Answers are grounded in this paper and cite their sources.
This study derives urban heat archetypes for Mediterranean historic cores strictly from spatial morphology. A graph convolutional autoencoder is trained on a heterogeneous network of building footprints and street segments in hillside Amman, Jordan, and flat Seville, Spain. The resulting archetypes are evaluated against held-out daytime and nighttime land surface temperature (LST), simulated pedestrian thermal comfort (Tmrt, UTCI), and rooftop photovoltaic yield. Incorporating relational context raises the daytime LST variance explained by building archetypes from η2 = 0.018 to 0.170 in Amman and from η2 = 0.005 to 0.051 in Seville—single realisations that vary across random initialisations. However, this advantage does not generalise across other environmental outcomes: archetype membership accounts for minimal variance in specific photovoltaic yield (η2 ≤ 0.034) and street-level radiant comfort (η2 ≤ 0.029) and fails to separate Amman’s building archetypes overnight (η2 < 0.001). Cross-city transfer is markedly asymmetric: Seville-trained archetypes recover 103% of Amman’s native thermal separation—achieving parity within initialisation variance—whereas Amman-trained models recover only 43% in Seville. This limitation stems from discrete partitioning rather than morphology itself, as the continuous canyon aspect ratio effectively separates street-level exposure. Consequently, archetypes validated solely against daytime LST cannot serve as general proxies for pedestrian comfort, nocturnal heat dynamics, or district solar capacity.
No takes yet. Share an insight, caveat, or question.
Al-Najjar et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: