Danger of NHC track LINES instead of CURVES between dots

[LarryWx]

The NHC has decided to keep the current format of "connecting the dots" for storm projections for 2005. The point of this post is not to argue against that decision, but to emphasize the problem with how these dots have recently been connected at the NHC website. I feel very strongly that the CONNECTING SEGMENTS SHOULD BE CURVED RATHER THAN STRAIGHT once the overall path projection becomes curved. (I thought that they used to be displayed as curved segments.)

Once a storm is projected to curve, the implication is a curve between two projection points as opposed to straight line segments between these points. This is not a big deal out in the middle of the ocean or even near land with a pretty straight path, but it can be a big problem one the track reaches land if the track indicates, say, a fairly sharp recurvature near the coast. I'll point to some actual archived images of 2004 storm projections on the internet to best illustrate my point:

1) Hermine 2004- Freeze the following movie on frame #5:

http://www.nhc.noaa.gov/archive/2004/HE ... hics.shtml

Note the sharp right turn shown at the first "S". That isn't realistic. It should have been DRAWN as a smooth right turn closer to, if not actually landfalling on, Cape Cod. (It did happen to make landfall near Cape Cod although that is beside the point.)

2) Jeanne 2004- I've got two images on which I'd like to focus:

http://www.nhc.noaa.gov/archive/2004/JE ... hics.shtml

a) Freeze on frame #52. The problem here can best be illustrated by talking about the sharp right turn just before it hits the FL coast. If one simply follows the lines, this track skirts the coast before a first real landfall near Daytona to St. Augustine. So, many in the media could easily have said that the NHC was calling for a DIRECT hit on St. Augustine at 2AM MON and that a place like Melbourne would be spared a direct hit.

However, one knowledgeable about hurricane tracks would realize that the points should be connected with curves, thus making the implied landfall SOUTH of Cape Canaveral near Melbourne before curving up the coast just INLAND and then moving to St. Augustine in a MUCH WEAKENED form. So, more knowledgeable people would realize that this track was implying a DIRECT hit somewhere around Melbourne between 2AM SUN and 2AM MON, and that Melbourne was then in much more danger than St. Augustine! Curves between projection points would prevent this kind of dangerous misinterpretation.

b) Freeze on frame #54. The main problem here is with the final line segment. The 2PM SUN position was inland just north of Orlando with an implied movement still somewhat west of north. Yet, the projection had it then suddenly moving a little EAST of north going offshore EAST of Jacksonville, just EAST of Brunswick, GA. and then giving a DIRECT hit on the GA coast just north of Brunswick! Then it projected it to end up inland a little north of Savannah. More knowledgeable people would realize that these forecast points were implying nothing like that, and instead, a smoother curve. This smoother curve would have kept it very close to the coast or just inland near Jacksonville (rather than offshore east) and then continuing a little INLAND of the GA coast to over or just west of Brunswick RATHER THAN OFFSHORE EAST of Brunswick, and then remaining INLAND the entire GA coast to a point a little WEST and then north of Savannah.

So, some media outlets could easily have interpreted this projection as saying DIRECT hit on the GA coast (i.e. a direct hit following the FIRST direct hit on FL) sometime MON prior to 2PM whereas more knowledgeable people would have realized that a second direct hit was not really being projected. Curved projection point connections would have prevented this kind of dangerous misinterpretation, since they wouldn't have even shown that second direct hit (on GA)!

[T'Bonz]

I prefer the lines.

But I'm not stupid enough to think that they are "gospel". The cone is what I watch.

One thing I LOVED from last season were the graphics at Skeeterbite. Those were very informative and gave me information that I found to be very useful as to what to expect.

[chadtm80]

I prefer the lines.

But I'm not stupid enough to think that they are "gospel". The cone is what I watch.

One thing I LOVED from last season were the graphics at Skeeterbite. Those were very informative and gave me information that I found to be very useful as to what to expect.

Skeeto partnered up with us last year, and I believe we will do the same this year

[LarryWx]

I prefer the lines.

