In the summer of 2023, an unusual phenomenon swept across the northeastern United States. Smoke from Canadian wildfires blanketed the sky, casting an eerie haze over cities and towns. While the immediate impacts on air quality and visibility were stark, the effects on solar energy production were equally significant, albeit less visible. This event highlighted a critical gap in the predictive capabilities of solar energy forecasts, particularly during severe wildfire conditions. Now, a groundbreaking study published in Environmental Research Letters, the English translation of the journal name, offers a solution that could revolutionize how we predict and manage solar energy output during such events.
Fenya Bartram, a researcher from the Department of Biological and Environmental Engineering at Cornell University, led a team that developed innovative machine learning models to improve the accuracy of solar photovoltaic (PV) generation forecasts during periods of heavy wildfire smoke. The study, which focuses on the northeastern U.S., comes at a time when the region is increasingly reliant on solar energy. The findings could have far-reaching implications for the energy sector, ensuring the reliability of power grids with high penetration of solar energy.
The summer of 2023 was a wake-up call for the energy industry. As smoke from Canadian wildfires spread, the New York Independent System Operator (NYISO) found its day-ahead forecasts for solar PV output significantly overpredicted. This discrepancy underscored the need for more accurate predictive models, especially during extreme weather events. “The challenge lies in the unpredictability of wildfires and the complex interplay between smoke and solar irradiance,” Bartram explained. “Our models aim to bridge this gap by leveraging advanced data products and machine learning techniques.”
The research team’s approach is twofold. First, they utilized data from the high-resolution rapid refresh smoke (HRRR-Smoke) weather forecasting system, which provides detailed predictions of aerosol optical depth (AOD) and downward shortwave radiation flux. This is the first time the HRRR-Smoke wildfire AOD product has been used in solar electricity forecasts, marking a significant advancement in the field. “The HRRR-Smoke data offers unprecedented granularity, allowing us to capture the nuanced impacts of wildfire smoke on solar irradiance,” Bartram noted.
Second, the team employed upsampling strategies to address the data imbalance issues inherent in wildfire events. Wildfires are infrequent but have a substantial impact when they occur. By enhancing the representation of these events in their models, the researchers achieved a remarkable R^2 value of up to 0.85 for severe wildfire periods, significantly outperforming NYISO’s R^2 value of 0.50 across six load zones.
The implications for the energy sector are profound. As solar energy becomes an increasingly vital component of the grid, the ability to accurately predict output during adverse conditions is crucial. “Our methodology can be readily adopted by power system operators to enhance predictions of solar electricity production during periods of wildfire smoke,” Bartram said. “This ensures the reliability of power grids and supports the integration of more renewable energy sources.”
The study published in Environmental Research Letters, not only addresses an immediate need but also paves the way for future developments in solar energy forecasting. As climate change continues to exacerbate extreme weather events, the ability to predict and manage their impacts on energy production will be paramount. Bartram’s work offers a blueprint for how machine learning and advanced data products can be harnessed to meet this challenge.
For energy professionals, the takeaway is clear: embracing innovative technologies and data-driven approaches is essential for building a resilient and sustainable energy future. As the northeastern U.S. and other regions grapple with the realities of climate change, the insights from this research will be invaluable in navigating the complexities of solar energy generation. The future of renewable energy may be bright, but it will also be challenging. With tools like those developed by Bartram and her team, the energy sector is better equipped to face whatever comes its way.
