Wahiduzzaman et al. (2025) Modelling of typhoon activities over the Western North Pacific using a generalised additive model and deep Q-learning model

Identification

• Journal: Climate Dynamics
• Year: 2025
• Date: 2025-11-20
• Authors: Md Wahiduzzaman, Xiang Wang
• DOI: 10.1007/s00382-025-07963-7

Research Groups

• State Key Laboratory of Climate System Prediction and Risk Management, Nanjing University of Information Science and Technology, Nanjing, China
• Key Laboratory of Meteorological Disaster, Ministry of Education, Nanjing University of Information Science and Technology, Nanjing, China
• Collaborative Innovation Centre On Forecast and Evaluation of Meteorological Disasters, Nanjing University of Information Science and Technology, Nanjing, China
• School of Atmospheric Sciences, Nanjing University of Information Science and Technology, Nanjing, China
• School of Artificial Intelligence/School of Future Technology, Nanjing University of Information Science and Technology, Nanjing, China

Short Summary

This study developed a statistical and deep Q-learning model framework, incorporating kernel density estimation and a generalized additive model (GAM), to predict typhoon genesis and trajectories over the Western North Pacific, demonstrating significant competency (73-76%) in replicating observed landfall trajectories.

Objective

• To develop and evaluate a statistical model framework, combining a Generalized Additive Model (GAM) and a Deep Q-learning model with Kernel Density Estimation (KDE), for predicting typhoon genesis locations and trajectories over the Western North Pacific.

Study Configuration

• Spatial Scale: Western North Pacific (WNP) basin.
• Temporal Scale: 50-year period (1975–2024).

Methodology and Data

• Models used: Generalised Additive Model (GAM), Deep Q-learning model, Kernel Density Estimation (KDE).
• Data sources: International Best Track Archive for Climate Stewardship-World Meteorological Organisation version (IBTrACS-WMO) global typhoon best-track data, recorded at 3-hour intervals.

Main Results

• The highest density of typhoon genesis points was observed from July to September, with August showing the peak, consistent with observational data.
• Simulated typhoon tracks exhibited westward or northward trajectories in June and July, transitioning to a northward progression with a tendency to recurve towards the north-east in August and September.
• The model framework demonstrated significant competency in predicting typhoon landfall locations, with approximately 73% (GAM) and 76% (Deep Q-learning) of simulated landfalls occurring within 500 kilometers of observed locations.
• These results were statistically significant at a 95% confidence level (Mann–Whitney U test), and the model showed an approximate 5% enhancement in skill over climatology.

Contributions

• Presents a novel integrated framework combining Kernel Density Estimation, Generalized Additive Models, and Deep Q-learning for predicting typhoon genesis and trajectories.
• Utilizes KDE to provide a non-parametric, smooth probability density estimate for intricate spatial distributions of typhoon formation and movement, enhancing spatial risk evaluation.
• Demonstrates the capability of GAM and Deep Q-learning to replicate seasonal typhoon trajectories by generating velocity fields.
• Achieves high accuracy in predicting typhoon landfall locations, with 73–76% of simulated landfalls within 500 kilometers of observed tracks, offering improved forecast precision and adaptability compared to traditional methods.

Funding

• Start-up funding from Nanjing University of Information Science and Technology, Nanjing, China.

Citation

@article{Wahiduzzaman2025Modelling,
  author = {Wahiduzzaman, Md and Wang, Xiang},
  title = {Modelling of typhoon activities over the Western North Pacific using a generalised additive model and deep Q-learning model},
  journal = {Climate Dynamics},
  year = {2025},
  doi = {10.1007/s00382-025-07963-7},
  url = {https://doi.org/10.1007/s00382-025-07963-7}
}