(maar alsnog)
Dr. Ryan Maue, WeatherBELL
During the past few weeks, I've been working with the new GFS upgrade model. To some it's a shiny new toy, to others it's simply a Mercury Sable with a new paint job. While that car was popular a decade ago, does that mean we should trust it today?
The new GFS model is more a "bunny hop" as I described it to the NY Times and Mashable ... or anyone that would listen. I'm pretty good with those buzz phrases in the print media. The horizontal resolution is vastly improved from T574 to T1534. The T is the number of spectral waves that fit around the Earth's sphere. For reference, the T574 reduced down to 27-km grid spacing in the middle latitudes while the new T1534 iteration is 13-km. However, the vertical resolution has not changed. Nope.
What about ECMWF? It runs at T1279 or about 14-16 km globally. WeatherBELL receives the full-resolution data from ECMWF -- for EVERYTHING. Not like Accuweather or WSI or whoever else. Their maps look like blocky Legos. But what makes ECMWF superior as a model is the vertical resolution. GFS has 62 levels while ECMWF has more than a 130. Those extra levels are spaced throughout the planetary boundary layer (PBL) and up near the tropopause and in the stratosphere. GFS could definitely stand to have a few more vertical levels. But that's very costly computationally. Also, the GFS model has been tuned and fudge-factored for a decade. It's hard to just reboot the thing from scratch and expect it to work out of the box spectacularly.
Recent research has discussed just why ECMWF succeeded in model Hurricane Sandy in 2012 while GFS sent it to Portugal at 6-days. Nick Bassill ran some very intriguing WRF simulations (paper: Link to GRL). Here's the abstract:
The extremely damaging Hurricane Sandy (2012) is noteworthy for the significant track bifurcation among several forecast models approximately 6–7?days before landfall. The operational versions of the ECMWF GFS models exemplify this difference over their runs early in Sandy's life cycle, as the former routinely forecast a storm which would make a U.S. landfall, while the latter routinely forecast a track toward the central North Atlantic. This was also generally true of their respective ensemble members. This paper demonstrates that these differences were caused not by resolution or initial condition differences but rather due almost exclusively to choice of cumulus parameterization. Simulations performed with the WRF model using an ECMWF-like cumulus parameterization in conjunction with GFS initial conditions yield forecasts whose accuracy is similar to that of ECMWF forecasts at extended time ranges (up to 1 week before landfall).
So, it's the physics stupid. Actually it's the convection.
Unfortunately, as far as I know, that component of the GFS model has not been changed. Thus, when the blizzard was supposedly forecast much better by GFS than ECMWF for NYC, it's premature to say it was right for the right reason. Also, it's not really very useful to compare two models in the short-range since they are ... different models. One doesn't know about the other and has no knowledge of how the other one is performing. However, humans surely do and put "all of their eggs in one basket" which was the ECMWF reinforced by (cringe) the NAM.
Now with the upcoming blizzard #3 potentially for New England come Thurs-Friday, we've got the GFS doing absolutely nothing with it while ECMWF 12z lowers the boom -- and then again with another Nor'easter Monday.
Here's the 00z GFS model forecast from Tuesday valid late Thursday evening. Nothing doing -- but very cold behind this Arctic front for New England. The 2-meter temperatures are hysterically cold from GFS, seemingly another bias that's been baked into the cake in this new "upgrade".
During the past few weeks, I've been working with the new GFS upgrade model. To some it's a shiny new toy, to others it's simply a Mercury Sable with a new paint job. While that car was popular a decade ago, does that mean we should trust it today?
The new GFS model is more a "bunny hop" as I described it to the NY Times and Mashable ... or anyone that would listen. I'm pretty good with those buzz phrases in the print media. The horizontal resolution is vastly improved from T574 to T1534. The T is the number of spectral waves that fit around the Earth's sphere. For reference, the T574 reduced down to 27-km grid spacing in the middle latitudes while the new T1534 iteration is 13-km. However, the vertical resolution has not changed. Nope.
What about ECMWF? It runs at T1279 or about 14-16 km globally. WeatherBELL receives the full-resolution data from ECMWF -- for EVERYTHING. Not like Accuweather or WSI or whoever else. Their maps look like blocky Legos. But what makes ECMWF superior as a model is the vertical resolution. GFS has 62 levels while ECMWF has more than a 130. Those extra levels are spaced throughout the planetary boundary layer (PBL) and up near the tropopause and in the stratosphere. GFS could definitely stand to have a few more vertical levels. But that's very costly computationally. Also, the GFS model has been tuned and fudge-factored for a decade. It's hard to just reboot the thing from scratch and expect it to work out of the box spectacularly.
Recent research has discussed just why ECMWF succeeded in model Hurricane Sandy in 2012 while GFS sent it to Portugal at 6-days. Nick Bassill ran some very intriguing WRF simulations (paper: Link to GRL). Here's the abstract:
The extremely damaging Hurricane Sandy (2012) is noteworthy for the significant track bifurcation among several forecast models approximately 6–7?days before landfall. The operational versions of the ECMWF GFS models exemplify this difference over their runs early in Sandy's life cycle, as the former routinely forecast a storm which would make a U.S. landfall, while the latter routinely forecast a track toward the central North Atlantic. This was also generally true of their respective ensemble members. This paper demonstrates that these differences were caused not by resolution or initial condition differences but rather due almost exclusively to choice of cumulus parameterization. Simulations performed with the WRF model using an ECMWF-like cumulus parameterization in conjunction with GFS initial conditions yield forecasts whose accuracy is similar to that of ECMWF forecasts at extended time ranges (up to 1 week before landfall).
So, it's the physics stupid. Actually it's the convection.
Unfortunately, as far as I know, that component of the GFS model has not been changed. Thus, when the blizzard was supposedly forecast much better by GFS than ECMWF for NYC, it's premature to say it was right for the right reason. Also, it's not really very useful to compare two models in the short-range since they are ... different models. One doesn't know about the other and has no knowledge of how the other one is performing. However, humans surely do and put "all of their eggs in one basket" which was the ECMWF reinforced by (cringe) the NAM.
Now with the upcoming blizzard #3 potentially for New England come Thurs-Friday, we've got the GFS doing absolutely nothing with it while ECMWF 12z lowers the boom -- and then again with another Nor'easter Monday.
Here's the 00z GFS model forecast from Tuesday valid late Thursday evening. Nothing doing -- but very cold behind this Arctic front for New England. The 2-meter temperatures are hysterically cold from GFS, seemingly another bias that's been baked into the cake in this new "upgrade".
Bericht laatst bijgewerkt: 03-02-2015 22:11
http://oakcreekwx.com