Research team uses satellite data and machine learning to predict typhoon strength

As climate change makes typhoons more difficult to predict, a team of researchers has developed a technique that uses real-time satellite data and deep learning capabilities to more accurately forecast typhoons.

A research team led by Professor Lim Jun-ho of UNIST's Department of Civil, Urban, Geospatial and Environmental Engineering has presented a deep learning-based forecasting model that combines geostationary meteorological satellite data and numerical model data in real time.

The results of the study are: GIScience and Remote Sensing and iScience They are scheduled for March and June 2024, respectively.

The hybrid convolutional neural network (hybrid CNN) model proposed by the research team effectively combines satellite-based spatial characteristics and the output of a numerical forecast model to objectively and accurately forecast tropical cyclone (TC) intensity with lead times of 24, 48, and 72 hours. The hybrid CNN model significantly reduces the uncertainties associated with numerical models, enabling more accurate typhoon forecasts.

Traditional methods of observing typhoons rely heavily on geostationary satellite data analyzed by forecasters, but this approach is hampered by the lengthy analysis times and the inherent uncertainties in numerical models. In contrast, the Hybrid-CNN model significantly reduces the uncertainties associated with numerical models, enabling more accurate typhoon forecasts.

The research team applied a transfer learning model to estimate typhoon intensity using satellite data from the Communications Ocean Weather Satellite (COMS), launched in 2010, and GEO-KOMPSAT-2A (GK2A), launched in 2019. The AI ​​improved the accuracy of typhoon forecasts by visualizing and quantitatively analyzing the automatic typhoon intensity estimation process.

It is expected that by objectively extracting the environmental factors that affect changes in typhoon strength and applying them to on-site forecasting systems, it will be possible to provide forecasters with fast and accurate typhoon information, which will make a significant contribution to disaster prevention and damage prevention.

“By providing more accurate prediction information, our deep learning-based typhoon prediction framework will enable forecasters to develop rapid and effective countermeasures,” said Professor Lim.