Researchers from Cambridge University’s Department of Architecture have developed a groundbreaking AI model that can identify and prioritize “hard-to-decarbonize” (HtD) houses for retrofitting and other decarbonization measures. This innovative approach has the potential to play a pivotal role in achieving net-zero emissions by addressing a significant source of housing-related emissions often overlooked by policymakers.

Hard-to-decarbonize houses: A barrier to net zero

HtD houses, which account for over a quarter of all direct housing emissions, have been a major stumbling block in the path to achieving net-zero emissions. Factors contributing to the difficulty in decarbonizing these houses include their age, structural characteristics, location, socioeconomic barriers, and the availability of relevant data. Traditionally, policymakers have focused on generic buildings or specific decarbonization technologies, neglecting HtD houses.

AI model identifies HtD houses with 90% precision

The Cambridge researchers, led by Dr. Ronita Bardhan and urban researcher and data scientist Maoran Sun, have trained a deep learning AI model to classify HtD houses with an impressive 90% precision rate. This accuracy is expected to improve further as more data is incorporated into the model, a process already underway.

Dr. Bardhan emphasized, “This marks the first time AI has been employed to identify hard-to-decarbonize buildings using open-source data. Policymakers often lack the resources for detailed audits of every house, but our model can help direct their efforts towards high-priority houses, saving them valuable time and resources.”

Geographical distribution insights

The AI model not only identifies HtD houses but also provides insights into their geographical distribution. This information empowers authorities to efficiently target and implement interventions where they are most needed.

Utilizing multiple data sources

The researchers trained their AI model using a variety of data sources, including Energy Performance Certificates (EPCs), street view images, aerial view images, land surface temperature data, and building stock data. They identified 700 HtD houses and 635 non-HtD houses within their home city of Cambridge, all using publicly available data.

Maoran Sun highlighted the model’s adaptability, stating, “We trained our model using the limited EPC data available. Now the model can predict for the city’s other houses without the need for any EPC data.” Moreover, this approach can be applied in countries with limited or patchy datasets, making it accessible worldwide.

Advancements in data layers

The researchers are actively enhancing their model by adding additional data layers related to factors such as energy consumption, poverty levels, and thermal images of building facades. These enhancements are expected to further boost the model’s accuracy and provide more detailed information about specific building features that require attention.

Expanding beyond Cambridge

While the initial model was developed using Cambridge as a study site, the researchers are already extending its application to other UK cities. They are also collaborating with a space products-based organization to access higher-resolution thermal images from new satellites. Dr. Bardhan has participated in the NSIP – UK Space Agency program, where she worked on using high-resolution thermal infrared space telescopes to monitor the energy efficiency of buildings globally.

Transforming decarbonization policy

This innovative AI model has the potential to transform decarbonization policy decisions. By harnessing AI’s capabilities and dealing with larger datasets, policymakers can make more informed decisions based on evidence, driving effective climate change adaptation strategies.

Dr. Bardhan emphasized, “Empowering people with their own data makes it much easier for them to negotiate for support.” She added, “These are simple datasets, and we can make this model very user-friendly and accessible for authorities and individual residents.”

Cambridge’s unique challenges

Cambridge, while relatively affluent, presents its own set of challenges for decarbonization. The city’s historic housing stock and building bylaws restrict retrofitting and the use of modern materials in some historically significant properties. These unique challenges make Cambridge an informative site for refining the AI model and assessing its applicability in diverse urban environments.

Future collaborations and building consensus

The researchers plan to discuss their findings with Cambridge City Council and continue working with colleagues at Cambridge Zero and the University’s Decarbonization Network. By making data more visible and accessible to the public, the researchers believe that building consensus around efforts to achieve net zero will become more attainable.

This groundbreaking AI model represents a significant step forward in addressing the challenge of hard-to-decarbonize houses, potentially accelerating progress toward achieving net-zero emissions in the housing sector. Its adaptability, precision, and potential for widespread use make it a powerful tool for policymakers and individuals committed to a sustainable and decarbonized future.

