Methods

Our study area was Oregon and Washington National Forest lands. We obtained historic Federal wildfire occurrence data (1980-2004) from Desert Research Institute (DRI), http://www.dri.edu. The DRI fire occurrence data are based on the National Interagency Fire Management Integrated Database (NIFMID) at the USDA Forest Service National Information Technology Center in Kansas City, Missouri. The data included information on the date, location, and size of the fire. The DRI version of the data had been subjected to a quality control procedure in which each fire occurrence record was flagged as usable or otherwise (Brown and others 2002). Fire locations are the latitude and longitude of the fire when first discovered.

We obtained average monthly temperature and Palmer Drought Severity Indices (PDSI) from U.S. Climate Divisions (NCDC 1994). The values at the climate-division level were projected onto a 1-km² grid to provide a monthly climate record for each grid cell. Variables at the climate division level may not be good estimates for local weather conditions, but they were used here to demonstrate how weather variables can be included in the model.

Spatial data on insect activity was obtained from aerial sketch maps created by the Forest Service Forest Health Protection program (http://www.fs.fed.us/r6/nr/fid). In particular, we obtained data on: (1) total number of trees (lodgepole pine, Pinus contorta, and ponderosa pine, P. ponderosa) killed by two bark beetle species (mountain pine beetle, Dendroctonus ponderosae, and western pine beetle, D. brevicomis) per year (1980-2004) in each km² grid of Forest Service lands (Region 6); and (2) total area defoliated by western spruce budworm (Choristoneura occidentalis) in each km² of Forest Service land per year.

In this study we did not use information on fire boundary. Consequently, the models described below are used for estimating associations between histories of fire sizes and insect infestations within 1 km of each other. Additional predictors, (e.g., distance to nearest infestation or fire) may improve estimation results. In this study, we are mainly concerned with describing the statistical framework given a set of predictors.