Approximating wildland fire risk or “fire weather” is a crucial step in preventing or mitigating potentially devastating fires.
In his latest research project, Kevin Granville, an assistant professor of mathematics and statistics, collaborated with the Ontario Ministry of Natural Resources and Forestry to investigate and develop methodology for the interpolation and visualization of fire weather variables in the province, prioritizing both accuracy and realism.
“I wanted to develop a decision support tool for the ministry,” says Dr. Granville.
Environmental observations, or weather data, are needed to calculate fire weather. Data such as temperature, humidity, wind speed, and precipitation are used daily to calculate Canada’s Fire Weather Index.
On the topic of visualization, Granville said “The big thing you want is access to more precise information, but to not have it take longer to read which could slow decision making.”
Interpolated fire weather is mapped using different colours to indicate risk classifications.
“I want to ensure there is still a sharp boundary between classes, while presenting more information within a class,” says Granville.
A unique aspect of Ontario’s weather stations is that there are few in the north compared to the south. Granville wanted to propose a solution that takes into consideration the density of weather station coverage across the province to get the best approximations.
“I went down a big rabbit hole looking at various methods and combinations of methods to determine what worked best while meeting the ministry’s requirements, and hence would be appropriate to recommend for use in practice.”
Granville used 30 years of environmental observations in his study. To measure and compare accuracy between methods, weather stations were left out one at a time while remaining data were used to approximate the Fire Weather Index at that location each day.
“Approximating at that location gives you a measure of performance, because you have a real observation to compare against,” he says.
Granville proposed the use of a “spatial ensemble” that combines multiple methods with mixing weights depending on station density.
“It is a way to balance things, so we maintain the nice aesthetics in the denser part of the province, but we are overall improving the performance in the sparser part of the province as well.”
Granville delves into his methods in the open source article “On the selection of an interpolation method with an application to the Fire Weather Index in Ontario, Canada,” published in the journal Environmetrics in September 2022.