Solar panels now account for over 10% of total electricity generation in some U.S. states, such as California. But policy-makers, utility companies, and engineers still find it difficult to put an accurate number on the country's total solar power installation, let alone to describe what factors make solar power thrive in certain areas and not others.
Now, researchers at Stanford University have developed a new tool and accompanying open access website that identifies solar panels from high-resolution satellite data using automated image analysis, giving them unprecedented insight into the societal trends that drive solar power adoption. Their work appears December 19 in the journal Joule.
The tool, dubbed DeepSolar by its developers, including co-first-author doctoral students Jiafan Yu and Zhecheng Wang, scans high-resolution images covering the entire United States for solar panels, registers their locations, and calculates their sizes.
"Previous algorithms were so slow that they would have needed at least a year of computational time to find every solar panel across the United States, but DeepSolar requires a fraction of that time," says co-senior author Ram Rajagopal, a civil engineering professor at Stanford.
"With these methods, we can not only maintain and update a high-fidelity database of solar installations, but also correlate them at the census-tract level with the amount of incoming solar radiation as well as non-physical factors such as household income and education level," adds co-senior author Arun Majumdar, a mechanical engineering professor at Stanford and co-Director of the Precourt Institute for Energy.
All told, the authors located 1.47 million individual solar installations nationwide, including rooftop setups, solar farms, and utility-scale systems. Before DeepSolar, Rajagopal and Majumdar say, the decentralization of solar power meant that there was no comprehensive way to catalog the photovoltaic panels strewn atop homes and businesses, limiting understanding of American solar deployment at an aggregate level.
One area where DeepSolar could make an immediate impact is in guiding upgrades meant to make the American power grid more compatible with solar sources, which are intermittent due to daily and seasonal fluctuations in incoming sunlight.
"Now that we know where the solar panels are, or are likely to be in the future, we can feed that information into questions of modeling the electricity system and predicting where storage units and substations should go," says Majumdar.
It could also come in handy for pointing out areas that are ripe for new solar deployment. The researchers used their results to extract correlations between solar installation levels and population density, household income, and other variables, creating a model that can predict which geographic regions are most likely to adopt solar technology based on socioeconomic factors.
"Utilities, companies that install solar panels, even community planners that are thinking about sustainability, they all can benefit from this high-resolution spatial data and a website where they can explore and analyze the different trends involved," Rajagopal says.
Moving forward, the researchers plan to expand the DeepSolar database to include solar installations in other countries with suitably high-resolution satellite images. They also intend to add in features that can calculate a solar panel's angle and orientation from image analysis alone, allowing for more complete and accurate estimation of power-generating capacity in addition to the basic location and size data already collected.