Forecast of sea surface temperature in the South China Sea based on multi-scale deep learning model
-
摘要: 海表温度是海洋最重要的物理量之一，提供了气候系统的基本信息，准确地预报海表温度有着广泛而重要的应用。近年来，基于人工智能的海温预报方法开始流行，并展现出巨大的潜力。基于卷积长短时记忆神经网络（ConvLSTM），本文研究了多尺度输入场对南海北部二维海表温度预报结果的影响。文章采用多元集合经验模态分解方法（MEEMD）将日均海表温度分解成多个尺度的空间主模态，并以不同的组合训练ConvLSTM模型进行预报实验。结果表明，采用前4个海表温度主模态数据训练模型时，预报1～7 d海表温度的均方根误差约为0.4～0.8℃，比仅用原始海表温度训练时减小了0.2～1.2℃；平均绝对百分比误差为1%～6%，减小了0.5%～10%；空间相关系数为99.5%～96.5%，提高了0.5%～3.5%。而且，随机实验也进一步证明该方法具有较高的普适性。基于深度学习的预报模型，需结合海温的物理特性，选择合适的数据进行训练，才能进一步提高其预报精度。本文初步探究了人工智能方法与物理概念在海温预报中的融合，可为以后的研究提供一定的参考。Abstract: Sea surface temperature (SST) is one of the most important physical variables of the ocean, which provides the basic information of the climate system. Accurately SST forecasting system has a comprehensive and essential application. In recent years, AI-based SST forecasting methods have become popular and shown great potential. Based on the convolutional long and short-term memory neural network (ConvLSTM), this paper studies the influence of multi-scale input fields on SST prediction in the northern South China Sea. Multi-dimensional ensemble empirical mode decomposition method (MEEMD) is used to decompose the average daily SST into the spatial eigenmodes of differentiated scales. Input different combinations of eigenmodes into ConvLSTM for training and prediction experiments. Results show that when using all four SST eigenmodes, the RMSE of the predicted SST in 1−7 days is 0.4−0.8℃, decrease 0.2−1.2℃ compared with the original SST alone; the MAPE is 1%−6%, decrease 0.5%−10%; the spatial correlation coefficient is 99.5%−96.5%, improve 0.5%−3.5%. Moreover, the randomized experiments also further proved the method has a high universality. The prediction model based on deep learning needs to select the appropriate training data in order to further improve its prediction accuracy. This paper preliminarily explores the integration of artificial intelligence methods and physical concepts in SST prediction, which can provide some reference for future research.
-
Key words:
- SST prediction /
- deep learning /
- ConvLSTM /
- MEEMD
-
图 1 基于MEEMD方法对南海北部海表温度分解的例子
左侧子图为原始日均海表温度，中间子图为经分解后得到IMF1、IMF2、IMF3、IMF4、 4个海温主模态，右侧子图为黑点（17.975°N，114.975°E）位置处温度随时间的变化趋势。其中，灰线表示日平均变化，黑线表示月平均变化，红线表示长期趋势
Fig. 1 SST spatial decomposition in the northern South China Sea based on the MEEMD method
The left subgraph shows the original daily average SST, the middle subgraph shows the four SST modes of IMF1, IMF2, IMF3, IMF4, after decomposition, and the right subgraph shows the trend of the SST amplitude over 30 years at the black point (17.975°N, 114.975°E). Gray lines indicate daily variation, black lines indicate monthly variation, and red lines indicate long-term trends
图 2 基于ConvLSTM模型进行海表温度预报的框架示意图
IMF1— IMFN代表第1至N个不同尺度的海表温度模态
Fig. 2 Schematic of the framework for 2D SST prediction based on the ConvLSTM model
IMF1-IMFN indicate SST modes of different scales from 1 to N
图 3 基于ConvLSTM模型，利用30年日均海表温度数据进行训练，预报南海北部7天海表温度的结果
第一行为实测SST；第二行为预报SST；第三行为两者之差
Fig. 3 The 7 days SST forecast results of 30 years daily SST data based ConvLSTM model
First row denote the observed SST; second row denote the forecast SST; third row denote the error between the observation and the prediction
图 4 利用不同组合的海表温度主模态进行训练，预报南海北部7 d海表温度的结果
预报所采用的初始海表温度场与图3一致。第1～4行分别表示采用前1～4个模态的结果，例如IMF1-3表示采用前3个模态作为模型的不同通道进行训练预报的结果
