With high pressure dominating Houston's weather, what's going on in the tropics?

[WeatherBot]

Originally published at: With high pressure dominating Houston’s weather, what’s going on in the tropics? – Space City Weather

In brief: In today’s post we briefly discuss the ongoing ridge of high pressure dominating Houston’s weather. We also take a look at what’s happening in the Atlantic tropics (not much), and more interestingly talk about the AI model that performed exceptionally well during the 2025 Atlantic hurricane season.

Tropics update and AI model review

The short answer is that the Atlantic tropics are quiet. This is often the case for late July, when Atlantic sea surface temperatures are still warming toward their peak later this summer, and wind shear is fairly high. The other factor working against tropical storms and hurricanes at this time of year is Saharan dust, and there is plenty of it across the Atlantic basin right now. I would expect the Atlantic tropics to remain quiet through the end of June, and probably at least a week into July, if not longer.

Before moving on to the rest of today’s post, I did want to note one thing about hurricane models. Earlier this year the National Hurricane Center published its annual verification report for the 2025 Atlantic hurricane season. One of the questions this report answers concerns the performance of the various forecast models for the “track” of tropical systems. And last year there was a clear winner, Google’s DeepMind model which is based on artificial intelligence. For forecast periods of 12 hours to 4 days out, the model’s performance was better even than that of the National Hurricane Center and various consensus models. It’s quite striking.

Our friend and colleague Michael Lowry has an excellent post explaining all of this on his “Eye on the Tropics” Substack, and it’s worth a read to put the performance of this AI model into perspective. Speaking for Matt and myself, we used Google’s model as part of our suite of tools last year, and obviously will continue to do so. I’m eager to see if it is a top performer again this year.

Wednesday, Thursday, and Friday

Back in Houston, our forecast has not changed much. Or rather, at all. A fairly strong ridge of high pressure remains in control of our weather for the time being. This will bring us mostly sunny skies, humidity, and hot temperatures. Highs for the rest of this week will be in the low- to mid-90s, with plenty of humidity. Overnight lows will generally drop into the mid- to upper-70s. Rain chances are basically zero. Winds will be calm in the morning, with moderately gusty conditions (up to 20 mph) during the afternoon. And that is basically it.

Saturday and Sunday

Are we expecting much change in the weather this weekend? In short, no. Perhaps we’ll see nighttime temperatures 1 to 2 degrees warmer, but it will be difficult to notice much difference.

Next week

Our sunny and hot weather continues into next week, but by Wednesday or so, some of our forecast models are indicating a turn toward a more typical summertime pattern, with a chance of afternoon showers and thunderstorms along the sea breeze. However, when we get into these persistent high pressure patterns it can be difficult to break them, so we’re offering no guarantees here.

[Blackhawks_Fan]

And just think - only three plus months of this…this…this… “stuff” left to go!

[Martin_Kuykendahl]

While it is nice to see the DeepMind model delivering accurate predictions, I have to say it makes me a bit sad how quickly AI has surpassed traditional, physics-based models. I mean how many years have AI models been consistently used for track predictions, not even 2 or 3? It’s crazy how fast AI has taken over everything

[dwight]

But if the AI models are significantly better than traditional models, isn’t that ultimately the point? That’s something to be glad about, not sad. That said, let’s see how AI models hold up over time, which is the ultimate test.

[Narillan]

Honestly as much as I don’t like AI in a lot of things, data crunching is an AI specially. And I am glad that they are confirming whether it is a good predictor or not.

[Hail]

For the first figure, why is peak accuracy at 2 days out? Why does accuracy fall off as you get closer to the present?

[Native]

Doesn’t a human intelligence have to write code for a computer to be able to produce what is being called AI? Or am I missing something?

[lee]

We sort of do. A human (often many of them!) has to design the AI model’s architecture, write the software to train it, decide what data it’s going to be trained on, and evaluate whether or not it’s performing well. But the actual forecasting done by these things isn’t really like they’ve got a decision tree of if-thens that a person programmed in. The model learns statistical relationships between and among massive amounts of historical weather data and uses those relationships to estimate the most likely outcome of a set of conditions.

Traditional weather models try to simulate the actual physics of weather—these are the big supercomputers running detailed atmospheric simulations, solving equations to try to figure out the behavior of the atmosphere. New-fangled AI models like the one I linked above learn statistical patterns from massive amounts of existing weather data, and then apply those patterns to current conditions. Both methods predict, but the traditional way predicts by attempting to simulate physics, and the new way that @Eric is talking about predicts by making inferences about what might happen based on what’s already happened.

The actual process involved is pretty wild, too—here’s a great 3Blue1Brown video laying out how a large language models specifically works, but the same machine learning techniques apply to the models doing weather prediction. You train the machine on a huge amount of data, helping it as it goes along and pointing it at the right answers (a training algorithm does the pointing, not a person), and at a certain point it develops enough statistical connections to be able to make inferences about new data.

Machine learning! It’s neat!

[Native]

Thank you. I’ll check out the links you referenced