The first analysis to be presented involved Oklahoma wildfire data from 2000 to 2012 obtained from the Office of the State Fire Marshal and analyzed with respect to weather and soil moisture data. Over 38,000 wildfires were involved and grouped according to growing (May through October) or dormant (November through April) season as well as five fire size classes (ranging from <=10 acres to >= 1000 acres burned). Weather and soil moisture data were obtained from the Oklahoma Mesonet, the state's automated weather station network of 120 sites. This analysis considered individual wildfires using the weather and soil moisture data from the start day of each fire (using the nearest Mesonet station to the fire location). Variables included were minimum relative humidity, maximum wind speed, maximum temperature, and soil moisture (expressed as fractional plant available water, scaled from 0 to 1, throughout a 40-cm soil column). The distribution of each weather/soil variable within a given fire size class and season was analyzed. Results showed the medians of maximum temperature and wind speed monotonically increased with increasing fire size class regardless of season. Medians of minimum relative humidity and fractional available water (FAW) monotonically decreased in value with increasing fire size class regardless of season. The most noteworthy finding was that 100% of large growing season wildfires (>= 1000 acres) occurred at FAW < 0.5 with 86% of them occurring at FAW < 0.2. Thus low values of FAW are seen as good predictors of large growing season wildfires.
The remaining analyses utilized a different wildfire database compiled by Karen Short of the Missoula Fire Sciences Laboratory. We extracted Oklahoma fires over the same 13-year period and working with Karen, added 111 wildfires >= 1000 acres from the Oklahoma Fire Marshal database that were deemed valid and not in the Short database. This combined database contains almost 26,000 wildfires. The next analyses to be discussed involved only large Oklahoma wildfires (those >= 1000 acres); there were 501 such fires in the database. We also investigated only two variables: FAW and 1000-h dead fuel moisture, the latter calculated as part of the operational fire danger model in OK-FIRE (http://okfire.mesonet.org), Oklahoma's wildland fire management system.
The second analysis looked at total number of large fires and total acres burned in each growing and dormant season through the 2000 to 2012 period. The 1000-h dead fuel moisture (DFM1000) and FAW values used in this analysis were the averages of the values for individual fires on their ignition dates throughout the given season. DFM1000 and FAW values for each fire were from the closest Mesonet station to the fire locations. Scatterplots for both numbers of growing season fires and total acres burned during the growing season showed highest values of each associated with FAW < 0.2 and DFM1000 < 8%.
The third analysis looked at acres burned from each individual large fire. DFM1000 and FAW values for each fire were taken from the closest Mesonet stations to the individual fires on their ignition dates. Similar to the last analysis, scatterplots for growing season large fires showed higher acres burned associated with FAW < 0.2 and DFM1000 < 8%.
The last analysis was a stepwise logistic regression looking at the probability of large fire occurrence (yes or no) somewhere in the state on each day of all growing seasons and each day of all dormant seasons during the 2000 to 2012 period. Only the growing season results will be presented here. Statewide average (all Mesonet sites) values for weather variables, dead fuel moisture, and soil moisture were calculated for each day of the period. The results of the logistic regression indicated the four most important variables were minimum relative humidity (RH), maximum wind speed (WS), maximum air temperature, and FAW seven days prior (or DFM1000 seven days prior). Plots of growing-season large fire probability versus FAW 7 days at various levels of RH and WS showed greatly enhanced probabilities of large wildfire occurrence at low FAW values given appropriate weather conditions. Similar plots of fire probability versus DFM1000 7 days showed greatly enhanced probabilities of a large growing season wildfire occurring as DFM1000 values decrease.
With soil moisture networks increasing in spatial density across the country, the above results indicate that soil moisture is a potentially good predictor of fire danger during certain times of the year and should be considered for inclusion in fire research studies as well as in operational fire danger systems locally (such as OK-FIRE in Oklahoma) and nationally.