Hurricane Ida Demonstrates Need for Artificial Intelligence in Improving Hurricane Forecasts
Written by Fredrik Borgström and Svante Henriksson
While writing this article, Hurricane Ida is not yet dead. It made landfall in Port Fourchon on Sunday August 29th as a major category 4 hurricane and has since then passed through Louisiana, Mississippi and North Alabama. Now it’s in Tennessee somewhere between Huntsville and Knoxville, with significantly reduced wind speeds and reduced attention. The aftermath, however, will go on for years.
The news flow started with some smaller side columns on Thursday, August 26 when a low pressure in the Caribbean Sea became Tropical Depression Nine. The amount of news articles increased at the same pace as the Tropical Depression continued its development towards a tropical cyclone (a hurricane when in the Atlantic) throughout Friday and Saturday to peak on Sunday August 29th when landfall happened near New Orleans in Louisiana USA. Ironically it was on the day, 16 years after Hurricane Katrina, a category 3 hurricane at landfall, which caused massive destruction in the very same area worth USD 125 billion and leaving 1800 dead. The second event in a relatively short timeframe was an awful reminder of scientists’ projections that these natural catastrophes will increase in severity with climate change.
The damage caused by Hurricane Ida is expected to be significantly less than with Hurricane Katrina. The landfall location was only about 40 miles west from where Katrina hit and the wind speed of Ida was higher, but the radius of hurricane level winds was smaller, there was less widespread storm surge, and the landfall angle was favorable compared to Katrina. The total number of deaths are not yet confirmed, but they are fortunately way fewer this time, and monetary losses are estimated to be somewhere between USD 15 to 20 billion.
Hurricanes often develop through the West African monsoon, low pressure areas travelling westwards just north of the equator over the Atlantic, or they can also develop from low pressures in the Caribbean Sea, like Hurricane Ida did. Throughout the storm’s development the world’s most advanced scientists, like the ones at the US National Hurricane Center, follow every possible detail of it. Some major factors that are monitored and that affect the development of the hurricane are pressure, humidity, windshear, sea surface temperature, movement speed of the low pressure, structure of the eye and eyewall as well as the presence of land, which have a declining effect on the intensity of the storm.
Technology used to monitor and forecast hurricanes
Hurricane data are gathered from a variety of sources including fixed and mobile weather stations, buoys, satellites and aircrafts dropping sondes and drones into the storm. Measurement data is used as input for weather models and meteorologists do their best to anticipate where the storm will travel and how it will behave in terms of location of landfall, wind intensity, rainfall and storm surge. This information is crucial for exposed areas where local governments, industries and households do their best to protect their inhabitants and assets. Primary insurers and re-insurers compensate a substantial share of the damages and hence, they too, try to estimate probabilities of different metrics which all affect the damage of the underlying insured assets. Human tragedies, fatalities and lost homes cause damages that are hard to put a price tag on.
Weather models are advanced and all available data make it possible to forecast hurricanes with impressive precision. However, some of their details are still unpredictable, and margins are small. For instance, 4 days in advance the average National Hurricane Center forecasted track error is in the range of 175 miles(1). Closer to landfall, even though uncertainty gets smaller, time is of critical essence. In the case of Hurricane Ida, the authorities were facing the challenging situation of deciding which areas to evacuate(2). Here artificial intelligence methods giving quick updates based on every incoming satellite image may help in the future.
To augment the weather models simulating the weather on supercomputers, advanced deep learning algorithms have recently been developed and trained with hundreds of thousands previously collected satellite images and other historical data. These data can be processed with convolutional neural networks(3) and more complex programming architectures that have been applied successfully in facial recognition, self-driving cars and traffic counting. Rapid intensifications as well as sudden changes in the direction can be detected quickly and with less effort than earlier. The technique can be applied globally and is thus equally beneficial when struggling against typhoons, tropical cyclones or hurricanes anywhere in the world.
Another area, where continuous improvement happens is novel instrumentation and improved data gathering. For example swarms of ultralight weather sondes, so called StreamSondes that are currently in the testing and piloting stage, are dropped from above the storm from aircrafts or solar gliders and fall through the complex, turbulent flows and vortices of a tropical cyclone providing completely new data of the structure of these storms for improved research and forecasting.
The way forward
As always, it is development in process and mankind becomes increasingly better at forecasting and preparing for hurricanes. As a rule of thumb, every dollar invested in disaster management is converted into 7 dollars in reduced damages. The authorities will have increasingly more timely and accurate information for informed evacuation and other decisions, which directly translates to saved money and lives.
About Skyfora
Skyfora’s AI-based Tropical Storm Tracker and ultralight sondes are at the forefront of next-generation weather intelligence solutions. The Tropical Storm Tracker identified parts of Hurricane Ida’s rapid intensification faster than any other algorithm in the world. It’s StreamSonde is currently being tested and piloted for hurricane launch in a project funded by the European Space Agency.
(1) https://www.nhc.noaa.gov/archive/2021/al09/al092021.discus.001.shtml?
(2) https://edition.cnn.com/2021/08/27/us/new-orleans-hurricane-ida-preparations/index.html
(3) https://en.wikipedia.org/wiki/Convolutional_neural_network
And finally, some additional interesting imagery related to Hurricane Ida:
NOAA Satellites — Public Affairs: https://twitter.com/i/status/1432028471551832065, https://twitter.com/i/status/1431718171455672327
National Hurricane Center: https://twitter.com/i/status/1432014084908609543
