Harnessing LSTM and XGBoost algorithms for storm prediction

Area of Science:

• Meteorology and climatology
• Data science and artificial intelligence
• Oceanography

Background:

• Storms pose significant risks to human life and infrastructure.
• Accurate storm prediction is hampered by the infrequent nature of these events.
• Advanced computational methods are needed to improve storm forecasting.

Purpose of the Study:

• To develop and evaluate a novel deep learning and machine learning approach for predicting storm characteristics and occurrence.
• To assess the efficacy of Long Short-Term Memory (LSTM) and Extreme Gradient Boosting (XGBoost) models in storm prediction.
• To apply these models to storm data from Western France.

Main Methods:

• Utilized a combination of buoy data and a storm database spanning 1996-2020.
• Trained and validated LSTM and XGBoost models using data from January 1996 to December 2015.
• Applied trained models to predict storm characteristics and occurrence from January 2016 to December 2020.

Main Results:

• The LSTM model demonstrated high accuracy in forecasting temperature and pressure but faced challenges with extreme wave height and wind speed.
• The XGBoost model exhibited excellent performance in predicting storm occurrence.
• The combined approach offers a robust framework for enhancing storm prediction capabilities.

Conclusions:

• Deep learning and machine learning models, specifically LSTM and XGBoost, can effectively contribute to storm prediction.
• The developed methodology shows potential for reducing the impact of storms on human populations and infrastructure.
• Further refinement may improve the prediction of extreme storm parameters.