A sub-meter resolution urban surface albedo dataset for 34 U.S. cities based on deep learning

Area of Science:

• Environmental Science
• Remote Sensing
• Urban Planning

Background:

• Surface albedo significantly influences urban heat islands (UHIs) by regulating solar energy absorption and reflection.
• Understanding the urban thermal environment at a hyperlocal level is challenging due to the lack of high-resolution albedo data.

Purpose of the Study:

• To create the first high-resolution urban albedo maps for 34 major U.S. cities.
• To enable detailed analysis of urban microclimates and thermal comfort.
• To provide valuable data for urban planning and environmental monitoring.

Main Methods:

• Utilized advanced deep learning models, specifically U-Net, for impervious surface classification and albedo prediction.
• Integrated multisource remote sensing data, including NAIP imagery, roof albedo data, building footprints, land cover classifications, and Sentinel-2 imagery.
• Achieved sub-meter resolution albedo mapping by differentiating between impervious surface albedo (ISA) and pervious surface albedo (PSA).

Main Results:

• Developed high-resolution urban albedo maps with validated accuracy: ISA R² of 0.9028 and PSA R² of 0.9538.
• Demonstrated high precision and reliability in predicting both impervious and pervious surface albedo.
• Made the comprehensive datasets publicly available for research and application.

Conclusions:

• The generated high-resolution albedo maps are crucial for a granular understanding of urban thermal dynamics.
• This research provides essential data for informed urban planning and effective environmental management strategies.
• The publicly available datasets will advance research in urban climatology and sustainable development.