Probability based 7-day fire potential forecasts

Fire managers at the Geographic Coordination Centers (GACC) across the United States use decision support tools from Predictive Services to assist with resource allocation and mobilization. Detailed information on when and where the greatest resource need is required to successfully mobilize resources within and between GACCs. To this end Predictive Services produces a 7-Day Significant Fire Potential Outlook. This product combines environmental conditions (fuel dryness), with weather conditions and expert knowledge to identify high risk days and location. The product is qualitative in that it issues low, average or high fire risk days but it does not forecast probabilities per se or expected numbers of large fire events. As technology advances however, more objective tools are being pursued to assist with these types of decisions. The present study uses historic fire occurrence data to quantify relationships between the frequency (probability) of large fire occurrence and historic environmental and weather conditions. The estimated probability models are then used to produce 7-day forecasts of probabilities of large fire occurrences using forecasted fire weather and fire danger indices. The probabilities are spatially and temporally explicit on a 1km x 1km x 1-day scale and use logistic regression with nonparametric spline functions to describe the potentially none linear relationships between the indices and fire probabilities. Expected large fire sizes are also estimated, using the Generalized Pareto distribution, the latter distribution having been found useful for modeling extreme events. One of the benefits of a probability based model is the ability to study the skill and utility of the product by comparing observed frequencies relative to those predicted by the model. This is not easily accomplished with the qualitative forecasts of fire risk levels. Another advantage of the probability models is the ability to study departure from observed data (residual analysis) and consequently to be able to introduce model improvements. For example, we have found our models do not give good estimates when there is a larger than average level of clustered lightning events. The model may be improved by including an additional explanatory variable describing forecasted lightning potential. Presently, 7-day forecasts of probabilities and numbers of large fires are being produced and tested for the Southern California GACC using forecasted fire weather and fire danger indices developed at the Dessert Research Institute (DRI). DRI uses linear regression equations to project National Fire Danger Rating System (NFDRS) indices such as Energy Release Component (ERC), Ten- and Hundred-Hour dead fuel moistures, as well as other variables out to ten days using the Global Forecast System (GFS) numerical weather model.