融合多模态气象信息与MJO时空演变特征的预测模型

徐政; 方巍

doi:10.12284/hyxb20250101

融合多模态气象信息与MJO时空演变特征的预测模型

doi: 10.12284/hyxb20250101

徐政^1,,
方巍^{1, 2, 3, 4, ,}

1.
南京信息工程大学计算机学院，江苏南京 210044
2.
中国气象科学研究院灾害天气全国重点实验室，北京 100081
3.
南京信息工程大学江苏省大气环境与装备技术协同创新中心，江苏南京 210044
4.
伦敦国王学院信息学系，英国伦敦 WC2B 4BG

基金项目: 国家自然科学基金面上项目（42475149）；中国气象局流域强降水重点开放实验室开放研究基金（2023BHR–Y14）；灾害天气国家重点实验室开放课题（2024LASW–B19）；广西重点研发计划（桂科AB25069126）。

详细信息

作者简介:
徐政（2000—），男，江苏省徐州市人，研究方向为深度学习、MJO预测。E-mail：2651479762@qq.com

通讯作者:
方巍，教授，研究方向为人工智能、大数据分析、机器学习和气象信息技术。E-mail：fangwei@nuist.edu.cn

中图分类号: P462.4
计量
- 文章访问数: 149
- HTML全文浏览量: 95
- PDF下载量: 14
- 被引次数: 0
出版历程
- 收稿日期: 2025-07-17
- 修回日期: 2025-10-28
- 网络出版日期: 2025-11-04
- 刊出日期: 2025-10-31

A predictive model integrating multimodal meteorological information and spatiotemporal evolution of MJO

Xu Zheng^1
,,
Fang Wei^{1, 2, 3, 4
, ,}

1.
School of Computer, Nanjing University of Information Science and Technology, Nanjing 210044, China
2.
State Key Laboratory of Severe Weather, Chinese Academy of Meteorological Sciences, Beijing 100081, China
3.
Jiangsu Collaborative Innovation Center of Atmospheric Environment and Equipment Technology (CICAEET), Nanjing University of Information Science and Technology, Nanjing 210044, China
4.
Department of Informatics, King’s College, London WC2B 4BG, UK

摘要

摘要: 马登–朱利安振荡（Madden-Julian Oscillation，MJO）作为热带季节内变率的主要模态，其准确预测对于提升次季节预测能力至关重要。然而，MJO具有多尺度演变特征和高度非线性动力过程，现有预测方法在捕捉其复杂时空结构方面仍存在不足。为此，本文提出了一种融合多模态数据与时空特征的MJO预测模型（Multimodal data and Integrated Spatiotemporal features for MJO prediction，MISM）。该模型以历史实时多变量MJO指数（Real-time Multivariate MJO index，RMM）和多个气象因子作为联合输入，通过压缩激励模块、卷积模块和Swin Transformer模块构建空间特征提取模块，并引入自回归注意力机制实现非线性时间序列建模。实验结果表明，MISM模型的预测技巧可延伸至30 d以上，并在25 d以上的长期预测阶段表现优于传统的动力学和统计学方法。此外，本文利用显著性图对气象因子贡献区域进行分析，结果显示西太平洋及印尼群岛在不同提前期均呈现较高敏感性，海洋区域贡献普遍强于陆地。水汽和海温异常在短期与中期作用更突出，而低层风场和对流活动在长期阶段贡献较强，高层环流则在各时效保持稳定影响，体现了模型对MJO演变机制的识别能力。
- MJO预测 /
- 次季节预测 /
- 多模态数据 /
- 时空建模 /
- 显著性图
Abstract: The Madden-Julian Oscillation (MJO), as the primary mode of tropical intraseasonal variability, plays a critical role in improving subseasonal prediction skill. However, due to its multi-scale evolutionary characteristics and highly nonlinear dynamical processes, existing prediction methods still struggle to effectively capture the complex spatiotemporal structure of MJO. To address this issue, we propose a novel prediction model named MISM (Multi-modal data and Integrated Spatiotemporal features for MJO prediction), which integrates multimodal inputs and spatiotemporal feature extraction. The model jointly leverages historical Real-time Multivariate MJO (RMM) indices and multiple meteorological variables as inputs. It constructs a spatial feature extraction module based on Squeeze-and-Excitation (SE) blocks, convolutional layers, and the Swin Transformer, as well as an autoregressive attention mechanism for nonlinear temporal modeling. Experimental results demonstrate that the MISM model extends predictive skill to beyond 30 d and shows overall superior performance compared with traditional dynamical and statistical methods in long-lead forecasts beyond 25 d. Furthermore, saliency maps are utilized to analyze the contribution regions of meteorological factors. The results reveal that the western Pacific and the Indonesian archipelago consistently exhibit high sensitivity across different lead times, with oceanic regions generally contributing more than land areas. Water vapor and sea surface temperature anomalies play a more prominent role in short- to medium-term forecasts, while low-level wind fields and convective activity contribute more significantly in longer-term forecasts. High-level circulation exerts a stable influence across all lead times, highlighting the model’s ability to capture the mechanisms of MJO evolution.
- MJO prediction /
- subseasonal forecast /
- multimodal data /
- spatiotemporal modeling /
- saliency map

