iiinova — where your digital self keeps evolving

James Franklin @franklinjamesl.bsky.social · 14/10/2026

The two (accuracy and consistency) are correlated, especially for large samples. Here's an old example for a slightly different consistency metric we were playing with a few years ago. For individual storms the correlations are much lower.

120 0

James Franklin @franklinjamesl.bsky.social · 14/10/2026

Lastly, note to self: next time don't try to do an analysis while the Panthers are playing. And especially while they're winning. (End)

080 0

James Franklin @franklinjamesl.bsky.social · 14/10/2026

After that things get a bit messy, but again, it's the hurricane models that are pulling up the rear with the HAFS being the least-consistent guidance. GDMI is kind of middle of the pack, but again I have to wonder if the rapid intensifiers are part of that. (7/8)

150 0

James Franklin @franklinjamesl.bsky.social · 14/10/2026

For intensity, the IVCN has the most consistent forecasts, followed by OFCL. I'm gonna speculate that the reason OFCL isn't the most consistent this year is because of all the rapid intensifiers that NHC's become more aggressive about lately. (6/8)

150 0

James Franklin @franklinjamesl.bsky.social · 14/10/2026

Next-most consistent is AEMI, also an ensemble mean, although it's a little surprising to me that AEMI is barely more consistent than GFSI, its "parent". Not sure what that means but I don't think it's good. Least consistent are the hurricane models HAFS, HMON, and HWRF. (5/8)

110 0

James Franklin @franklinjamesl.bsky.social · 14/10/2026

The three most consistent track models are TVCN, GDMI, and HCCA. This also really isn't a surprise because these are all consensus models (GDMI being an ensemble mean), and averaging individual forecasts together tends to reduce variability. (4/8)

130 0

James Franklin @franklinjamesl.bsky.social · 14/10/2026

The plot on the left shows that for 2026, OFCL track forecasts have more cycle-to-cycle consistency than any of the guidance models, which is no surprise since NHC explicitly tries to make its forecast consistent. (3/8)

130 0

James Franklin @franklinjamesl.bsky.social · 14/10/2026

The metric shown here is obtained by looking at the last 24 h of forecasts all verifying at the same time, and then calculating the mean change in the predicted storm position or intensity from forecast to forecast. Consistent forecasts will show little change from forecast to forecast. (2/8)

120 0

James Franklin @franklinjamesl.bsky.social · 14/10/2026

With some chatter going around about how consistent the Google DeepMind forecasts were for Isaias, I decided to apply some work Andy Penny, a former NHC colleague of mine, and I did a few years back. Basically, we came up with a few measures of forecast consistency. (1/8)

2176 0

James Franklin @franklinjamesl.bsky.social · 13/10/2026

Looking at my two Panthers Stanley Cup rings and thinking there ain’t gonna be no more. This team looks too old, too slow, too injury prone, and just not good enough.

010 0

James Franklin @franklinjamesl.bsky.social · 12/10/2026

That’s the number of track cases and number of intensity cases, respectively.

020 0

James Franklin @franklinjamesl.bsky.social · 11/10/2026

Quite sure we’re stuck with it for at least another six years. There’s only one other 4-syllable name on the Atlantic lists (Sebastien), but that one seems much easier to say even though it’s not English, because it’s got a nice balance of vowel and consonant sounds.

010 0

James Franklin @franklinjamesl.bsky.social · 11/10/2026

I didn’t remember Olga. I’m not sure that was my best work. ;-)

010 0

James Franklin @franklinjamesl.bsky.social · 11/10/2026

Oh, I know the list. But Isaias has four vowel sounds, three consecutively, with one barely voiced consonant. For me at least, it’s hard to say right without slowing down to think about it. Sadly, I probably don’t have enough trips around the list left in me to get it right.

120 0

James Franklin @franklinjamesl.bsky.social · 11/10/2026

Wasn't planning on this since the sample is tiny, but had two requests, so here are the track and intensity verifications for Hurricane CantPronounce. GDMI big winner for track. Mixed bag for intensity, but the physical hurricane models were positive contributors. (Name credit Marty Merzer.)

2204 0

James Franklin @franklinjamesl.bsky.social · 10/10/2026

Sequence of OFCL and Google DeepMind forecasts verifying ~at the time of landfall for Isaias. GDMI was never more than ~50 nm to the left of landfall and never more than ~4 h slow - very consistent and impressive. OFCL started with larger left and slow biases but corrected them very quickly.

0122 0

James Franklin @franklinjamesl.bsky.social · 10/10/2026

I think the real question on everyone's minds about Isaias is - was it bad enough to retire that name?

230 0

James Franklin @franklinjamesl.bsky.social · 10/10/2026

They probably had no choice, since it was C3 until right before landfall and there was no way to be sure C3 sustained wouldn’t be observed.

050 0

James Franklin @franklinjamesl.bsky.social · 10/10/2026

Not sure if it’s still the case, but when it was introduced the EWW was only for Cat 3 sustained winds. Which technically wouldn’t exist any more.

130 0

James Franklin @franklinjamesl.bsky.social · 08/10/2026

And the Gulf of Mexico the Gulf of America.

030 0

James Franklin @franklinjamesl.bsky.social · 08/10/2026

Also worth pointing out that NHC doesn't actually try to be the most accurate forecast each time, because their forecasters are trying to maintain some consistency from forecast to forecast. But still, what GDMI has done to the rest of the guidance is almost unimaginable.

0110 0

James Franklin @franklinjamesl.bsky.social · 08/10/2026

Forgot the caveat that this sample doesn't include ECMWF, since it's not in the public ATCF decks in real time.

