Global Renewables Watch: A Temporal Dataset of Solar and Wind Energy Derived from Satellite Imagery

TL;DR

Global Renewables Watch addresses the need for up-to-date, spatially explicit data on solar PV and onshore wind installations and their development history. The authors implement a data-centric pipeline that cleans OpenStreetMap labels, trains solar PV and wind detectors on PlanetScope basemaps, and conducts global temporal inference to produce a temporally resolved geospatial dataset. They validate capacity implications by comparing country-level estimates to IRENA 2023 data, achieving high correlations with $r^2$ up to $0.960$ for solar and $0.932$ for wind, and quantify land-cover changes associated with siting. The resulting dataset, comprising 86,410 solar installations and 375,197 wind turbines, supports planning, policy, and SDG tracking, while noting limitations from label noise and regional coverage gaps.

Abstract

We present a comprehensive global temporal dataset of commercial solar photovoltaic (PV) farms and onshore wind turbines, derived from high-resolution satellite imagery analyzed quarterly from the fourth quarter of 2017 to the second quarter of 2024. We create this dataset by training deep learning-based segmentation models to identify these renewable energy installations from satellite imagery, then deploy them on over 13 trillion pixels covering the world. For each detected feature, we estimate the construction date and the preceding land use type. This dataset offers crucial insights into progress toward sustainable development goals and serves as a valuable resource for policymakers, researchers, and stakeholders aiming to assess and promote effective strategies for renewable energy deployment. Our final spatial dataset includes 375,197 individual wind turbines and 86,410 solar PV installations. We aggregate our predictions to the country level -- estimating total power capacity based on construction date, solar PV area, and number of windmills -- and find an $r^2$ value of $0.96$ and $0.93$ for solar PV and onshore wind respectively compared to IRENA's most recent 2023 country-level capacity estimates.

Figures (6)

• Figure 1: (Left) Distribution of 86,410 solar PV installations and 375,197 onshore windmills detected by our models in 2024 Q2. (Right) Close-up visualizations of solar and wind installations in the village of Farmsum in the Dutch province of Groningen.
• Figure 2: Inference workflow. We run separate solar PV and windmill semantic segmentation models over PlanetScope quarterly satellite imagery from 2017 Q4 through 2024 Q2 to create a solar PV polygon dataset and windmill point dataset with estimated construction dates from each feature.
• Figure 3: Examples of noisy labels of solar PV installations and windmills from OpenStreetMap. Solar PV installations may be labeled according to the total area of the larger installation they are in (A, C) or might not be aligned with satellite imagery in locations where rapid expansion of renewable energy is occurring (B). Similarly, windmills point labels may be misaligned with satellite imagery (D), missing from wind farms that are in development (E) or out-of-date (F). Deep learning models trained on such labels will not generalize well as they overfit to the noise.
• Figure 4: Progression of the predictions from a solar PV segmentation model trained on unfiltered versus filtered datasets.
• Figure 5: Effect of repowering a wind farm in Clear Lake, Iowa, United States (A) and solar expansion from 2019 Q3 to 2023 Q4 in Ordos City, China (B). Repowering of wind farms involves replacing older, less efficient wind turbines with newer, more powerful models to increase energy production and extend the lifespan of a wind farm, often using existing infrastructure. Solar farms footprints also tend to grow rapidly over time. These are common practices that make labels for single time stamp obsolete over time. Repowering along with continuous and rapid development of new solar and wind installations are some of the factors that make temporal monitoring crucial.