PhD Oral Exam - Ahmed Marey, Building Engineering

When studying for a doctoral degree (PhD), candidates submit a thesis that provides a critical review of the current state of knowledge of the thesis subject as well as the student’s own contributions to the subject. The distinguishing criterion of doctoral graduate research is a significant and original contribution to knowledge.

Once accepted, the candidate presents the thesis orally. This oral exam is open to the public.

Abstract

Extreme heat is intensifying across the world's cities, and how severe it becomes locally is governed as much by urban morphology – the three-dimensional arrangement of buildings and streets – as by the regional climate. Yet the tools meant to anticipate it fall short: global climate models operate at 50–100 km, far coarser than the 100 m–1 km detail adaptation requires, and they typically hold the urban surface fixed, ignoring how a city's form will change. This dissertation builds an integrated framework spanning understanding, validation, efficient prediction, and forward projection, so that morphology-informed prediction of urban climate extremes becomes physically credible and computationally feasible.

It begins by consolidating what is known: a systematic synthesis of 111 artificial-intelligence studies shows that a handful of descriptors dominate urban thermal outcomes – building density explains up to 75% of land-surface-temperature variance and sky view factor over 67% of heat-exposure variation – while exposing a concentration of research in temperate cities and a neglect of extreme heat. Because prediction must rest on validated physics, the second study tests two operational models against the 2018 Montreal heatwave: the Global Environmental Multiscale (GEM) model attains higher near-surface accuracy (mean absolute error 1.35–2.63 °C) while the Weather Research and Forecasting (WRF) model better resolves urban-heat-island structure, yet both are too costly to capture morphology at the scales adaptation demands. That tension motivates the third study, Context-Aware Structural Prior Enhanced Resolution (CASPER), a data-efficient deep-learning framework that downscales atmospheric fields to kilometer scale while preserving extremes. Its central result reframes the economics of downscaling: held-out error is set by the climatological distance between training and target periods rather than by data volume (leave-one-month-out R² = 0.89), so representative training matches state-of-the-art accuracy with four times less simulation and reproduces heat-event observations to within 1.8 K. The fourth study then restores the dimension the others hold fixed – time – projecting urban morphology to 2100 across Canadian metropolitan areas under five Shared Socioeconomic Pathways, and showing that assuming a static city biases future heat-risk assessment by an amount comparable to the emissions scenario itself, a bias that accumulates over the century and peaks overnight.

The overarching contribution is an integrated framework in which morphology quantification specifies what the models must resolve, physically consistent artificial intelligence delivers that resolution efficiently without sacrificing extremes, and projected urban evolution feeds back into climate assessment – a coherent, morphology-aware basis for predicting urban climate extremes and for climate-responsive planning in cities that will not stand still.