A Fire Weather Index (FWI) is created at Stony Brook University (SBU) using only near-surface weather conditions to statistically predict the probability of wildfire initiation. The SBU-FWI can vary between 0 and 3 with each index value representing an increasing probability of wildfire formation within a 100 km radius. The probabilities are as follows:
• FWI = 0 has a wildfire initiation probability of less than 30%.
• FWI = 1 has a wildfire initiation probability between 30% and 40%.
• FWI = 2 has a wildfire initiation probability between 40% and 50%.
• FWI = 3 has a wildfire initiation probability greater than 50%.
The FWI is calculated from the Short Range Ensemble Forecast (SREF) ensemble and represents a forecast for the future state of the atmosphere. This allows for the advanced detection of atmospheric conditions that might be favorible for wildfire development.
The Specifics:
The FWI is a statistical index developed using a binomial logistic regression to predict the probability of wildfire occurrence. It was developed testing several near-surface atmospheric variables and verified against a 10 year observed wildfire occurrence dataset between 1999 and 2008. A logistic regression is similiar to a linear regression in that multiple predictors can be used to optimize the statistical fit. However, the predictand for a logistic regression is a probability value, in this case the probability that a wildfire will occur.
Several different near-surface weather variables were tested to optimize the model fit, including temperature, wind speed, relative humidity, specific humidity and dew point. The most significant variable was minimum daily relative humidity, with maximum daily temperature also being statistically significant. Therefore, the standardized values of relative humidity and temperature are the only two predictors used in the logistic regression formulation as shown below:
where p is the probability of wildfire occurrence, RELH is the standardized minumum daily relative humidity, TEMP is the standardized maximum daily temperature, i indicates daily value, and b are coefficients determined by the regression.
After the coefficients are determined using 10 years of wildfire occurrence data, the logistic regression is adapted to a grid using the 2007-2014 climatology of the Rapid Update Cycle (RUC) and Rapid Refresh (RAP) analysis field. Here we apply the FWI logistic regression to the SREF 26-member ensemble four times a day. FWI values are adjusted to zero where snow cover* is currently on the ground. The result is a FWI ensemble forecast over the Northeast United States. Additional information on the development of the FWI and details on its adaption to a model or analysis grid will be available soon by publication.
* Snow cover data are obtained from the NOAA/NESDIS Interactive Multisensor Snow and Ice Mapping System (IMS) Daily Analysis at 4 km resolution.
