Diagnosing Regional Low-Skill Forecasts in the FV3-Based GFS

While operational numerical forecast continue to become more skillful, episodic periods of extremely low mid-latitude forecast skill still occur. The evaluation and documentation of such events thus far has been primarily Eurocentric with several studies diagnosing low-skill European forecasts through individual European Centre for Medium-Range Weather Forecasts (ECMWF) operational case studies or composite analysis using 20 years of forecasts from a fixed 2006 version of the ECMWF deterministic model. Here, we expand past this Eurocentric view and use 15 years of forecasts from the newly developed FV3-based Global Forecast System (GFS; i.e., GFS version 15.0) to assess low-skill forecasts occurring in 10 different regions of the Northern Hemisphere. For each region, low-skill forecasts are identified where regionally averaged anomaly correlation coefficient values for the 120-h forecast 500-hPa height field are less than 0.5 and regionally averaged root mean square errors are larger than 60 m. Ocean basins and polar regions are found to have an increased number of low-skill forecasts than continental regions, with the majority of regions favoring these forecasts in the summer and autumn over the winter and spring. All regions do not have a strong correlation with any particular phase of the prominent teleconnection indices, suggesting the synoptic patterns associated with low-skill events may differ from these well-known setups. Therefore, composites of the 500-hPa height fields are used to identify the synoptic patterns associated with these events. A combination of composite analysis and individual case studies are also utilized to identify the likely feature(s)/process(es) which allow these poor forecasts to occur.