Physics-Guided Machine Learning Identifies 5 Optimum Test Locations to Predict Global PV Energy Yield for Arbitrary Farm Topologies

The photovoltaics (PV) technology landscape is evolving rapidly. To gauge the relative merit of emerging PV technologies and their scalable deployability, the global performance of these systems must be understood. Historically, most experimental and computational studies have focused on PV performance in specific regional climatic conditions; however, it has been difficult to translate these isolated regional studies to a global scale. Here, we present a physics-guided machine learning (PG-ML) scheme to demonstrate that: (a) analogous to Köppen–Geiger classification, the world can be divided into just a handful of PV-specific climate zones, and (b) the monthly energy yield (YM) data from a few locations (only 5!) is sufficient to predict the yearly energy yield (EY) of over 250,000 locations with a high spatial resolution ( $0.5^{\circ}\times 0.5^{\circ}$ ) and accuracy with root mean square error (RMSE) less than just 8 kW·h·m ⁻² . The map reveals that physically relevant meteorological conditions are shared across continents allowing pan-continental geographical extrapolation. Moreover, the scheme is agnostic to PV technology and farm topology, and thus, can be extended to novel PV technology/farm topology. Our results will lead to data-driven collaboration between national policymakers and research organizations to build efficient decision support systems for accelerated PV qualification and deployment across the world.