12z ECMWF shows wintry potential

[WeatherX]

Mucho spread to the deep side with the diving h5 low on the 18Z GEFS.....just tossing that out. 0Z in the batters box now....

[am19psu]

For those interested...a couple of us on wikipedia are improving the Numerical weather prediction page at...

http://en.wikipedia.org/wiki/Numerical_weather_prediction

For reference, this edting is not being done on government time. We're adding information, as well as trying to source some of the unreferenced lines, but it's a good introduction to the guidance in general, and why many of us don't inherently trust an individual piece of deterministic guidance more than five days into the future, no matter how well/poorly it verifies at that time range. There are references to a couple useful links regarding ensemble forecasting (why it's being done at all) as well as generic information regarding the history of atmospheric modeling. If you want to help out, feel free. Just make sure to reference your additions. =)

DR

Thanks, DR. If this takes off, I'll split it off into its own thread.

I found this quote particularly interesting, since I am often guilty of this misconception.

A common misconception is that low spread amongst ensemble members necessarily implies more confidence in the ensemble mean. Although a spread-skill relationship sometimes exists, the relationship between ensemble spread and skill varies substantially depending on such factors as the forecast model and the region for which the forecast is made.

Can you elaborate on that passage more (from an operational forecasting perspective), if you have time?

[baroclinic_instability]

For those interested...a couple of us on wikipedia are improving the Numerical weather prediction page at...

http://en.wikipedia....ther_prediction

For reference, this edting is not being done on government time. We're adding information, as well as trying to source some of the unreferenced lines, but it's a good introduction to the guidance in general, and why many of us don't inherently trust an individual piece of deterministic guidance more than five days into the future, no matter how well/poorly it verifies at that time range. There are references to a couple useful links regarding ensemble forecasting (why it's being done at all) as well as generic information regarding the history of atmospheric modeling. If you want to help out, feel free. Just make sure to reference your additions. =)

DR

Nice info. I posted this in the ensemble thread already, but ECMWF also has an awesome user guide with a lot of good information regarding all sorts of numerical prediction topics including ensembling.

http://www.ecmwf.int/products/forecasts/guide/index.html

[Cyclonicjunkie]

Yea this storm is not happening.

Best post all day

12z model comparison at 168

GFS

CMC

ECM

[baroclinic_instability]

Best post all day

12z model comparison at 168

GFS

CMC

ECM

Comparing the models at that time alone won't do it. The way the ECM gets there is completely different, and there is a big difference between the ECM coastal and the weak OTS development of the GFS/CMC you have posted there. In this case, JI is right.

[Cyclonicjunkie]

Comparing the models at that time alone won't do it. The way the ECM gets there is completely different, and there is a big difference between the ECM coastal and the weak OTS development of the GFS/CMC you have posted there. In this case, JI is right.

There is big differences in all of them, and personally I believe the pattern looks more favorable closer to the middle of the month, but to say this storm wont happen a week out with most models showing a storm of some type in this timeframe is just plain silly.

[baroclinic_instability]

There is big differences in all of them, and personally I believe the pattern looks more favorable closer to the middle of the month, but to say this storm wont happen a week out with most models showing a storm of some type in this timeframe is just plain silly.

Read my earlier posts bud.

The ECM storm at 168 as it has it won't be happening unless the stars align for the reasons I mentioned. Threats exist later in the period though with the potential for some Miller B action with that northern vortex.

[Avdave]

????

GFS loving it too

Talking about the euro ensembles you were mentioning.

[Anyweather]

How the H--- can we trust the models, when they didn't come to an agreement 18 hrs before the last Noreaster?

C'mon.

Really?

[PhineasC]

00Z GFS has nothing.

[WxLover]

00Z GFS has nothing.

Just where we want it.

[CooL]

00Z GFS has nothing.

The PV has a date with the northeast on the gfs.

