Jump to content
  • Member Statistics

    17,588
    Total Members
    7,904
    Most Online
    LopezElliana
    Newest Member
    LopezElliana
    Joined

Do We have one More in Us?


Damage In Tolland

Recommended Posts

  • Replies 2.5k
  • Created
  • Last Reply

Again and I want to stress this again... looking at deterministic solutions expect to discover a trend is just flat out silly... all deterministic solutions do is capture one potential solution of a probability distribution function (PDF).

Each individual solution, sure, but we sometimes see a signal emerge over a period of several successive runs, and it takes time for that to manifest itself.

We do see trends imo.

Link to comment
Share on other sites

Each individual solution, sure, but we sometimes see a signal emerge over a period of several successive runs, and it takes time for that to manifest itself.

We do see trends imo.

 

The problem is, how do you know its a real trend vs. the deterministic solution just jumping around between the left or right side of the PDF of possible solutions. Since we don't completely know what the initial conditions of the atmosphere are, any given deterministic solution can fall on one tail end of the ensemble distribution or the other and can easily sides from one model cycle to the next. 

 

We have ensembles to show this uncertainty to show whether or not the deterministic solution has support from the ensemble mean or if its an outlier. Just because the deterministic solution "trends" in one direction does not mean the ensemble distribution follows suit. In my opinion the ensemble PDF is a much more meaningful quantity to watch evolve over time. The ECMWF ensembles are superior because there are more members which fulfill a more complete PDF. The GFS ensembles tend to be underdispersive which doesn't give you the full degrees of freedom the atmosphere often provides in the medium range. 

Link to comment
Share on other sites

Anyways, back to weather. I did a few comparisons between model runs and found that the removal of the secondary LP from the CMC did nothing for placement on the original LP, which seems completely illogical. I would think that a denial of secondary and the emergence of a Miller A would at least allow the storm to hug the coast a bit more upon strengthening and axis tilt.

Link to comment
Share on other sites

The problem is, how do you know its a real trend vs. the deterministic solution just jumping around between the left or right side of the PDF of possible solutions. Since we don't completely know what the initial conditions of the atmosphere are, any given deterministic solution can fall on one tail end of the ensemble distribution or the other and can easily sides from one model cycle to the next. 

 

We have ensembles to show this uncertainty to show whether or not the deterministic solution has support from the ensemble mean or if its an outlier. Just because the deterministic solution "trends" in one direction does not mean the ensemble distribution follows suit. In my opinion the ensemble PDF is a much more meaningful quantity to watch evolve over time. The ECMWF ensembles are superior because there are more members which fulfill a more complete PDF. The GFS ensembles tend to be underdispersive which doesn't give you the full degrees of freedom the atmosphere often provides in the medium range. 

Well, what is why we wait for a signal to emerge over several successive runs before deeming it a "trend".

 

One or two runs do not constitute a trend.

Link to comment
Share on other sites

I was discussing it with Forky earlier, I do think it comes west, just a gut feeling I guess.  Ridge is located pretty far west, and little to no blocking downstream.  Hell, if this thing phased really hard it would barrel right into SNE lol.  

 

I'm thinking just E of  CQX.  

 

Big interior hit coming.

Link to comment
Share on other sites

Well, what is why we wait for a signal to emerge over several successive runs before deeming it a "trend".

 

One or two runs do not constitute a trend.

 

But neither do 5 or 6 runs in a row due to the same logic I described above. My argument is that there is no way to know if the deterministic run is staying in the same place on the ensemble PDF. There is no rule that says the model can't slowly shift from the right side of a ensemble PDF to the left side over the course of 5 or 6 runs. It doesn't imply a trend, it just implies the semi-random uncertainty of choosing a deterministic solution in the ensemble PDF. 

Link to comment
Share on other sites

But neither do 5 or 6 runs in a row due to the same logic I described above. My argument is that there is no way to know if the deterministic run is staying in the same place on the ensemble PDF. There is no rule that can't say the model can't slowly shift from the right side of a model PDF to the left side over the course of 5 or 6 runs. It doesn't imply a trend, it just implies the inherent uncertainty of choosing a deterministic solution in the ensemble PDF. 

I've gone on this crusade before...but here you've described it more succinctly and eloquently than I ever could.  

Link to comment
Share on other sites

But neither do 5 or 6 runs in a row due to the same logic I described above. My argument is that there is no way to know if the deterministic run is staying in the same place on the ensemble PDF. There is no rule that says the model can't slowly shift from the right side of a ensemble PDF to the left side over the course of 5 or 6 runs. It doesn't imply a trend, it just implies the semi-random uncertainty of choosing a deterministic solution in the ensemble PDF. 

Sure, I can't grab a calculator and mathematically prove that its a trend, but you are never going to get the word abolished from the meteorological lexicon.

 

For instance, the GFS has had a strong southern bias on coastal systems in the past.....more often than not, the low would not jump from near Bermuda to the benchmark in one run.....but rather at a more deliberate pace, protracted through a series of a successive runs.

Now, is that not a trend because the ensemble envelope moved in concert with it?

Maybe not, but I'm not sure that many folks care, though you are scientifically sound.

Link to comment
Share on other sites

Sure, I can't grab a calculator and mathematically prove that its a trend, but you are never going to get the word abolished from the meteorological lexicon.

 

For instance, the GFS has had a strong southern bias on coastal systems in the past.....more often than not, the low would not jump from near Bermuda to the benchmark in one run.....but rather at a more deliberate pace, protracted through a series of a successive runs.

Now, is that not a trend because the ensemble envelope moved in concert with it?

Maybe not, but I'm not sure that many folks care, though you are scientifically sound.

 

Haha you are certainly right, but that won't keep me from trying ;)

 

What you describe in the second paragraph below are systematic biases. When a model tends to always under-do a certain feature (e.g. subtropical ridging over oceanic basins are often undersampled and as a result get stronger with each successive run). The key thing to note here is how the initial conditions of a model are created. In fact 70-85% of the initial conditions produced through data assimilation are a result of the previous model cycle. Only 15-30% of the new initial conditions come from the inclusion of new observations, both in-situ (radiosondes, planes, surface obs) and remote sensing (satellite observations). So lets say a model tends to dampen out subtropical ridging too quickly. The inclusion of new observations over a 24 hour period slowly modify the initial conditions that are primarily derived from the previous model grid of the GFS. This is why you can often seen small but noticeable shifts which can have a lasting impact in later forecast times.

 

So yes in these instances such a pattern can result in a model trend over a period of time. However, its still difficult to partition the changes in the model due to a common systematic bias vs. semi-random changes of the vast majority of the model's initial conditions. In many cases these changes can trump the typical biases of a model and make it difficult to depict a model trend without serious investigation to which observations drove what.

 

There are sophisticated studies that go into trying to diagnosis which observations end up creating the largest spread in model solutions, and just which regions need to be sampled better in order to reduce model uncertainty. The north Pacific (close to the Gulf of Alaska) is often a space where errors start and grow as they propagate downstream across the globe. The feature we need to be focusing on is currently located in that part of the globe (the shortwave currently over far NW Canada). Instead of trying to diagnosis a model trend happening in the 96-120 hour range, I'd rather focus on what might be happening with that shortwave the next 12-36 hours that might be driving these changes.

Link to comment
Share on other sites

The reason why the ECMWF tracked further west is related to the shortwave I described in far NW Canada. More of the energy shifts east (not as much goes back into the Aleutian Low. Again the next 24 hours could still see significant shifts because this shortwave is in the process of rounding the omega block over Alaska and is in a relatively data sparse region in terms of observations. 

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...