Information-Theoretic Modeling of Wildfire size Using Spatiotemporal Land Cover Data

This study employs advanced geospatial analysis and information-theoretic modeling to evaluate the relationship between multi-scale vegetation patterns, temporal change, and wildfire size over 19 ecoregions in the western United States. More specifically, the study quantifies and extends established understanding about the relationship between vegetation patterns and fire behavior while demonstrating how information-theoretic modeling and GIS data can provide quantitative insights into complex spatiotemporal relationships. Using a custom-developed concentric ring buffer aggregation approach, we extracted vegetation characteristics from National Land Cover Database (NLCD) rasters at three nested spatial scales around the 2015 to 2020 wildfire ignition points: an inner ring (0-90m from ignition), middle ring (90-150m from ignition), and outer ring (150-210m from ignition) at the 30m resolution of the NLCD data. The entire yearly land cover raster catalog from 1985 to 2020 was sampled using this algorithm in five year increments. This results in a 30-year record for each fire, consisting of the most common land cover in each ring, for each 5 year increment, for a total of 7 time samples per fire. Categorical data from these spatiotemporal lags were analyzed using Variable-Based Reconstructability Analysis (VB-RA), an information-theoretic modeling framework that identifies statistically significant structural relationships among categorical variables. Our analysis reveals that vegetation that has developed for 15 to 30 years is a strong predictor of larger fires, as measured by the overall size of final destruction, not by other metrics such as destruction of buildings or severity. Grassland and shrubland dominance, particularly older, well-established vegetation, were consistently associated with large fires, The temporal consistency of vegetation patterns from 15 to 30 years appears particularly important, suggesting that stable grassland and shrubland configurations are highly predictive of large-sized fire events.