Successes and Challenges using Ensemble-Based Tools to Forecast and Communicate a High-Impact Cool Season Precipitation Event in Northern Arizona
Andrew Taylor, NOAA/National Weather Service, Flagstaff Weather Forecast Office, Bellemont, AZ
Tracey Dorian, NOAA/National Centers for Environmental Prediction, Environmental Modeling Center
 ,
 ,
 ,

Abstract
On 27-28 February 2017, widespread precipitation fell over northern Arizona, leading to wintry road conditions at high elevations as well as flooded low water crossings at low elevations.  This event provided an opportunity to put science into service.  The process held plenty of challenges.

     The general idea of a trough over the western U.S. in the 27-28 February time frame with subtropical moisture transport into Arizona was forecast well roughly a week in advance.  An email discussing the potential for a high-impact winter precipitation event was sent to partners during the day shift on 22 February.  From that point through the 12Z model cycle on 24 February, the deterministic Global Forecast System (GFS) and its entire ensemble (GEFS) shifted much drier.  This caused Arizona offices to trend the precipitation forecast drier, but not as strongly as the GFS or GEFS mean.  At 12Z 25 February, the GFS and GEFS once again resolved the subtropical moisture connection and began forecasting substantial precipitation amounts over northern Arizona.

     Archived data leading up to the event from a number of ensemble-based tools will be analyzed and compared with observed snowfall and rainfall reports.  In addition, the National Centers for Environmental Prediction (NCEP) Model Evaluation Group (MEG) will be consulted for their viewpoint regarding the behavior of the GFS and GEFS for this particular case.  Based on experiences during this event, it is recommended that probabilistic model output using as many members as possible be used in lieu of deterministic forecasts for partner briefings in the day 4-7 range.

     To deliver high-quality decision support, we need reliable ensemble-based probabilistic guidance that reasonably estimates the probability density function (PDF).  Ensembles are by their very nature underdispersive since they do not contain an infinite number of members.  This case demonstrates a very undesirable ensemble behavior.