Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Enhancing Tropical Cyclone Path Forecasting with an Improved Transformer Network

Nguyen Van Thanh, Nguyen Dang Huynh, Nguyen Ngoc Tan, Nguyen Thai Minh, Nguyen Nam Hoang

TL;DR

This work tackles the challenge of tropical cyclone trajectory forecasting by leveraging a Transformer-based model combined with a coordinate-grid mapping to predict six-hour-ahead storm positions. The method processes sequential storm data on a 1°×1° grid, training on NOAA-derived Atlantic basin records (1944–2022) to map 12-step inputs to the next grid cell, using an encoder with three layers and an MLP output. It achieves a mean squared error of $0.0086$ and an accuracy of $0.783$, outperforming GPRA benchmarks and NOAA methods while offering faster training times on standard GPUs. The approach demonstrates potential for scalable, rapid forecasts and could inform operational decisions, with room for addressing overfitting and computational resource requirements in deployment.

Abstract

A storm is a type of extreme weather. Therefore, forecasting the path of a storm is extremely important for protecting human life and property. However, storm forecasting is very challenging because storm trajectories frequently change. In this study, we propose an improved deep learning method using a Transformer network to predict the movement trajectory of a storm over the next 6 hours. The storm data used to train the model was obtained from the National Oceanic and Atmospheric Administration (NOAA) [1]. Simulation results show that the proposed method is more accurate than traditional methods. Moreover, the proposed method is faster and more cost-effective

Enhancing Tropical Cyclone Path Forecasting with an Improved Transformer Network

TL;DR

This work tackles the challenge of tropical cyclone trajectory forecasting by leveraging a Transformer-based model combined with a coordinate-grid mapping to predict six-hour-ahead storm positions. The method processes sequential storm data on a 1°×1° grid, training on NOAA-derived Atlantic basin records (1944–2022) to map 12-step inputs to the next grid cell, using an encoder with three layers and an MLP output. It achieves a mean squared error of and an accuracy of , outperforming GPRA benchmarks and NOAA methods while offering faster training times on standard GPUs. The approach demonstrates potential for scalable, rapid forecasts and could inform operational decisions, with room for addressing overfitting and computational resource requirements in deployment.

Abstract

A storm is a type of extreme weather. Therefore, forecasting the path of a storm is extremely important for protecting human life and property. However, storm forecasting is very challenging because storm trajectories frequently change. In this study, we propose an improved deep learning method using a Transformer network to predict the movement trajectory of a storm over the next 6 hours. The storm data used to train the model was obtained from the National Oceanic and Atmospheric Administration (NOAA) [1]. Simulation results show that the proposed method is more accurate than traditional methods. Moreover, the proposed method is faster and more cost-effective
Paper Structure (5 sections, 2 equations, 9 figures, 1 table)

This paper contains 5 sections, 2 equations, 9 figures, 1 table.

Table of Contents

  1. Introduction
  2. Transformer Model
  3. Proposed Transformer Model Combined with Coordinate Grid
  4. Results and Evaluation
  5. Conclusion

Figures (9)

  • Figure 1: Overview of the original Transformer architecture, consisting of encoder and decoder components.
  • Figure 2: Storm occurrence points from 1944 to 2022
  • Figure 3: A coordinate grid with a resolution of 1° longitude × 1° latitude.
  • Figure 4: Trajectories of the 5 longest-lasting storms
  • Figure 5: Proposed Transformer model architecture
  • ...and 4 more figures