How to Predict a Hurricane
Hurricanes are the most violent weather patterns on the planet. Outside North America's oceans, such storms are called tropical cyclones. Like a giant engine, these cyclones pull warm air from the surface of an ocean. As a result, the air pressure at sea level drops. The air around it quickly fills this pressure drop. As this new air warms, it too fuels the cyclone engine in a continuous cycle.
The National Oceanic and Atmospheric Administration (NOAA) predicted the 2023 hurricane season to have a 40% chance of near-normal, a 30% probability of above-normal, and a 30% chance of above-normal. These calculations come from the Hurricane Analysis and Forecast System (HAFS), a new system that models atmospheric conditions with 10 to 15% improvements over previous technologies [NOAA 2023].
Hurricane Patricia
On October 11, 2015, a tropical disturbance crossed the southern part of Central America on its way to the eastern Pacific. Several hundred miles south, a tropical wave moved westward over the Caribbean Sea, reaching Central America a few days later [Kimberlain 2015]. As the warm wave merged with the tropical wind, it injected cyclonic vorticity, forming an area of low pressure several hundred miles in size. By October 17, the resulting cyclone intensified.
Hurricane Patricia continued to strengthen and, on October 22, was about 200 miles south of Acapulco, Mexico. Over the next day, the cloud-top temperatures dropped to -90 degrees Celsius, resulting in hurricane-strength winds. A day later, on October 23, the winds sustained 180 miles per hour. By October 23, Patricia landed in the Mexican state of Jalisco to become the most powerful hurricane in recorded history.
Prediction Methods
Meteorologists look for thunderstorms forming over oceans as a first watch for hurricane formation. While most never result in a tropical cyclone, mixing wind with warm water is enough to warrant further investigation [IEEE 2020]. Factors like water salinity are hypothesized to be critical parameters in predicting wind speed. However, the means to accurately study hurricanes is limited by the large amount of space they take up. Hundreds of miles of oceans are hard to monitor with sensor data.
As a result of these difficulties, researchers are recreating the conditions of a hurricane inside of a computer model to determine how the actual storm will behave [IEEE 2020]. On June 27, 2023, the Hurricane Analysis and Forecasting System (HAFS) entered operational status to do just that [HFIP 2023]. Giving a 7-day forecast with a 10 to 15% improvement over previous models is a great accomplishment. However, it still relies on many of the traditional sample-taking methods of meteorology.
The Future
Giant technological leaps occur after people have let go of the old ideas. Ray Dalio explains this phenomenon as "the ability to reject good alternatives in order to pursue even better ones." For example, Deep Learning was a significant breakthrough in machine learning that allowed a computer to beat human masters in Chess and other strategy games. While impressive, Deep Learning is still just curve-fitting [Pearl 2018]. Meaning it relies on correlations and probabilities over actual cause and effect.
Judea Pearl, the father of Causal AI, developed the mathematical tools to teach machines cause and effect. Based on counterfactuals or what-if scenarios, this technology provides a giant leap toward creating machines that think like humans. Pearl explains, "The algorithmization of counterfactuals invites thinking machines to benefit from this ability and participate in this (until now) uniquely human way of thinking about the world."
Modeling the World
Estimand has created a Causal AI based on Pearl's work. A system that sorts events by cause and effect, learns from them and provides what-if scenarios for the future. For example, if the temperature in the Atlantic Ocean increases by 1 degree while top-level clouds cool by 1 degree, how will it affect the weather over the next five days?
We have expanded it further by incorporating over 500 million real-time global events to provide a continuous model of the globe from an economic, environmental, geopolitical, social, and technological perspective. The goal is to provide organizations with the means to quickly identify how changes in the world affect their operations and the things around them. The purpose of Estimand is to provide instant answers to global questions.
References:
- HFIP (2023) Hurricane Analysis And Forecast System (HAFS). HFIP. https://hfip.org/hafs
- IEEE (2020) Hurricane Prediction Technology. IEEE Public Safety Technology. https://publicsafety.ieee.org/topics/hurricane-prediction-technology
- Kimberlain, T., Blake, E., Cangialosi, P. (2015) HURRICANE PATRICIA. National Hurricane Center. https://www.nhc.noaa.gov/data/tcr/EP202015_Patricia.pdf
- NOAA (2023) NOAA predicts a near-normal 2023 Atlantic hurricane season. NOAA. https://www.noaa.gov/news-release/2023-atlantic-hurricane-season-outlook
- Pearl, J. (2018) The Book of Why: The New Science of Cause and Effect. Basic Books