Typhoons are among the most severe natural disasters impacting the southeastern coastal regions of China,and accurately predicting their tracks and intensities is crucial for effective disaster prevention and mitigation. Traditional numerical forecasting methods, while comprehensive in their physical mechanisms,face challenges such as high computational costs, sensitivity to initial meteorological conditions, and limitations in capturing the rapid evolution of typhoons. In recent years,data-driven deep learning methods have introduces a novel paradigm for typhoon prediction. This study presents an end-to-end model based on the Fourier Neural Operator(FNO)for the joint prediction of typhoon tracks and intensities upon landfall in southeastern China. By leveraging the unique capability of FNO to model global dependencies in continuous function spaces,this method effectively captures the non-local spatiotemporal dynamics of meteorological fields,addressing the inherent limitations of traditional convolutional and recurrent neural networks in long-range dependency modeling. The experiments utilized the 1949-2024 best-track dataset of tropical cyclones from the China Meteorological Administration(CMA),focusing on typhoon samples that entered the region between 105°E-130°E and 15°N-35°N. Additionally, multi-source reanalysis meteorological fields were integrated as environmental covariates. The results show that the proposed FNO model achieves an average 24-hour track prediction error of 128. 6km and an intensity prediction root mean square error(RMSE)of 6. 4 m/s,significantly outperforming state-of-the-art deep learning models such as LSTM and Transformer.
为全面验证所提TyphoonFNO 模型的预测性能,选取多种代表性深度学习方法作为对比基线,具体包括:经典循环神经网络模型长短期记忆神经网络(long short term memory,LSTM)[8]、双向长短期记忆神经网络(bi-directional long short-term memory,BiLSTM)[21];自注意力机制架构模型Transformer[22];以及近年来提出的全连接类模型Mlp-Mixer[23]、卷积类模型时间卷积神经网络(temporal convolutional neural network,TCN)[24]。所选对比基线涵盖循环(RNN)、卷积(CNN)、注意力(Transformer)、全连接(MLP) 四大主流深度学习范式,且均在时空序列预测、轨迹建模等相关领域有成熟应用案例,能够充分覆盖不同建模思路的性能差异。为保障实验对比的公平性,所有基线模型均在与本文 TyphoonFNO 模型完全一致的数据划分、数据预处理条件下重新训练,统一训练超参数(如学习率、批量大小、训练轮次等),确保实验结果的可比性与可靠性。TyphoonFNO 模型的训练损失(loss)曲线如图2所示。
JIANGGuoqing, XUJing, WEIJun. A deep learning algorithm of neural network for the parameterization of typhoon-ocean feedback in typhoon forecast models[J]. Geophysical Research Letters,2018,45(8):3706-3716.
[4]
LEER S T, LIUJ N K. Tropical cyclone identification and tracking system using integrated neural oscillatory elastic graph matching and hybrid RBF network track mining techniques[J]. IEEE Transactions on Neural Networks, 2000, 11(3): 680-689.
[5]
ALIM M, KISHTAWALC M, JAINS. Predicting cyclone tracks in the north Indian Ocean: An artificial neural network approach[J]. Geophysical Research Letters, 2007, 34(4): L04603.DOI:10.1029/2006GL028353 .
[6]
MORADI KORDMAHALLEHM, GORJI SEFIDMAZGIM, HOMAIFARA. A sparse recurrent neural network for trajectory prediction of Atlantic hurricanes[C]//Proceedings of the Genetic and Evolutionary Computation Conference 2016. Denver, Colorado, USA:ACM, 2016: 957-964.
[7]
ALEMANYS, BELTRANJ, PEREZA, et al. Predicting hurricane trajectories using a recurrent neural network[C]//Proceedings of the AAAI Conference on Artificial Intelligence. Palo Alto:AAAI Press, 2019: 468-475.
[8]
HOCHREITERS, SCHMIDHUBERJ. Long short-term memory[J]. Neural Computation, 1997, 9(8): 1735-1780.
[9]
KOLENJ F, KREMERS C. Gradient flow in recurrent nets: The difficulty of learning long-term dependencies[C]//A Field Guide to Dynamical Recurrent Networks.New York:Wiley-IEEE Press, 2001: 237-243.
[10]
BENGIOY, SIMARDP, FRASCONIP. Learning long-term dependencies with gradient descent is difficult[J]. IEEE Transactions on Neural Networks, 1994, 5(2): 157-166.
[11]
ZHANGYanfei, CHANDRAR, GAOJunbin. Cyclone track prediction with matrix neural networks[C]//International Joint Conference on Neural Networks (IJCNN). Piscataway,NJ, USA:IEEE, 2018: 1-8.
[12]
ZHANGY, LIH, CHENJ. Typhoon track prediction based on convolutional neural network and long short-term memory network. Advances in Atmospheric Sciences, 2020, 37(1): 69-82.
[13]
JIANGWenjun, HUGang, WUTiantian T, et al. DMANet_KF: Tropical cyclone intensity estimation based on deep learning and Kalman filter from multispectral infrared images[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2023, 16: 4469-4483.
[14]
LITongfei, CHEKaihua, LUJiadong, et al. Tropical cyclone ensemble forecast framework based on spatiotemporal model[J]. Earth Science Informatics, 2024, 17(5): 4791-4807.
[15]
YanfangLÜ, ZHAOXiaoyu, GONGZhiqiang, et al. Multi-fidelity prediction of fluid flow based on transfer learning using Fourier neural operator[J]. Physics of Fluids, 2023, 35(7): 077118. DOI:10.1063/5.0155555 .
[16]
KURTHT, SUBRAMANIANS, HARRINGTONP, et al. FourCastNet: Accelerating global high-resolution weather forecasting using adaptive Fourier neural operators[C]//Proceedings of the Platform for Advanced Scientific Computing Conference. Davos, Switzerland:ACM, 2023: 13.
[17]
WEIWei, FULiyun. Small-data-driven fast seismic simulations for complex media using physics-informed Fourier neural operators[J].Geophysics,2022,87(6): T435-T446.
[18]
LIGuancheng, CHENGLijing, ZHUJiang, et al. Increasing ocean stratification over the past half-century[J]. Nature Climate Change, 2020, 10(12): 1116-1123.
[19]
YINGMing, ZHANGWei, YUHui, et al. An overview of the China meteorological administration tropical cyclone database[J]. Journal of Atmospheric and Oceanic Technology, 2014, 31(2): 287-301.
[20]
ZHANGChangjiang, WANGXiaojie, MALeiming, et al. Tropical cyclone intensity classification and estimation using infrared satellite images with deep learning[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2021, 14: 2070-2086.
[21]
SIAMI-NAMINIS, TAVAKOLIN, NAMINA S. The performance of LSTM and BiLSTM in forecasting time series[C]//IEEE International Conference on Big Data (Big Data). Piscataway,NJ, USA:IEEE, 2020: 3285-3292.
[22]
VASWANIA, SHAZEERN, PARMARN, et al. Attention is all you need[C]// Advances in Neural Information Processing Systems,2017,30:5998-6008.
[23]
TOLSTIKHINI O, HOULSBYN, KOLESNIKOVA, et al. MLP-mixer: An all-MLP architecture for vision[EB/OL].(2021-06-11)[2025-12-16].
[24]
HEWAGEP, BEHERAA, TROVATIM, et al. Temporal convolutional neural (TCN) network for an effective weather forecasting using time-series data from the local weather station[J]. Soft Computing, 2020, 24(21): 16453-16482.