Solar panels largely confined to wealthy Americans

When a new consumer technology makes its debut, whether it’s a smartphone or an electric car, its adoption rate typically follows a predictable path. The first buyers come from a narrow slice of high-income users or tech enthusiasts who are willing to pay high prices. Over time, as prices fall and economies of scale kick in, sales climb sharply and the technologies become mass-market products. Eventually, the market becomes saturated, and the number of users reaches a plateau.

Solar panel array on the rooftop of an apartment building in Brooklyn, New York. (Image credit: Bright Power Inc.)

That hasn’t happened with residential solar power in the United States, however. Even though solar power has increased dramatically over the past decade, and the cost of photovoltaic panels has fallen significantly, lower-income communities have been much slower than high-income users to ramp up.

In a new study that used artificial intelligence to interpret a decade’s worth of satellite images, Stanford University researchers find that low-income communities were far less motivated than high-income communities by the kinds of tax incentives that state and local governments often offer. Even when lower-income communities initially installed some solar, their adoption rate plateaued at comparatively low levels.

By contrast, the researchers found, less commonly used performance-based incentives appeared to spur the adoption of solar in lower-income communities. Interestingly, the performance incentives, which reward customers based on how much solar they produce or how much less electricity they buy from the grid, appeared to have little or no impact on higher-income areas.

“This is the first study to examine the adoption of solar power across the United States in such a granular way over a long period of years,” says Ram Rajagopal, associate professor of civil and environmental engineering at Stanford and one of the paper’s senior co-authors.

“Our conclusion is a little surprising. Low-income communities didn’t accelerate their adoption of solar power,” he added. “They either didn’t start at all or they reached saturation at a very low level. By contrast, think about smartphones, which were initially so expensive that they were only for very high-income people or super tech nerds. But in less than 10 years, they became cheaper and are now owned by hundreds of millions of people.”

Deep solar

The study, published in today’s print issue of Joule, employed a machine-learning model – named DeepSolar++ by the researchers – that analyzed satellite images to identify where solar panels are and when they were installed in more than 400 counties across the United States. The researchers compiled images from 2006 through 2017 and then combined that data with information about each community’s demographics as well as local financial incentives for solar power.

The study was led by Zhecheng Wang, a doctoral student in Stanford’s Department of Civil and Environmental Engineering. Wang helped build DeepSolar++.

In 2018, the Stanford team published an initial study that analyzed the number of solar installations at a single point in time. That study confirmed that solar arrays were much less common in low-income communities, but it didn’t offer much insight about the trend over time. Were those communities starting later but then catching up, as had been the case with smartphones, big-screen televisions, and other new technologies?

To map the long-term trajectory, the researchers had to deal with the fact that commercial satellite imagery in the mid-2000s was low-resolution. That made it hard to recognize solar panels. The solution, Rajagopal says, was to have DeepSolar++ compare the more recent high-resolution pictures with older images of the same locations. That comparison provided enough information for the system to pick out solar panels even though the earlier images were much blurrier.