[thegreatdr]

Sure. For the record, I didn't add the passage, although if I did, it would be for a different reason. I've seen times where there was significant ensemble clustering within the ensemble guidance, only to see a fringe solution (or worse, a solution outside the ensemble spread) verify. Other times going in the dead center of a huge ensemble envelope works out perfectly. There doesn't seem to be any way of knowing ahead of time which situation will occur, so you play the averages and go with the solution with enough (or the most) support, depending upon how much of a risk you are comfortable making as a forecaster. Even though 150 ensemble members (50 ECMWF, 20 Canadian, 40 GEFS from two GEFS runs, and 40 SREF from two SREF runs) are available to NWS forecasters in various forms (depending on your office) every 12 hours, who's to say that those 150 solutions cover all possible solutions? The way ensemble guidance makes tiny changes to the initial conditions does not take into account where the error is potentially coming from (figuring that out takes significant computational resources all on its own), it is much more random. Human forecasters can help determine this and task recon to go out and investigate a system which subjectively appears to be the problem. Meteorological organizations, NWS or otherwise, do not have unlimited financial resources to send reconnaissance planes on a routine basis to places where the error/forecast uncertainty seems to be originating from, though we try to do so where we can. Last time I checked, the NWS gets something like $1-1.50 from every tax payer and NOAA as a whole gets about $5 per tax payer. Satellite and radar systems are expensive to implement, with the latter expensive to operate and maintain, so there's not a tremendous amount of money left for much else.

Another thing to keep in mind is that the ensemble guidance is no different than the deterministic guidance. The physics remain non-exact, and the initial conditions are not exactly known either, even with the input of satellite data. Using convective parameterizations can lead to all sorts of errors, especially if the convection does not form as expected. Until we get model resolution down to an order of meters per grid square, some small detail could be missed, whether it is a barrier jet near an individual mountain, or how thunderstorms on the near microscale contort the QPF pattern on the mesoscale, then synoptic scale. Once we get down to that scale, chances are we'll have to incorporate traffic patterns (human or animal migratory patterns) in order to better know where emissions of water vapor could be coming from, whether it is from exhalation or hydrogen fuel cell cars. Maybe at some point during the next generation global model resolution will be able to get down to that scale, which will partially solve the initial value problem. But you can see how complex it already appears. As long as we use equations that are partial differential equations (there doesn't seem to be any other choice), error will still be introduced into the model guidance as you go out farther in time. As I read in a recent version of the Ayers meteorological textbook (usually used for intro to meteorology, a 100 or 1000 level course which the board posters should be able to comprehend with no problem), there are times that lower resolution guidance performs BETTER because it has smoothed out the potential problem due to its larger grid scale. It's more complicated than most people realize. I wish it were as simple as going with the model with the best resolution. If that were true, we'd all be using a 4 km version of the MM5/WRF for short range, and the DGEX for medium range, with either GFS or ECMWF used for boundary conditions and never look back. No matter what the resolution of any piece of guidance (deterministic or ensemble), there will be some bias that is detectable. At least that's how it has played out so far during my college and professional career anyway.

For the past 6 1/2 years (not very long compared to some forecasters) that I've been dealing with a day 3-7 forecast, I've found the nature of the flow pattern anticipated to have more bearing on forecast certainty than ensemble spread. If you look on a run-to-run or day-to-day basis, you'll see spread in the ensemble guidance wax and wane, and sometimes completely shift, just like the deterministic runs. Sometimes the shift follows the deterministic runs, but especially if it all shifts at once. So if there is an overall trend among multiple pieces of deterministic guidance, older global ensemble members need to be discarded. This last storm was a good example of this about 60 hours out from its impact on the Northeast. Many of us knew the forecast would be difficult because we were dealing with multiple streams which were progressive while trying to phase, which the models normally have problems with in short range, let alone medium range. Personally, I would have had more confidence in the more northwestern/more coastal solutions if ANY of the Canadian/ECMWF ensemble members or any of the deterministic guidance 3.5 days out supported it, AND this was merely the case of a southern stream shortwave embedded in zonal flow rounding an already established trough in the upper Ohio Valley. This just wasn't the case during the last forecast scenario. When two or more systems interact, whether one shears another out or it is binary interaction, you're bound to have run-to-run and day-to-day changes in the guidance, even when those systems aren't very progressive. One would expect the same issues during zonal flow. The most confident forecasts are those which deal with the amplification of single, unified full latitude stream of the Westerlies (no separate northern and southern branches). Amplification slows down system progression (a big plus for forecast confidence), and if there is no flow undercutting the upstream ridge, it's bound to remain slow moving for several days, until zonal flow upstream knocks down the upstream ridge. Then you just have to hope that no unforecasted shortwave appears out of nowhere to accelerate a cold front's progression and dampen the wave expected along your front. But that's another complication entirely...