But I'm not stupid enough to think that they are "gospel". The cone is what I watch.

One thing I LOVED from last season were the graphics at Skeeterbite. Those were very informative and gave me information that I found to be very useful as to what to expect.

Skeeto partnered up with us last year, and I believe we will do the same this year

Do they show projection track maps? If so, please post a link so that I can see if/how they connect the dots.

[cycloneye]

http://www.skeetobite.com/weather/

Check it out Larry.

[Guest]

The bottom line is what happened. The NHC forecast was good, and even though at those times mentioned, the emphasis was on the east coast of Florida and Georgia. As we now know, the track shifted further west, and left Georgia out of the scenerio.
The key is to follow the forecasts as their released, and not to try to decipher what they say, but to use the information, without changing it, to inform the public of what is highly possible. This, being the OFFICIAL word and forecast should not be changed, recalculated, construed as only one forecast, or even debated. This should be left to the NHC and NOAA.
When we look at the NHC advisories and forecasts, then turn on the television and see something completely oppisite of what the NHC is forecasting, it can and more then likely has and will cause problems with evacuations and preparedness, let alone, it has and will be a major factor on whether some people decide to make the preparations or to evacuate from an area that is forecast to be affected.
The margin of error is clearly shown on the forecast maps, and it is clearly explained that "IF" you are in the Bubble, or Cone, that you should be prepared to, or to take action to make preparations.
So, when reanalysing the past, is it actually helpful or is it just showing that the prior forecast tracks were off?
Keep in mind, not the TV mets, Radio, or any other form of weather outlets even The Weather Channel, Accuweather, or even our sites, are the official word. Only the NHC/NWS and NOAA!!!!!

[LarryWx]

http://www.skeetobite.com/weather/

Check it out Larry.

Thanks cycloneye!

I found a map at the following link at "skeetobite" that illustrates my concern:

http://www.skeetobite.com/weather/maps/NHC_ADV29.gif

This shows the NHC projection of Frances as of 11PM 8/31. The track has the center passing barely to the east of Orlando whereas the implied path (using curved segments) would have had it just WEST of Orlando, i.e. higher impact at Orlando. The point is that assuming the NHC's implied path is to the west of Orlando, why not show it that way IF some exact path is going to have to be displayed?

[LarryWx]

This, being the OFFICIAL word and forecast should not be changed, recalculated, construed as only one forecast, or even debated. This should be left to the NHC and NOAA. When we look at the NHC advisories and forecasts, then turn on the television and see something completely oppisite of what the NHC is forecasting, it can and more then likely has and will cause problems with evacuations and preparedness, let alone, it has and will be a major factor on whether some people decide to make the preparations or to evacuate from an area that is forecast to be affected.
The margin of error is clearly shown on the forecast maps, and it is clearly explained that "IF" you are in the Bubble, or Cone, that you should be prepared to, or to take action to make preparations.

Bob, I agree with just about all you are saying here and I feel the NHC, indeed, does a good job overall.

1) I agree there should just be one "official" forecast, the one from the NHC, to limit confusion. However, the way the dots are connected can cause diferent media outlets to interpret the NHC projection points differently imo. (per my examples)

2) Keeping in mind the inherent uncertainities in forecasting the tropics, I also strongly believe in emphasizing the idea of cones of uncertainty as opposed to a single track. If I had my way, there would be no track displayed past 3 days and instead just the cones at most if there were a clear way to display them. However, that is outside of the point I'm trying to make. I'm saying that IF the NHC must show a single track and it is, say, recurving near the coast, why not show it in a way that best reflects the true NHC feelings, i.e. with curved segments?

