UNIVERSITY PARK, Pa. — Destructive wildfires in California cause the loss of homes, businesses and human life each year. A Penn State statistics professor is working to help mitigate the effects of these natural disasters by applying complex statistical models to better understand the likeliness that extreme forest fires will develop.

Assistant Professor of Statistics Benjamin Shaby, a Penn State Institute for CyberScience co-hire, has been working with California Department of Forestry and Fire Protection (Cal Fire) since 2014. The collaboration began in 2009, shortly after strong winds knocked over telephone poles, leading to several destructive wildfires. Cal Fire needed to know where utility companies should install stronger poles that wouldn’t blow down as easily, and brought in a team of fire experts to pinpoint the locations.

“That’s where I came in,” said Shaby. “I am not a fire risk expert; I am a statistician. What I did was look at historical weather data to try to determine where the weather would be extreme and where fire risk was most likely, geographically speaking.”

To begin to understand and interpret all of the information surrounding fire risk, Shaby focused on risk due to weather, critical factors in wind speed and wind direction, dryness, relative humidity, precipitation and temperature. His primary data source was output from a high-resolution numerical weather model. He wanted to understand the weather variables over the entire state of California, so they compiled the data from specific points, obtained from wind gauges, thermometers and observation locations.

“The idea was to use the numerical weather model to assimilate all the observations from these point locations to get a fuller picture of the weather,” Shaby said. “This was done by replicating the weather from over an entire decade.”

Dave Sapsis, research program specialist at Cal Fire, said that Shaby presented novel inputs into assessing rare, significant fire and wind events that are crucial for understanding fire risks.

The nitty gritty

Rare events such as these large, destructive fires are hard to study, difficult to model, and behave in fundamentally different ways than normal events.

Shaby's work is very computationally intensive and cannot run on an average computer, so Shaby uses the Institute for CyberScience Advanced CyberInfrastructure (ICS-ACI), Penn State’s high-performance research cloud. Shaby and his students write a code, determine the code’s time complexity, debug it and then fit it to their data. On the ICS-ACI supercomputer, it takes a day or two to fit this model on a single data set, whereas a regular computer could take up to a month.

To test these models, Shaby and his students create “fake” datasets, using a dataset that has an answer they already know; in this way they are able to compare this correct answer with the answer they receive. They run thousands of these simulations to ensure that the models are working properly and giving them the correct answer. ICS-ACI allows these kinds of tests to run easily and quickly.

By using ICS-ACI, Shaby is able to study extreme events. He is particularly interested in the “dependence” of these events, which is taking information from the most extreme winds and trying to figured out how long they will last or how much area the winds can cover.

“In a nutshell, that is the game we play with dependence and extreme events,” Shaby said.

Tricky solutions

Estimating fire risk requires analyzing many factors, including the wind variables, heat variables and more. Shaby says that scientists are starting to have good models for a single variable, but he hopes to develop an effective way to model all variables at the same time.

“If we can do this, then we want to know how can we translate this into risk analysis,” Shaby said. “We really haven't thought about how to tackle this in a systematic way. Understanding the probabilities of these extreme events is only the first part of the puzzle.”

That’s not the only issue his team is trying to understand — they also are trying to figure out the downstream impact on infrastructure, human life and the surrounding areas. For example, the implications of a fire happening in a heavily populated area will vary from a fire occurring in the mountains, with little to no population.

To tackle these tricky problems, the team has have been looking at additional risk factors, such as winds — for example, hot, dry and fast winds in the Santa Ana region of California move south, toward areas of greater population. One of Shaby’s students, Mauricio Nascimento, is looking at wind gauges to study wind speed, strength and duration.

Shaby said studying winds is important for two reasons: because of knocking over power poles and starting fires, and because the strong winds can then spread the fire across the state. If the winds last for a long time, then there is a greater chance of fire spread.

Building the field

Shaby and his students’ involvement with Cal Fire will continue.

“Cal Fire has a very robust research staff that really wants to understand risks,” Shaby said. “Their motivations are closely aligned with mine. I'm looking for substantive problems where I can apply these mathematical models.”

Sapsis said that while downscaling techniques for weather forecasting have been in place for a while, only recently has his Cal Fire team started to address critical estimates of low-probability wind events that are the conditional requirement for seeing wholesale, house-to-house fire spread in dense, urbanized areas.

“Ben has helped us fit scant data and provided error estimates such that we can have an idea on the confidence of those estimates,” Sapsis said.

Shaby, through this continuing collaboration, wants to have a better understanding of extreme events to get to the root of risks and to advance this field.

“We’re trying to come up with statistical methods that are really good at understanding rare events and the dependence in the rare events,” Shaby said. “This field is in its infancy, so we're starting to get okay at it, but I think there's a long way to go.”