AI Forecasting Case Study: Predicting Hurricane Ian

WindBorne’s state-of-the-art AI global forecast model, named WeatherMesh, produces the most accurate global weather forecasts available as measured by RMSE. Our deep learning numerical weather prediction (DLNWP) model surpasses the previous record-holder, an AI-based model, by a significant 11% margin in accurately predicting a primary forecasting metric known as 500 millibar geopotential height, a measure of pressure that describes the movement of weather systems.

WeatherMesh has also surpassed the gold-standard for traditional physics-based weather modeling, known as numerical weather prediction (NWP), held by the European Centre for Medium-Range Weather Forecasts (ECMWF). WeatherMesh surpasses ECMWF’s medium-range forecast (ENS) by an even greater margin.

Accurate weather forecasts are essential to enabling both climate change prevention and adaptation. More precise predictions of the path, timing, and intensity of extreme weather can save lives and mitigate hundreds of billions of dollars in damages every year. Better forecasts also help curb climate change by informing shifts that reduce emissions: more intelligent forecasts mean more efficient transport and logistics routing and smarter use of energy grids, accelerating the switch to renewables.

Hurricane Ian Case Study

To investigate how WeatherMesh’s performance translates into real world impact, we selected Hurricane Ian for our first case study. We analyzed Ian first given WindBorne’s deep familiarity with the event. In 2022, we flew a sounding balloon into its eye wall to collect in-situ data. Furthermore, Ian represents exactly the type of extreme weather for which better predictions are critical: it peaked as a deadly Category 5 hurricane in September 2022 after making landfall in Cuba and Florida, causing $112 billion in damages.

To compare the forecasting accuracy of WindBorne WeatherMesh with the National Weather Service (NWS), we plotted predictions using only the information available at 12am UTC on September 26, 2022 (see: “Ground Track for Hurricane Ian” chart above). This time is roughly 70 hours before Ian first made landfall on Florida’s Gulf Coast at Caya Costa, a southwestern barrier island near Fort Myers. In addition, WeatherMesh was trained on data up through July 2022 (ie, the pre-Summer 2022 data set) in order to run a fair comparison.

WindBorne WeatherMesh produced its forecast using solely data that would have been available at the time, namely the Global Forecast System (GFS) and European Centre for Medium-Range Weather Forecasts (ECMWF) analyses (estimates of the current state of the Earth system). The NWS’s prediction at that same time stamp was archived in the National Hurricane Center.

In the “Ground Track” chart above, the marked dots show times for which a forecast was made; these are in six-hour increments in each forecast’s track up until Ian’s landfall. After the point of landfall, the prediction shows forecasts in 12-hour increments.¹ The blue shaded cone shows the NWS’ uncertainty cone, which represents the NWS’ estimated range in the probable track of the center of a storm. The below “Track Error” chart plots the distance between the forecasted location and the actual location of Hurricane Ian vs the forecast lead time; lower is better as it represents a closer match with reality.

In these figures, we see WindBorne WeatherMesh predicting Ian’s landfall 200km more accurately than the NWS at a lead time of about 70 hours. WeatherMesh also continued to predict Ian’s eastward path across Florida between 300km and 400km more accurately than the NWS throughout the next 50 hours, a striking improvement.

Improving Global Forecasts

Both forecast modeling technology and observation data are key to advancing global forecast accuracy. Weather is a global system in which weather patterns often originate over remote areas such as oceans and then propagate thousands of miles to more populated areas. As a result, in-situ data collection is critical on a global, ongoing basis in order to understand weather from its earliest stages.

Hurricane Ian, for example, originated from a tropical wave off the coast of Africa near the Cabo Verde Islands, the source region for many hurricanes that impact North America. This month, WindBorne is setting up a permanent launch site on Cabo Verde to observe even more critical atmospheric data well before weather reaches land. The new site expands WindBorne’s permanent launch presence to three continents.

Case studies like Ian demonstrate the ability for groundbreaking forecasts like WeatherMesh to have tangible, real-world impact. They make our team even more motivated to continue to advance our global in-situ data collection and deep learning forecast modeling. More accurate forecasts mean people everywhere can make more informed, safer decisions every day.

This blog will be the first in a series of AI forecasting case studies, which will also include statistical analysis demonstrating WeatherMesh’s accuracy across major storms. Across every single tropical cyclone thus far investigated, we’ve seen comparable improvements as we did with Hurricane Ian. We’ll also share deep dives into WeatherMesh’s modeling technology, observation data, and diverse applications.

Get in touch to let us know what you think or learn more by sending us a message at contact@windbornesystems.com. We look forward to hearing from you.

‍¹ To preserve, disk space we typically shift to 12-hourly outputs after 72hrs when running historical analyses.

‍