Jump to content
  • Member Statistics

    17,609
    Total Members
    7,904
    Most Online
    Vesuvius
    Newest Member
    Vesuvius
    Joined

What IF?


Recommended Posts

The ECMWF were run 4 times a day?

-It seems the GFS has gotten better this year, with the recent upgrade.

-A lot of people like to accuse the GFS of flip-flopping.

-BUT, is this because of the GFS being a bad model or is it because its run double the amount of times as the ECMWF?

-IF the ECMWF were run 4 times a day, would it flip flop a lot too?

Thoughts?

Link to comment
Share on other sites

The ECMWF were run 4 times a day?

-It seems the GFS has gotten better this year, with the recent upgrade.

-A lot of people like to accuse the GFS of flip-flopping.

-BUT, is this because of the GFS being a bad model or is it because its run double the amount of times as the ECMWF?

-IF the ECMWF were run 4 times a day, would it flip flop a lot too?

Thoughts?

it would be a disaster :axe:

Link to comment
Share on other sites

The Euro is, on paper, the better of the two models, but it has nothing to do with the fact that it is not run 4x daily. It has a better data pre-processing (assimilation) technique (4D-VAR). This is the key to the Euro's success. It has been shown (can't remember the paper, consult NCEP) that if you take data that has been initialized by the Euro and plug it into the GFS as initial conditions, then the GFS is virtually as good as the Euro.

Unfortunately, NCEP has neither the money nor the resources to use 4D VAR (it takes about as long to run the 4D-VAR as it does to run the actual model!).

Link to comment
Share on other sites

The Euro is, on paper, the better of the two models, but it has nothing to do with the fact that it is not run 4x daily. It has a better data pre-processing (assimilation) technique (4D-VAR). This is the key to the Euro's success. It has been shown (can't remember the paper, consult NCEP) that if you take data that has been initialized by the Euro and plug it into the GFS as initial conditions, then the GFS is virtually as good as the Euro.

Unfortunately, NCEP has neither the money nor the resources to use 4D VAR (it takes about as long to run the 4D-VAR as it does to run the actual model!).

The first part is debatable, and people keep trying to reference a conference paper that I take great exception to. In fact, the only thing they were able to show is that they were able to mitigate some (not all) of our worst forecasts....they were never able to show that this made the (old version of) GFS "nearly as good as the Euro" on a consistent basis.

As you know, initial conditions are created by updating a model forecast / first guess with information from observations (through variational, OI, ensemble, or some other method). There is no way to separate out (completely) the influence of the forecast model from the initial condition....on top of the fact that these things are cycled and have "memory" of previous updates.

NCEP (we) are actively pursuing the development of both 4DVAR and hybrid var-ensemble (this development is much closer to being implemented than 4DVAR) techniques for future upgrades; and it's going to be a matter of being creative with resource usage.

Lastly, there are plenty of aspects of the EC model that are probably superior to ours (outside of their 4DVAR, which is better than our 3DVAR, I won't argue that). They run at higher resolution, use a different method for solving the dynamics (semi-lagrangian, which developers here have been working on implementing into the GFS), and have substantially different physics.

Link to comment
Share on other sites

I'm fine with the 2x per day runs of the ECMWF. Ideally I'd like it about 1-2 hrs earlier to give me some more time to peruse the data before my forecast is due but we don't operate under ideal circumstances all the time.

And this is one issue we face as we move toward higher resolution and more advanced (expensive) data assimilation techniques. We will never be able to push back delivery times of GFS products, so we have to be more creative about how we implement new things.

Link to comment
Share on other sites

The first part is debatable, and people keep trying to reference a conference paper that I take great exception to. In fact, the only thing they were able to show is that they were able to mitigate some (not all) of our worst forecasts....they were never able to show that this made the (old version of) GFS "nearly as good as the Euro" on a consistent basis.

As you know, initial conditions are created by updating a model forecast / first guess with information from observations (through variational, OI, ensemble, or some other method). There is no way to separate out (completely) the influence of the forecast model from the initial condition....on top of the fact that these things are cycled and have "memory" of previous updates.

NCEP (we) are actively pursuing the development of both 4DVAR and hybrid var-ensemble (this development is much closer to being implemented than 4DVAR) techniques for future upgrades; and it's going to be a matter of being creative with resource usage.

Lastly, there are plenty of aspects of the EC model that are probably superior to ours (outside of their 4DVAR, which is better than our 3DVAR, I won't argue that). They run at higher resolution, use a different method for solving the dynamics (semi-lagrangian, which developers here have been working on implementing into the GFS), and have substantially different physics.

A question I always had is how will the modeling community change as we continue to flatten out the rate of improvements in models with the persistent advancement towards the theoretical ceiling. As these developments continue, will we ever reach a day when the GFS is also a "premium" service? While I know law requires it to be in public domain, will that possibly change? Will the ECMWF ever reach a day when they will need to branch into different modeling sub-branches including mesoscale modeling, etc. to keep an overall competitive advantage to justify the premium costs?

Link to comment
Share on other sites

A question I always had is how will the modeling community change as we continue to flatten out the rate of improvements in models with the persistent advancement towards the theoretical ceiling. As these developments continue, will we ever reach a day when the GFS is also a "premium" service? While I know law requires it to be in public domain, will that possibly change? Will the ECMWF ever reach a day when they will need to branch into different modeling including mesoscale modeling, etc. to keep an overall competitive advantage to justify the premium costs?

These are legit questions (and way above my role/pay grade). I do not think you will ever see the GFS (or any NCEP products) become a "premium service", but that is just my feeling. I don't have a feeling as to whether or not this is even good or bad, it's just the reality of the situation.

Based on my understanding in how the EC works (and the agreement signed by member states), I don't think they're moving in that direction (I think they're going in the opposite direction, i.e. more emphasis on seasonal/climate prediction). Also, why even bother with a mesoscale model if in the foreseeable future they may be running their global model at higher resolution than we run our regional models?

Link to comment
Share on other sites

These are legit questions (and way above my role/pay grade). I do not think you will ever see the GFS (or any NCEP products) become a "premium service", but that is just my feeling. I don't have a feeling as to whether or not this is even good or bad, it's just the reality of the situation.

Based on my understanding in how the EC works (and the agreement signed by member states), I don't think they're moving in that direction (I think they're going in the opposite direction, i.e. more emphasis on seasonal/climate prediction). Also, why even bother with a mesoscale model if in the foreseeable future they may be running their global model at higher resolution than we run our regional models?

A thought I had, but I don't think they would want to be running a 4 km global anytime soon. Also, I would have to believe high resolution models in a less aggressively filtered state to preserve things important in microscale/mesoscale (for instance high frequency mountain induced waves) forecasting have a place. No global model, at least to my understanding, would be able to get away with that without increasing the noise levels substantially. I just wasn't sure if ECMWF may someday want to use their model expertise to filter into other model subsets. The increased emphasis on longer range forecasting makes perfect sense though.

Link to comment
Share on other sites

Archived

This topic is now archived and is closed to further replies.

  • Recently Browsing   0 members

    • No registered users viewing this page.
×
×
  • Create New...