This repository contains code for statistical analysis of wildfires in California (CalFire) from 2004--2024.
Californias Department of Forestry and Fire Protection provides data on historical wildfires available via arcgis-server. The dataset is filtered for the time period 2004--2024 and for 'OBJECTIVE' 'Suppression (Wildfire)'.
Meterological data is obtained from PRISM-Group at Orgegon State University. The variables considered are precipitation(ppt), maximal air temperature(tmax) and vapor pressure deficit(vpd).
Making use of satellite images provided by NASA(MODIS) LAI (Leaf Area Index), a measure of vegitational/biomass coverage and EVI (Enhanced Vegitational Index), a measure of fuel potential are included.
Other covariates include Elevation, obtained from NASAs DEM and derived from this aspect and slope. To take human activity(arson) into account pixelwise road density from californias road network was constructed.
The aim is to setup three consecutive point process models: a baseline inhomogeneous Poisson Process, a Log-Gaussian-Cox-Process(LGCP) and a Hawkes Process to model wildfire intensity. The statistical model framework is a bayesian hierarchical structure that is estimated using INLA.