基于领域先验知识的时空神经网络模型在MJO预报中的应用

Application of Domain Prior Knowledge-Based Spatio-temporal Neural Network Model in MJO Forecasting

摘要

针对目前人工神经网络方法无法准确预报季节性气候现象Madden-Julian振荡(MJO)的问题,提出一种基于领域先验知识的时空神经网络模型.首先,该方法结合气候环流数据的特性,融入领域先验知识进行数据预处理;其次,采用预训练-微调架构,利用次季节-季节的模式数据进行模型预训练,并通过再分析数据(ERA5)完成微调;最后,通过时空建模,选用卷积神经网络和长短期记忆网络结合的框架,将先验知识嵌入预训练过程并优化预报.实验结果表明,该模型能实现23d的MJO准确预报,其性能优于其他人工神经网络方法及国内数值预报方法.

Abstract

Aiming at the problem that current artificial neural network methods could not accurately forecast the Madden-Julian oscillation (MJO), a seasonal climate phenomenon, we proposed a domain prior knowledge-based spatio-temporal neural network model. Firstly, we integrated domain prior knowledge into data preprocessing according to the characteristics of climate circulation data. Secondly, a pretraining-finetuning architecture was adopted, model data from the subseasonal-to-seasonal scale were used for pretraining model, and we completed finetuning by using the ERA5 reanalysis data. Finally, through spatio-temporal modeling, a framework combining convolutional neural networks and long short-term memory networks was selected to embed the prior knowledge into the pretraining process and optimize the forecasting. Experimental results show that the proposed model can achieve accurate MJO forecasting for 23d, and its performance is superior to other artificial neural network methods and domestic numerical forecasting methods.

关键词

领域先验知识 / 时空神经网络 / Madden-Julian振荡预报 / 预训练模型

Key words

domain prior knowledge / spatio-temporal neural network / Madden-Julian oscillation forecasting / pretraining model

参考文献

[1] ZHANG C D. Madden-Julian Oscillation: Bridging Weather and Climate[J]. Bulletin of the American Meteorological Society, 2013, 94(12): 1849-1870.

[2] ZHANG C D. Madden-Julian Oscillation[J]. Reviews of Geophysics, 2005, 43(2): 1-36.

[3] HE S J, LI X Y, DELSOLE T, et al. Sub-seasonal Climate Forecasting via Machine Learning: Challenges, Analysis, and Advances[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2021, 35(1): 169-177.

[4] 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.

[5] LORENZ A C. Analysis Methods for Numerical Weather Prediction[J]. Quarterly Journal of the Royal Meteorological Society, 1986, 112: 1177-1194.

[6] SCHULTZ M G, BETANCOURT C, GONG B, et al. Can Deep Learning Beat Numerical Weather Prediction?[J]. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, 2021, 379: 0097-1-0097-22.

[7] VITART F, ROBERTSON A W. The Sub-seasonal to Seasonal Prediction Project (S2S) and the Prediction of Extreme Events[J]. NPJ Climate and Atmospheric Science, 2018, 1: 3-1-3-7.

[8] HAM Y G, KIM J H, LUO J J. Deep Learning for Multi-year ENSO Forecasts[J]. Nature, 2019, 573(7775): 568-572.

[9] MARTIN Z, BARNES E, MALONEY E. Predicting the MJO Using Interpretable Machine-Learning Models[EB/OL]. (2022-11-23)[2024-10-15]. https://doi.org/10.1002/essoar.10506356.1.

[10] CHEN L, ZHONG X H, ZHANG F, et al. FuXi: A Cascade Machine Learning Forecasting System for 15-Day Global Weather Forecast[J]. NPJ Climate and Atmospheric Science, 2023, 6(1): 190-1-190-11.

[11] HERSBACH H, BELL B, BERRISFORD P, et al. The ERA5 Global Reanalysis[J]. Quarterly Journal of the Royal Meteorological Society, 2020, 146(730): 1999-2049.

[12] 谭奕鑫, 詹永照, 刘洪麟. 基于显著特征和时空图网络的视频异常事件检测[J]. 江苏大学学报(自然科学版), 2025, 46(2):179-188.

[13] (TAN Y X, ZHAN Y Z, LIU H L. Video Abnormal Event Detection Based on Salient Features and Space-Time Fusion Graph Network[J]. Journal of Jiangsu University (Natural Science Edition), 2025, 46(2):179-188.)

