New Tool Predicts Solar Drop in Smoke

By Blake Jackson

Wildfires cause more than just damage to nature, property, and livelihoods. They also affect electricity generation by reducing sunlight, especially in areas using solar panels. Smoke from wildfires can block sunlight across large regions, limiting the performance of photovoltaic (PV) solar systems.

To address this problem, researchers at Cornell University have created a machine learning-based model that predicts how wildfire smoke will impact solar electricity production. This model gives more accurate day-ahead forecasts compared to current systems. With better predictions, power operators can manage electricity supply more efficiently and avoid using expensive backup generators.

Max Zhang, a professor at Cornell Engineering, led the project. He noticed the problem in 2023 when smoke from Canadian wildfires reduced solar power output in the northeastern U.S. While others focused on air quality, Zhang wondered about the energy side of the issue. The standard forecasts by the New York Independent System Operator (NYISO) failed to predict the drop in PV output correctly.

The new model uses data from the National Oceanic and Atmospheric Administration’s High-Resolution Rapid Refresh Smoke (HRRR-Smoke) system. It includes predictions of aerosol impacts and smoke density, which are key to understanding how sunlight is blocked.

To deal with limited local data on wildfire events in New York, the team used a method called "upsampling," which puts extra focus on the rare smoke days during model training. Testing showed that this tool outperformed existing NYISO forecasts, and it works hourly, not just daily, which is a first.

The New York State Energy Research and Development Authority supported the research. The model is fully operational and can be adopted by any power system operator. It helps maintain grid reliability, especially as climate change increases wildfire frequency.

“This is just the start. We are improving the model while creating pathways for adoption by system operators,” said Zhang. “The better the forecast, the more reliable the power system.”

Photo Credit: cornell-university