HTML全文

图 1 MISM模型结构

Fig. 1 Model structure of MISM

下载: 全尺寸图片幻灯片

图 2 SE模块结构

Fig. 2 SE module architecture

下载: 全尺寸图片幻灯片

图 3 自回归注意力机制Token生成

Fig. 3 Autoregressive attention Token generation

下载: 全尺寸图片幻灯片

图 4 MISM模型预测性能

Fig. 4 MISM model prediction performance

下载: 全尺寸图片幻灯片

图 5 多模型预测技巧对比

Fig. 5 Comparison of multi-model forecast skills

下载: 全尺寸图片幻灯片

图 6 各气象因子的显著性分布图

Fig. 6 The saliency maps of the meteorological variables

下载: 全尺寸图片幻灯片

表 1 消融实验结果

Tab. 1 Results of the ablation experiments

模型	不同提前期τ/d
模型	τ = 1	τ = 3	τ = 5	τ = 10	τ = 15	τ = 20	τ = 25	τ = 30	τ = 35
MISM	0.963	0.896	0.820	0.720	0.656	0.626	0.584	0.527	0.472
气象因子	0.906	0.823	0.780	0.703	0.642	0.604	0.555	0.475	0.467
历史指数	0.943	0.837	0.697	0.354	0.245	0.223	0.219	0.167	0.100
w/o Swin	0.942	0.856	0.750	0.519	0.342	0.324	0.298	0.228	0.157
w/o Auto	0.867	0.539	0.694	0.465	0.580	0.288	0.390	0.431	0.292
注：加粗的数据是该指标下最优的情况。

下载: 导出CSV

参考文献(27)

