Mapping global onshore wind turbines using multi-source remote sensing images and hybrid learning approaches

Shujun Li, Jianchuan Qi, Yongze Song, Peng Wang2025 · DOI

Abstract

Wind power is a vital zero-carbon alternative to fossil fuels, but its rapid expansion requires extensive land resources, raising concerns about ecological impacts. Quantifying these effects requires near real-time geospatial data. We developed an integrated framework combining OpenStreetMap (OSM) data with multi-source remote sensing (Google Earth, Sentinel-1/2) and hybrid machine learning (ResNet-18, Random Forest) to map global onshore wind turbines. Our models achieve >97% validation accuracy. We established a 2024 dataset of 379,595 wind turbines—a tenfold expansion over 2020 inventories. Analysis shows 80% of turbines are on cropland and grassland. This dataset facilitates research on renewable energy land management and ecological impact analysis.

Introduction

To meet global climate targets, onshore wind power is expanding rapidly. However, wind farms require significant land, leading to concerns about land-use competition and habitat fragmentation. Researchers need precise, facility-level geospatial data to study these impacts and help grid operators forecast power generation. Current resources, such as crowdsourced OSM data, often contain unverified entries or significant omissions. This paper introduces a unified, computationally efficient methodology for accurately tracking worldwide onshore wind turbine deployments.

Methodology

The study employs a two-phase hybrid pipeline to clean and augment crowdsourced data.

The hybrid learning framework integrates deep learning for data cleaning and traditional ML for global search.

Phase 1: Filtering Erroneous Data

The researchers started with raw OSM coordinates and extracted high-resolution image patches from Google Earth. They trained a ResNet-18 Convolutional Neural Network (CNN) to classify these patches as "turbine" or "non-turbine," effectively pruning false positives from the crowdsourced data and reducing manual verification needs by 86%.

Phase 2: Identifying Missing Turbines

To find turbines missing from OSM, a Random Forest classifier was trained using Sentinel-1 (Synthetic Aperture Radar) and Sentinel-2 (optical) data via Google Earth Engine. Sentinel-1's radar bands were particularly effective; metallic towers act as "corner reflectors," creating bright signatures that optical sensors might miss due to cloud cover or background interference.

Results and Discussion

The final 2024 Global Onshore Wind Turbine Dataset contains 379,595 verified turbines, representing a 10x increase over widely used 2020 inventories.

Global distribution and density of the 379,595 onshore wind turbines identified in the 2024 dataset.

Global Distribution

China dominates the global landscape with 122,602 turbines (32.3% of the total). The top five countries—China, the US, India, Germany, and Spain—account for the majority of global installations. The total impacted area is estimated at approximately 242,940 km².

Land Use Insights

Globally, 80% of turbines are situated on agricultural land (42% cropland, 38% grassland). However, land-use patterns vary by region. Germany sites over 90% of its turbines on agricultural land, while China has a notably higher percentage in forested areas (15%) compared to the global average, highlighting different ecological considerations for different nations.

Conclusion

This hybrid learning approach provides a cost-effective and timely method for updating global renewable energy inventories. The resulting open-access dataset and training benchmarks (37,285 annotated image patches) serve as a foundation for future research in wildlife disturbance tracking, habitat fragmentation analysis, and optimized turbine siting.

The codes and dataset are available at: https://doi.org/10.5281/zenodo.16759861

Summary updated

Abstract

Wind power serves as a vital zero-carbon alternative to fossil fuels for climate change mitigation. Nevertheless, the vast expansion of wind turbine installation requires extensive terrestrial resources, raising wide concerns regarding land use competition and ecological impacts. Quantifying these effects necessitates near real-time geospatial data on wind turbine placement and density. However, current methods remain inadequate monitoring for the fast-growing wind turbine deployment. Here, we developed an integrated framework that combines OpenStreetMap (OSM) data with multi-source remote sensing images (Google Earth and Sentinel-1/2) and deep learning and traditional machine learning models (ResNet-18 and Random Forest) to map global onshore wind turbines. Our models achieve validation accuracy >97 % while enabling cost-effective, timely updates of global onshore wind turbines. Eventually, we established a geographical dataset covering a total of 379,595 wind turbines globally by 2024. This dataset represents a tenfold expansion over currently available global wind turbine inventories as of 2020. In addition, we found that 80% wind turbines are situated on cropland and grassland, followed by forest and bare ground. This dataset facilitates essential studies on renewable energy land management, ecological impact analysis, and data-driven energy transition policies. The codes and dataset of the global onshore wind turbines is available at Zenodo link: https://doi.org/10.5281/zenodo.16759861 (Shujun et al., 2025).
BibTeX
@misc{li2025mapping,
  author = {Shujun Li and Jianchuan Qi and Yongze Song and Peng Wang},
  title = {{Mapping global onshore wind turbines using multi-source remote sensing images and hybrid learning approaches}},
  year = {2025},
  doi = {10.5194/essd-2025-512},
}
Figure