2024 Global Model Runs Discussion (Out thru day 16)

[Teban54]

More improvements to the ECMWF. The ALFS goes to 360 hours.

https://twitter.com/TropicalTidbits/status/1763344186571632865

https://twitter.com/TheSteveCop/status/1763358239125623014

Great, more La La Land 300+ hour model canes for us to talk about!

[cycloneye]

Reactions to the ALFS.

 https://twitter.com/pppapin/status/1763358064592445690




 https://twitter.com/TropicalTidbits/status/1763362148061958550




 https://twitter.com/TropicalTidbits/status/1763364000929296888

[tolakram]

More improvements to the ECMWF. The ALFS goes to 360 hours.

https://twitter.com/TropicalTidbits/status/1763344186571632865

https://twitter.com/TheSteveCop/status/1763358239125623014

Great, more La La Land 300+ hour model canes for us to talk about!

AI lala land! It has to be true.

[cycloneye]

GFS getting better.

 https://twitter.com/RyanMaue/status/1763644265899360508

[zzzh]

Very impressive monsoon trough extending from 70E all the way to the dateline in southern hemisphere (thanks to the strong MJO). Models are showing ~4 storms out of this.

[kevin]

12z GFS has a subtropical-esque system of the coast of Florida at +200 hrs. There are also some other model runs which show some weak subtropical activity in the long range, albeit all from different origins and thus not really consistent. Almost certainly not gonna happen considering the time range, but the fact that stuff like this occurs in some model runs does show how anomalously high the SSTs already are (especially taking into account that there has only been a single TC in the month of March since 1851).

[MetroMike]

12z GFS has a subtropical-esque system of the coast of Florida at +200 hrs. There are also some other model runs which show some weak subtropical activity in the long range, albeit all from different origins and thus not really consistent. Almost certainly not gonna happen considering the time range, but the fact that stuff like this occurs in some model runs does show how anomalously high the SSTs already are (especially taking into account that there has only been a single TC in the month of March since 1851).

https://i.imgur.com/GjtfyO7.png

Well If something like that going to form in March this is the year for it to happen. A non zero chance as it has happened once before.

[JetFuel_SE]

12z GFS has a subtropical-esque system of the coast of Florida at +200 hrs. There are also some other model runs which show some weak subtropical activity in the long range, albeit all from different origins and thus not really consistent. Almost certainly not gonna happen considering the time range, but the fact that stuff like this occurs in some model runs does show how anomalously high the SSTs already are (especially taking into account that there has only been a single TC in the month of March since 1851).

https://i.imgur.com/GjtfyO7.png

Well If something like that going to form in March this is the year for it to happen. A non zero chance as it has happened once before.

It was a C2 in 1908

[TheAustinMan]

12z GFS has a subtropical-esque system of the coast of Florida at +200 hrs. There are also some other model runs which show some weak subtropical activity in the long range, albeit all from different origins and thus not really consistent. Almost certainly not gonna happen considering the time range, but the fact that stuff like this occurs in some model runs does show how anomalously high the SSTs already are (especially taking into account that there has only been a single TC in the month of March since 1851).

https://i.imgur.com/GjtfyO7.png

At this range, there's fairly decent consensus among the deterministic and ensemble guidance that throughout next week the subtropical jet will be roaring across the Gulf due to a blocking ridge setting up over the western U.S. this weekend. As one might expect, a few upper level shortwave troughs are probably going to speed along that jet, and with the comparatively reduced atmospheric stability over the Gulf (especially over the >25C waters of the Loop Current), that might help ignite some localized areas of stronger rising motion and convection when a subtle trough swings by, as depicted in that particular GFS run for example.

That might produce localized areas of heavier precipitation and gusty or potentially damaging winds, but would probably not lead to something particularly noteworthy from a tropical or even subtropical perspective. Any vortex that develops in that sort of environment is going to fly across the Gulf in 24 hours (or less) and would be subjected to more than 80 kt of shear (and that's low-balling it; model outputs right now depict over 100 kt of deep-layer wind shear across the region).

[zzzh]

EC and GFS are showing a potential storm in western subtropical Atlantic in day ~9 range. Note that the origin is different on the 2 models.

[Teban54]

EC and GFS are showing a potential storm in western subtropical Atlantic in day ~9 range. Note that the origin is different on the 2 models.

