Artificial Intelligence: The cat’s meow for weathercasting?


Take a look at the above cover photo I used (thanks, and see if you can figure out what it is.

Got it?

Okay, I’m using this example to step through exactly how Artificial Intelligence can work to forecast weather, and, in this case, the current El Nino event. AI can perform what is called “Deep Learning”. For example, taking a cluster of pixels like the cover shot above and using available data, the computer can figure out that this is the edge of a cat’s ear. In fact, not just any cat, but a British Shorthair cat. Using the available data and a series of steps, Artificial Intelligence can determine that the pixels belong to a true object and, eventually, get to the answer. Here’s the full picture:


So let’s talk about El Nino and forecasting such an event. That extremely warm Pacific Water has profound global impacts. We talk about how the increased warm water creates windy conditions over the Caribbean to knock down developing tropical storms, but a warmer environment leads to more moisture in the air which leads to more flooding and more drought, more wildfires, more disease, and impacts coastal fisheries across the globe. The impacts add up into the trillions of dollars.

Knowing when to expect climate changing disruptions has been limited to about a year, but now scientists led by Yoo-Guen Ham out of South Korea have used Artificial Intelligence to make those forecasts as much as 18 months in advance, more lead time than eight other forecasting models. How it works is like the cat’s ear. The available data for climate and El Nino events is there, just like the small pixels of the ear--small warming events in different oceans that signal an onset of El Nino, for example. Compiling years of past and current data to make sense of it is the challenge.

Using those current ocean measurements and past data from the 19th century, Ham and his colleagues created a model to predict El Nino events from 1871 to 1973 and also gave the computer the right answers. That trained the model to know how to make future forecasts. Then, they gave the computer data from 1984 to 2017, but withheld the answers as to when an El Nino event occurred. The model successfully predicted the events more than a year and half in advance, beating out any other statistical model. You can learn more about this right here.

No one, not even Ham, seems to know the upper limit of AI weather forecasts. But as we know, the sooner you learn about big changes that are coming, the better.

Stay cool and safe this week, we are back in the Dangerous Heat “cat”-egory!


Email me with comments and questions.

About the Author:

KPRC 2's chief meteorologist with four decades of experience forecasting Houston's weather.