Fig. 4 Using different combinations of SST eigenmodes to forecast the 7-day SST
The initial SST field employed in the prediction is consistent with Figure 3. Rows 1−4 denote the results of the first 1−4 modes (IMF1-1, IMF1-2, IMF1-3 and IMF1-4) respectively
图 6 不同实验预报效果的量化比较
黑线（No IMF）表示未采用模态分解的结果，蓝线（IMF1-1）表示采用前1个模态的结果，黄线（IMF1-2）表示采用前两个模态的结果，绿线（IMF1-3）表示采用前3个模态的结果，红线（IMF1-4）表示采用前4个模态的结果，紫线代表海表温度的持续性预报结果
Fig. 6 Quantitative comparison of the prediction effects of different experiments
The black line (No IMF) indicates the results of no mode decomposition, the blue line (IMF1-1) represents the results of the first mode, the yellow line (IMF1-2) represents the results of the first two modes, the green line (IMF1-3) represents the results of the first three modes, the red line (IMF1-4) represents the results of the first four modes, and the purple line represents the results of the SST persistence forecast
图 7 采用IMF1-4进行7 d 海表温度训练预报的随机实验
本文任意选取了30年中90个时段的海表温度进行测试，计算每个时段的3种量化指标，取IMF1-4与No IMF的差进行比较。图中黑色实线表示等值线为0值的位置
Fig. 7 Randomized experiment with 7-day SST prediction using IMF1-4
We arbitrarily selected 90 different periods in the 30 years, calculated the difference between IMF1-4 and No IMF of three quantitative indexes for each period. The black solid lines indicate the zero values position
-
[1] 方长芳, 张翔, 尹建平. 21世纪初海洋预报系统发展现状和趋势[J]. 海洋预报, 2013, 30(4): 93−102. doi: 10.11737/j.issn.1003-0239.2013.04.013Fang Changfang, Zhang Xiang, Yin Jianping. Development status and trends of ocean forecasting system in the 21st Century[J]. Marine Forecasts, 2013, 30(4): 93−102. doi: 10.11737/j.issn.1003-0239.2013.04.013 [2] 韩鹏, 李宇航, 揭晓蒙. 国际全球海洋环流预报系统的现状与展望[J]. 海洋预报, 2020, 37(3): 98−105. doi: 10.11737/j.issn.1003-0239.2020.03.012Han Peng, Li Yuhang, Jie Xiaomeng. The status and prospect of global ocean circulation forecasting system in foreign countries[J]. Marine Forecasts, 2020, 37(3): 98−105. doi: 10.11737/j.issn.1003-0239.2020.03.012 [3] 王兆毅, 李云, 王旭. 中国近岸海域基础预报单元海温预报指导产品研制[J]. 海洋预报, 2020, 37(4): 59−65. doi: 10.11737/j.issn.1003-0239.2020.04.007Wang Zhaoyi, Li Yun, Wang Xu. Development of forecast guidance product for sea temperature of basic forecast units in the Chinese coastal waters[J]. Marine Forecasts, 2020, 37(4): 59−65. doi: 10.11737/j.issn.1003-0239.2020.04.007 [4] Reichstein M, Camps-Valls G, Stevens B, et al. Deep learning and process understanding for data-driven Earth system science[J]. Nature, 2019, 566(7743): 195−204. doi: 10.1038/s41586-019-0912-1 [5] Li Xiaofeng, Liu Bin, Zheng Gang, et al. Deep-learning-based information mining from ocean remote-sensing imagery[J]. National Science Review, 2020, 7(10): 1584−1605. doi: 10.1093/nsr/nwaa047 [6] 贺圣平, 王会军, 李华, 等. 机器学习的原理及其在气候预测中的潜在应用[J]. 大气科学学报, 2021, 44(1): 26−38.He Shengping, Wang Huijun, Li Hua, et al. Machine learning and its potential application to climate prediction[J]. Transactions of Atmospheric Sciences, 2021, 44(1): 26−38. [7] Dong Changming, Xu Guangjun, Han Guoqing, et al. Recent developments in artificial intelligence in oceanography[J]. Ocean-Land-Atmosphere Research, 2022, 2022: 9870950. [8] Liu Yingjie, Zheng Quanan, Li Xiaofeng. Characteristics of global ocean abnormal mesoscale eddies derived from the fusion of sea surface height and temperature data by deep learning[J]. Geophysical Research Letters, 2021, 48(17): e2021GL094772. doi: 