We predict the expected number of fires for upcoming seasons. The main goal of this project is to estimate the probability of large fires (>200 ha) for one to six months ahead. We produce forecasted probability estimates of large fire (>200 ha) occurrence, odds ratios (normalized by historic risk). We also estimate expected area burned in large fires and suppression costs, but those values are not yet shown in the figures posted here.
The model utilizes the history of climate variables through the month preceding the date each forecast is made. The first forecast is made at the start of April each year, using data observed through March as explanatory variables, with forecasts made at that time for each month from April to September. At the start of each subsequent month, forecasts for the remaining months are updated using the climate observations from the previous month. So, the probability estimates are evaluated for one to six months ahead using history of climate data up to the present. Forecasts for the later part of the fire season can thus change significantly over the season, as they incorporate more recently observed climate. All of the forecast maps are maintained online to allow comparisons as the season progresses.
Retrospective forecasts back to 1980 are made available here as well to allow assessment of the models' forecast skill.
1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
These reliability plots below show the goodness of-fit of the estimated models by plotting observed fractions of (≥200 ha fires) responses against predicted probabilities for 4 different models:
Figure a: Forecasts based on historic model that includes only month of year and site characteristics.
Figure b: Forecasts done end of March using lagged climate data up to end of March.
Figure c: Forecasts done end of April with lagged climate data up to end of april.
Figure c: Forecasts done end of May with lagged climate data up to end of May.
- Dashed lines indicate 95% confidence intervals.
Preisler, H.K., A.L. Westerling, K. M. Gebert, F. Munoz-Arriola: "Spatially Explicit Forecasts of Large Wildland Fire Probability and Suppression Costs for California," International Journal of Wildland Fire, submitted