Latest Forecast
Red Flag Warning Potential
|
||||||||||||||||||||||||||||||||||||||||||||||||
 |
||||||||||||||||||||||||||||||||||||||||||||||||
Background -
A Model to Predict Red Flag Warning Days
*Randall P. Benson, *Greg C. Carbin
ABSTRACT
Red Flag Warnings and Fire Weather Watches are issued by local National Weather Service offices when wildfire danger is high and weather conditions will adversely affect the ability to manage a wildfire if it occurs. The issuance of a Red Flag Warning initiates an increased demand for fire-fighting resources at the federal, state, and/or county level. We have developed a model to predict potential red flag conditions to 10 days based on historical weighted red flag weather conditions and a predicted Energy Release Component (ERC). Critical thresholds for temperature, relative humidity, wind speed, and the potential for dry thunderstorms determine the red flag weather variables. Dry thunderstorm probability is calculated from the probability difference between the likelihood of thunder minus the likelihood of a wetting rain (>0.10 inch). The locations involved in our study include Casper, Wyoming, Rapid City, South Dakota, and Scottsbluff, Nebraska. Output from the Map Analog Retrieval System (MARS) forecast system developed at the Storm Prediction Center (SPC) in Norman, Oklahoma serves as the basis for the predictor variables used in this analysis. MARS is based on the NCEP GFS ensemble mean. MARS uses the mean of the 850mb, 500mb, and precipitable water fields at 0000 UTC to search for 0000 UTC analogs from the North American Regional Reanalysis. Only 3 components go into the MARS prediction system; 500 height gradient, 850 height gradient, and precipitable water. MARS uses the NCEP Reanalysis data with a somewhat course grid resolution of 2.5 degrees. The level of detail is less than what is required to discern typical mesoscale features that contribute to fire behavior, however, synoptically apparent cases may be captured if the ensemble forecast contains good analog fits. To construct the site-specific Red Flag Warning potential, the top 10 analogs or matches are identified from the MARS output and ranked according to Root-Mean-Squared (RMS) error. The red flag probabilities are constructed from the MARS analogs using archived Automated Surface Observation Stations (ASOS) climate data containing the red flag criteria from 1979 to 2004. The MARS RMS error numbers describe the essence of the match between the MARS output and the Reanalysis data. A computer program ingests the twice-daily (00z and 12z) MARS forecasts and user input forecast ERC values and computes the 10-day red flag potential based on a weighted average of the red flag weather and the input ERC percentile rank. Testing and of the model will occur during the 2007 wildfire season and verification will be completed at the conclusion of the coming fire season.
An example calculation is provided to provide additional detail on the prediction method used herein. A 10-day probability of red flag conditions is constructed for each of the three MARS variables (500mb height, 850mb height, and precipitable water) while the final summary red flag probability is an average of the three MARS variable probabilities. To calculate the daily red flag potential for each MARS variable, a weighted system is used for the four red flag ASOS weather variables in the form of the following: 0.35*probability of relative humidity less than 15% + 0.30*probability of dry lightning + 0.25*probability of wind speed >25 mph + 0.10 * probability of temperature > 80 F.
The user input ERCG values are based on GFS 10-day model predicted ERCG values for RAWS locations in closest proximity to Rapid City, Casper, and Scottsbluff. The final daily red flag potential is calculated based on a weighted average in the form of: 0.3 * user input ERCG percentile rank + 0.7* red flag weather probability.
Cautionary Note: As with any forecast, there are inherent uncertainties involving model skill and model bias, and initial analysis error. Since this research method involves a perfect prog approach, it assumes that each of the 10 analogs are perfect matches to the MARS SPC model output. Ultimately, each of the 10 analogs used to construct the daily variable probabilities from archived surface weather observations are not perfect matches.
*Randall P. Benson is the State of South Dakota Fire Meteorologist and also is Adjunct Faculty for the South Dakota School of Mines & Technology.
*Greg C. Carbin is with the NOAA/NWS/Storm Prediction Center, Norman, Oklahoma.