Stochastic Simulation of Typhoon in Northwest Pacific Basin Based on Machine Learning

Area of Science:

• Meteorology
• Machine Learning
• Climate Science

Background:

• Typhoons cause significant global economic losses and casualties.
• Stochastic simulation of large tropical cyclone samples is crucial for evaluating typhoon hazard risk.
• Typhoon full-track models are commonly used for stochastic simulation.

Purpose of the Study:

• To improve the accuracy of typhoon track and intensity prediction in stochastic simulations.
• To integrate machine learning, specifically a backpropagation neural network (BPNN), into empirical typhoon track models.
• To generate a 1000-year dataset of full-track typhoon events for the Northwest Pacific basin.

Main Methods:

• Utilized a backpropagation neural network (BPNN) to replace the traditional regression model in empirical typhoon track models.
• Developed a novel neural network model for predicting typhoon track and intensity.
• Generated a comprehensive 1000-year dataset of typhoon events using stochastic simulation for the Northwest Pacific.

Main Results:

• The backpropagation neural network (BPNN) demonstrated improved accuracy in predicting both typhoon track and intensity.
• The study successfully constructed a large-scale dataset of full-track typhoon events.
• The findings validate the effectiveness of machine learning in enhancing typhoon simulation models.

Conclusions:

• The integration of BPNN significantly enhances the precision of typhoon track and intensity prediction.
• This approach offers a more robust method for evaluating typhoon hazard risks.
• The developed model provides a valuable tool for climate research and disaster preparedness in typhoon-prone regions.