NAM Phase Shift

[baroclinic_instability]

The NAM can be good at certain times, and it has its uses if used carefully by the forecaster, but for the most part, it is just plain bad. I talk about the NAM "Phase Shift" issue a lot, but I constantly see threads where members are asking why it is slow. I think most experienced mets are well aware of this problem, but it arises under certain weather patterns where the entire height field is "phase shifted" westward, resulting in a far too slow progression of weather features. Sometimes it can rear its ugly head as early as 6-12 hours into the forecast with errors growing with each successive forecast hour through 84 hours. This is not uncommon, but it has been especially atrocious this storm.

Problems? There are differing theories, but a logical explanation is its early runtime therefore it relies on the previous run GFS to initialize its boundary conditions (i.e. a 12Z NAM run is initialized on the boundaries with the 06Z GFS). That still can't explain it all though. Something else is obviously wrong...if anyone has ideas post here.

Of course, it can do some things quite well including rapid and intense marine cyclogenesis and cyclogenesis associated with deep slow progressing and deep PV Anomalies over intense baroclinic zones, surface wind fields, mountain drag, downslope wind storms, lee waves, etc.

Compare the NAM @ 0Z Friday :

GFS at the same forecast hour:

Note in the NAM, compared to the GFS, how far behind the shortwave over the northern plains is...and how far behind the shortwave over Nevada is.

Then look at the NAM this morning at the same forecast time. Note how it is "corrected" and everything is in phase (faster):

Here are the HPC NAM 500 hpa verification fields (red is the current model run, blue was the forecast):

84 Hours:

72 Hours:

60 Hours:

48 Hours:

36 Hours:

24 Hours:

12 Hours:

Here is the HPC verification page for those interested.

http://www.hpc.ncep....tml#diagnostics

[Chicago WX]

Good info. But I can assure you that most posters on here (central/western) do not believe (or hope) that the NAM is usually accurate at 84 hours...well at least the sensible ones don't.

[JoMo]

Everyone knows the NAM sucks. People are just hoping it will be right so they don't have to face the fact that they won't get snow.

[baroclinic_instability]

Everyone knows the NAM sucks. People are just hoping it will be right so they don't have to face the fact that they won't get snow.

It does suck 90 % of the time, but it can do some things well, i.e. strong and intense cyclogenesis with slow propagating and deep PV Anomalies, for instance.

[TrekkerCC]

baroclinic_instability wrote:

Problems? There are differing theories, but a logical explanation is its early runtime therefore it relies on the previous run GFS to initialize its boundary conditions (i.e. a 12Z NAM run is initialized on the boundaries with the 06Z GFS). That still can't explain it all though. Something else is obviously wrong...if anyone has ideas post here.

I have a couple of questions below:

Would using another model's boundary conditions, such as the ECMWF model, really help correct this behavior? I read from various board members in the past that the GFS forecast improves, if the ECMWF initial conditions are used.

How much error remains when the NAM "corrects" itself? Will the NAM forecast be noticeably different from the GFS model even after the NAM model corrects itself. What are the forecast implications down the road?

Is this similar to how the ECMWF holds back "energy" too long, or is it a different issue than phase shifting?

Thank you for shedding light on this NAM quirk. When I examine the NAM in the future, this will be in the back of my mind.

[baroclinic_instability]

baroclinic_instability wrote:

I have a couple of questions below:

Would using another model's boundary conditions, such as the ECMWF model, really help correct this behavior? I read from various board members in the past that the GFS forecast improves, if the ECMWF initial conditions are used.

How much error remains when the NAM "corrects" itself? Will the NAM forecast be noticeably different from the GFS model even after the NAM model corrects itself. What are the forecast implications down the road?

Is this similar to how the ECMWF holds back "energy" too long, or is it a different issue than phase shifting?

Thank you for shedding light on this NAM quirk. When I examine the NAM in the future, this will be in the back of my mind.

Well the NAM has boundary conditions on its outer edges because it isn't a global model. Here is an example of the NAM domain. http://www.emc.ncep.noaa.gov/mmb/namgrids/g212.12kmexp.jpg

Either way, I should be careful when saying that because it doesn't seem to be the main problem with the model. There seems to be something else wrong in the model dynamics, somewhere in the parameterizations, etc. Usually the NAM can initialize ok and is somewhat reliable for the first 24 hours, then it begins to slowly phase shift westward with increasing forecast hours. This is especially noticeable in fast zonal flow with embedded low amplitude shortwaves where the NAM can be 6-9 hours too slow by the end of the forecast. I always compare the GFS/NAM height fields through the forecast and look for this phase shift, and I also look at the HPC model verification height fields (the images above) to see how well the NAM has done lately. As for the Euro/GFS, they are both global models and don't need "boundary conditions". In terms of forecast implications down the road, the NAM has proven to be unreliable and statistically (by far) the worst verifyoing model amongst all the well known numerical models. It is very erratic and has many errors, but it can be used intelligently at times and it has its uses. This severe wx event over Illinois/WI today is a good example of that.

[Powerball]

Bluntly put, the NAM sucks.

Then again it was never meant to be a macroscale/long range model like the GFS. A lot of people only use it to better define the short term meso/microscale features in a weather event (that's where it shines and where the GFS sucks).

[TrekkerCC]

Well the NAM has boundary conditions on its outer edges because it isn't a global model. Here is an example of the NAM domain. http://www.emc.ncep....212.12kmexp.jpg

Either way, I should be careful when saying that because it doesn't seem to be the main problem with the model. There seems to be something else wrong in the model dynamics, somewhere in the parameterizations, etc. Usually the NAM can initialize ok and is somewhat reliable for the first 24 hours, then it begins to slowly phase shift westward with increasing forecast hours. This is especially noticeable in fast zonal flow with embedded low amplitude shortwaves where the NAM can be 6-9 hours too slow by the end of the forecast. I always compare the GFS/NAM height fields through the forecast and look for this phase shift, and I also look at the HPC model verification height fields (the images above) to see how well the NAM has done lately. As for the Euro/GFS, they are both global models and don't need "boundary conditions". In terms of forecast implications down the road, the NAM has proven to be unreliable and statistically (by far) the worst verifyoing model amongst all the well known numerical models. It is very erratic and has many errors, but it can be used intelligently at times and it has its uses. This severe wx event over Illinois/WI today is a good example of that.

Thanks for the reply. You did a great job in answering my questions. I do try to keep up with the weather (examining the models, looking at surface and upper obs, satellite, etc), especially in the wintertime, and so I enjoy learning about the models and their biases/errors. It helps me to look at their output with more critical eye to what I'm seeing.

Bluntly put, the NAM sucks.

Then again it was never meant to be a macroscale/long range model like the GFS. A lot of people only use it to better define the short term meso/microscale features in a weather event (that's where it shines and where the GFS sucks).

True. However, I do remember that the NAM nailed a North Texas post-frontal light snow event (it may have been in 2007 or 2008?) when the GFS had nothing. Now, the NAM caught onto this within 24 hours out -- but it did impress me a little bit - given North Texas snow is always very difficult to forecast.