Performance of the GFS versus the ECMWF

AJ from the NWS has suggested that it might be useful to start a thread on the blog focused on the mid-range performance of the GFS in comparison to that of the ECMWF. So this post is in response to his request.

First, some qualifiers:

I am not a model expert and all of my comments result from subjective appraisal of the various forecasts. I have not attempted to keep any kind of quantitative verifications, and the degree to which I examine the model forecasts at 3 to 7 days depends upon my level of interest and time available. Thus, my inputs are purely informal. I do not have access either to the ECMWF quantitative precipitation forecasts at this time, nor to many of the products from the GFS that are available at NWS offices. I tend to examine products at the NCEP web page and at the PSU electronic wall. I usually examine the 00Z runs, since sometimes the GFS goes into a flip-flop mode where the forecasts are quite different at 12 GMT than at 00Z GMT – I never bother to examine the runs at the 6-hour off times. I usually look most closely at the model forecasts over the Southwest and am thus regionally biased. All of that said, I note the following:

During this winter, and especially last winter, the mid-range ECMWF has seemed to catch important systems, and moisture influxes, better than the GFS. This has been a distinct enough difference that I routinely examine the ECMWF out to 168 hours before I check the GFS forecasts. I have noticed this winter that the GFS operational run is often quite similar to the ECMWF operational run (I don’t have access to, or know where, to examine the ECMWF ensemble members). However, the runs from last night at 00 GMT on January 21, 2009 are VERY different. Thus, today is a good day to try to start a new blog thread.

Note that at 168 hours from last evening (valid time 00 GMT on January 28th) the GFS indicates a quite cold cutoff west of southern California with a distinct 500 mb ridge over western Canada.. The GFS indicates significant precip over much of the West and Southwest and does not indicate precip over the central and northern Plains. On the other hand, the ECMWF indicates the two most important short-waves to be over Colorado and western Canada, with trailing vorticity across southern California into the east Pacific. The ECMWF indicates a frontal like band of precip across the Southwest. But the ECMWF forecasts an intense lee low and strong baroclinic zone over the Plains. Indeed the ECMWF indicates that near blizzard conditions would be impacting the central and high Plains at 168 hours.

I suspect that when the ECMWF performs better than the GFS that it is because its initialization over the Pacific, and also at low latitudes, is capturing features of key importance more accurately. This is sometimes especially noticeable with subtropical features and moisture fields. I think that it is well known that the GFS tends to be too dry over the Southwest.

I have also noticed that if the GFS ensemble members come into sync and consistently forecast precip over southeast Arizona in 10 to 12 of the members, then precip is almost certain to occur. This is true even at low elevations (the 00Z run indicates precip over SE Arizona in only 5 of the 12 members at 168 hours). Note that I have seen a couple of cases during the past two winters where the GFS forecasts were better than those of the ECMWF, but my subjective conclusion remains that I should examine the ECMWF first and then consider and contrast the other models.

What do readers think?