This study evaluated the ability of the High Resolution Rapid Refresh (HRRR) modeling system to forecast the characteristics of mesoscale atmospheric boundaries arising from thunderstorm outflows, gust fronts, and downburst winds (referred collectively as convective outflows) within the contiguous United State and Alaska. Such convective outflows in the vicinity of wildfires can lead to rapid changes in fire behavior and growth that increase risks to firefighters. Our objective was to develop and evaluate diagnostic tools based on HRRR model output that could improve situational awareness within the operational fire weather community of the ability of the HRRR model to nowcast and forecast convective outflows at lead times of 18 hours or less.

Through the development of the only publicly-accessible archive of HRRR model output at the start of the project, we were able to apply novel data analytical methods applied to the HRRR fields at high temporal (hourly) and spatial (3 km resolution) resolution across the nation. A benefit of our creation of this archive was that over a thousand other researchers were able to access the HRRR model output for other operational and research applications, including initializing fine-scale models for simulating wildfire and other hazardous weather conditions. Beginning in late 2020, the NOAA Open Data Program has created similar archival capabilities using Google and AWS cloud resources. Hence, we, and other researchers, have a sustainable path to migrate existing codes that relied on our archive to those cloud environments.

In addition to the extensive online resources developed as part of this project (see http://hrrr.chpc.utah.edu(link is external)), the results of our research have been disseminated through 4 peer-reviewed publications, over a dozen presentations at regional and national conferences, and direct discussions with Incident Meteorologists and Fire Behavior Analysts. This work confirmed our original hypothesis that the HRRR model and fire behavior tools that rely upon it are not able to provide highly specific forecast guidance on convective outflows in complex terrain. However, the HRRR can facilitate nowcasting at lead teams less than 6 h and does improve situational awareness for the potential for convection at lead times less than 18 hours, particularly in synoptic-mesoscale situations for which the model is well initialized as part of the model’s data assimilation procedures. In addition, the GOES-Lightning Mapper sensors onboard GOES-16 and GOES-17 satellites were shown to be useful for evaluating the HRRR forecasts of lightning potential as a proxy for forecasts of intense convection.

GOES-16 and GOES-17 fire products for automated alerts to new fire starts have been successful enough in some parts of the country to suggest potential for earlier warnings of fire threats. However, assessment of the utility of the HRRR products in combination with other available resources as part of this study highlighted that automated alerts for action in the absence of well-designed fire metrics beyond those commonly used (e.g., fire danger, red flag) will require greater effort by trained personnel to understand local trends and thresholds applied to HRRR-derived or other model-derived products in order to minimize false positive and false negative errors. Such automated alerts, and the criteria they are derived from, will require extensive verification, calibration and validation. Criteria based on objective model guidance is useful for situational awareness, but is not of sufficient accuracy for actionable decisions.