Global Solar Siting from Geospatial Embeddings

Global solar photovoltaic (PV) capacity is rapidly growing, helping decrease our emissions and dependence on fossil fuels. However, optimally siting a solar plant remains challenging, requiring anticipation of geographical, urban, and infrastructure needs. Current methods rely on multi-criteria decision analysis (MCDA) using hand-picked geophysical layers and solar-resource products, however these strategies are not geographically transferable and do not capture the urban requirements of solar sites. We use geospatial embeddings, dense vectors representing geophysical properties of a location, to analyse 38,591 existing solar sites and the land they were built on to learn the latent solar suitability signal. We assess a set of linear and non-linear models, selecting a Random Forest (RF) as the optimal model for accuracy and large scale deployment. The RF accurately selects solar sites when validated against future installations, with a ROC-AUC of 0.89, and 56% of existing solar sites captured in the top 10% of ranked land. Additionally, we compare our RF model to published MCDA strategies, demonstrating an increase in ROC-AUC of up to 0.26 in predicting solar sites across several countries. We show that our model performs well in capturing sites with high generation (annual energy yield) and capacity (installed power): the top 20% of suitable land holds 78% of estimated generation and 55% of installed capacity. The global 10 m resolution solar suitability dataset can be found here: (made available upon acceptance)