[Guest]

As in this gif:
http://www.skeetobite.com/weather/maps/NHC_ADV29.gif
It shows straight lines between each forecast period. To curve the projected forecast track could create more confusion with the media. This would be an excellant question to ask during the upcoming shows.
Mike Watkins will have Stacy Stewart on his show Tuesday, and I will have Tom Terry on my show April 7.
But, if the NHC was to show a curved plot between referanced times, it could take out of the projected path some areas along a coast.
Lets say, that in that image, this would have left Orlando in a weaker wind field, and the coast in a higher wind field. This is based on the winds being stronger on the right side, and weaker on the left side.
As we know, the actual path was further west then north.
http://www.skeetobite.com/weather/maps/ch_fr_jnpath.gif
If the forecast track in the advisory 29 had occured, there would have been a sharp right turn, similar to Erin as it approached South Florida in 95(?).
The trough didn't dig, and the forecast shifted west.
We know, and the media knows, that prior to landfall, these forecast tracks can and more then liely change, and sometimes drastically. Looking at the advisory 29, how would you have prepared or told your listeners to prepare if you lived in Jacksonville, Tampa, Orlando, or Miami?
Some of the media outlets in tampa actually stated not to worry, we won't be affected. We were discussing this just prior to landfall with Frances.
The forecast track showed, there was a turn expected, but when was the key. This was even stated from Max Mayfield.

[MWatkins]

Larry...

First of all great topic...it was well-thought out with really good examples...and I have often wondered the same thing...in fact I used to do that ALL the time when I graphed all of these tracks in XL back in the day...in fact I used curved/smoothed lines in my graphing tool rather than linier displays between points.

With that in mind, I think there would be the same problems with a curved line vs. a straight line between points. For example, how would you show the curve? If a storm was forecast to move westward for 9 hours, then due north for three, then the curve should be sharper and later in the grid than a storm that moves W for 6 hours and N for 6 hours.

Also, we know for a fact that storms wobble along track (see 1.54 million posts from last year on which was Ivan was moving (NW vs. WNW)) so how does that get communicated? After thinking about this all day, I believe I've rested on the opinion that you don't solve any of the problems the line causes with a curve...the same uncertainty is still there.

Furthermore...I wonder if any more detail is put into the line itself (in terms of how it's displayed) then wouldn't that tend to put more focus on the line and less on the dreaded cone of uncertainty.

However...perhaps a compromise can be reached. Perhaps more data points should be included when displaying the track. Perhaps adding 6 hour plots for the first 36 hours, then going to 12 through 120 would be the way to go. The forecast plots are determined at the 60, 84 and 108 hour intervals anyway...why not just include them in the track forecast instead of leaving a 24 hour gap in the forecast after 48 hours? Unless these gaps are intentional late in the period to convey uncertainty. It's also just as likely that the 12 to 24 hour track time spacing is done that way because that's the way it's always been done. Anyway more datapoints would help your concern about the shape of the track especially in the short term. But it may add to the line vs. cone concern.

Anyway...great post. If nothing else...it's a great example of how inexact and difficult hurricane track forecasting can be...

MW

[StormChasr]

I think the collective opinion seems to be that it was an extremely foolish decision to keep the black line. The "zone of probability" would have been far more effective, and far more accurate. As regards how well the NHC did in terms of forecasting, the jury's out on that one---their conclusions were based on blind adherence to bad data in the computer models from which they were basing their forecasts.

[caneflyer]

The forecast plots are determined at the 60, 84 and 108 hour intervals anyway...why not just include them in the track forecast instead of leaving a 24 hour gap in the forecast after 48 hours?

Actually, not true, if I understand you correctly. No official forecast, even internally or informally, is made at 60, 84, or 108 h. Model guidance is available at those times, but that's it.

(It stands to reason that no such official forecast would exist - why go to the trouble of coming up with a 60 h position, for example, if it is not going to be disseminated?)

[vacanechaser]

Nice reply Mr. Watkins. And good topic Larry.

