Hurricane Forecasts - Why The Inaccuracies?

[Bill Tabor]

Ok, before you all slam me for how great a job we do, and considering all things they are very accurate - I'll say YES I agree. But there are some things I've seen with Rita and Katrina that I don't quite understand and it seems odd that it hasn't been caught by the models or the forecast teams.

1) With both Rita and Katrina they were both very strong initially up to 175 mph before landfall, but then they appeared to entrain dry air primarily as shown on water vapor which weakened them. This is simplistic explanation of course but can get the ball rolling. Anyway, it appears that the models did a poor job of anticipating this.

2) I've looked at the strength forecasts for both canes before and after strengthening and in both cases the models were underforecasting abrupt strengthening by a category or two. They also seem to be missing such as now I believe - the amount of weakening. The further in time you go the error gets much greater.

3) For some reasons the forecast models keep clustering around the wrong area. Originally for instance Rita was headed further west, and then to Galveston area, and now supposedly almost to the Tx/La border. Does this bother anyone or is this at acceptable levels in your opinion?

4) How do the models apparently miss the dry air, etc when it is so vastly seen on water vapor? Also it doesn't seem like the forecasters are looking at this and making their own conclusions regarding the impact. In other words they don't seem to be deducing that the dry air, or ERC, SST or whatever are going to have an effect apart from what the models are saying and changing their forecast appropriately.

Ok, with all that said this is not a bash session. I realize we've come a long way and we do the best we can, and I appreciate what we do have and how hard folks are working. Part of this is I suppose I want to learn more what is going on? Is this just the state of the art, and about all we can do? What are your observations regarding these types of items?

 

[Scott Olson]

Ok, before you all slam me for how great a job we do, and considering all things they are very accurate - I'll say YES I agree. But there are some things I've seen with Rita and Katrina that I don't quite understand and it seems odd that it hasn't been caught by the models or the forecast teams.

1) With both Rita and Katrina they were both very strong initially up to 175 mph before landfall, but then they appeared to entrain dry air primarily as shown on water vapor which weakened them. This is simplistic explanation of course but can get the ball rolling. Anyway, it appears that the models did a poor job of anticipating this.

2) I've looked at the strength forecasts for both canes before and after strengthening and in both cases the models were underforecasting abrupt strengthening by a category or two. They also seem to be missing such as now I believe - the amount of weakening. The further in time you go the error gets much greater.

3) For some reasons the forecast models keep clustering around the wrong area. Originally for instance Rita was headed further west, and then to Galveston area, and now supposedly almost to the Tx/La border. Does this bother anyone or is this at acceptable levels in your opinion?

4) How do the models apparently miss the dry air, etc when it is so vastly seen on water vapor? Also it doesn't seem like the forecasters are looking at this and making their own conclusions regarding the impact. In other words they don't seem to be deducing that the dry air, or ERC, SST or whatever are going to have an effect apart from what the models are saying and changing their forecast appropriately.

Ok, with all that said this is not a bash session. I realize we've come a long way and we do the best we can, and I appreciate what we do have and how hard folks are working. Part of this is I suppose I want to learn more what is going on? Is this just the state of the art, and about all we can do? What are your observations regarding these types of items?

I don't know why anyone would bash you these are good questions. The models are much more accurate on track than in intensity. However, the GFDL has done pretty well with some of the abrupt strengthning this year.
Certainly more refined models and more ocean data is necessary to determine environments conductive to weakening/strengthning. As far as clustering moving around that's not suprising with all the wobbles and slight shifts in course would affect small consensus changes. Though it's pretty impressive that at 100+ hrs out the models could predict a region where this would occur and then they usually work around 40+ miles. I think forecasters do take SST's and TCHP into context when wondering whether to believe intensity forecasts and dry air entrainment is harder to forecast IMO.

 

[rdale]

Barons model (BAMS or VipirCAST) had the storm coming onshore near the TX/LA border for days... I guess it's all in the dynamics of the model, versus data input, versus luck? There's no one thing you can put on a model that explains why it works (or doesn't work.)

- Rob

 

[Jeff Snyder]

It's important to note that the water vapor imagery only show mid to upper level imagery. You could still plenty of "moisture" in the low-level to drive a storm. There may be cloud-scale entrainment on the periphery of the convection, but it is important to remember that the water vapor doesn't tell you anything about low-level moisture, which is where the bulk of moisture is locate anyways... With hurricanes, moisture means latent heating... Perhaps precipitable water would be a better indicator of where the "dry air" is located.

