基于PredRNN++模型对南海中尺度涡旋的预测研究

doi:10.11978/2023060

摘要/Abstract

摘要：

基于26年的海表面高度异常、海表面风速异常、海表面温度异常资料, 利用时空序列预测模型PredRNN++, 本文预报 1~28d时效的南海中尺度涡旋轨迹和南海西部偶极子活动。结果表明, PredRNN++模型能从整体上考虑整个南海区域时空演变特征和环境风场、温度场的作用, 在短期(1~2周)、中期(3~4周)预报上具有良好的性能。该模型具备一定预报涡旋产生、消亡的能力, 且能将涡旋轨迹4周预报误差控制在42.1km, 对于生命时长小于100d的涡旋生命中期的位置、振幅预报误差小。此外模型在8—11月份的月平均、4天平均下的任意时间点和任意预报时效下均能较好地追踪到偶极子结构的演变、强度变化, 偶极子涡旋相关属性预报误差最小且存在年际、类型差异, 2017年涡旋1~4周振幅位置、预报、半径误差最小, 分别为40~60km、3~5cm、20~40km, 且气旋涡位置预报效果优于反气旋涡。

关键词: 中尺度涡旋, 越南偶极子, 海洋预报, 深度学习

Abstract:

Based on 26 years of data on sea level anomalies, sea surface wind speed anomalies, and sea surface temperature anomalies, using the spatiotemporal series prediction model PredRNN++, this paper predicts the trajectory of mesoscale eddies in the South China Sea and dipole activity in the western South China Sea over a period of 1 to 28 days. The results indicate that the PredRNN++model can comprehensively consider the spatiotemporal evolution characteristics of the entire South China Sea region and the role of environmental wind and temperature fields, and has good performance in short-term (1~2weeks) and medium-term (3~4weeks) forecasting. This model has the ability to predict the generation and disappearance of eddies to a certain extent, and can control the 4-cycle prediction error of eddy trajectories to 42.1 km. For eddies with a lifespan of less than 100 days, the mid-term position and amplitude prediction error are small. In addition, the model can better track the evolution and intensity change of dipole structure at any time point under the monthly average, 4-day average and any forecast time effect in August-November. The prediction error of dipole eddy related attributes is the smallest and there are interannual and type differences. In 2017, the amplitude position, prediction and radius error of eddy 1-4 cycles are the smallest, which are 40~60 km, 3~5 cm and 20~40 km respectively, and the prediction effect of cyclone position is better than that of anticyclone.

Key words: mesoscale eddies, dipole off eastern Vietnam, ocean forecast, deep learning

赵杰, 林延奖, 刘燃, 杜榕. 基于PredRNN++模型对南海中尺度涡旋的预测研究[J]. 热带海洋学报, 2024, 43(1): 16-27.

ZHAO Jie, LIN Yanjiang, LIU Ran, DU Rong. Prediction of mesoscale eddies in the South China Sea based on the PredRNN++ model[J]. Journal of Tropical Oceanography, 2024, 43(1): 16-27.

