Determining the right wind level to design storm-resistant buildings is a difficult task, but artificial intelligence can solve this.
Based on 100 years of hurricane data and modern AI techniques, researchers at the US National Institute of Standards and Technology (NIST) have developed a new method to simulate hurricanes. Results published in the journal Artificial Intelligence for the Earth Systems demonstrate that simulations can accurately represent the trajectory and wind speed of actual storms. This result may contribute to the development of guidelines for building design in storm-prone areas.
Legislation governing the design and construction of buildings (building codes) is directing designers towards standard maps. On the map, engineers can determine how much wind a building must withstand, based on the location and importance of the building (hospitals are more wind resistant than self-managed warehouses, for example). . The wind speeds in the map are taken from scores of hypothetical storms simulated by computer models.
“Imagine you have a second Earth, or a thousand Earths, where you can watch the storms for 100 years and see where on the coast they hit, their intensity. If those simulated storms behave like they do in real life, then we can use them to generate data almost directly in the map,” said NIST mathematical statistician Adam Pintar and study co-author. know.
The researchers developed the latest maps by simulating the complex activity inside the storm, which is influenced by physical parameters such as sea surface temperature and Earth’s surface roughness. However, data on these factors are not always available.
The team chose a new approach – using machine learning to develop a model that simulates real storm data. With a quality source of information, machine learning algorithms can build models based on patterns they detect in datasets that other methods might miss.
For the new study, the authors used the National Hurricane Center’s Atlantic Hurricane Database (HURDAT2), which contains information about hurricanes from more than 100 years ago, such as track coordinates. and their wind speeds. They split the data on 1,500 hurricanes into datasets to train and test the model. In the simulation test that simulates the trajectory and wind strength of storms that are not included in the training data, the model scores quite well. Pintar commented, “[Mô hình] works very well. Depending on the location, it is difficult to distinguish between a simulated storm and a real one.”
However, there are still some limitations. In the Northeast coastal states, for example, HURDAT2 data are sparse, so the model for generating hurricanes is less realistic. “The less data you have, the greater the uncertainty about your prediction,” said co-author Emil Simiu.
As a next step, the team intends to use simulated storms to develop coastal maps of extreme wind speeds, as well as quantify the uncertainty in estimated wind speeds.
Because the model’s understanding of hurricanes is currently limited to historical data, it cannot simulate the effects of climate change on future storms. The traditional approach of simulating storms from the bottom up would be more appropriate. In the short term, however, the authors are confident that the map wind based on their model – less dependent on elusive physical parameters than other models – will better reflect the real situation. They aim to develop and propose new wind maps for inclusion in building codes.