The sea-level pressure (SLP) errors are quantified for cyclones within the NCEP GFS and NAM (Eta) for the 2002-2006 cool seasons (October to March), as well as the NCEP 15-member SREF modeling system for the 2004-2006 cool seasons. The average position and magnitude of the GFS and NAM cyclones were used for verification. The model and observed data were interpolated to a common 80 km grid, and the cyclones were identified using an automated cyclone finding and tracking system originally developed by NCEP. The observed and simulated cyclones were matched using the closest pair within 800 km of each other. This approach was ~90% effective in identifying cyclone events and matches in our dataset.
This presentation will illustrate how the cyclone errors vary by region and modeling system. The cyclone are separated into all events as well as the major (high impact) cases (those central pressure < 1.5 stnd deviation less than mean). For all events the largest mean absolute errors (MAEs) in central pressure are over the eastern Pacific in the short-term (< 60-h), likely because of the more data sparse Pacific Ocean. In contrast, the MAEs are half as large over the southern Plains, while the skill along the U.S. East Coast is between these two extremes. In many regions, the NAM errors are 10-30% larger than the GFS, with the largest difference over the eastern Pacific, where the NAM errors are 40% larger than GFS after hour 18. The NAM cyclones are 2-4 mb too weak on average over the Pacific Ocean. The GFS cyclones tend to be located east of the observed position along the U.S. West Coast and too deep, which suggests not enough terrain-induced cyclolysis and retardation of cyclones in the GFS. The GFS underpredicts major cyclone events by 6-8 mb on average for the > 4 day forecast over the western Atlantic, which for these extended periods results in much larger errors than the eastern Pacific. The mean SREF cyclone strengths and positions have larger errors on average than the GFS and NAM, which results from apparent overdispersion of cyclone positions and magnitudes during initialization, which can result large forecast errors (10-20 mb) by hours 36-48 in many SREF members.
This talk will also show how some errors are related to the large-scale flow. For example, many underdeepened GFS cyclones along the East coast are favored when short-waves troughs from the northwest merge with the southern stream, while overdeepened cyclones occur for more progressive troughs across the central southern United States.