Grainger College of Engineering, All Events

Intra-Urban Heat Assessment, Exposure & Understanding

Advisor: Professor Lei Zhao

Abstract

Cities are widely recognized as global hotspots of climate risks, where urbanization both elevates

near-surface temperatures and concentrates large populations. Despite its criticality, credible

impact-relevant intra-urban heat assessment is severely limited by both sparse observations and

inadequate model representation, resulting in a poor understanding of the human-perceived

thermal environment in cities. Consequently, much of our knowledge is derived from satellite landsurface

temperature (Ts) or from inaccurately modeled near-surface air temperature (Ta), neither

of which reliably captures human-experienced heat. This persistent data gap has hindered urban

climate and weather research for decades.

This dissertation aims to improve the intra-urban heat assessment and advance our understanding

of urban human-perceived thermal environment both within and across cities. I achieve this goal

by developing a novel urban transfer learning (U-TL) framework to generate an urban highresolution

air temperature (U-HAT) data at a large scale, and by using U-HAT to analyze

spatiotemporal patterns of heat hazard, exposure, and vulnerability across the 384 largest cities in

the contiguous United States (CONUS).

First, I develop the U-TL framework for urban high-resolution air temperature prediction. U-TL

leverages a combination of physics-informed transfer learning (TL) principles, a fusion of satellitederived

variables, and fundamental knowledge about the urban surface energy budget to overcome

the fundamental barrier of sparse true urban Ta observations. Through thorough validation, I show

that U-TL exhibits high accuracy in predicting urban Ta across scales, and this strong overall

predictive power is achieved with very limited urban training labels. This work provides a

generalizable framework to tackle the long-standing challenge of estimating urban climate

variables under severe data sparsity and high spatial heterogeneity.

Next, I provide, for the first time, a comprehensive pixel-to-pixel assessment of the discrepancies

between Ta and Ts within cities at continental scale using the U-HAT data developed from U-TL.

U-HAT is, to our knowledge, the first urban Ta dataset that accurately reproduces observed urban

climatology. It reproduces the observed diurnal, daily, and spatial variabilities of urban Ta. I reveal

substantial Ta–Ts differences both within and across cities, which is further amplified in

population-scale exposure to urban heat extremes. This work highlights the limitations of the

widespread use of Ts as a proxy for Ta in intra-urban exposure and impact assessments.

Finally, using U-HAT, I conduct a comprehensive assessment of intra-urban, human-experienced

heat across the 384 largest cities in the CONUS. I quantify spatial gradients in Ta and exposure to

extreme heat, evaluate how these burdens vary across demographic and socioeconomic groups,

and assess the effectiveness of heat mitigation strategies. I find a small variability in Ta both across

landcover classes and sociodemographic groups. Instead, disparities in heat exposure within cities

are driven primarily by population distribution rather than by temperature differences alone. UHAT

reveals severe heat exposure in many suburban neighborhoods that Ts-based analyses have

largely overlooked. This work highlights the importance of using impact-relevant variables when

drawing risk-related implications.

Through the development of a novel modeling framework, construction of a data product, and

comprehensive continental-scale analyses, this dissertation enables robust, impact-relevant

assessment of intra-urban heat. The U-TL framework and the U-HAT dataset have broad

applicability beyond this dissertation in supporting accurate prediction of other data-poor

variables in highly heterogenous environments, improving weather and climate models, and

enabling high-resolution epidemiological and socioeconomic studies.