Researchers at Brookhaven National Laboratory and Stony Brook University integrate physics into data-driven models to advance solar forecasting and help generate solar energy more efficiently.
Stony Brook, NY, Mar 26, 2025 — Solar energy is a clean, renewable source of energy that can help us build a sustainable future. But predicting how much solar energy would be available at a given time is tricky — especially because clouds play such a significant role in blocking or letting sunlight through.
A study conducted by researchers at Brookhaven National Laboratory, in collaboration with researchers from Stony Brook University, Nanjing University of Information Science and Technology, and National Renewable Energy Laboratory, sheds light on how different cloud types can impact solar forecasting, advancing our ability to predict how much solar energy is available.
The way the sun’s radiation interacts with a cloud varies widely depending on cloud type. Clouds like cumulus, cirrus, stratus, and deep convective clouds exhibit differences in macrophysical properties (e.g., cloud thickness, cloud fraction), microphysical properties (e.g., cloud droplet size and concentration), and optical characteristics (e.g., cloud albedo, optical depth). These factors determine how clouds scatter and absorb sunlight, either enhancing or diminishing solar irradiance reaching a surface. This challenge is further compounded by the fact that clouds change rapidly, creating significant uncertainty in solar forecasts.
Less solar irradiation reaching panels
While previous research has explored cloud impacts on solar forecasts, most of it has been limited to broad, overall cloudy conditions or short-term scenarios and has only considered a few cloud types.
Yangang Liu, Senior Scientist at BNL and former Adjunct Professor, School of Marine and Atmospheric Sciences, Stony Brook, and Department of Applied Math and Statistics , said, “Thanks to the decade-long, high-quality data collected by the U.S. Department of Energy’s Atmospheric Radiation Measurement (ARM) Program, our study offers a detailed evaluation of solar forecasting models across varied cloud conditions.”
Using data from 2001 to 2014, the researchers systematically analyzed how eight distinct cloud types affect solar irradiance predictions. These included cumulus, stratiform clouds, congestus, deep convective clouds, altostratus, altocumulus, cirrostratus/anvil, and cirrus. The study built upon the team’s earlier work on physics-informed data-driven models, which integrated cloud-radiation physics to improve solar forecasting accuracy. These models were tested against real-world measurements of solar radiance and cloud types from the ARM South Great Plain (SGP) Central Facility site.
The results showed a clear hierarchy in the models’ accuracy based on cloud type. They performed best with weak convective clouds (like cirrus), followed by stratiform clouds, and worst with strong convective clouds, such as deep convective clouds.
Left to right: Yangang Liu and Shinjae Yoo
Shinjae Yoo, Adjunct Assistant Professor at Stony Brook University and Distinguished Scientist with Brookhaven Lab’s Computing and Data Sciences directorate, said, “By categorizing clouds into stratiform, weak, and strong convective types, we were able to identify where our models performed best and where they needed improvement. The trends we saw highlighted the complexity of forecasting under certain cloud conditions. For example, in the case of deep convective clouds — which have more complex spatial structures with dynamic and unpredictable nature — we noticed a significant uncertainty in the results.”
The study also showed that including information about cloud types in forecasting models can improve predictions. By factoring in the physics of how clouds interact with sunlight, the models performed better than previous models that didn’t consider cloud types. The researchers saw an improvement of 12% to 33% in forecast accuracy, which could lead to more reliable solar energy predictions.
“These advancements are important because they can help us better predict solar energy availability under cloud conditions,” added Liu. “As solar power becomes a bigger part of the energy grid, having more accurate forecasts will help optimize how we use solar energy.”
The researchers are looking for ways to further refine these models by upgrading cloud-radiation physics formulation and using advanced machine learning systems to improve predictions. They hope that by deepening our understanding of the fundamental physical processes governing solar radiation, they can make solar forecasting more accurate and reliable.
Ankita Nagpal
Communications Assistant