Will the Atlantic Break its Hurricane Drought? AI Models Discussion 9-14-2026

The Atlantic basin remains remarkably quiet for peak season thanks to the ongoing El Niño event, which has led to record shear across the basin. We have set the record for the latest that a season has gone with no hurricanes forming in the satellite era, and there are no immediate prospects for hurricane development.

The only real system of note is an area over the Central Atlantic subtropics that seems to be a piece of an old trough, that Google DeepMind is keying on for development. Both the currently operational WeatherNext 2 (Figure 1) and next WeatherNext 3 (Figure 2) ensembles show a strong signal for development and intensification of this system.

The 1000-member ensemble from Google Deepmind also shows a fairly robust signal for development, with about 50% of the members currently forming a TC. The median formation time (per TC-ATLAS) is about 5 days from now (Figure 3).

It is worth noting, however, that almost all deterministic forecasts (both physics-based and AI-based) do not show much development of the system. A few runs of AIFS yesterday did show some development, and the AIFS ensemble also has a few members that develop the system (Figure 4).

Looking at a few different AI models, there is one that shows some signal for development. PanguWeather shows a TC forming in 5-7 days from this system (Figure 5).

One reason for more development on PanguWeather might be the more favorable upper pattern, with a pocket of light shear over the system and outflow into two separate troughs helping to lower pressures (Figure 6).

In contrast, AIFS seems to show more northerly shear over the region, which might be why it’s not showing much development (Figure 7).

In addition to the shear, it looks like there will be a lot of dry air in the region, which could limit development (Figure 8).

This will be an interesting case to follow. NHC currently has a 40% chance of development in the next 7 days, likely due at least partially to the DeepMind signal. This will be an interesting case, to see if this model is able to sniff something out before all the other models, or perhaps has a bit of an aggressive bias in some cases. We’ll keep an eye on things and be back with an update soon!