[1]	徐邦琪, 臧钰歆, 朱志伟, 等. 时空投影模型(STPM)的次季节至季节(S2S)预测应用进展[J]. 大气科学学报, 2020, 43(1): 212−224. Xu Bangqi, Zhang Yuxin, Zhu Zhiwei, et al. Subseasonal-to-seasonal (S2S) prediction using the spatial-temporal projection model (STPM)[J]. Transactions of Atmospheric Sciences, 2020, 43(1): 212−224.
[2]	伍继业, 谢欣芮, 罗京佳. 基于改进版NUIST CFS1.1的热带大气季节内信号及其对中国气温降水影响的预测评估[J]. 大气科学学报, 2024, 47(2): 284−299. Wu Jiye, Xie Xinrui, Luo Jingjia. Prediction of tropical intraseasonal oscillations and their impacts on air temperature and precipitation in China using the upgraded version of NUIST CFS1.1[J]. Transactions of Atmospheric Sciences, 2024, 47(2): 284−299.
[3]	White C J, Domeisen D I V, Acharya N, et al. Advances in the application and utility of subseasonal-to-seasonal predictions[J]. Bulletin of the American Meteorological Society, 2022, 103(6): E1448−E1472. doi: 10.1175/BAMS-D-20-0224.1
[4]	李力锋, 陈雄, 李崇银, 等. 2019年2−3月中国南方持续性降水成因及其与MJO的关系[J]. 气象, 2022, 48(9): 1090−1100. Li Lifeng, Chen Xiong, Li Chongyin, et al. Causes of persistent precipitation over southern China during February-March 2019 and the relationship with MJO[J]. Meteorological Monthly, 2022, 48(9): 1090−1100.
[5]	Ratna S B, Sabeerali C T, Sharma T, et al. Combined influence of El Niño, IOD and MJO on the Indian Summer Monsoon Rainfall: case study for the years 1997 and 2015[J]. Atmospheric Research, 2024, 299: 107214. doi: 10.1016/j.atmosres.2023.107214
[6]	Cowan T, Wheeler M C, Marshall A G. The combined influence of the Madden-Julian oscillation and El Niño-Southern Oscillation on Australian rainfall[J]. Journal of Climate, 2023, 36(2): 313−334. doi: 10.1175/JCLI-D-22-0357.1
[7]	Zhou Wenyu, Yang Da, Xie Shangping, et al. Amplified Madden-Julian oscillation impacts in the Pacific-North America region[J]. Nature Climate Change, 2020, 10(7): 654−660. doi: 10.1038/s41558-020-0814-0
[8]	Yao Junchen, Liu Xiangwen, Wu Tongwen, et al. Progress of MJO prediction at CMA from phase I to phase II of the sub-seasonal to seasonal prediction project[J]. Advances in Atmospheric Sciences, 2023, 40(10): 1799−1815. doi: 10.1007/s00376-023-2351-z
[9]	Wheeler M C, Hendon H H. An all-season real-time multivariate MJO index: development of an index for monitoring and prediction[J]. Monthly Weather Review, 2004, 132(8): 1917−1932. doi: 10.1175/1520-0493(2004)132<1917:AARMMI>2.0.CO;2
[10]	Zhu Zhiwei, Li T, Hsu P C, et al. A spatial-temporal projection model for extended-range forecast in the tropics[J]. Climate Dynamics, 2015, 45(3/4): 1085−1098.
[11]	Wu Jiye, Li Yue, Luo Jingjia, et al. Prediction and predictability of boreal winter MJO using a multi-member subseasonal to seasonal forecast system of NUIST (NUIST CFS 1.1)[J]. Climate Dynamics, 2024, 62(5): 3003−3026. doi: 10.1007/s00382-023-07047-4
[12]	Dey A, Chattopadhyay R, Sahai A K, et al. MJO prediction skill using IITM extended range prediction system and comparison with ECMWF S2S[J]. Pure and Applied Geophysics, 2020, 177(10): 5067−5079. doi: 10.1007/s00024-020-02487-z
[13]	Yuval J, O’Gorman P A. Stable machine-learning parameterization of subgrid processes for climate modeling at a range of resolutions[J]. Nature communications, 2020, 11(1): 3295. doi: 10.1038/s41467-020-17142-3
[14]	Toms B A, Barnes E A, Ebert-Uphoff I. Physically interpretable neural networks for the geosciences: applications to earth system variability[J]. Journal of Advances in Modeling Earth Systems, 2020, 12(9): e2019MS002002. doi: 10.1029/2019MS002002
[15]	Weyn J A, Durran D R, Caruana R. Improving data-driven global weather prediction using deep convolutional neural networks on a cubed sphere[J]. Journal of Advances in Modeling Earth Systems, 2020, 12(9): e2020MS002109. doi: 10.1029/2020MS002109
[16]	杨淑贤, 零丰华, 应武杉, 等. 人工智能技术气候预测应用简介[J]. 大气科学学报, 2022, 45(5): 641−659. Yang Shuxian, Ling Fenghua, Ying Wushan, et al. A brief overview of the application of artificial intelligence to climate prediction[J]. Transactions of Atmospheric Sciences, 2022, 45(5): 641−659.
[17]	Kim H, Ham Y G, Joo Y S, et al. Deep learning for bias correction of MJO prediction[J]. Nature Communications, 2021, 12(1): 3087.
[18]	Delaunay A, Christensen H M. Interpretable deep learning for probabilistic MJO prediction[J]. Geophysical Research Letters, 2022, 49(16): e2022GL098566. doi: 10.1029/2022GL098566
[19]	Shin N Y, Kim D, Kang D, et al. Deep learning reveals moisture as the primary predictability source of MJO[J]. npj Climate and Atmospheric Science, 2024, 7(1): 11. doi: 10.1038/s41612-023-00561-6
[20]	Shin N Y, Kang D, Kim D, et al. Data-driven investigation on the boreal summer MJO predictability[J]. npj Climate and Atmospheric Science, 2024, 7(1): 248. doi: 10.1038/s41612-024-00799-8
[21]	Chen Lei, Zhong Xiaohui, Li Hao, et al. A machine learning model that outperforms conventional global subseasonal forecast models[J]. Nature Communications, 2024, 15(1): 6425. doi: 10.1038/s41467-024-50714-1
[22]	胡家晖, 陆波, 李昊, 等. 人工智能模型“风顺”对中国区域降水技巧检验[J]. 大气科学学报, 2025, 48(3): 366−376. Hu Jiahui, Lu Bo, Li Hao, et al. Skill test of the artificial intelligence model “Fengshun” for precipitation forecasting in China[J]. Transactions of Atmospheric Sciences, 2025, 48(3): 366−376.
[23]	张浩睿, 周磊. 热带季节内振荡在印尼海区域东传路径的分布及其机制[J]. 海洋学报, 2023, 45(10): 13−30. Zhang Haorui, Zhou Lei. The distribution of eastward propagating pathways of the Tropical Intraseasonal Oscillation and its mechanism in the Maritime Continent[J]. Haiyang Xuebao, 2023, 45(10): 13−30.
[24]	Dai Kuai, Li Xutao, Ma Chi, et al. Learning spatial-temporal consistency for satellite image sequence prediction[J]. IEEE Transactions on Geoscience and Remote Sensing, 2023, 61: 4104517.
[25]	Chen Guosen. Deciphering chaos in the Madden-Julian oscillation[J]. npj Climate and Atmospheric Science, 2024, 7(1): 311. doi: 10.1038/s41612-024-00870-4
[26]	Li Hongliang, Zhang Nong, Xu Zhewen, et al. DK-STN: a domain knowledge embedded spatio-temporal network model for MJO forecast[J/OL]. Expert Systems with Applications, Forthcoming, 2023[2023−09−18][2025−07−17]. https://dx.doi.org/10.2139/ssrn.4574792.
[27]	Liu Yumin, Duffy K, Dy J G, et al. Explainable deep learning for insights in El Niño and river flows[J]. Nature Communications, 2023, 14(1): 339. doi: 10.1038/s41467-023-35968-5