[14] LI H L, XU Z W, FANG Z L, et al. MJOFormer: An Adaptive Land-Ocean Spatio-temporal Transformer for Madden Julian Oscillation Forecasting[J]. Computers & Geosciences, 2025: 106097-1-106097-11.

[15] CHAI T, DRAXLER R R. Root Mean Square Error (RMSE) or Mean Absolute Error (MAE)? — Arguments against Avoiding RMSE in the Literature[J]. Geoscientific Model Development, 2014, 7(3): 1247-1250.

[16] 王绮, 黄国民, 卢恩琪, 等. 基于地理探测器的旅游生态安全时空特征及影响因素研究[J]. 东北师大学报(自然科学版), 2024, 56(4):133-143.

[17] (WANG Q, HUANG G M, LU E Q, et al. Study on the Spatial-Temporal Characteristics and Influencing Factors of Tourism Ecological Security Based on Geodetector[J]. Journal of Northeast Normal University (Natural Science Edition), 2024, 56(4): 133-143.)

[18] 贺金鑫, 张涵雅, 周俊宏, 等. 基于知识图谱的火星季节性斜坡纹线时空特征分析[J]. 吉林大学学报(地球科学版), 2026, 56(1):252-265.

[19] (HE J X, ZHANG H Y, ZHOU J H, et al. Analyzing Spatial and Temporal Features of Martian Recurring Slope Lineae Based on Knowledge Fraph[J]. Journal of Jilin University (Earth Science Edition), 2026, 56(1): 252-265.

[20] DEVLIN J, CHANG M W, LEE K, et al. Bert: Pre-training of Deep Bidirectional Transformers for Language Understanding[C]// Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. [S. l.]: ACL, 2019: 4171-4186.

[21] ROY S K, NICOLSON A, PALIWAL K K. A Deep Learning-Based Kalman Filter for Speech Enhancement[C]// Interspeech. [S. l.]: ISCA, 2020: 2692-2696.

[22] TANG Y M, YU B. MJO and Its Relationship to ENSO[J]. Journal of Geophysical Research: Atmospheres, 2008, 113(14): 2007JD009230-1-2007JD009230-40.

[23] LI Z W, LIU F, YANG W J, et al. A Survey of Convolutional Neural Networks: Analysis, Applications, and Prospects[J]. IEEE Transactions on Neural Networks and Learning Systems, 2022, 33(12): 6999-7019.

[24] DEY R, SALEM F M. Gate-Variants of Gated Recurrent Unit (GRU) Neural Networks[C]// 2017 IEEE 60th International Midwest Symposium on Circuits and Systems. Piscataway, NJ: IEEE, 2017: 1597-1600.

[25] SHERSTINSKY A. Fundamentals of Recurrent Neural Network (RNN) and Long Short-Term Memory (LSTM) Network[J]. Physica D: Nonlinear Phenomena, 2020, 404: 132306-1-132306-28.

[26] ZHOU J X, LI X, DING T Y, et al. On the Optimization Landscape of Neural Collapse under MSE Loss: Global Optimality with Unconstrained Features[C]// International Conference on Machine Learning. [S.l.]: PMLR, 2022: 27179-27202.

[27] HO Y, WOOKEY S. The Real-World-Weight Cross-Entropy Loss Function: Modeling the Costs of Mislabeling[J]. IEEE Access, 2020, 8: 4806-4813.

[28] PEZZANO G, RIPOLL V R, RADEVA P. CoLe-CNN: Context-Learning Convolutional Neural Network with Adaptive Loss Function for Lung Nodule Segmentation[J]. Computer Methods and Programs in Biomedicine, 2021, 198: 105792-1-105792-13.

[29] FAN M, IMRAN O, SINGH A, et al. Using CNN-LSTM Model for Weather Forecasting[C]// 2022 IEEE International Conference on Big Data (Big Data). Piscataway, NJ: IEEE, 2022: 4120-4125.

[30] SHI X J, CHEN Z R, WANG H, et al. Convolutional LSTM Network: A Machine Learning Approach for Precipitation Nowcasting[J]. Advances in Neural Information Processing Systems, 2015, 28: 1-9.

基金资助

吉林省自然科学基金面上项目(20230101062JC)

国家自然科学基金(42175052)