Dual-Branched Spatio-temporal Fusion Network for Multi-horizon Tropical Cyclone Track Forecast

TL;DR

This study introduces DBF-Net, a dual-branched architecture for multi-horizon tropical cyclone track forecasting that jointly leverages (i) inherent TC features and (ii) reanalysis 2D geopotential height (GPH) fields. The TC branch encodes temporal dynamics via a two-layer LSTM, while the pressure-field branch uses a 2D-CNN encoder-decoder to model spatio-temporal information in the surrounding GPH, with a fusion-decoder integrating both modalities to predict relative track changes $\\mathcal{Y}_\\tau = \{\\mathbf{Y}_i\}$ over multiple horizons. Training proceeds in three stages culminating in an end-to-end objective $L_{final} = L_{loc} + \\alpha L_{GPH} + \\beta L_2$, enabling robust multi-modal feature fusion. Evaluated on CMA-BST/CFSR data for the Northwest Pacific, DBF-Net achieves substantial MDE reductions across 6h–24h horizons compared with statistical baselines and competitive performance versus NWP, while offering higher efficiency. The work demonstrates that efficient fusion of temporal inherent TC data and nearby pressure-field dynamics yields accurate, timely multi-horizon TC forecasts with lower computational cost than traditional NWP approaches.

Abstract

Tropical cyclone (TC) is an extreme tropical weather system and its trajectory can be described by a variety of spatio-temporal data. Effective mining of these data is the key to accurate TCs track forecasting. However, existing methods face the problem that the model complexity is too high or it is difficult to efficiently extract features from multi-modal data. In this paper, we propose the Dual-Branched spatio-temporal Fusion Network (DBF-Net) -- a novel multi-horizon tropical cyclone track forecasting model which fuses the multi-modal features efficiently. DBF-Net contains a TC features branch that extracts temporal features from 1D inherent features of TCs and a pressure field branch that extracts spatio-temporal features from reanalysis 2D pressure field. Through the encoder-decoder-based architecture and efficient feature fusion, DBF-Net can fully mine the information of the two types of data, and achieve good TCs track prediction results. Extensive experiments on historical TCs track data in the Northwest Pacific show that our DBF-Net achieves significant improvement compared with existing statistical and deep learning TCs track forecast methods.

Figures (5)

• Figure 1: The overall structure of the proposed DBF-Net for multi-horizon TC track forecast.
• Figure 2: The LSTM-encoder module in TC Features Branch.
• Figure 3: The pressure field branch architecture.
• Figure 4: The LSTM-based decoder module in TC Features Branch with dual-branched features fusion.
• Figure 5: Exmaple of TCs track forecast results. The blue line represents the ground truth track. The red line represents the forecast results.