AI4UP: Artificial Intelligence for low-carbon Urban Planning is a project aiming to use advances in machine learning and geospatial data science to provide novel urban planning solutions helping city mitigate and adapt to climate change. AI4UP is developed by the working group Land Use, Infrastructure, Transport at MCC and the chair of Sustainability Economics of Human Settlements at TU Berlin.

The motivation for AI4UP is the lack spatial granularity and of specific solutions existing in global climate change mitigation scenarios. Local policy makers are often left disoriented on which measures are adequate and impactful, and how everyday decisions related to infrastructure investments or urban planning can be modified to adjust to a low-carbon future. In addition, global climate change mitigation requires solutions that can scale across cities and regions, while most existing research looks at city-specific case studies.

AI4UP aims at developing tools addressing these issues and providing climate solutions at street and building level for municipalities and human settlements. For this, we use geospatial data, such as remote sensing imagery, GPS, government data or volunteer geographic information, which are now available at high-resolution and at large scales. We are applying machine learning to learn features of urban form that are predictive of relevant metrics for urban sustainability. For example, we developed proof-of-concepts that buildings heights or commute distances are related to the specific configurations of neighborhoods and the larger spatial structure of cities.

Our goal is to provide solutions integrating different sectors and relevant dimensions of urban climate change mitigation and adaptation. Our current work focuses on the building and transportation sectors, and on urban heat.

We presented our main architecture in Milojevic-Dupont & Creutzig (2020), presented in the figure below.



A key example is the prediction of building heights (Milojevic-Dupont et al. 2020). In this work, we trained several machine-learning algorithms to learn building heights from 150 features representing several dimensions of urban form (building stock, street network, etc.) engineered with domain knowledge. The results revealed that it is possible to predict building heights with an average error well below the typical floor height across several European countries, demonstrating that urban form data contains useful information to predict important metrics for urban sustainability.

In the future, we aim to apply this methodology for building attribute predictions across the whole European Union to generate a building-level database that will include more than 200 million buildings. The database is intended to be generated by aggregating and harmonizing existing data, and using machine learning to fill data gaps. Such a database would enable us to explore new opportunities for scaling existing building energy demand models, for example to investigate how to retrofit the current building stock in the EU to achieve decarbonization targets.





Milojevic-Dupont, Nikola, and Felix Creutzig. "Machine learning for geographically differentiated climate change mitigation in urban areas." Sustainable Cities and Society (2020): 102526.

Milojevic-Dupont, Nikola, et al. "Learning from urban form to predict building heights." PLOS ONE 15.12 (2020): e0242010.

Wagner, Felix et al. "Understanding the impact of the built environment on travelled vehicle kilometres in Berlin" Proceedings of the 28th International Seminar of Urban Form (2021).

Rolnick, David, et al. Tackling climate change with machine learning. ACM Computing Surveys (in press)

Creutzig, Felix, et al. "Upscaling urban data science for global climate solutions." Global Sustainability 2 (2019).

Creutzig, Felix, et al. "Leveraging digitalization for sustainability in urban transport." Global Sustainability 2 (2019).