Publication Type

Journal Article

Version

acceptedVersion

Publication Date

8-2025

Abstract

Achieving carbon neutrality is a critical yet elusive goal for many cities, hindered by limited understanding of the relationship between building emissions and their surroundings. To address this challenge, we present a generalizable open science framework that integrates building energy-consumption data, multi-modal geospatial inputs and graph deep learning to quantify building operating emissions and their links to urban form and socio-economic factors. Applying this approach to five cities with diverse climates and planning contexts—Melbourne, New York City (Manhattan), Seattle, Singapore and Washington DC—we demonstrate that our models explain 78.4% of the variation in building operating carbon emissions across cities, achieving state-of-the-art accuracy for urban-scale energy modelling. Our findings reveal strong connections between a city’s planning history and its building carbon profile, alongside stark inequalities where wealthier areas often exhibit the highest per capita emissions. Additionally, the relationship between urban density and building emissions is complex and city specific, with emissions extending beyond dense urban cores into suburban areas. To design effective decarbonization strategies, cities must consider how their planning histories, urban layouts and economic conditions shape current emissions patterns.

Keywords

Climate Change, City Science, Machine Learning, GeoAI, Urban Analytics

Discipline

Energy Policy | Engineering

Research Areas

Integrative Research Areas

Publication

Nature Sustainability

Volume

8

Issue

10

First Page

1199

Last Page

1210

ISSN

2398-9629

Identifier

10.1038/s41893-025-01615-8

Publisher

Nature Research

Additional URL

https://doi.org/10.1038/s41893-025-01615-8

Download

Share

COinS