The University of Connecticut has partnered with Eversource, the state’s lead energy provider, in a research effort to predict extreme weather events and their effects on energy systems and aquatic environments.
Israt Jahan, a Ph.D. student studying environmental engineering, is one of the researchers working on the project alongside Dr. Marina Astitha, an associate professor of civil and environmental engineering at UConn. Their research focuses on improving the predictability of wind gusts using machine learning and deep learning with an artificial intelligence (AI) model.
“I have developed hybrid models that integrate ML [machine learning] and DL [deep learning] approaches with physics-based numerical weather prediction [NWP] outputs to improve prediction accuracy and reliability,” Jahan said. “I have also applied various explainable AI techniques to demonstrate the drivers behind the AI model predictions and uncertainties for better interpretability and transparency. Currently, I am analyzing severe wind events across the contiguous United States under both historical climate and pseudo–global warming conditions to assess how extreme wind hazards may evolve in a warming climate.”
The use of AI was not initially proposed in the research program until 2016. Astitha’s work was founded in physical modeling. However, with the introduction of AI, there have not been many challenges with capturing nonlinear relationships and uncertainties in physics-based models, according to Astitha.
She’s found herself integrating AI in all areas of her research on extreme weather prediction, renewable energy climate assessment and aquatic ecosystem health.
“We use AI to reduce forecast bias, quantify uncertainty, fuse multiple data streams and build hybrid physics-ML models that outperform either approach on its own,” Astitha said. “And increasingly, we’re developing trustworthy AI frameworks that emphasize interpretability and reliability, which is essential when predictions influence engineering decisions or emergency operations.”
The intelligence has not only cut down the computational time but has also shown a decrease in prediction error in comparison to a NWP model, according to Jahan.
“Overall, AI expands what is scientifically and operationally possible. It doesn’t replace physical modeling; it enhances it, filling gaps and increasing trustworthiness,” Astitha said.
Center. Photo courtesy of Peter Morenus/UConn Photo, UConn Today
This approach in Astitha’s research group unites physics-based models with AI to improve extreme weather predictions.
“I lead the Atmospheric and Air Quality Modeling Group, where we blend physics-based atmospheric models with state-of-the-art AI and machine learning to improve forecasts of high-impact events such as windstorms, snowstorms, and tropical cyclones,” Astitha said. “These storms have enormous societal consequences, from power outages to infrastructure damage, so improving prediction is an urgent need.”
The group is currently developing the next generation of AI frameworks that combine physical knowledge with machine learning to capture storm dynamics, according to Astitha.
Along with these models, they are also determining the impact climate change will have on wind resources and the offshore wind industry by using “high-resolution modeling and downscaled climate projections to understand how hub-height winds will change over time, which directly informs energy planning in the region,” Astitha said.
She continued to say that the group had also built physics-guided ML systems for freshwater ecosystems, integrating “weather, hydrology, air quality and fertilizer-application data to forecast chlorophyll-a, phosphorus and dissolved oxygen dynamics in lakes.”
Astitha began working with Eversource Energy at UConn in 2015, prior to the inauguration of the Eversource Energy Center.
This program was created to “bring academic research and utility expertise together to reduce storm-related power outages and improve emergency preparedness,” according to Astitha.
Together, forecasting methodologies, data sharing, modeling infrastructure and translations of research innovations into operational tools used are created.
The research group led by Astitha “focuses on improving the predictions of extreme weather events using advanced modeling and machine learning approaches,” Jahan said. “My specific role in this collaboration has been to develop and validate machine learning (ML) and deep learning (DL) models for wind gust predictions, ensuring that they outperform benchmark Numerical Weather Prediction (NWP) models.”
Astitha and Jahan aren’t the only ones conducting weather forecast research. They said a second research group is running a power outage prediction model that inputs various weather variables to determine the number of power outages there may be in Eversource’s service area with upcoming storms.
“The goal overall is to provide Eversource with more accurate power outage predictions by improving extreme weather forecasts to support operational preparedness and outage restoration efforts,” Jahan said.
This long-standing partnership has grown through UConn’s research programs but also through support from outside organizations, according to Asitha.
“This work has also catalyzed broader support from NSF’s IUCRC program and DOE, enabling us to expand the research to climate resilience, renewable energy, and next-generation AI systems,” Asitha said.