As the climate crisis intensifies, switching to renewable energy remains a critical piece of the solution to ensure rapid decarbonization. However, renewable energy generation is highly reliant on the ambient environmental conditions, making it difficult to predict output in the short-term, as well as determine the optimal locations of sites in the long-term, given climate change is likely going to impact local environmental conditions. Accounting for the impact of climate change is particularly difficult, as there remains uncertainty related to the magnitude of climate change within the mid- and long-term in addition to the relatively unknown impacts of climate change on generation capacity of renewable energy technologies, particularly solar. In this work, we aim to fill this gap by leveraging machine learning to investigate the impact of climate change and associated uncertainty on solar power generation. Using data from the National Renewable Energy Laboratory (NREL), we investigate the solar power output and the predictability of that output under climate change across 10 different solar panel technologies. Our goal is to answer the questions: (a) How will climate change impact solar power generation; and (b) Do those impacts differ across panel technology? To do this, we will leverage a complex neural network to predict the optimized output of various solar panel technologies, given the climatic conditions in the surrounding area. We expect that the optimal panel for total power output will change over different climate scenarios. Further, this panel might not be the most predictable (e.g., have the least error in our model), which would create issues of uncertainty for developers. To test these hypotheses, we will present results from a study focused on Puerto Rico. Puerto Rico represents a smaller test site, yet the island has experienced a multitude of climate-related disasters recently that have devastated their energy system. As the island looks to rebuild, decentralized renewable energy has been investigated as possible option for a more resilient system under climate change. This presents an opportunity to investigate how solar power output could evolve over time on the island and which panels might be optimal, given their unique climate conditions. It will also provide a baseline for extending the modeling framework to other regions in the US. Ultimately, the results will provide critical insights into the sustainability of solar power and the viability of certain technologies over the long-term.

Wildfires are a significant threat in California, burning more than 1.7 million acres per year and costing the state billions of dollars. In addition to directly damaging homes and infrastructure, fires can destabilize soil, leading to debris flows when significant precipitation events occur in recently burned areas. Electric utilities own and operate infrastructure located in areas that are vulnerable to both wildfires and post-fire debris flows. While in recent years there has been a focus on understanding the risks of wildfire ignition caused by power infrastructure and identifying the responsible electric utilities to mitigate these risks, there has been little work to understand the vulnerability of utilities themselves to wildfires and post-fire debris flows. This is partly due to the complex nature of these hazards, defined by multiple disciplines and dynamics. In this work we assess the vulnerability of transmission lines, electric substations, and power generation facilities in California to wildfires and post-fire debris flows. We assess wildfire risk by overlaying geospatial power infrastructure data with wildfire probability from Cal Adapt and post-fire debris flow threat levels from recent modeling efforts. We assess risk in today’s climate and in the future. To understand uncertainty in future climate impacts, we use two climate models: the Canadian Earth Systems Model, which produces an average prediction of future climate and the Hadley Centre Global Environment Model, which produces a warmer, drier estimate. In conjunction with both climate models, we use two representative concentration pathways (RCPs): RCP 4.5, representing a future in which greenhouse gas emissions begin to decrease in the mid-21st century, and RCP 8.5, in which emissions increase through the end of the century. We assess risks at the state level and identify vulnerable electric utility companies. We find that under current conditions, electric utility assets in Northern California are most vulnerable, being located in areas with up to 40% fire probability compared to the state average of roughly 10%, and high risk for post-fire debris flow. Under future conditions, we find that fire risk to assets may increase substantially in the Sierra Nevada and Northern Coast regions, and post-fire debris flow risk may increase substantially in the coastal ranges and North Central California. However, there is large uncertainty in future risks across climate scenarios. While many electric utility companies have primarily low-risk power infrastructure assets, we find that some smaller utilities may be particularly vulnerable due to the majority of their transmission lines and substations being located in high risk areas. Power generation is also vulnerable to wildfire and post-fire debris flow, with geothermal, hydro, and nuclear power plants in the state facing the highest risks in current and future climate scenarios. These results provide a basis for decision-making around the allocation of resources for infrastructure resilience to wildfire impacts.