For GIS professionals, obtaining high-precision, global distribution data for energy facilities—particularly emerging sources like photovoltaics (PV) and wind power—is often challenging. While open-source maps like OpenStreetMap provide some data, they frequently fall short in terms of timeliness, coverage, and attribute detail required for rigorous scientific research or commercial analysis. Recently, Microsoft, in collaboration with organizations like The Nature Conservancy (TNC), launched a significant open-source project on GitHub called the Global Renewables Watch (GRW). The research team utilized high-resolution satellite imagery and deep learning image segmentation models to conduct a quarterly analysis of global high-resolution satellite images from Q4 2017 to Q2 2024. This process automatically identified PV and wind power installations worldwide, accompanied by estimated construction dates and pre-construction land use information.

Data Overview

Coverage: Globally processed
Data Volume: Over 13 trillion pixels
Detections:

  • PV Power Plants: 86,410
  • Wind Turbines: 375,197
    Data Format: gpkg (GeoPackage)
    Temporal Range: Q4 2017 to Q2 2024

Data Download

Download Links:

https://github.com/microsoft/global-renewables-watch

Application Scenarios

The project's significant value lies not only in telling you "where" facilities are but also "when they appeared" and "what was there before." There are two primary application scenarios:

  1. Construction Timeline Analysis: For each detected PV panel or wind turbine, the model provides an estimated construction date, allowing for easy retrospective analysis of global renewable energy growth trajectories over the past 7 years.
  2. Land Use Change Analysis: The dataset includes information on previous land use types. This makes it straightforward to determine whether, for example, PV panels were built on desert land or if they occupied farmland or forest. This is an invaluable tool for Environmental Impact Assessments (EIA) and sustainability research.

Advanced Applications: Training Your Own Data and Models

Microsoft has open-sourced not only the resulting data but also the inference code and pre-trained models. If you wish to run identification on your own imagery (e.g., high-resolution images of specific regions) or fine-tune models based on theirs, you can refer to the official project documentation.

Official Project Page:

https://github.com/microsoft/global-renewables-watch

After spot-checking several locations within China, the identification accuracy appears quite satisfactory.

Conclusion

The release of the Global Renewables Watch significantly lowers the barrier to monitoring the global energy transition. It transforms "global renewable energy infrastructure mapping + temporal evolution + land use information" from an improbable task into an "open-source reality." It also highlights how, in the AI era, traditional data acquisition methods are becoming marginalized, giving way to inevitable new approaches. The project is currently released under the MIT License, meaning you are free to use it for academic research and even commercial development. GIS practitioners with related interests should not miss this opportunity.