These two new forecast tools augment the information provided by Predictive Services, by giving a quantitative forecast of the probability a fire that ignites today will subsequently grow to significant size, and by forecasting the number of significant fire events during the upcoming week. Below, the two new forecast tools are described. These forecasts may be combined with the 7-Day Significant Fire Potential Outlook issued by Predictive Services to gain additional information about the likelihood of a large number of significant fires.
Map of probability of a significant fire given ignition
Probability of an ignition becoming a significant fire event, based on today’s ERC values and historic fire occurrence, as well as topography, vegetation, and human population factors. The cool colors (grey and blue) show areas where the likelihood of a significant fire starting today is low (less than 10%), and the warm colors (red) show areas where the likelihood of a significant fire is high (greater than 20%).
The map together with the color-coded boxplots is a useful way of visualizing forecasted levels and expected uncertainties around the forecasts. For example, PSAs with forecasted level in the orange category (15-20% chance of a significant fire starting today) would expect the variability around this average over various years and GACCs to be between 0 - 40% with an interquartile range between 6 – 20 %. The boxplots are based on data on fire occurrence and size during years of the study (2007-2012).
Map of forecasted number of significant fire events
Forecasted number of significant fire events for each PSA during the upcoming week, based on today’s ERC values and historic fire occurrence, as well as topography, vegetation, and human population factors. The fire size threshold at which a fire is considered significant depends on the PSA and ranges between 50 – 1000 acres (refer to Figure S2 in our paper). Cool colors show areas where the forecasted number of significant fire events during the upcoming week is low (where the map is grey or blue, the expected number is less than 1), and warm colors show areas where the expected number is high (with red signifying that greater than or equal to 10 fires are expected).
While the map gives a categorical information about the forecasted number of fires for each PSA, additional information about the expected number of fires can be drawn from the boxplots of the historic distributions of significant fire events for each risk level based on 6 years of fire data (2007 -2012). For example, PSAs with a red forecast (3-10 significant fires) should expect more than 50% chance of having at least one significant fire and 20% chance of at least 5 significant fires, based on the quantiles of the of the empirical distributions of events during the study years. On the other hand, PSAs coded at the lowest grey level (0 – 1.1 fires) should expect less than 5% chance of getting one significant fire event. Note that the variability (uncertainty) around the forecasts (given by the boxplots) are quite large. Therefore, it is important to consider both the map of forecasts and the distributions (as given by the boxplots) when using these maps.
Combining the probabilistic forecasts with SFPO forecasts
The overall percentage of outliers (more significant events than expected) in the boxplots for number of significant fires during the upcoming week was about 6% (reference to paper for definition of outlier). The table below gives the percentage of outliers at the various SFPO forecasts levels. For example, while managers could expect 6% chance of an outlier on a given week, if the SFPO forecast is level{4}, lightning, then the chance of an outlier (in boxplots for number of significant events) increases to 28%. In other words, if the SFPO forecasts lightning, then the chance of a large number of significant fires increases.
SFPO levels* {1} {2} {3} {4} {5,6,7,9}
Probability of outlier 0.03 0.09 0.17 0.28 0.16
* See (reference to our table 1) for definitions of SFPO levels.
|