Here is the thing I think you guys are missing all together, the line does not mean crap. Period. Everyone here should know that if a hurricane's center is forecast to stay offshore by 5,10,15 miles or so, that the eyewall, which really matters here could still impact the coastal area near that center point. The media has a hard time for whatever reason getting this point across. The unfortunate thing here is that most people dont pay any attention to what is being said anyway. The problem with Charley, which this discussion is stemming from, came at the coast at an angle that exaggerated the "turn to the northeast" more so than if it was moving due west at the coast. I think more attention needs to be payed to explaining to the folks out there that are dense and have a hard time understanding what the point on a map really means. The ceneter after all means the calm part of the storm. The "point" as it were seems to be lost here.


Jesse V. Bass III
http://www.vastormphoto.com
Hurricane Intercept Research Team

[vacanechaser]

One thing I forgot to mention, maybe the NHC should consider putting out the area of maximum winds and hurricane force winds and tropical storm force winds with the "cone and point" this way, people would know if they need to be ready for the worst, or can relax just a bit. However, that could bring about other problems. People in the area of a landfalling hurricane or T.S. should always be prepared for the worst case.


Jesse V. Bass III
http://www.vastormphoto.com
Hurricane Intercept Research Team

[MWatkins]

The forecast plots are determined at the 60, 84 and 108 hour intervals anyway...why not just include them in the track forecast instead of leaving a 24 hour gap in the forecast after 48 hours?

Actually, not true, if I understand you correctly. No official forecast, even internally or informally, is made at 60, 84, or 108 h. Model guidance is available at those times, but that's it.

Although officially, I do concede, 60, 84 and 108H positions are not maintained (even in ACTF...which I though were but those are the interopolated positions not the Offical Forecast), they have to be resolved somewhere along the way if for no other reason than forecast verification. I suppose I'm making an inductive leap that they're calculated by the forecaster as the track is assembled (I have no knowledge about this either way)...and since those numbers almost hae to be dealt with why not just keep the same 12 hour resolution all the way through the process? Unless, as I mentioned, you want to switch to 24 hour timeframes in the extended period to convey the uncertanity.

That time period switch, to the lay person who doesn't get that the 120 hour position is actually 117 hours from forecast time, makes it look like the storm is speeding up in many cases when there are just less data points along the way.

Really the main point is that the curve an be represented by more datapoints to address Larry's concern. That's a solution...perhaps not a good one if the goal is to get away from looking at the line in the first place.

(It stands to reason that no such official forecast would exist - why go to the trouble of coming up with a 60 h position, for example, if it is not going to be disseminated?)

I think you almost have to come up with it. Since the motion and bearing for a storm are used to calculate or validate the next lat/long pair I think it's almost a given that you have to come up with the number even if it isn't used in an official forecast, again even if you're only using it for verification.

Again...just wading in on the matter...

MW

[caneflyer]

Although officially, I do concede, 60, 84 and 108H positions are not maintained (even in ACTF...which I though were but those are the interopolated positions not the Offical Forecast), they have to be resolved somewhere along the way if for no other reason than forecast verification.
MW

This is not a big deal, but it seems like you are confusing the forecast process with the best-tracking process. All verifications are done against the best track, created at 6 h intervals after the storm is over. The forecasts that were issued in real time don't have anything to do with the creation of the best track, which is created weeks later. So I don't see why a 60 h forecast would be needed for "verification" or for any other purpose. A 48 h forecast is made, and verified against the best track 48 h (okay, 45 h) later. Then you make a 72 h forecast and verify it against the best track 72 h later. Where is the need to make a 60 h forecast?

[MWatkins]

So I don't see why a 60 h forecast would be needed for "verification" or for any other purpose. A 48 h forecast is made, and verified against the best track 48 h (okay, 45 h) later. Then you make a 72 h forecast and verify it against the best track 72 h later. Where is the need to make a 60 h forecast?

Ok. I agree it's not a big deal (it's a good discussion though, no?) and it doesnt have to be done for track verification for the public product (OFCL forecast vs Best Track), but I'm not speaking specifically about the public product. If you are going to keep the models at that resolution (12 hour intervals), it would stand to reason that you would have no other choice than to keep the baseline you're using for comparision (in this case the official forecast) in the same scale if for no other reason than to see where the bigger jumps in error rates begin (72 vs 84 vs 96 etc).

