The researchers found new insights into the dynamics of wildfire probabilities across California. Isaac Park of the University of California, Santa Barbara, and colleagues presented their approach and findings in the November 3, 2021 open-ended journal Plos One. Recent wildfires in California and nearby states suggest the need for a better understanding of the dynamics that determine where and when wildfires occur.
To improve their understanding of these relationships, Park and colleagues used a statistical method known as a generalized additive model to explore and plot annual wildfire probabilities across California from 1970 to 2016. This work builds on previous studies that used the same technique for longer timescales. In this case, the researchers tailored the method for the annual probability by incorporating information about local climate change, human activity, and the time since the last fire event at each site and year—all within a geographical range of 1 kilometer.
This analysis uncovered some new insights into the probability of wildfires in California. For example, researchers have found that local climate and human activities play a key role in determining the probability of wildfires across the state.
In addition, in some environments, the timing since the last fire has had an important impact; so does short-term climate change involving extreme conditions, particularly in southern California's fire-prone bushlands and forests.
The researchers also note that their large-scale, state-wide approach to predicting the probability of wildfires is superior to statistical models developed for certain localized areas. The researchers say this work and further refinements of its modeling methods could prove valuable for a variety of studies and practical applications, such as wildfire emissions and hazard maps for implementing fireproof building codes.
The researchers added: "This study presents a powerful tool for mapping wildfire probabilities across California under various historical climate regimes. By utilizing machine learning methods, it demonstrates the unique contribution of local climate, human development, and previous fire histories to annual wildfire risk in space and time. ”