Google's Solar API Expansion: Utilizing Satellite Imagery and Machine Learning for Global Solar Potential Estimation

• Introduction:
  • Energy demand is projected to rise significantly, with solar photovoltaics expected to generate nearly 28% of the total demand by 2035.
  • The Google Maps Platform (GMP) Solar API has been expanded to include areas in the Global South, using machine learning (ML) on satellite imagery to generate digital surface models (DSMs) and roof segmentation for solar assessments.
• Solar API Insights:
  • The Solar API, launched in 2023, simplifies solar panel installation by providing comprehensive building solar data, rooftop imagery, weather, and financial data.
  • Aims to help businesses and individuals assess solar potential, optimize panel layouts, and accelerate solar adoption.
• Global Expansion with Satellites:
  • ML techniques are applied to satellite imagery to overcome challenges like lower resolution and scarcity of elevation maps.
  • Enables more frequent data updates and expands coverage to areas previously limited by reliance on aerial imagery.
  • Adds 125 million new buildings across 23 countries, with a potential coverage increase of 1.9 billion buildings.
• ML for High-Quality DSMs and Roof Segments:
  • New ML models generate high-quality nadir (overhead) DSMs and planar roof segment instances from single-view satellite imagery.
  • A two-stage model (base and refinement) is used, employing a U-Net–style architecture with a Swin Transformer encoder.
  • Models are evaluated using metrics like DSM mean absolute error (MAE), roof pitch error, and roof segment instance intersection over union (IOU).
• Model Performance:
  • Results show that the model performs well with RGB-only inputs, making it suitable for regions with only satellite RGB imagery.
  • Error variation between countries is small, indicating adaptability to various building styles and landscapes.
• Visualizations:
  • Model predictions are shown for locations in India, Malaysia, and Australia, demonstrating accurate capture of roof details and obstacles.
  • Comparisons with high-quality aerial data are presented, along with visualizations of annual solar flux predictions.
• Future Work:
  • Ongoing research aims to improve accuracy and explore new directions like obstacle detection, roof material detection, and existing solar panel identification.
• Accessibility:
  • This experimental data is available through the Solar API Expanded Coverage Testing Program.
• Publication:
  • The methodology is detailed in the paper "Satellite Sunroof: High-res Digital Surface Models and Roof Segmentation for Global Solar Mapping," published at the Climate Change and AI workshop at NeurIPS 2024.
• Collaboration:
  • Google invites users to explore the data and contribute to a sustainable future by signing up for the program.