Happy 2011,

DR

Thanks, DR. If this takes off, I'll split it off into its own thread.

I found this quote particularly interesting, since I am often guilty of this misconception.

Can you elaborate on that passage more (from an operational forecasting perspective), if you have time?

[Absolute Humidity]

144 GGEM is quite interesting

[showmethesnow]

06Z GFS shows some very good potential at 204 for the east coast. Has a positive tilted trough going to nuetral around the Mississippi tapping into the gulf with a 1045 high dropping down from Canada behind it. A strong 50/50 low is present with the -NAO right on top of Greenland. At this point no short waves are present rotating around the polar vortex but I think this is due to the GFS being smoothed out at this time range. I would expect that if this were at higher resolution we would see several available for a potential phase. At present the 50/50 low is to far south and would keep any potential system south effecting the southern mid atlantic. But move that a couple hundred miles north and I would think the whole east coast would be in the game.

[showmethesnow]

After following the models the last several days and the uncertanty of the ejection of the trough I have come to the conclusion that until we actually see the trough in the Southwest moving east we are going to keep seeing completely different solutions from model to model and from run to run. Think there is potential when it does move out but it's basically a crap shoot until then and what players are on the board at that time. So until the models start showing the event 72 to 96 hours away, or the time it takes for the trough to move east when it is ejecting, I will take any storm they show as a fantasy storm.

[Confuzzled]

I think consistency is the key. People really have been down on the models lately...but the EC & GFS actually did an excellent job keying in on the Dec 26-27 threat a good 180+ hours out. They both showed several runs in a row at that time frame of a decent threat on the east coast. So don't s*** on it just cause its so far out...although feel free to do so because its only one run. We see a couple more runs still showing a storm, then maybe its worth noting.

I thought that we want to **** on it PRECISELY because it is so far out? Generally a non linear and chaotic system with sampling errors gets worse with time.

It is a nine day weather model. That is reason enough to **** on it. The Euro also wavered and flip flopped so much on the boxing storm that the EC fanboys would come in here and opine that it accurately predicted whatever happened. That sort of anecdotal nonsense is not scientific and has no place in meteorology.

1) Its one run.

2) Its nine days out.

3) No model consistency yet.

4) Multiple model attestaton lacking.

I'd say all four are legitimate reasons to **** on it at this point. There is nothing to discuss per the subject of this thread. Just something to watch for a few days.

[Ender]

Thanks, DR. If this takes off, I'll split it off into its own thread.

I found this quote particularly interesting, since I am often guilty of this misconception.

A common misconception is that low spread amongst ensemble members necessarily implies more confidence in the ensemble mean. Although a spread-skill relationship sometimes exists, the relationship between ensemble spread and skill varies substantially depending on such factors as the forecast model and the region for which the forecast is made.

Can you elaborate on that passage more (from an operational forecasting perspective), if you have time?

I too found this interesting.

When I see a family, say the GFS and GFFS, or the EC and ECE (I don't want to set off that bitter argument again), clustered around a similar solution that the remaining global models don't seem to "like" I tend to chalk it up to "stupid is as stupid does" IOW, something with the model's physics or resolution is making it, and its children, susceptible to this this type of (likely) error. When I see "clusters of solutions" in the ensemble members I tend to view it as, exceptionally small perturbations in initial conditions and/or very fine timing differences are accounting for one of the two or three clusters of solutions...and, assuming the event is outside the 120hr or so timeframe, it's unlikely that any model, even the ECMWF, has a huge advantage in having had a "better idea" as to the true initial conditions.

As for model spread and mean...I've tended to view it as, high spread = ignore the mean and low spread = compare to the other globals....does this "high confidence" low spread mean represent a likely scenario or merely a weakness in the physics of all its ensemble members. When we're dealing with the situation of multiple clusters of solutions, inside a model family and across the meta ensemble of globals, I do tend to look at those clusters as "one of these clusters of relatively low spread solutions is the likely outcome...now, which one makes the most sense?"

I'm now not even sure the above has much merit. Does it?

As for the coming event, and the prior three ECOP runs, it's interesting that the higher resolution OP run has had repeated problems with cyclogenesis along the SE to MA coast this season while the majority of lower resolution ECE members never really buy into such a solution.

