The National Fire Danger Rating System (NFDRS) indices deduced from the monthly to seasonal predictions of a meteorological climate model at 50-km grid space from January 1998 through December 2003 were used in conjunction with a probability model to predict the expected number of fire occurrences and large fires over the U.S. West. The short-term climate forecasts are ongoing experimental products from the Experimental Climate Prediction Center at the Scripps Institution of Oceanography. The probability model uses non-parametric logistic regression with spline functions for evaluating relationships between covariates and probabilities of fires. The 2-meter relative humidity and the Forsberg fire weather index, along with NFDRS indices of the Keetch-Byram drought index and energy release, were previously found to produce more significant information for the observed big fire events than all the other stand-alone fire weather variables. Utilizing this previously determined regression relationship between historical fire information and the nowcast fire indices, these predicted indices were skillful in generating fire severity forecasts at monthly and seasonal time-scales. However, certain meteorological model biases, due to a known drying-up defect of the climate model, needed to be removed from the predicted indices before being used as input to the probability model. It was shown that the probability model using the bias-corrected fire danger indices outperformed the one with historic information only. The inter-annual fire frequency variability was predicted particularly well. This dynamical-statistical hybrid climate forecast application demonstrates a potential predictive capability (with specified precision) for the resulting economic impacts with a lead-time varying from a month to a season.