Evaluation of fire weather indices predicted based on real-time high-resolution MM5 forecasts over central and upper Great Plains

Fire weather indices, such as the Haines Index and Fosberg index, are simple tools for evaluating the potential influence of weather on a wildland fire. Traditionally, these indices are estimated using vertical profiles of temperature, relative humidity, and for some indices, wind speed from standard rawinsonde soundings. Because of the sparse spatial and temporal resolution of the standard rawinsonde network, fire weather forecasting has grown more dependent on model predictions of these indices to fill large gaps between rawinsonde stations and time intervals between soundings. Although the performances of real-time mesoscale model forecasts have been extensively validated against observations, the focus has been generally on precipitation and surface temperature. Little is known pertaining to the accuracy of the predicted fire weather indices based on mesoscale model forecasting.

In the current study, the most widely used fire weather index, the Haines index (HI), predicted using real-time MM5 model runs were validated against observed HI values at large number of upper air sounding locations and for a relatively long period of time. The MM5 predictions were made twice daily for a 12-km coarse domain over the central and upper Great Plains and a 4-km fine domain over the Great Lakes region. The validation was performed for the 2003 fire season from May through October. The results show that the predicted HI values tend to be lower than the observed values and this trend exists for most locations and throughout the fire season. This lower risk bias in MM5 prediction is attributed to a general cold and wet bias and weaker stability in the model forecasting. The results also show little difference between the 12-km and 4-km forecasting, indicating that an increase in model horizontal resolution does not necessarily translate into more accurate fire weather prediction.