[Ender]

...and I should add, as for this pattern, and anyone left of a RIC to PHL to IPT, if not NYC, we're not in a good spot. The next 10 - 15 days looks abysmal and any model that shows exciting promise outside of the true SE and NJ through New England should be viewed with considerable suspicion, and that's being generous to NJ/PHL...

[usedtobe]

I too found this interesting.

When I see a family, say the GFS and GFFS, or the EC and ECE (I don't want to set off that bitter argument again), clustered around a similar solution that the remaining global models don't seem to "like" I tend to chalk it up to "stupid is as stupid does" IOW, something with the model's physics or resolution is making it, and its children, susceptible to this this type of (likely) error. When I see "clusters of solutions" in the ensemble members I tend to view it as, exceptionally small perturbations in initial conditions and/or very fine timing differences are accounting for one of the two or three clusters of solutions...and, assuming the event is outside the 120hr or so timeframe, it's unlikely that any model, even the ECMWF, has a huge advantage in having had a "better idea" as to the true initial conditions.

As for model spread and mean...I've tended to view it as, high spread = ignore the mean and low spread = compare to the other globals....does this "high confidence" low spread mean represent a likely scenario or merely a weakness in the physics of all its ensemble members. When we're dealing with the situation of multiple clusters of solutions, inside a model family and across the meta ensemble of globals, I do tend to look at those clusters as "one of these clusters of relatively low spread solutions is the likely outcome...now, which one makes the most sense?"

I'm now not even sure the above has much merit. Does it?

As for the coming event, and the prior three ECOP runs, it's interesting that the higher resolution OP run has had repeated problems with cyclogenesis along the SE to MA coast this season while the majority of lower resolution ECE members never really buy into such a solution.

Personally,

I think it does have merit and would prefer to have all the operational models and ensembles in a tight cluster like we had last feb 5th. I'm not sure I agree with all Dave's comments concerning ensembles. However, ensembles in the logest time ranges often don't have enough spread during those times when they are tightly clustered. However, usually as you get into the 3 or 4 day range, those times when they are tightly clustered is usually telling you something and the times whent he models are jumping all over the place is a clear signal that the pattern is not one that's going to be fun to forecast.

[usedtobe]

...and I should add, as for this pattern, and anyone left of a RIC to PHL to IPT, if not NYC, we're not in a good spot. The next 10 - 15 days looks abysmal and any model that shows exciting promise outside of the true SE and NJ through New England should be viewed with considerable suspicion, and that's being generous to NJ/PHL...

Shane, I agree and was going to post something longer on it. It's a cooler than normal pattern but not a good one for us.

[Rainshadow]

...and I should add, as for this pattern, and anyone left of a RIC to PHL to IPT, if not NYC, we're not in a good spot. The next 10 - 15 days looks abysmal and any model that shows exciting promise outside of the true SE and NJ through New England should be viewed with considerable suspicion, and that's being generous to NJ/PHL...

We're ahead of the climo norm and well ahead of the nina climo norm here, no peeps from me about misses to the wide right.

From your other post, I noticed this too about the Euro, it looks as though its giving way too much love to the gulf stream next to the Carolinas even if its idea of a hit along the east coast was ultimately correct on Boxing Day.

[Rainshadow]

Shane, I agree and was going to post something longer on it. It's a cooler than normal pattern but not a good one for us.

Wes,

Happy New Year!!!

I just forwarded Dave's comments to the whole staff at PHI, Walt's probably is typing something back to me right now.

I always felt more comfort in clustering even if it doesn't work all of the time (nothing really does).

Maybe we have the models right where we want them in the medium/long range, showing nothing.

[usedtobe]

Wes,

Happy New Year!!!

I just forwarded Dave's comments to the whole staff at PHI, Walt's probably is typing something back to me right now.

I always felt more comfort in clustering even if it doesn't work all of the time (nothing really does).

Maybe we have the models right where we want them in the medium/long range, showing nothing.

I wish, i do think patterns are important and I don't like the lack of a west coast ridge, that lack makes it hard to dig a vort south of my house. This pattern looks to me like one where you might get northwest flow clippers or flat waves that don't amplify enough to give dc or Phl much if anything. The D+8 and D+11 ensemble mean analogs suggested the snow potential is no better than climo. for Jan 7-14 period with no big snowstorms in the mix at least for DC.

