Changes in wildfire occurrence and risk to homes from 1990 through 2019 in the Southern Rocky Mountains, USA

Abstract

Abstract Wildfires and housing development have increased since the 1990s, presenting unique challenges for wildfire management. However, it is unclear how the relative influences of housing growth and changing wildfire occurrence have altered risk to homes, or the potential for wildfire to threaten homes. We used a random forests model to predict burn probability in relation to weather variables at 1‐km resolution and monthly intervals from 1990 through 2019 in the Southern Rocky Mountains ecoregion. We quantified risk by combining the predicted burn probabilities with decadal housing density. We then compared the predicted burn probabilities and risk across the study area with observed values and quantified trends. Finally, we evaluated how housing growth and changes in burn probability influenced risk individually and combined. Fires burned 9055 km2 and exposed more than 8500 homes from 1990 to 2019. Observed burned area increased 632% from the 1990s to the 2000s, which combined with housing growth, resulted in a 1342% increase in homes exposed. Increases continued in the 2010s but at lower rates; burned area by 65% and exposure by 32%. The random forests model had excellent fit and high correlation with observations (AUC = 0.88 and r = 0.9). Observed values were within the 95% uncertainty interval for all years except 2016 (burned area) and 2000 (exposure). However, our model overpredicted in years with low observed burned area and underpredicted in years with high observed burned area. Overpredictions in risk resulted in lower rates of change in predicted risk compared with change in observed exposure. Increases in risk between the 1990s and 2000s were primarily due to warmer and drier weather conditions and secondarily because of housing growth. However, increases between the 2000s and 2010s were primarily due to housing growth. Our modeling approach identifies spatial and temporal patterns of wildfire potential and risk, which is critical information to guide decision‐making. Because the drivers behind risk shift over time, strategies to mitigate risk may need to account for multiple drivers simultaneously.

Keywords

• exposure
• hazard
• machine learning
• modeling
• risk
• wildland fire