AI model maps building emissions to support fairer climate policies
Open-source approach uses publicly available data and machine learning to identify carbon hotspots and guide targeted urban decarbonisation.
image:
Department of Architecture PhD student and Lead Author Winston Yap (left) with Asst Prof Filip Biljecki (right) who led the research.
view moreCredit: College of Design and Engineering at NUS
An open-source artificial intelligence model to accurately map the carbon emissions of buildings across multiple cities could become a powerful new tool to help policymakers plan targeted and equitable decarbonisation strategies.
The model, developed by researchers at the College of Design and Engineering (CDE) at the National University of Singapore (NUS), offers city planners a detailed picture of how building carbon emissions are distributed and what drives them, with a view to helping authorities design smarter, fairer strategies to cut emissions.
The model is the result of research led by Assistant Professor Filip Biljecki from the Department of Architecture at CDE. The team’s findings were published on 15 August 2025 in the journal Nature Sustainability.
“Our model estimates operational carbon emissions of individual buildings at the scale of entire cities,” said Department of Architecture PhD student Winston Yap, Lead Author of the study.
“Unlike previous approaches that rely on proprietary data, our open approach is designed to be transferable across cities, including those with different data availability conditions.”
Applied to data mapping over half a million buildings in five cities - Singapore, Melbourne, New York City (Manhattan), Seattle and Washington DC - the researchers say their model explained up to 78 per cent of the variation in emissions. The results revealed significant differences in how emissions are distributed within cities and identified key factors that influence building energy use, including urban form, planning history, and income levels.
“Building emissions are not just about size or density, they’re deeply shaped by the unique context of each city, from its planning legacy to climate and economic conditions,” said Asst Prof Biljecki. “By using only open data, we’ve built a flexible framework that cities around the world can use to better understand their carbon footprint and plan more effective responses.”
One of the key insights from the study is the complex relationship between building density and carbon emissions. While taller buildings tend to be more energy-efficient per unit area due to economies of scale, dense urban cores may also experience higher cooling demands due to urban heat island effects. Suburban areas, typically associated with detached low-rise buildings, were found to be significant contributors to total emissions, sometimes rivalling those of city centres.
The research also uncovered stark inequalities. In most cities studied, wealthier neighbourhoods were found to have disproportionately high per capita emissions. In Manhattan, for example, more than half of total building emissions were attributed to just a handful of large buildings.
“Uniform carbon pricing or blanket regulations risk placing an unfair burden on lower-income communities that may already be struggling with older, less efficient infrastructure,” said Asst Prof Biljecki. “Our findings highlight the need for place-based strategies that take both emissions intensity and socioeconomic vulnerability into account.”
The framework integrates diverse data sources including satellite imagery, street view photos, population maps, road networks, and local climate data using graph neural networks, a form of deep learning that captures spatial relationships between urban elements.
By making their approach entirely open, the researchers say they want to support global efforts to reduce emissions from the built environment and to help cities meet their climate targets.
“This work demonstrates the potential of open science and AI to accelerate urban sustainability,” said Asst Prof Biljecki. “It’s not just about understanding where emissions come from, but also ensuring that climate action is both effective and fair.”
Journal
Nature Sustainability
Method of Research
Computational simulation/modeling
Subject of Research
Not applicable
Article Title
Revealing building operating carbon dynamics for multiple cities
Article Publication Date
15-Aug-2025
Advancing disaster response with the EBD dataset
Journal of Remote Sensing
image:
Illustration of the proposed SS-FT framework. (A-B) show the overall SS-FT and the model’s dataflow. (C-D) elaborates on the supervised fine-tuning, and the self-supervised contrastive learning processes. For each mini-batch, Lcontra is calculated on positive ”queries” and negative representations stored in the category-wise memory bank on a pixel level.
view moreCredit: Journal of Remote Sensing
A new dataset, the Extensible Building Damage (EBD) dataset, offers significant improvements in disaster response mapping by combining satellite imagery and deep learning techniques. This dataset, covering 12 natural disasters, uses semi-supervised fine-tuning (SS-FT) to reduce the time and effort traditionally required for manual damage labeling, speeding up disaster recovery efforts globally.
Building damage assessments (BDA) are crucial for post-disaster recovery, as they help in identifying areas most in need of urgent assistance. However, current BDA methods suffer from slow dataset development, largely due to manual labeling requirements. The new Extensible Building Damage (EBD) dataset addresses this by leveraging deep learning for semi-automated labeling, improving the speed and accuracy of damage assessment in disaster zones. Based on these challenges, or due to these problems, there is a need for further research into semi-automated disaster response technologies.
Researchers from Zhejiang University and the RIKEN Center for Advanced Intelligence in Japan, with collaboration from various international institutions, have introduced the EBD dataset, published (DOI: 10.34133/remotesensing.0733) in Journal of Remote Sensing. This dataset represents a leap in disaster mapping by using machine-driven annotation to assist human experts in quickly categorizing building damage post-disaster. The SS-FT method it uses provides an innovative solution to the traditionally slow and labor-intensive task of damage classification.
The EBD dataset includes over 18,000 image pairs from 12 major natural disasters, with labels for over 175,000 buildings. Unlike earlier efforts, the dataset uses a semi-automatic annotation process, drastically reducing the manual workload by 80%. The SS-FT method not only utilizes a small amount of manually labeled data but also incorporates large sets of unlabeled samples for improved accuracy. This breakthrough provides faster, more reliable damage assessment results, particularly in areas with limited human resources.
The process begins with a pre-trained model using a historical dataset, which is then fine-tuned on disaster-specific data through the SS-FT method. By comparing pre- and post-disaster images, the model automatically classifies damage into four categories: No Damage, Minor Damage, Major Damage, and Destroyed. The SS-FT method has proven to improve model accuracy especially in situations with limited labeled samples. This capability is demonstrated through disaster events such as Hurricane Ian and the Turkey Earthquake, where the model showed significant improvements over pre-trained only setting and supervised fine-tuning setting.
"By reducing the reliance on manual labeling, the EBD dataset represents a major step forward in how we can use artificial intelligence in disaster response," said Dr. Zeyu Wang, a leading researcher on the project. "This system not only accelerates post-disaster recovery but also makes it more scalable, meaning it can be used globally to address future disaster events."
The research used high-resolution satellite imagery from the Maxar Open-Data Program, processing bi-temporal images to assess building damage. The SS-FT method was implemented using the PyTorch framework, with the model optimized on NVIDIA GPUs. The process involved multiple rounds of fine-tuning, using both labeled and unlabeled data to improve damage classification accuracy.
The EBD dataset has the potential to transform emergency response by providing rapid, accurate damage assessments. As this dataset continues to grow, it could be integrated into broader global disaster monitoring systems, offering valuable insights for climate change-related disasters. Additionally, its semi-automated labeling system can be applied to new disaster scenarios, making it an indispensable tool for disaster management worldwide. The future of disaster response relies on datasets like EBD, offering more timely and precise interventions to save lives.
###
References
DOI
Original Source URL
https://spj.science.org/doi/10.34133/remotesensing.0733
Funding Information
This work was supported by the National Key Research and Development Program of China under495Grants 2019YFE0127400.
About Journal of Remote Sensing
The Journal of Remote Sensing, an online-only Open Access journal published in association with AIR-CAS, promotes the theory, science, and technology of remote sensing, as well as interdisciplinary research within earth and information science.
Journal
Journal of Remote Sensing
Subject of Research
Not applicable
Article Title
Constructing an Extensible Building Damage Dataset via Semi-supervised Fine-Tuning across 12 Natural Disasters
Article Publication Date
17-Aug-2025
