Hopkins Assistant Professor Seth Guikema and a geographer at Texas A&M University led a team that predicted the power outages that Sandy caused within 15 percent. Their model, based on weather predictions, real-time data and data from 11 previous hurricanes, could help companies and emergency-response teams better prepare for hurricanes in the future.
The researchers predicted 10 million people would be without power while the storm was still over the Bahamas. They ultimately adjusted this estimate to 8-10 million. These results were close to the 8.5 million people without power, as released from the Department of Energy on Oct. 31.
“This estimate days before landfall provides a basis for utility companies and government agencies to better plan their response to the storm in terms of resources and personnel,” Guikema said.
Guikema is leading the research along with Texas A&M geography professor, Steven Quiring. They have been working on this model since 2006 and plan to continue developing it.
“I'm not sure the model will ever really be completed,” Guikema said. “We are constantly trying to improve the predictive accuracy of the model to better support emergency response and utility response planning.”
Guikema has been at Hopkins since 2008.
“JHU has provided a very strongly interdisciplinary place for this interdisciplinary research, and the students here have been very strong,” Guikema said.
Guikema also works in industrial economics, risk management and planning for the University of Stavanger in Norway.
Guikema’s interest in this type of modeling started as a postdoc at Cornell. He then went on to start at Texas A&M in 2005, where he met Professor Quiring. They started working on a model after a utility company approached them about developing a power outage forecast system. Guikema and Quiring continue to receive funding from the company in return for letting the company use the models.
The model, which can predict outages three to five days in advance, is based on data from 11 historic hurricanes. This data, consisting of power outages in small grid cells, was used to train a statistical model to make future outage predictions given a wind field estimated from Quiring’s wind field model.
“Our goal is to have a model that is useful in practice and helps society be better prepared for and respond to hurricanes,” Guikema said.
The team initially worked with data from the sponsoring utility company; however, since 2011 they have been working with publicly available data to create a model that can predict outages all along the US coastline. Guikema and Quiring are still working to improve this later model and hope to add factors such as soil moisture, which are included in the utility model.
This broad-area model was first tested with Irene in 2011, but the results from this model were not released until the predictions made for Sandy last week. These predictions were within eight percent of the Department of Energy’s predictions of the fraction of customers without power in New York, Rhode Island, Virginia, Massachusetts and Pennsylvania. The model overestimated the number of people without power in Maryland and Delaware because of the lower-than-expected winds.
The researchers are still working to assess why the model underestimated outages in Connecticut and New Jersey.
