The Potentially Deleterious Impact of Choice of Sea-Level Pressure Calculation in the GFS Model When Determining Cyclone Intensity and Track

Numerous methods have been developed for extrapolating sea-level pressure (SLP) from the atmosphere above, with most of them using the lapse rate in the lower troposphere as part of the calculation. Typically, such an algorithm is needed when the base of the atmosphere does not reside at sea level due to terrain. Generally, differences in the output of these algorithms are not large and mostly result from the choice of lapse rate used during the extrapolation – along with the strength of the LaPlacian applied (if any) to the “underground” areas. A discussion of many of these aspects can be found in Harrison (1970) and more recently, Meisinger and Treadon (1995).

However, for the NCEP GFS model uniquely, there are two SLP fields output by the post-processor. The first, MSLET, is of the type described above. The second, PRMSL, takes the alternative approach of spectral filtering down to T80 (approximately 150km effective resolution) everywhere – even over water -- before the SLP is calculated. The consequence of this is of course a much smoother sea level pressure field than obtained by MSLET. Unfortunately, the consequence of the spectral filtering is that the analyzed intensity (minimum SLP) of cyclones using PRMSL can be significantly weaker than that obtained by MSLET, the latter being closer to “truth” by definition. These areas are maximized for smaller and stronger vortices, such as tropical cyclones. Further, the analyzed tracks of cyclones can be significantly in error when using PRMSL, leading to biases in human and derived guidance issued using this field.

In this poster presentation, we demonstrate those errors through several case studies, including: two tropical cyclones (one deep TC, one higher latitude TC over a strong SST gradient), a forecast case of TC genesis, and two extratropical cyclones. For one of these extratropical cyclones (in 2016), there is evidence that the forecasters tailored short-term blizzard predictions based on erroneous cyclone track guidance generated using PRMSL output from the GFS ensemble system.

Finally, we will show how the use of PRMSL vs. MSLET can significantly impact subsequent value-added guidance that relies on the SLP field. One example is the TC genesis guidance provided by Halperin et al. (2017), which uses as part of its logistic regression the PRMSL field. The negative impact of switching from PRMSL (which was used during the training phase of the guidance) to the more correct MSLET on the genesis guidance will be shown.

It should be noted as a postscript that although the SLP field is provided by many operational models under the standard “PRMSL” GRIB identifier, the approach of spectral filtering to T80 over water is unique to the GFS. Thus, the deficiencies discussed do not necessarily apply to the PRMSL field as output by other models.