Happy new year to you too! I went golfing yesterday and did well...hadn't played in a month....

[mdsnowlover]

Wes,

Happy New Year!!!

I just forwarded Dave's comments to the whole staff at PHI, Walt's probably is typing something back to me right now.

I always felt more comfort in clustering even if it doesn't work all of the time (nothing really does).

Maybe we have the models right where we want them in the medium/long range, showing nothing.

the blizard of 66 just happened on a saturday, with no advance noticeF!!!!My first really big snowstorm.It could remain cold and dry like 77/78

[easternsnowman]

HM says it's looking good around mid Jan.....at least for Philly and NYC areas.

[mdsnowlover]

HM says it's looking good around mid Jan.....at least for Philly and NYC areas.

this would not surprise me at all, considering the last storm .

[Ender]

Personally,

I think it does have merit and would prefer to have all the operational models and ensembles in a tight cluster like we had last feb 5th. I'm not sure I agree with all Dave's comments concerning ensembles. However, ensembles in the logest time ranges often don't have enough spread during those times when they are tightly clustered. However, usually as you get into the 3 or 4 day range, those times when they are tightly clustered is usually telling you something and the times whent he models are jumping all over the place is a clear signal that the pattern is not one that's going to be fun to forecast....

...Shane, I agree and was going to post something longer on it. It's a cooler than normal pattern but not a good one for us.

[/b]

From your other post, I noticed this too about the Euro, it looks as though its giving way too much love to the gulf stream next to the Carolinas even if its idea of a hit along the east coast was ultimately correct on Boxing Day.

It's reminiscent of most the other cold phases of previous Ninas. Ninas seem to support a blow torch scenario, with GL to NNE events or highly amplified trofs that put us in a bad spot and the SE or eastern NJ through New England in a relatively good spot. IOW, cold and dry or warm and dry for the lower and central MA outside the eastern fringes.

I think one part we're, or I, am probably taking for granted is the relative "simplicity" in last winter's forecast scenarios. The December 19th and February 5th events featured a really potent southern stream system and an almost classic H5 setup for the MA. Basically, with such a powerful southern stream and such a phenomenal feed of moisture, it all came down to a matter of "when" rather than "if". IOW, it was already cold and there was an established feed of more cold air....the system was fairly mature 48 - 72hr in advance and there weren't too many places for said system to head. This year things are a bit different...in fact all potential events have come down to phasing and timing of that phasing. This is a far more difficult problem than simply timing when the wall of moisture will run into the wall of cold. I seem to recall that the 2/10/10 Miller-B wasn't as well handled as were the earlier 12/19/09 and 2/4/10 events, though I'm not certain of that recollection.

As for the Boxing Day storm...well, I guess it comes down to what's the definition of "correct". Personally I see the two (or was it three?) Euro runs that showed flawless run-to-run consistency for an inside HAT 996mb low that tracked to a just east of DE cutoff (at FH 126) as "failure". Especially since none of the other deterministic models and extremely few of their ensemble members liked the idea. Ultimately the EC OP had to speed up its southern stream system by about 6hr and that in turn led to later phasing and a track that was considerably closer to the ECE mean. Earlier I'd suggested that the EC OP was having trouble with systems phasing along the east coast. Perhaps, however, its real issue is in over estimating impact/strength of the baroclinic zone along the land/sea interface, particularly WRT to steep temperature/moisture flux gradient that takes place along the western edge of the gulf stream. This at least would seem like something that may be a fallout of having upgraded to a considerably higher resolution model (as the EC did last year). If the EC's problems are stemming from this, then that'd also explain why the lower resolution ECE members have given the OP run so little support on these events. It'd also explain why later, further offshore, bombogenesis has been the typical result...as that account for the "need to wait" for proper alignment of northern and southern stream shortwaves rather than having taken advantage of the western edge of the gulf stream early on...

Heck if I know.

[thegreatdr]

Within 5-5 1/2 days, using ensemble clustering on the whole has improved my pressure forecast verification, so I do use it. But from 6-7 1/2 days...it's a mixed bag. Some days it works fine, other days it doesn't. You can say the same thing with a deterministic model consensus. Within 5 days (if consistent) some compromise solution works out fine. After that, it depends.

