UCLA

Spatio-temporal point processes are a common method to analyze data that involves event occurrences in space and time, such as wildfires. Model parameters for a point process are typically fitted using maximum likelihood estimation, which finds parameter values that maximize the probability of observing the data according to the specified model. This method, however, often involves finding a complex and non-closed-form integral. The Stoyan-Grabarnik (SG) statistic is a way to find model parameters for a spatial point process that is faster and easier than maximum likelihood estimation and does not require computing or approximating a computationally intensive integral. This work uses the SG statistic methodology to estimate model parameters for forest fire ignitions occurring in National Forest System lands in California between 2008-2012. The models utilize covariates such as precipitation, wind speed, temperature, and evaporation and are evaluated for a variety of subsets of the data, including size and cause over northern and southern California. The results show that modeling accuracy is not compromised while also revealing interesting patterns in the relationship between fire ignitions and weather conditions. Results in this work could help advance modeling efficiency and provide insights pertinent to fire risk management.