Mapping facade materials utilizing zero-shot segmentation for applications in urban microclimate research.

Area of Science:

• Urban Climatology
• Computer Vision
• Remote Sensing

Background:

• The Urban Heat Island (UHI) effect poses a significant challenge to urban climate.
• Accurate urban microclimate modeling requires detailed surface property data, including facade materials.
• Current methods struggle with large-scale facade material mapping due to data and segmentation complexities.

Purpose of the Study:

• To develop a high-fidelity facade material extraction method using advanced computer vision.
• To overcome the limitations of traditional supervised learning by employing zero-shot learning paradigms.
• To enable efficient and adaptable mapping of urban facade materials for UHI and energy studies.

Main Methods:

• Utilized state-of-the-art zero-shot learning computer vision models.
• Focused on high-fidelity facade material extraction from urban imagery.
• Tested the algorithm across architecturally diverse cities: Dubai, Amsterdam, and Boston.

Main Results:

• The algorithm achieved 68% accuracy in detecting the predominant facade material.
• Successfully identified the top three material classes in 85% of cases.
• Demonstrated the link between material coverage and outdoor thermal comfort, showing shifts in heat/cold stress hours.

Conclusions:

• Zero-shot learning offers a powerful solution for facade material mapping, reducing reliance on extensive labeled datasets.
• Accurate facade material data is vital for improving urban microclimate models and assessing thermal comfort.
• This method facilitates adaptation to various urban contexts and material types, advancing UHI research.