80%. Tree canopy coverage also demonstrates to be a more effective way for heat resilience in cities lowering the air, surface and mean radiant temperature, compared to white painted surfaces, which only lowered the surface temperature.