Publication Date

Fall 2022

Degree Type

Thesis

Degree Name

Master of Science (MS)

Department

Meteorology and Climate Science

Advisor

Patrick T. Brown

Subject Areas

Meteorology

Abstract

Wildfires have become a major environmental, social, and economic problem in California. The consequences can be especially detrimental when they exhibit behavior like very large daily growth (an individual fire burning >10,000 acres over a 24-hour period). Environmental conditions influencing the risk of large daily growth include weather variables such as temperature, wind, relative humidity, and precipitation; fuel variables such as type, loading, availability, and moisture content; as well as topographic variables such as slope, aspect, elevation, and shape. However, there remains great uncertainty in the importance of these variables relative to each other and the existence of any threshold values in these variables. Our study applied random forest modeling using multivariate and high spatiotemporal data for 16,013 wildfire days in California from 2003 to 2020 to determine feature importance for the task of predicting whether a fire would burn >10,000 acres over a 24-hour period. Shapely Additive Explanations indicate that 100-hour dead fuel moisture, maximum daily air temperature, and soil moisture provide the highest predictive power for large daily growth. Additionally, our study identifies thresholds where the probability of large daily growth significantly increases. These thresholds include a 100-hour dead fuel moisture value of <10%, a maximum air temperature of >75 F, and a 0-10 cm soil moisture of <12%. Finally, we establish the number of days per year that these thresholds are being crossed has increased substantially over the last four decades.

Share

COinS