NAM Proposed Termination

[Boris Konon]

See PDF attachment for all the details. This is a good thing overall, as the RRFS is in-line to replace the NAM fully, and is much better!

The last update/upgrade to the NAM was in 2017 and frozen since due to its legacy status.

I am not a big fan of the NAM. It doesn't do well after about 36 hr, and can't handle TCs correctly at all (never has). The 3km NAM has a big bias about overdoing precip, esp. when orographics are involved, and its simulated radar fcst is inconsistent run to run w/ convection.

The HRRR does much better in all of the above for a comparison.

No reason why the NAM would not be terminated sooner than later, just NWS policy is to give plenty of time for users to provide comments and adjust.

One thing though, many still use NAM MOS (I do), will there be a RRFS equivalent? I will ask about it!

And this is important, b/c MOS is better than 2 m temps. No ECMWF MOS, so ppl use the model 2 m temps...NO! Same for GFS, but we do have GFS MOS.

MOS incorporates station climo and other things to make it more correct. 2 m temps do not. In fact, for global models, the lowest model layer is 10 m, and that's b/c that's the standard for wind measurements. So models have to do some quirky things to estimate temps at 2 m, and it is not always accurate. 2 m temps get really bad in the longer ranges (too hot). And many mets do not know this still I have found, and post ridiculous model fcst temps when a heat wave is coming up!

 

[Dan Robinson]

Looks like this is a controversial issue with many opposed to it.

https://twitter.com/x/status/1938442892386668650

I started integrating the RRFS into my normal chase forecasting earlier this year, but just gave up on it due to it consistently over-convecting on nearly every setup.

I am not an expert in models by any means, but I know it takes a long time to get acclimated to a new model as a forecaster and chaser. You really have to see how it performs day in and day out on a broad range of setups to get a feel for its biases and where it is useful. I felt like the RRFS issues with over-convecting made it too problematic in that context. I have yet to use it for winter weather.

By contrast, I find the 3KM NAM to be among the best with winter weather, including things like precip type and shallow cold air intrusions.

 

[Sean Ramsey]

Other than using them a little bit in the spring I have no other knowledge about how these things work (which requires me by internet law to comment, of course).

Hopefully someone can help answer these questions in which I'm genuinely interested. Why would any model be developed and given up on as far as continuous improvement? It has to take tremendous resources to get a model online for use (not to mention all the calculations to bring it about), so if it has tendencies or biases, why not work to iron them out instead of putting it to bed? If it has served a purpose up until a certain point, why can't that simply be extended with some TLC?

Are other models (especially all the new CAM's) based on different equations, therefore different outputs? If so, are these models really more accurate or just another arrow in the quiver hoping to get closer to the holy grail of the perfect forecast?

Lots of questions but it would be interesting if someone had some insight into this for us casual users.

 

[JamesCaruso]

I have some of the same questions as Sean and more. I’m ashamed to say I am woefully ignorant about models, even after nearly three decades of chasing. I have a hard enough time keeping up with developments in my own profession, let alone trying as a non-meteorologist to keep up with what’s happening with models and all their biases (and Boris mentions in the OP that even many meteorologists are unaware of certain biases).

It’s not even clear to me what the RRFS is. Is that a particular model, or is it a suite of models that includes the HRRR? Is the NAM being replaced by higher-resolution CAMs? The Roger Edwards X post above mentions FV3, is that the HRW FV3? The PDF posted by Boris says that the HREF is also being discontinued. With the new landscape of models, including all the new experimental model families like MPAS, I’m more confused than ever!

Is there a good resource that summarizes all the current biases, and what each model is best for, in one place? If not, I’m going to try some research with the Perplexity AI application.