Harmonised global datasets of wind and solar farm locations and power

Abstract

Energy systems need decarbonisation in order to limit global warming to within safe limits. While global land planners are promising more of the planet's limited space to wind and solar photovoltaic, there is little information on where current infrastructure is located. the majority of recent studies use land suitability for wind and solar, coupled with technical and socioeconomic constraints, as a proxy for actual location data. Here, we address this shortcoming. Using readily accessible OpenStreetMap data we present, to our knowledge, the first global, open-access, harmonised spatial datasets of wind and solar installations. We also include user friendly code to enable users to easily create newer versions of the dataset. Finally, we include first order estimates of power capacities of installations. We anticipate these data will be of widespread interest within global studies of the future potential and trade-offs associated with the global decarbonisation of energy systems.

Earth transformed: detailed mapping of global human modification from 1990 to 2017

“…Stressors that are particularly important to improve include effects of grazing (currently coarse data and very broad expanse), pasture land, invasive species, and climate change (especially wildfire and effects of sea-level rise), and we encourage future work to focus on developing appropriate datasets and approaches to include or better capture these stressors. Key datasets we believe should be improved include transportation networks, including logging roads (e.g., van Etten, 2019), that are comparable through time; livestock grazing, rangelands, croplands, timber plantations, and pasturelands and their intensity of use; resource extraction (especially mining footprints); and temporal trends in gas flares, utility-scale solar and wind installations (Dunnett et al, 2020), and electrical substations.…”

A harmonised, high-coverage, open dataset of solar photovoltaic installations in the UK

“…However, validation identified an important limitation of radius-based clustering for geographic data such as ours, having geographical objects of widely varying sizes. The radius threshold of 300 m, though stricter than the 400 m threshold selected by Dunnett et al 22 , was appropriate for larger installations but could often inappropriately group many smaller installations together. Hence in a second iteration we introduced an area-adaptive clustering, in which the clustering threshold is selected dynamically based on the geographical area of the two objects being compared:…”

“…Dunnett et al applied DBSCAN to solar farm data, a widely-used clustering algorithm based on first connecting items that lie within a fixed radius of each other, and then pursuing transitive links to agglomerate small clusters into larger ones 22 . They evaluated a variety of distance thresholds for the radius-based clustering, finding no clear cutoff but selecting 400 m. For the present work, we used our own clustering implementation, likewise using a distance threshold and then pursuing the closure of the transitive links, but with a slightly stricter threshold of 300 m.…”

“…OSM data is not constrained to any particular application domain; it includes detailed geographic mapping of power infrastructure, thanks to OSM contributors with industrial or personal interest in that information. Recently Dunnett et al 22 used a snapshot of OSM data as a global listing of solar and wind power installations, reported as a simplified dataset of geolocation points with capacity values. They also merged some official open data sources including REPD into this data, using a spatial join to connect spatially overlapping entries, and spatial clustering to aggregate solar arrays into their corresponding farms.…”

“…We validated our dataset for internal consistency-for example, whether groupings were meaningful, or whether the indicated capacities appeared sensible given the object size-as well as external consistency with the SS dataset. Since alternative clustering of UK PV were available from 22 (with partial overlap in the source data), we also inspected correspondences between our dataset and theirs (henceforth the SD dataset). Differences in grouping or matching were visualised in the interface, and manually inspected for correctness.…”