120 0

James Franklin @franklinjamesl.bsky.social · 08/10/2026

A cool verification tidbit: since the Google DeepMind (GDMI) went "live" last year at NHC, it's been the most accurate track forecast 2-3 times as often as OFCL and 4-5 times as often as any of the physics models. For intensity, OFCL is still most frequently the best forecast, except at 5 days.

2258 0

James Franklin @franklinjamesl.bsky.social · 07/10/2026

Where would you place the seasonal cutoff? I'm thinking maybe March 1st?

110 0

James Franklin @franklinjamesl.bsky.social · 07/10/2026

Yeah, there probably are quite a few names that might differ a bit from what’s on the list. Cristobal, Ana, Gaston, Tobias, Brianna, Sebastien (oops, another one w/ 4 syllables), Fernand - are going to commonly be pronounced differently from the guide. But the guide is the target.

030 0

James Franklin @franklinjamesl.bsky.social · 07/10/2026

When WMO added the name in 2009 they also provided the pronunciation. I was actually NHC’s caretaker of the pronunciation guide at the time. Each name has one “official” pronunciation; some are English, some are Spanish, some are French to reflect the RA-IV region.

040 0

James Franklin @franklinjamesl.bsky.social · 07/10/2026

I'd deleted my previous post, but I'll risk another try; Isaias is four syllables, the only Atlantic basin name that has that many. That, along with the three consecutive vowel sounds, makes it a poor communications tool. There's no name like it on the AL, EP, or CP lists, for good reason IMO.

171 0

James Franklin @franklinjamesl.bsky.social · 06/10/2026

Once the president hears the pronunciation of the next tropical storm, I expect an executive order within about a half hour renaming it "America".

0110 0

James Franklin @franklinjamesl.bsky.social · 03/10/2026

Franklin, the coach, neither the storm nor the forecaster…

030 0

James Franklin @franklinjamesl.bsky.social · 01/10/2026

True - I was thinking mainly of screens - first TV and then digital. For 45 years we’ve had a Times print subscription and for most of that time the Herald as well. And we were spoiled by that same advertising model in the early days of digital.

110 0

James Franklin @franklinjamesl.bsky.social · 01/10/2026

It’s a tough change for those of us who’ve spent their lives consuming news content paid for by advertising. I have some subscriptions, but one has to pick and choose. What drives me nuts is to subscribe to something and then there are *still* ads.

100 0

James Franklin @franklinjamesl.bsky.social · 30/09/2026

Been a lot of storms since I ran 2026 AL/EP/CP verifications in early September. Results haven't changed much: GDMI still beating all the guidance for track and slightly better than OFCL on D4-5. For intensity, HCCA best, followed by OFCL and GDMI. Reminder that ECMWF is not in public ATCF decks.

164 0

James Franklin @franklinjamesl.bsky.social · 25/09/2026

My last NOAA flight was Mitch at 905 mb. Definitely broke my personal g-force limitations.

110 0

James Franklin @franklinjamesl.bsky.social · 23/09/2026

Will Hawaii have to contend with Nolo? It's no contest.

051 0

James Franklin @franklinjamesl.bsky.social · 23/09/2026

Surely this must be the first time NHC ever wrote simultaneous advisories on rhyming storms?

030 0

James Franklin @franklinjamesl.bsky.social · 23/09/2026

It’s kind of a hobby. And annoyingly so; I once gave each of the NHC Hurricane Specialists a copy of Garner’s Modern English Usage. A gift I’m sure each of them treasured.

000 0

James Franklin @franklinjamesl.bsky.social · 23/09/2026

Yeah, that works, I guess, but it’s inelegant. Might have been coined that way because “RI” is the common abbreviation and in that context “extreme RI” sounds okay.

110 0

James Franklin @franklinjamesl.bsky.social · 23/09/2026

It has to be modifying “rapid”. If it were only modifying “intensification” then rapid is unnecessary and it’s a measure of net intensity change and not intensification rate.

111 0

James Franklin @franklinjamesl.bsky.social · 23/09/2026

I'm not sure who came up with the term "extreme rapid intensification", but did they have to leave out the "ly"?

360 0

James Franklin @franklinjamesl.bsky.social · 23/09/2026

Might be a fairer comparison to interpolate between best track times and compute the hours between 35 kt and 140 kt?

140 0

James Franklin @franklinjamesl.bsky.social · 21/09/2026

Florida Panthers arena replay operator is in mid-season form. Horrible as ever.

020 0

James Franklin @franklinjamesl.bsky.social · 20/09/2026

I don’t always do pre-season hockey, but when I do, I prefer to avoid watching the Dolphins.

040 0

James Franklin @franklinjamesl.bsky.social · 12/09/2026

I'm usually cautious about laws/rules against poor decision making; tailgating is optional. Plus you can't prohibit it at the last minute. But if you're a player or band member or employee, you've got less say about being in dangerous conditions, so I'm not opposed to smarter scheduling.

020 0

James Franklin @franklinjamesl.bsky.social · 12/09/2026

Lightning can pass fairly quickly but heat doesn’t, so not sure how practical it is to delay games for heat. Unless you’re talking about wholesale rescheduling.

120 0

James Franklin @franklinjamesl.bsky.social · 09/09/2026

Just to clarify, are you talking about the explicit tropical cyclone forecast model tracks (e.g., AVNI, GDMI, etc.)? I didn’t see that in the “what’s new”, nor any way to invoke them.

000 0