by Clarence Oxford
Los Angeles CA (SPX) Apr 23, 2026
A computer vision system developed at Columbia Engineering can determine how much energy a solar panel will generate over a year – and how much more it could generate if repositioned – using nothing more than a single 360-degree photograph taken from the panel’s location.
Graduate student Jeremy Klotz and professor Shree Nayar, T.C. Chang Professor of Computer Science at Columbia Engineering, developed the technique and tested it on solar-powered bike docking stations across upper Manhattan. At each station, the researchers raised a spherical camera above the panel, snapped a photo, and received an energy forecast on a laptop within seconds. The results showed that many panels across New York City and other urban areas are leaving significant energy untapped simply due to suboptimal orientation. By reorienting some of the panels they surveyed, Klotz and Nayar estimate energy harvest could increase by up to 30% over the course of a year.
“Our research result makes it possible to make important decisions such as where to place solar panels, how to angle them, and what the return on investment of an installation will be in the long run,” Nayar said.
The challenge in cities is that urban solar panels contend with what the industry calls urban canyons – environments where buildings, utility poles, water towers, and signs block direct sunlight. Existing methods for predicting energy output in these settings are time-consuming, expensive, and notoriously inaccurate. They struggle to account for three distinct light sources that reach a panel: direct sunlight, diffuse light scattered from a partially visible sky, and light reflected off surrounding structures.
That third source is easily underestimated. The Nayar lab found that reflections from surrounding buildings account for roughly 12% of a panel’s total annual energy on average. “If a building in the panel’s field of view is being directly lit by the sun while the panel is in shadow, that reflected light will account for most of the energy the panel receives,” Nayar explained. Small nearby objects – HVAC units, parapets, chimneys – compound the problem further, often absent from 3D city models yet capable of casting significant shadows.
The method works by extracting multiple layers of information from a single spherical image. Shadows in the photo reveal the sun’s direction; straight architectural lines indicate gravity orientation; a segmentation algorithm maps the visible portion of sky; and the appearance of surrounding buildings provides cues about their geometry and surface materials. From this, the system forecasts sun movement across the full year, estimates light bouncing off adjacent structures, and integrates historical weather data to extend predictions across cloudy and overcast conditions – not just clear days. The computation runs in near-real time on standard laptop hardware and works for rooftop panels, pole-mounted installations, and vertical wall surfaces.
The technique is described in a paper published in the journal Solar Energy. Klotz and Nayar have filed for a patent on the technology.
Vijay Modi, professor of mechanical engineering and earth and environmental engineering at Columbia, sees the system as enabling solar deployment at scale. “If you don’t have a tool to assess, you can’t deploy at scale,” Modi said. His own team used the method to measure the east-facing wall of a building on Columbia’s campus and found solar radiation levels high enough to meaningfully offset both the building’s energy use and peak electricity demand – a result obtained in hours rather than the month of work and expensive instrumentation previously required.
The solar price collapse has shifted the economics of installation so that permitting, labor, and site assessment now exceed the panel cost itself. Knowing in advance what energy a surface will actually yield has become the decisive factor in whether an installation makes financial sense. “At the price panels are today, virtually any surface in the world has the potential to be an energy harvester,” Nayar said. “The question is which ones are actually worth it.”
Research Report:Forecasting solar energy using a single image
Related Links
Columbia University School of Engineering and Applied Science
All About Solar Energy at SolarDaily.com