I think a lot of it just comes down to the fact that we just don't know a lot about the internal dynamics of hurricanes. Much of the time, we can't even forecast ERCs beyond 4-6 hours, which is remarkable considering how important ERC seems to be for intensity. There is an awful lot that just isn't understood yet. If it isn't understood very well, you can't model it very well.

I think a some of it comes down to the lack of observation on which models initialize. This isn't as important in the Gulf, since the gulf is surrounded by surface and upper-air obs, and the Gulstream IV usually makes synoptic upper-level dropsonde missions when a big hurricane is setting its sights on the US. However, this would seem to be considerably more important for those storms that are still in the open atlantic (east of the bahamas/antilles). Yeah, there's likely some ACARS data, and ship reports, but I'd think most of the other obs come from satellite estimates. Without many obs, again, you can't expect a quality model guidance suite.

What comes out of a model is only as good as what goes into it.

 

[Bill Tabor]

Some good points guys. I suppose we have made big strides and improvement, but there is still a certain amount of it that is still 'art'. Guess it just bugged me and made me wonder when I saw some of these sudden (supposedly unanticipated) changes. Perhaps as far as models are concerned that says something about their ability and tendency to trend as opposed to noticing and utilizing some absolute rules that cause changes - albeit even if they are initially minor changes. As some of you say it is a complicated system and our modeling rules just don't cover all the bases and exceptions yet.

 

[Dan Dawson]

Ok, before you all slam me for how great a job we do, and considering all things they are very accurate - I'll say YES I agree. But there are some things I've seen with Rita and Katrina that I don't quite understand and it seems odd that it hasn't been caught by the models or the forecast teams.

1) With both Rita and Katrina they were both very strong initially up to 175 mph before landfall, but then they appeared to entrain dry air primarily as shown on water vapor which weakened them. This is simplistic explanation of course but can get the ball rolling. Anyway, it appears that the models did a poor job of anticipating this.

2) I've looked at the strength forecasts for both canes before and after strengthening and in both cases the models were underforecasting abrupt strengthening by a category or two. They also seem to be missing such as now I believe - the amount of weakening. The further in time you go the error gets much greater.

3) For some reasons the forecast models keep clustering around the wrong area. Originally for instance Rita was headed further west, and then to Galveston area, and now supposedly almost to the Tx/La border. Does this bother anyone or is this at acceptable levels in your opinion?

4) How do the models apparently miss the dry air, etc when it is so vastly seen on water vapor? Also it doesn't seem like the forecasters are looking at this and making their own conclusions regarding the impact. In other words they don't seem to be deducing that the dry air, or ERC, SST or whatever are going to have an effect apart from what the models are saying and changing their forecast appropriately.

Ok, with all that said this is not a bash session. I realize we've come a long way and we do the best we can, and I appreciate what we do have and how hard folks are working. Part of this is I suppose I want to learn more what is going on? Is this just the state of the art, and about all we can do? What are your observations regarding these types of items?

Keep in mind that the vast majority of the models you are referring too are fairly coarse resolution and do not resolve the inner core of the hurricane very well. The GFDL is the only operational model that is really capable of that right now. Even if you can resolve the inner core, as Jeff stated, we cannot observe on that scale except with a limited array of instruments, such as radar and satellite-derived observations. So, the initialization of the hurricane vortex is poorly done. You need this information to accurately describe how the inner core is going to interact with such things as dry air intrusions and eyewall replacement cycles. A high resolution model run of a hurricane may indeed produce ERC's in the model storm, but there is no guarantee that they will have any relation to the real ERC's present in the real storm. We have a long way to go before we can faithfully represent anything approaching a one-to-one correspondence of features on the scale of a hurricane inner core between models and observations, similar to the problem of trying to predict individual supercell thunderstorms during a severe weather outbreak. I haven't even talked about the representation of physics in the models. You have to make sure you have a good handle on such things as cloud microphysics, surface flux processes, and turbulence, and in our current models, there is a lot of room for improvement in each of these areas. Without accurate representations of such processes (we need lots more observations to validate our simulations in this regard), your model representation of a hurricane, for example, may not faithfully reflect the actual processes in a real hurricane in many respects.