Researchers from Cambridge University’s Department of Architecture have developed a groundbreaking AI model that can identify and prioritize “hard-to-decarbonize” (HtD) houses for retrofitting and other decarbonization measures. This innovative approach has the potential to play a pivotal role in achieving net-zero emissions by addressing a significant source of housing-related emissions often overlooked by policymakers.

Hard-to-decarbonize houses: A barrier to net zero

HtD houses, which account for over a quarter of all direct housing emissions, have been a major stumbling block in the path to achieving net-zero emissions. Factors contributing to the difficulty in decarbonizing these houses include their age, structural characteristics, location, socioeconomic barriers, and the availability of relevant data. Traditionally, policymakers have focused on generic buildings or specific decarbonization technologies, neglecting HtD houses.

AI model identifies HtD houses with 90% precision

The Cambridge researchers, led by Dr. Ronita Bardhan and urban researcher and data scientist Maoran Sun, have trained a deep learning AI model to classify HtD houses with an impressive 90% precision rate. This accuracy is expected to improve further as more data is incorporated into the model, a process already underway.

Dr. Bardhan emphasized, “This marks the first time AI has been employed to identify hard-to-decarbonize buildings using open-source data. Policymakers often lack the resources for detailed audits of every house, but our model can help direct their efforts towards high-priority houses, saving them valuable time and resources.”

Geographical distribution insights

The AI model not only identifies HtD houses but also provides insights into their geographical distribution. This information empowers authorities to efficiently target and implement interventions where they are most needed.

Utilizing multiple data sources

The researchers trained their AI model using a variety of data sources, including Energy Performance Certificates (EPCs), street view images, aerial view images, land surface temperature data, and building stock data. They identified 700 HtD houses and 635 non-HtD houses within their home city of Cambridge, all using publicly available data.

Maoran Sun highlighted the model’s adaptability, stating, “We trained our model using the limited EPC data available. Now the model can predict for the city’s other houses without the need for any EPC data.” Moreover, this approach can be applied in countries with limited or patchy datasets, making it accessible worldwide.

Advancements in data layers

The researchers are actively enhancing their model by adding additional data layers related to factors such as energy consumption, poverty levels, and thermal images of building facades. These enhancements are expected to further boost the model’s accuracy and provide more detailed information about specific building features that require attention.

Expanding beyond Cambridge

While the initial model was developed using Cambridge as a study site, the researchers are already extending its application to other UK cities. They are also collaborating with a space products-based organization to access higher-resolution thermal images from new satellites. Dr. Bardhan has participated in the NSIP – UK Space Agency program, where she worked on using high-resolution thermal infrared space telescopes to monitor the energy efficiency of buildings globally.

Transforming decarbonization policy

This innovative AI model has the potential to transform decarbonization policy decisions. By harnessing AI’s capabilities and dealing with larger datasets, policymakers can make more informed decisions based on evidence, driving effective climate change adaptation strategies.

Dr. Bardhan emphasized, “Empowering people with their own data makes it much easier for them to negotiate for support.” She added, “These are simple datasets, and we can make this model very user-friendly and accessible for authorities and individual residents.”

Cambridge’s unique challenges

Cambridge, while relatively affluent, presents its own set of challenges for decarbonization. The city’s historic housing stock and building bylaws restrict retrofitting and the use of modern materials in some historically significant properties. These unique challenges make Cambridge an informative site for refining the AI model and assessing its applicability in diverse urban environments.

Future collaborations and building consensus

The researchers plan to discuss their findings with Cambridge City Council and continue working with colleagues at Cambridge Zero and the University’s Decarbonization Network. By making data more visible and accessible to the public, the researchers believe that building consensus around efforts to achieve net zero will become more attainable.

This groundbreaking AI model represents a significant step forward in addressing the challenge of hard-to-decarbonize houses, potentially accelerating progress toward achieving net-zero emissions in the housing sector. Its adaptability, precision, and potential for widespread use make it a powerful tool for policymakers and individuals committed to a sustainable and decarbonized future.