Biases in modeled vs. observed fire danger weather variables

Fire danger predictions are routinely made from operational numerical weather models, but those predictions are only as good as the predicted variables used in their computation. Biases and other metrics of model performance are often computed for individual models, but there are typically not systems in place to compare several models against the same observational data. As part of a larger project funded by the Joint Fire Science Program, we have developed a data acquisition system that automatically ingests surface and upper air observational data, and model output for four weather models: the GFS, NAM, and NDFD from NCEP, and the Pacific Northwest WRF from the Northwest Modeling Consortium. Using these data, we concentrate on four surface weather variables used for fire danger prediction – temperature, relative humidity, wind speed, and precipitation - collected over the fire season. This study will present time series of biases and other performance metrics for all four models at several ASOS and RAWS stations that are located within all model domains.