Modeling the evolution of wildfire : an analysis of short term wildfire events and their relationship to meteorological variables

Abstract: Wildfire events are expected to increase with the changing climate; thereby increasing atmospheric, economic, and anthropogenic impacts. Gaining a real-time understanding of the evolution of wildfire events can benefit regional meteorology models, global atmospheric models, and hazard warning systems. As a result, an attempt at modeling the evolution of wildfire was undertaken utilizing 8 years of daily meteorological variables and fire radiative power (FRP) provided by the European Centre for Medium-Range Weather Forecasting from 2003 to 2010. Fire Radiative Power (FRP) measures the rate of radiative energy emitted by active wildfire events across the globe. FRP is observed from MODIS sensors aboard the sun-synchronous Terra and Aqua satellites and provides a unique way to incorporate active wildfire information into climate models and weather forecasting. Observations have shown that the amount of FRP is related to the rate at which fuel is being consumed, linking it directly to the fuel load of ecosystems. The impact of meteorological variables on the behavior of FRP, on a daily basis, is expected to show an observable relationship. Modeling this relationship is the primary objective of the thesis. Meteorological variables and a time-delayed FRP value were established as independent variables for linear regression modeling. The relative change in FRP (ΔFRP) functioned as the dependent variable. Three distinct ecosystems (Equatorial, Warm Temperate, and Boreal) were included to account for vegetation structure and fuel load. Ecosystem selection was performed using the climate based Köppen-Geiger Climate Classification which created approximately homogeneous ecosystems based off of observed temperature and precipitation values. Covariate analysis showed no significant correlation between the independent variables and ΔFRP. Mann-Whitney U Tests identified ecosystems where statistically significant trends were observed and suggested opportunities for successful linear regression modeling. Both linear and non-linear relationships were accounted for in the application of a Bayesian Information Criteria to the linear regression modeling. The linear regression results did not produce a successful model and the impact of meteorological variables on FRP was not observable. This body of work can be improved by incorporating the magnitude of FRP in the calculation of ΔFRP. Additionally, identifying threshold behavior of meteorological variables can improve the identification of significant relationships. Finally, focusing on smaller spatial scales and including actual fuel load values along with anthropogenic mitigation practices is poised to improve the linear regression modeling of the evolution of wildfire.