10.1029/2021GL094772 [9] Xu Guangjun, Xie Wenhong, Dong Changming, et al. Application of three deep learning schemes into oceanic eddy detection[J]. Frontiers in Marine Science, 2021, 8: 672334. doi: 10.3389/fmars.2021.672334 [10] Zhang Xudong, Zhang Tao, Li Xiaofeng. Satellite-data-driven propagation speed model for internal solitary waves in the shallow and deep oceans[C]//Proceedings of 2021 IEEE International Geoscience and Remote Sensing Symposium IGARSS. Brussels: IEEE, 2021: 7402−7405. [11] Zhang Xudong, Li Xiaofeng, Zheng Quanan. A machine-learning model for forecasting internal wave propagation in the Andaman Sea[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2021, 14: 3095−3106. doi: 10.1109/JSTARS.2021.3063529 [12] Xiao Changjiang, Chen Nengcheng, Hu Chuli, et al. Short and mid-term sea surface temperaure prediction using time-series satellite data and LSTM-AdaBoost combination approach[J]. Remote Sensing of Environment, 2019, 233: 111358. doi: 10.1016/j.rse.2019.111358 [13] Wei Li, Guan Lei, Qu Liqin, et al. Prediction of sea surface temperature in the China seas based on long short-term memory neural networks[J]. Remote Sensing, 2020, 12(17): 2697. doi: 10.3390/rs12172697 [14] Yu Xuan, Shi Suixiang, Xu Lingyu, et al. A novel method for sea surface temperature prediction based on deep learning[J]. Mathematical Problems in Engineering, 2020, 2020: 6387173. [15] Zhou Shuyi, Xie Wenhong, Lu Yuxiang, et al. ConvLSTM-based wave forecasts in the South and East China Seas[J]. Frontiers in Marine Science, 2021, 8: 680079. doi: 10.3389/fmars.2021.680079 [16] Liang XiangSan, Xu Fen, Rong Yineng, et al. El Niño Modoki can be mostly predicted more than 10 years ahead of time[J]. Scientific Reports, 2021, 11(1): 17860. doi: 10.1038/s41598-021-97111-y [17] Sun Wenjin, Zhou Shuyi, Yang Jingsong, et al. Artificial intelligence forecasting of marine heatwaves in the South China sea using a combined U-Net and ConvLSTM system[J]. Remote Sensing, 2023, 15(16): 4068. doi: 10.3390/rs15164068 [18] Zhou Lu, Zhang Ronghua. A self-attention–based neural network for three-dimensional multivariate modeling and its skillful ENSO predictions[J]. Science Advances, 2023, 9(10): eadf2827. doi: 10.1126/sciadv.adf2827 [19] Guo Yanan, Cao Xiaoqun, Liu Bainian, et al. El Niño index prediction using deep learning with ensemble empirical mode decomposition[J]. Symmetry, 2020, 12(6): 893, doi: 10.3390/sym12060893 [20] Zheng Gang, Li Xiaofeng, Zhang Ronghua, et al. Purely satellite data-driven deep learning forecast of complicated tropical instability waves[J]. Science Advances, 2020, 6(29): eaba1482, doi: 10.1126/sciadv.aba1482 [21] Zhou Lu, Zhang Ronghua. A hybrid neural network model for ENSO prediction in combination with principal oscillation pattern analyses[J]. Advances in Atmospheric Sciences, 2022, 39(6): 889−902, doi: 10.1007/s00376-021-1368-4 [22] Gao Chuan, Zhou Lu, Zhang Ronghua. A transformer-based deep learning model for successful predictions of the 2021 second-year La Niña condition[J]. Geophysical Research Letters, 2023, 50(12): e2023GL104034, doi: 10.1029/2023GL104034 [23] Wan Zhongyi, Sapsis T P. Machine learning the kinematics of spherical particles in fluid flows[J]. Journal of Fluid Mechanics, 2018, 857: R2. doi: 10.1017/jfm.2018.797 [24] Mashayek A, Reynard N, Zhai Fangming, et al. Deep ocean learning of