[ChrisH-UK]

The GFS, ECMWF and CMC models do show something happening in the atlantic but it looks like it be a bit of a cyclonic mess and likely come to not much. Upper level winds look to be fairly good, little wind shear. Waters are not warm enough yet to drive anything tropical, the GFS does have it heading south towards which might be something to key a eye on.

[JetFuel_SE]

GFS showing a storm in the CPAC, probably just your average phantom thing.

[cycloneye]

The new era of models=AI

 https://twitter.com/BenNollWeather/status/1771252343151239183




Here is his complete post.

Here are my thoughts on the new ECMWF Artificial Intelligence Forecasting System (AIFS) over the last 3 weeks:

1) Given the relatively coarse resolution, I find it to be most useful at medium-range pattern/storm detection (after day 5).

2) During periods of higher forecast uncertainty, I look to it for a confidence boost or to provide another possible outcome that physics-based modelling might not be sniffing out.

3) I don't find it particularly useful at short-range for the reason described above (resolution). Some model fields have a smoothed output that looks non-physical.

4) I have created products that extract forecast rainfall for key locations (e.g., New Zealand's hydro catchments) every AIFS run, emailed to me every six hours; I've generally found these forecasts to be less changeable than traditional physics-based modelling.

5) Because AIFS ensembles are not yet available, the output is purely deterministic. AIFS ensembles would be particularly helpful as Atlantic hurricane season approaches. In the absence of ensembles, time-lagged products or presenting the last several runs side-by-side is an alternative.

Three additional thoughts on AI modelling:

1) Things are moving quickly. New, higher resolution AI models are being developed constantly. Some will provide more realistic depictions of mesoscale processes like convection.

2) It's new: understanding bias or lack thereof will take time. Artificial intelligence-based modelling is part of our forecasting future, so it's best to learn about it, look at it, and to try to stay ahead of the curve.

3) AI modelling in the year 2034 or 2044 will be better than 2024 - models can and will be re-trained on the latest data, becoming more skillful.

I can't imagine what the state of the science will be in 5-10 years. It's an amazing time to be a meteorologist.

[Teban54]

 https://twitter.com/AliciaMBentley/status/1772764188772900929



 https://twitter.com/AndyHazelton/status/1772969865021678079

[USTropics]

https://twitter.com/AliciaMBentley/status/1772764188772900929?s=19
https://twitter.com/AndyHazelton/status/1772969865021678079?s=19

Been looking forward to this being released after seeing some of the products at AMS, some observations from last season:

Regarding absolute track error for North Atlantic tropical cyclones last season, the global models are tightly clustered through 72 hours. The ECMWF continues to excel here though, although the UKMET at 5 days (~120 hours) did slightly better.


For intensity, the GFS actually outperformed the other models in this category. Looking at intensity bias, the global models typically undercast intensity.


Comparing this to the hurricane/mesoscale models, the HFS models performed the best here. Again, there is a bias towards over intensifying tropical systems, especially after 72 hours.

[Teban54]

Not exactly what the thread is for, but this is about official forecast improvements:
 https://x.com/BMcNoldy/status/1777333028815245456




Reposting some figures from Brian McNoldy's blog post. NHC cone size (i.e. track errors) over time:


Intensity errors over time:

[USTropics]

Not exactly what the thread is for, but this is about official forecast improvements:
 https://x.com/BMcNoldy/status/1777333028815245456




Reposting some figures from Brian McNoldy's blog post. NHC cone size (i.e. track errors) over time:
https://i.postimg.cc/503G7Dfz/image.png

Intensity errors over time:
https://i.postimg.cc/Kcq2N3qY/image.png

I would agree to an extent, I don't think we will see any 'large' jumps in improved forecasting skill in terms of track (like we saw from 2004 to 2014), but smaller incremental improvements I would still expect. Models can still be improved, and with AI/machine learning just now being introduced, there is room for improvement there (in terms of track). Intensity is another animal though, modeling rapid intensification and the evolution of the inner mechanisms of a developing hurricane is extremely difficult. I'm not sure 'throwing more computer power' at the problem will solve it either, there is just too much entropy/unpredictability to really nail those evolutions down, even with improvements to mesoscale modeling.

[Hurricane2022]

 https://twitter.com/BackpirchCrew/status/1782908971424088220