This scientific finding is based on a highly complex approach. But how could it contribute to solving the world’s energy problems? Especially since the study’s authors do not advocate installing a solar system on every rooftop in the world. The goal is a different one: First, rooftop PV systems represent a low-carbon, highly available, and inexpensive technology. Second, 800 million people worldwide still live without electricity.
This is where the study comes in, as a useful tool for all governments, organizations, and companies that are committed to building PV infrastructure. Essentially, this work is about nothing less than accurately scanning 130 million square kilometers of land for its solar potential. Obviously, such a mammoth project requires a high level of methodical creativity and tangible support from AI applications.
The researchers’ core issue is satellite image evaluation
The researchers analyzed cadastral data, crowd sourced data, and satellite based data, among other things, to determine building layouts, climate conditions, solar radiation, and technology specific information related to photovoltaics. All in all, this involved petabytes of data, which clearly puts the study in the realm of big data.
One core issue for the researchers was the analysis of the satellite images. There is no program that can precisely distinguish rooftop areas from green or fallow spaces. The key to the solution: If you relate data on population size to the length of roads and the boundaries of built-up areas, you can infer the existing rooftop areas. To do this, the scientists developed a machine learning model that can independently assess unknown data based on comprehensive training data. To do so, they divided the entire global land mass into assessment units of ten square kilometers each – adding up to exactly 3,521,120 squares, all of which had a unique identifier and were assigned to a unique country.
The learning program itself contained a gigantic amount of samples, corresponding to about 11.4 million square kilometers of land area and 300 million individual buildings. This was used as the foundation for calculating the assessment units – with a data set of the respective building area for each individual assessment cell. But that was not the end of it – after all, the rooftop areas still had to be converted into solar potential. The values for climate and solar radiation, as well as the technology costs, also had to be taken into account.