Ensemble Verification and Post-processing of Fire Weather Indices Using the NCEP SREF over the Northeast United States

The use of ensemble modeling systems has become commonplace in many operational numerical weather prediction applications. Ensembles help quantify the probability of future forecast outcomes given the uncertainty in the initial atmospheric state or model physics. However, ensembles are not often used in the preparation of operational short-range fire weather forecasts. Furthermore, the impact of atmospheric model biases on fire weather forecasts can lead to an underdispersed (members too clustered) ensemble. For this reason, the impact of post-processing an atmospheric ensemble is examined for the purpose of generating accurate model-derived fire-weather forecasts.

The ensemble used in this study is the 21-member Short Range Ensemble Forecast (SREF, 32 to 45-km grid spacing) run at the National Centers for Environmental Prediction (NCEP) from 2008 to 2012. The SREF includes different initial conditions and physical parameterizations (convective parameterization, boundary layer, and microphysics). The 13-km grid spacing Rapid Update Cycle (RUC) analysis fields are used to verify the ensemble predictions.

Two fire weather indices are calculated in this study. The first is the well-known Haines Index which considers the low-level lapse rate and dew point depression. The second, called the High Fire Threat Index (HFTI), is new index that considers the state of the near-surface atmosphere. In the absence of snow-cover (determined from the Interactive Multisensor Snow and Ice Mapping System Snow and Ice Analysis) or rainfall during the past 48 hours (from the Stage IV precipitation database), the HFTI exceeds zero only if the relative humidity is within the bottom 2.5 percentile of the 2008-2012 RUC climatology. Thereafter, HFTI is calculated by considering two subcategories; 1) a 2-m relative humidity component assigning an index value of 1, 2, or 3 given that relative humidity falls in the bottom 2.5, 1, or 0.5 percentile, respectively and 2) a 10-m wind speed component assigning a value of 1 or 2 given that wind speed exceeds 10 or 15 knots, respectively. The total HFTI, which ranges between zero and five, is summation of the relative humidity and wind speed components.

Results will show that the SREF consistently under predicts both HFTI and the Haines Index, especially for more extreme events. This is largely the result of a persistent cool and wet bias in the ensemble, particularly in the lower levels of the planetary boundary layer. Applying a simple additive bias correction to each model state variable significantly improves these biases and the overall error in the ensemble. This suggests that post-processing should be applied to model output before attempting to make operational fire threat forecasts.