Deep-learning climate emulator ACE2 reveals a global decrease in tropical cyclone frequency in the 15th Century under an El Niño-like sea surface temperature pattern
Using the deep-learning-based ACE2 emulator, this study reveals that an El Niño-like sea surface temperature pattern in the 15th Century drove a global decrease in tropical cyclone frequency through enhanced vertical wind shear and a drier midtroposphere, thereby demonstrating the utility of deep learning in paleoclimate analysis to improve future projections.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine the Earth as a giant, swirling bathtub of air and water, where the temperature of the ocean surface acts like the thermostat for a massive, chaotic weather machine. When the water gets warmer or cooler in specific patterns, it doesn't just change the temperature; it changes how the atmosphere breathes, stretches, and twists. One of the most dramatic things this machine produces are tropical cyclones—those swirling storms we call hurricanes or typhoons. For a long time, scientists have been trying to figure out exactly how the ocean's temperature patterns control the number of these storms. The problem is that we only have a very short "video recording" of modern storms, maybe about 80 years, which isn't long enough to see the full story of how climate and storms dance together. To solve this mystery, researchers are using a new kind of tool: artificial intelligence. Think of these AI tools not as robots that think, but as incredibly fast, super-smart students who have studied the weather of the last few decades so thoroughly that they can predict what the weather would look like if the rules of the game (the ocean temperatures) were slightly different. This paper uses that AI to look back in time, simulating a period long before we had weather satellites, to see what happened to global storms when the ocean looked a bit like an "El Niño" year.
The story begins with a team of scientists who wanted to know what happened to tropical cyclones during the 15th Century, specifically between the years 1351 and 1550. This was a time when the Earth was cooling down, entering a period often called the "Little Ice Age." While the planet was getting colder overall, the ocean didn't cool down evenly. Instead, the Pacific Ocean developed a pattern that looked a lot like an "El Niño," where the water near the equator was warmer than usual, while other parts of the ocean were cooler. The researchers used a deep-learning climate emulator called ACE2 (which is like a super-fast weather simulator trained on modern data) to run a virtual experiment. They fed the AI the ocean temperatures from that era and watched what happened to the storms.
The results were clear and surprising. In these simulations, the total number of tropical cyclones around the entire globe dropped significantly. Specifically, the number of storms decreased by about 13.1% when comparing the most active 20-year period (1401–1420) to the least active 20-year period (1461–1480). This drop wasn't just a fluke; it matched up with what a very powerful, traditional computer model (HiRAM) predicted and even lined up with clues found in ocean sediments along the Atlantic coast, which act like a fossil record of past storms. The paper suggests that this decrease happened because the "El Niño-like" pattern of the ocean changed the way the atmosphere moved.
To understand why the storms disappeared, the scientists used a clever trick. They broke down the storm-making process into two steps: first, the birth of a "seed" (a small disturbance in the air that could become a storm), and second, the "probability" of that seed growing into a full-blown hurricane. They found that the number of seeds didn't change much; the air was still making plenty of little disturbances. However, the "probability" that these seeds would survive and grow into big storms plummeted. It was as if the nursery was full of babies, but the conditions for them to grow up were terrible.
The culprit was the "El Niño-like" ocean pattern. Because the equatorial Pacific was warmer, it caused the air to rise more strongly there. This strengthened a giant atmospheric conveyor belt called the Hadley circulation. As this belt got stronger, it pushed dry air down and increased the wind shear (the change in wind speed and direction with height) in the regions where storms usually form. Imagine trying to build a sandcastle while a strong wind is blowing sideways and the sand is bone dry; that's what the atmosphere became for these storms. The drier air and the stronger winds tore the storms apart before they could get strong. The paper explicitly notes that this wasn't just because the whole world got colder; in fact, when they ran a separate experiment where the ocean cooled evenly without the "El Niño" pattern, the storms didn't drop as much. It was the specific shape of the temperature change that mattered most.
The researchers also tested the opposite scenario: a "La Niña-like" pattern, where the equator is cooler. In that simulation, the storms actually increased in number. This suggests that the pattern of the ocean's temperature is a master switch for global storm activity. While the study relies on computer simulations and not direct observation from the 1400s, the fact that the AI emulator, the traditional model, and the sediment records all tell the same story gives the findings a lot of weight. The paper concludes that to predict how many storms we might face in the future, we can't just look at how hot the planet gets; we have to understand exactly how the ocean's temperature patterns will shift, because that pattern dictates whether the atmosphere becomes a storm factory or a storm graveyard.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.