In short, we have a long way to go in multiple areas: model initialization, model physics, model resolution, in order to faithfully represent the inner core dynamics of hurricanes in models (even in idealized simulations, much less in real data prediction cases). Remember it is the inner core dynamics that largely controls the intensity changes of a hurricane. This is why models don't handle intensity too well.

Dan

 

[Jeff Lawson]

With the exception of the rapid deepening (which is always a possibility with an intensfying storm), has Rita behaved all that unexpectedly? Even when the storm was sitting at 897mb, forecasts called for gradual weakening leading up to landfall. So far, that's exactly what's happened.

As for the models "missing" the landfall point, perhaps we should wait for the storm to actually make landfall before making a final evaluation of their performance. For the time being, however, the models seem to be performing about as well as expected. I haven't seen any truly screwy solutions.

 

[Mike Johnson]

Part of the problem is this: If you read (carefully) the NHC discussions, you will notice subtle hints of what they are thinking -- but do not dare mention to the public. I wrote to this forum two days ago after reading into the NHC discussions that they might be hinting of the storm weakening before landfall. The last thing they need to do is tell the public that there is a better than 50/50 chance the storm will poop out before landfall. This would send thousands back home. The other problem are the Accu-WX types who think they know everything and go on the air spreading opinions not based on any sound meteorology. Only last night the Bastardi guy was continuing to claim Galveston was going to take a direct hit by a Cat-5. What if -- for example only -- the decision to drive that bus North was based on his (wrong) forecast.. I almost get the feeling that some forecasters are building all this up so if its big -- they look like heros. I think the NHC did an overall good job.

Mike

 

[Andrew Khan]

Originally posted by Bill Tabor

3) For some reasons the forecast models keep clustering around the wrong area. Originally for instance Rita was headed further west, and then to Galveston area, and now supposedly almost to the Tx/La border. Does this bother anyone or is this at acceptable levels in your opinion?

Yes, sure it's quite bothersome, but the fact remains that there path changes constantly. I still think, with that said, that sometimes the forecasters do not concentrate on the right things sometimes. Sometimes I think they do not look to far into what will change/alter it's directed path and such.

 

[Dick McGowan]

[/quote]

Yes, sure it's quite bothersome, but the fact remains that there path changes constantly. I still think, with that said, that sometimes the forecasters do not concentrate on the right things sometimes. Sometimes I think they do not look to far into what will change/alter it's directed path and such.[/quote]

Andrew, could you be a little more specific? What "right things" are you referring to? I guess I just don't understand WHAT you are trying to say. I think the forecasters have considered every possibility with Rita and every other hurricane for that matter.

 

[Glen Romine]

I think it is fair to say here that the forecast quality has not been as great as was enjoyed with Katrina. If you look at the model forecasts from a few days ago - all of the solutions were west of where landfall is going to occur. So at what point did the models fail to capture the 'real' evolution? Dropsondes have proven very helpful in improving hurricane track forecasts - but perfection is still way out of reach. It costs ~ 1 million per mile of coastline evacuated - so do the math on this one and it will carry a hefty price tag for the eastward shift in track. Rita largely is following a simple Beta drift - even this isn't really all that simple.

As previsouly mentioned - one of the greatest weaknesses in hurricane forecasting is intensity. Unfortunately - the intensity also controls the amount of air 'pumped' through the hurricane, and subsequent subsidence that then occurs around it which feeds back to the hurricane track. Since the hurricane actively interacts with the ocean beneath it - a truely accurate model will also need to be coupled with the ocean below - which requires not only very accurate atmospheric state estimates, but also exact ocean state estimates. Also, all of the physical processes need to be accurately represented, as well as having the resolution needed to capture all of the important scales. Reality is probably that substantial improvements in intensity forecasts are unlikely to happen anytime soon.

Nevertheless, trying to understand the controlling factors of hurricane intensity, and particularly what leads to extreme intensification episodes as was seen with Rita and Katrina is still an active area of research - much like understanding tornadogenesis. There just haven't been many events well studied - and it takes time to transfer knowledge from research to operations. I'm hopeful actually that folks remain unsatisfied with the quality of forecasts - because so long as that is the case there will be jobs around to try and make them better.

Glen