I can assure you I'm not confused about the process...I have never thought about the wheher or not keeping 12 hour tracks past 48 hours was really done before this topic came up anyway (I just always thought it should be and was)...sounds like you may know this so I'll ask (and I'll hopefully be clear about the question).

We know, without argument, that the NHC writes a forecast every 6 hours. We also know that part of the data used is the interopolated forecast from the last track (in other words, the 12Z forecast coordinates are interpolated for 18Z using the previous 6 hours of motion to calculate datapoints for the next 120H (starting at 18Z position).

We also know that the 12 hour spacing is maintained for the interpolated tracks through 120 (see ATCF).

So then how are the interoplated points at forecast time + 66 hours derived if they are not maintained in the original set? They have to be there somewhere, right? Unless the mid point of the 48(+6) and 72(+6) hour position is a calculation based on the exact mid-point between the new positions...which although could be handled in a calculation...seems like more work then actually working at 12 hour resolution all the way out to 120. I think I have proably answered my own question there...I don't see any other way to do it.

Also, from a science standpoint, why would you not care if your 84 hour forecast is any better than the 96h and not any worse than the 72h? Why would you introduce the chance for error when shifting scales after 48 hours? I dont see the scientific benifit of saving time by calculating 3 less data points when they could be used to determine error/skill/etc. Seems like the effort is well worth it to me.

This really is more of an acedemic discussion anyway since my basic position on the matter is that why would you mess around with the line when you want people to focus on the cone anyway?

MW

[caneflyer]

I think you've figured it out, so hopefully the following won't confuse the issue.

The way the interpolated tracks are constructed is a little difficult (well, more than a little difficult) to explain. This description is from the NHC web site:

"Fortunately, a simple technique exists to take the latest available run of a late model and adjust its forecast to apply to the current synoptic time and initial conditions. In the example above, forecast data for hours 6-126 from the previous (06Z) run of the GFS would be adjusted, or shifted, so that the 6-h forecast (valid at 12Z) would exactly match the observed 12Z position and intensity of the tropical cyclone. The adjustment process creates an “early” version of the GFS model for the 12Z forecast cycle that is based on the most current available guidance. The adjusted versions of the late models are known, for historical reasons, as interpolated models."

In the case of making an OFCI out of an OFCL, however, it is a little more complicated than that, because OFCL is only available at 0, 12, 24, 36, 48, 72, 96, and 120 h. First, as in the example above, you shift the whole of the previous OFCL forecast to make sure the new OFCI starts in the right place. Think of this as a simple lifting of the previous forecast off the map and placing it back down so that the 6-h (interpolated) forecast position is on the actual current position of the storm. Then, you have to use interpolation of the shifted forecast to fill in points wherever they are needed. So, for example, you would get your 24-h OFCI by interpolating a 30-h position from the (shifted) previous 24 and 36 h OFCL positions.

I understand that cubic splines (a kind of curve) were used in the past to do the interpolation, rather than straight lines, but this method gave peculiar results and was abandoned.

The bottom line, as it were, of course, is to ignore the line and focus on the cone!

[MGC]

Using a smoth line will do little with reguards to the novice hurricane tracker, whom just recently retired to Florida (or any coastal area) that has never experienced a hurricane decide wether to stay or flee. Sorry, but I can still recall the interviews of the Hurricane Charley survivors whom thought they were out of the woods because the thin line the NHC uses bypassed them and instead implied a Tampa Bay area landfall. A cone the diameter of hurricane force winds would be appropriate....MGC

[MWatkins]

The bottom line, as it were, of course, is to ignore the line and focus on the cone!

Of course!

Gotcha...I didn't do a good job filling in some gaps there...but we are on the same page. I was just curious how those spaces were filled for the lat/long pairs where the 24 hours gaps weren't filled in...but I think you've answered that.

Thanks for the discussion...

MW