re: New thread started for Beryl

I'll write some stuff up if we get some gulf action looming, dont worry.

quote:

---

Must be a new model the doomcasters are using to hype up their YouTube stream counts.

How did AI come to that conclusion? Was it monitoring the entails of Central American pigs, or the tried and true use a blind fold and throw a dart at a map?

---

ECMWF has been running several AI models for a while now and I haven't been overly impressed. Some of the AI models do show improved skill when using ERA5 reanalysis for initial conditions. Basically, taking historical data and feeding it into the model and then comparing the model out put to a 10 day stretch of past weather.

quote:

---

with the rapid improvement of AI in the past year, do you think it will have a significant impact on future storm prognostication?

---

I think we are approaching the upper limits of model accuracy until there are significant scientific & modeling breakthroughs in the area of convective parametrization (how models handle thunderstorms). The best CAMs still struggle with convective parametrization at even short lead times. It's not uncommon to flip through numerous CAMs and not find any that nailed a 3 hr thunderstorm forecast.

So, IMO, the current AI models are just another version of our current models with the same underlying issues. Even if they eventually improve at H5 or temp forecasting or whatever they will still struggle with one of the single most important drivers of extreme weather events.

quote:

---

I think we are approaching the upper limits of model accuracy until there are significant scientific & modeling breakthroughs in the area of convective parametrization (how models handle thunderstorms). The best CAMs still struggle with convective parametrization at even short lead times. It's not uncommon to flip through numerous CAMs and not find any that nailed a 3 hr thunderstorm forecast.

---

I was just thinking the same thing.

It means AI isn't fixing anything yet because it isn't effective being used to actually improve the models in part because the understanding of what causes things to happen (or not) is still so imperfect.

quote:

---

I think we are approaching the upper limits of model accuracy until there are significant scientific & modeling breakthroughs in the area of convective parametrization (how models handle thunderstorms). The best CAMs still struggle with convective parametrization at even short lead times. It's not uncommon to flip through numerous CAMs and not find any that nailed a 3 hr thunderstorm forecast.

---

We need more super computers. Though, there are some promising machine learning projects out there right now. Atmosphere Analytics and their STORM-Net has done better than I expected with severe/tornadic storms.

Loading Twitter/X Embed...
If tweet fails to load, click here.

GFS still has something popping up in the gulf six days out. Currently has it plowing into Florida as slow moving rain maker.

Don’t look too far out something else might be behind it, 7 days later headed to the usual hotspot.

quote:

---

Conditions become more favorable in that time frame for typical mid-end of June mischief. Euro ensembles seem to be the one model that is being aggressive with it. Other models are showing something just not as active as the Euro

---

The biggest concern for people along the Gulf is heavy rain. Some of the telltale signs of an extreme rainfall event are coming together. Hopefully, it stays mostly out over the Gulf.

That's a lot of rain for a five day span.


And the rest of the week could add even more.