DR

I too found this interesting.

When I see a family, say the GFS and GFFS, or the EC and ECE (I don't want to set off that bitter argument again), clustered around a similar solution that the remaining global models don't seem to "like" I tend to chalk it up to "stupid is as stupid does" IOW, something with the model's physics or resolution is making it, and its children, susceptible to this this type of (likely) error. When I see "clusters of solutions" in the ensemble members I tend to view it as, exceptionally small perturbations in initial conditions and/or very fine timing differences are accounting for one of the two or three clusters of solutions...and, assuming the event is outside the 120hr or so timeframe, it's unlikely that any model, even the ECMWF, has a huge advantage in having had a "better idea" as to the true initial conditions.

As for model spread and mean...I've tended to view it as, high spread = ignore the mean and low spread = compare to the other globals....does this "high confidence" low spread mean represent a likely scenario or merely a weakness in the physics of all its ensemble members. When we're dealing with the situation of multiple clusters of solutions, inside a model family and across the meta ensemble of globals, I do tend to look at those clusters as "one of these clusters of relatively low spread solutions is the likely outcome...now, which one makes the most sense?"

I'm now not even sure the above has much merit. Does it?

As for the coming event, and the prior three ECOP runs, it's interesting that the higher resolution OP run has had repeated problems with cyclogenesis along the SE to MA coast this season while the majority of lower resolution ECE members never really buy into such a solution.

[Rainshadow]

It's reminiscent of most the other cold phases of previous Ninas. Ninas seem to support a blow torch scenario, with GL to NNE events or highly amplified trofs that put us in a bad spot and the SE or eastern NJ through New England in a relatively good spot. IOW, cold and dry or warm and dry for the lower and central MA outside the eastern fringes.

I think one part we're, or I, am probably taking for granted is the relative "simplicity" in last winter's forecast scenarios. The December 19th and February 5th events featured a really potent southern stream system and an almost classic H5 setup for the MA. Basically, with such a powerful southern stream and such a phenomenal feed of moisture, it all came down to a matter of "when" rather than "if". IOW, it was already cold and there was an established feed of more cold air....the system was fairly mature 48 - 72hr in advance and there weren't too many places for said system to head. This year things are a bit different...in fact all potential events have come down to phasing and timing of that phasing. This is a far more difficult problem than simply timing when the wall of moisture will run into the wall of cold. I seem to recall that the 2/10/10 Miller-B wasn't as well handled as were the earlier 12/19/09 and 2/4/10 events, though I'm not certain of that recollection.

As for the Boxing Day storm...well, I guess it comes down to what's the definition of "correct". Personally I see the two (or was it three?) Euro runs that showed flawless run-to-run consistency for an inside HAT 996mb low that tracked to a just east of DE cutoff (at FH 126) as "failure". Especially since none of the other deterministic models and extremely few of their ensemble members liked the idea. Ultimately the EC OP had to speed up its southern stream system by about 6hr and that in turn led to later phasing and a track that was considerably closer to the ECE mean. Earlier I'd suggested that the EC OP was having trouble with systems phasing along the east coast. Perhaps, however, its real issue is in over estimating impact/strength of the baroclinic zone along the land/sea interface, particularly WRT to steep temperature/moisture flux gradient that takes place along the western edge of the gulf stream. This at least would seem like something that may be a fallout of having upgraded to a considerably higher resolution model (as the EC did last year). If the EC's problems are stemming from this, then that'd also explain why the lower resolution ECE members have given the OP run so little support on these events. It'd also explain why later, further offshore, bombogenesis has been the typical result...as that account for the "need to wait" for proper alignment of northern and southern stream shortwaves rather than having taken advantage of the western edge of the gulf stream early on...

Heck if I know.

I pretty much agree with you, if I was a college professor I guess the students would love me because of my grading curve. From a medium range perspective and a pass/fail as to whether the euro solution would affect the mid atlantic vs an offshore scrape, yeah it passed, even though its details and arriving at it left alot to be desired. Even in our CWA I'm guessing about 1/3 to 1/4 of it did not meet warning criteria.

Yes last winter I felt as if we were fishing in a barrel, this winter with no go to model and late phasing/bombing problems it looks like we're going to be working much harder for the money.