Ensemble modeling of the climate-energy nexus for renewable energy generation across multiple US states

Abstract The effects of climate change on renewable energy generation are of growing concern, as shifts in weather patterns and extreme events can significantly impact energy production. This study aims to leverage machine learning models to predict renewable energy generation based on the surrounding climate. We analyze data from four key states: California, New York, Florida, and Georgia, and focus on three critical renewable energy sources: hydroelectric, solar, and wind power. To determine the optimal model, we test six primary machine learning techniques, as well as an ensemble and a mean-only baseline. The results indicate that the ensemble approach improves the predictive accuracy of the model. Using this ensemble, we projected the changes to climate-sensitive portion of the renewable energy generation under climate change. Our results indicated that there was a wide variation of possible futures, depending on the state, source, and season. For example, the model projected a reduction in California’s monthly total renewable energy generation in the summer by 0.5%, or about 30 000 MWh, under SSP5-8.5, the worst-case scenario, but an increase of 0.5%, or about 25 000 MWh, in New York’s monthly total summer renewable energy generation under the same scenario. The modeling techniques detailed in this study can be applied across new regions, sources, or time periods. Ultimately, by understanding the influence of climate on renewable energy generation, we can improve the long-term planning process for the electricity grid, while building resilience and ensuring sustainable climate change mitigation and adaptation.