China maps 319,972 solar sites and 91,609 turbines with AI

Researchers at Peking University and Alibaba Group’s DAMO Academy used deep-learning on 7.56 terabytes of sub-meter satellite imagery to identify 319,972 solar photovoltaic facilities and 91,609 wind turbines across China. The work produced a county-level national inventory and was published in Nature.

The team trained a neural network to detect installations ranging from rooftop panels in coastal cities to utility-scale farms on the Mongolian plateau. The processing covered imagery for 1,915 counties and produced a consistent geospatial map. The dataset and the analytical code have been posted on Zenodo.

The paper reports that pairing solar and wind resources across wider geographic areas reduces overall generation variability. The authors find that complementarity between wind and solar output becomes more pronounced as the distance between paired facilities increases, and they give examples showing local cloud cover can suppress solar output without affecting distant wind corridors.

The study notes that China currently manages much renewable integration at the provincial level. The paper’s authors argue that national-scale coordination could help match complementary sources and reduce curtailment, the loss of generated renewable power that has affected the country’s clean-energy rollout.

Demand for electricity from data centers and artificial intelligence computing has risen sharply. Data reported by the China Electricity Council show data-center consumption increased 44 percent year-over-year in the first quarter of 2026 to 22.9 billion kilowatt-hours. The paper notes many new data centers are expanding into northern and western provinces where land is cheaper and wind and solar resources are abundant.

The International Energy Agency projects global data-center electricity consumption could approach 1,000 terawatt-hours by the end of the decade. The authors write that national-scale mapping and coordination tools can help system planners match generation assets with transmission, storage and dispatch policies.

Technically, the project required methods that work across varied installation types, terrain and image quality. DAMO’s model identified installations under diverse conditions and produced a nearly 412,000-facility inventory that can be analyzed at county level. The paper suggests the approach could be replicated by other countries seeking similar visibility.

Peking University professor Liu Yu described the inventory as a ‘God’s-eye view’ of China’s new-energy landscape. The researchers and the published dataset aim to provide a tool for operators, planners and policymakers to manage renewable assets and address curtailment as demand from AI and other digital services grows.