A Multi-View Sequence Fusion Network for Improving Tropical Cyclone Intensity Estimation

Accurate estimation of tropical cyclone (TC) intensity in real time is crucial for TC forecasting and disaster management. Conventional intensity estimation methods like the Dvorak technique are subjective, and most deep learning approaches ignore the natural variability and time delay between the actual intensity and the remote sensing cloud pattern imagery. This paper introduces the Multi-View Sequence Fusion Network (MSFN), a novel deep learning framework that enhances TC intensity estimation by combining multichannel cloud pattern imagery, sea surface temperature (SST), and statistical factors. The MSFN features three key innovations: a rotation-equivariant convolutional neural network (CNN) encoder to extract robust cloud pattern features, a Transformer Interactor module to capture temporal and cross-view dependencies, and a Gated Memory Fusion module to dynamically integrate multi-view data while improving resilience to noise. Tested on TC data from the western North Pacific basin (2016–2019), the MSFN achieves a coefficient of determination (R²) of 0.845 and root mean square error (RMSE) of 4.6–5.2 m s-1. Evaluated on a strictly homogeneous dataset from the western North Pacific basin (2016–2019), MSFN achieves an R2 of 0.848 and RMSE of 4.3 m s-1 . Statistical analyses demonstrate that MSFN significantly outperforms the Advanced Dvorak Technique (ADT) and is comparable to the Satellite Consensus (SATCON) technique in overall performance, while exhibiting superior accuracy for weaker systems (< 45 kt). Interpretability analyses and ablation studies confirm the model’s focus on meteorologically significant features and the synergy of its components. The MSFN offers a reliable, interpretable tool for objective TC intensity estimation, advancing forecasting capabilities.