small scale turbulence[J]. Geophysical Research Letters, 2022, 49(15): e2022GL098039. doi: 10.1029/2022GL098039 [25] Zhu Yuchao, Zhang Ronghua, Moum J N, et al. Physics-informed deep-learning parameterization of ocean vertical mixing improves climate simulations[J]. National Science Review, 2022, 9(8): nwac044. doi: 10.1093/nsr/nwac044 [26] Zhang Qin, Wang Hui, Dong Junyu, et al. Prediction of sea surface temperature using long short-term memory[J]. IEEE Geoscience and Remote Sensing Letters, 2017, 14(10): 1745−1749. doi: 10.1109/LGRS.2017.2733548 [27] Xiao Changjiang, Chen Nengcheng, Hu Chuli, et al. A spatiotemporal deep learning model for sea surface temperature field prediction using time-series satellite data[J]. Environmental Modelling & Software, 2019, 120: 104502. [28] Sarkar P P, Janardhan P, Roy P. Prediction of sea surface temperatures using deep learning neural networks[J]. SN Applied Sciences, 2020, 2(8): 1458. doi: 10.1007/s42452-020-03239-3 [29] Xie Jiang, Zhang Jiyuan, Yu Jie, et al. An adaptive scale sea surface temperature predicting method based on deep learning with attention mechanism[J]. IEEE Geoscience and Remote Sensing Letters, 2020, 17(5): 740−744. doi: 10.1109/LGRS.2019.2931728 [30] Hao Peng, Li Shuang, Song Jinbao, et al. Prediction of sea surface temperature in the South China sea based on deep learning[J]. Remote Sensing, 2023, 15(6): 1656. doi: 10.3390/rs15061656 [31] Wei Li, Guan Lei. Seven-day sea surface temperature prediction using a 3DConv-LSTM model[J]. Frontiers in Marine Science, 2022, 9: 905848. doi: 10.3389/fmars.2022.905848 [32] Good S, Fiedler E, Mao Chongyuan, et al. The current configuration of the OSTIA system for operational production of foundation sea surface temperature and ice concentration analyses[J]. Remote Sensing, 2020, 12(4): 720. doi: 10.3390/rs12040720 [33] Wu Zhaohua, Huang N E, Chen Xianyao. The multi-dimensional ensemble empirical mode decomposition method[J]. Advances in Adaptive Data Analysis, 2009, 1(3): 339−372. doi: 10.1142/S1793536909000187 [34] Fang Guohong, Chen Haiying, Wei Zexun, et al. Trends and interannual variability of the South China Sea surface winds, surface height, and surface temperature in the recent decade[J]. Journal of Geophysical Research:Oceans, 2006, 111(C11): C11S16, doi: 10.1029/2005jc003276 [35] Wang Chunzai, Wang Weiqiang, Wang Dongxiao, et al. Interannual variability of the South China Sea associated with El Niño[J]. Journal of Geophysical Research:Oceans, 2006, 111(C3): C03023, doi: 10.1029/2005jc003333 [36] Chow C H, Liu Qinyu. Eddy effects on sea surface temperature and sea surface wind in the continental slope region of the northern South China Sea[J]. Geophysical Research Letters, 2012, 39(2): L02601, doi: 10.1029/2011gl050230 [37] Liu Yingjie, Yu Lisan, Chen Ge. Characterization of sea surface temperature and air‐sea heat flux anomalies associated with mesoscale eddies in the South China Sea[J]. Journal of Geophysical Research:Oceans, 2020, 125(4): e2019JC015470, doi: 10.1029/2019jc015470 [38] Shi Xingjian, Chen Zhourong, Wang Hao, et al. Convolutional LSTM network: a machine learning approach for precipitation nowcasting[C]. In Proceedings of the 28th International Conference on Neural Information Processing Systems. Montreal, MIT Press, 2015: 802−810. -
下载:
计量
- 文章访问数: 113
- HTML全文浏览量: 46
- PDF下载量: 24
- 被引次数: 0