图/表 10

图1

图2

图3

表1

表2

图4

图5

图6

图7

图8

参考文献 35

[1]	韩玉康, 周林, 吴炎成, 2016. 基于HYCOM的南海中尺度涡数值模拟[J]. 海洋通报, 35(3): 299-316.
	HAN YUKANG, ZHOU LIN, WU YANCHENG, 2016. Numerical simulation of the mesoscale eddy in the South China Sea based on HYCOM[J]. Marine Science Bulletin, 35(3): 299-316 (in Chinese with English abstract).
[2]	李佳讯, 张韧, 陈奕德, 等, 2011. 海洋中尺度涡建模及其在水声传播影响研究中的应用[J]. 海洋通报, 30(1): 37-46.
	LI JIAXUN, ZHANG REN, CHEN YIDE, et al, 2011. Ocean mesoscale eddy modeling and its application in studying the effect on underwater acoustic propagation[J]. Marine Science Bulletin, 30(1): 37-46 (in Chinese with English abstract).
[3]	CHELTON D B, SCHLAX M G, SAMELSON R M, 2011. Global observations of nonlinear mesoscale eddies[J]. Progress in Oceanography, 91(2): 167-216. doi: 10.1016/j.pocean.2011.01.002
[4]	CHEN GENGXIN, HOU YIJUN, CHU XIAOQING, 2011. Mesoscale eddies in the South China Sea: mean properties, spatiotemporal variability, and impact on thermohaline structure[J]. Journal of Geophysical Research, 116(C6): C06018.
[5]	CHEN GENGXIN, HOU YIJUN, ZHANG QILONG, et al, 2010. The eddy pair off eastern Vietnam: interannual variability and impact on thermohaline structure[J]. Continental Shelf Research, 30(7): 715-723. doi: 10.1016/j.csr.2009.11.013
[6]	CHOW C H, LIU QINYU, 2012. 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, 39(2): L02601.
[7]	DAS M, GHOSH S K, 2020. Data-Driven approaches for spatio-temporal analysis: a survey of the state-of-the-arts[J]. Journal of Computer Science and Technology, 35(3): 665-696. doi: 10.1007/s11390-020-9349-0
[8]	DIPPNER J W, NGUYEN K V, HEIN H, et al, 2007. Monsoon-induced upwelling off the Vietnamese coast[J]. Ocean Dynamics, 57(1): 46-62. doi: 10.1007/s10236-006-0091-0
[9]	DONG CHANGMING, MCWILLIAMS J C, LIU YU, et al, 2014. Global heat and salt transports by eddy movement[J]. Nature Communications, 5(1): 3294. doi: 10.1038/ncomms4294
[10]	KUO N-J, ZHENG QUANNAN, HO C-R, 2000. Satellite observation of upwelling along the Western Coast of the South China Sea[J]. Remote Sensing of Environment, 74(3): 463-470. doi: 10.1016/S0034-4257(00)00138-3
[11]	LI JIAXUN, WANG GUIHUA, XUE HUIJIE, et al, 2019. A simple predictive model for the eddy propagation trajectory in the northern South China Sea[J]. Ocean Science, 15(2): 401-412. doi: 10.5194/os-15-401-2019
[12]	LI JIAXUN, WANG GUIHUA, ZHAI XIAOMING, 2017. Observed cold filaments associated with mesoscale eddies in the South China Sea[J]. Journal of Geophysical Research: Oceans, 122(1): 762-770. doi: 10.1002/jgrc.v122.1
[13]	LI YUANLONG, HAN WEIQING, WILKIN J L, et al, 2014. Interannual variability of the surface summertime eastward jet in the South China Sea[J]. Journal of Geophysical Research: Oceans, 119(10): 7205-7228. doi: 10.1002/jgrc.v119.10
[14]	LOTTER W, KREIMAN G, COX D, 2017. Deep predictive coding networks for video prediction and unsupervised learning[J/OL]. https://doi.org/10.48550/arXiv.1605.08104.
[15]	MA CHUNYONG, LI SIQING, WANG ANNI, et al, 2019. Altimeter observation-based eddy nowcasting using an improved Conv-LSTM network[J]. Remote Sensing, 11(7): 783. doi: 10.3390/rs11070783
[16]	MARTÍNEZ-MORENO J, HOGG A M, ENGLAND M H, et al, 2021. Global changes in oceanic mesoscale currents over the satellite altimetry record[J]. Nature Climate Change, 11(5): 397-403. doi: 10.1038/s41558-021-01006-9
[17]	QIU CHUNHUA, YI ZHENHUI, SU DANYI, et al, 2022. Cross-Slope heat and salt transport induced by slope intrusion eddy’s horizontal asymmetry in the Northern South China Sea[J]. Journal of Geophysical Research: Oceans, 127(9): e2022JC018406.
[18]	REICHSTEIN M, CAMPS-VALLS G, STEVENS B, et al, 2019. Deep learning and process understanding for data-driven Earth system science[J]. Nature, 566(7743): 195-204. doi: 10.1038/s41586-019-0912-1
[19]	ROBINSON A R, CARTON J A, MOOERS C N K, et al, 1984. A real-time dynamical forecast of ocean synoptic/mesoscale eddies[J]. Nature, 309(5971): 781-783. doi: 10.1038/309781a0
[20]	SHAW P-T, 1999. The seasonal variation of the intrusion of the Philippine Sea water into the South China Sea[J]. Journal of Geophysical Research: Oceans, 96(C1): 821-827.
[21]	SHI XIBNGJIAN, GAO ZHIHAN, LAUSEN L, et al, 2017. Deep learning for precipitation nowcasting: a benchmark and a new model[J]: 5622-5632. https://arxiv.org/abs/1706.03458
[22]	SHI XINGJIAN, CHEN ZHUORONG, WANG HAO, et al, 2015. Convolutional LSTM Network: a machine learning approach for precipitation nowcasting[J/OL]. https://arxiv.org/abs/1506.04214v2
[23]	THOPPIL P G, FROLOV S, ROWLEY C D, et al, 2021. Ensemble forecasting greatly expands the prediction horizon for ocean mesoscale variability[J]. Communications Earth & Environment, 2(1): 89.
[24]	WANG GUIHUA, CHEN DAKE, SU JILIAN, 2006. Generation and life cycle of the dipole in the South China Sea summer circulation[J]. Journal of Geophysical Research, 111(C6): C06002.
[25]	WANG LIPING, KOBLINSKY C J, HOWDEN S, 2000. Mesoscale variability in the South China Sea from the TOPEX/Poseidon altimetry data[J]. Deep Sea Research Part I: Oceanographic Research Papers, 47(4): 681-708. doi: 10.1016/S0967-0637(99)00068-0
[26]	WANG XIN, WANG HUIZAN, LIU DONGHAN, et al, 2020. The prediction of oceanic mesoscale eddy properties and propagation trajectories based on machine learning[J]. Water, 12(9): 2521. doi: 10.3390/w12092521
[27]	WANG YUNBO, GAO ZHIFENG, LONG MINGSHENG, et al, 2018. PredRNN++: towards a resolution of the Deep-in-Time dlemma in spatiotemporal predictive learning[J/OL]. https://arxiv.org/abs/1804.06300
[28]	WANG YUNBO, LONG MINGSHENG, WANG JIANMIN, et al, 2017. PredRNN: Recurrent Neural Networks for predictive learning using spatiotemporal LSTMs[C]// GUYON I, LUXBURG U V, BENGIO S, et al. Advances in Neural Information Processing Systems 30. Long Beach: Curran Associates, Inc.:879-888.
[29]	XIA QIONG, HE YIJUN, DONG CHANGMING, et al, 2018. Prediction of the South China Sea Dipole using SSA-MEM[J]. Atmosphere-Ocean, 56(4): 240-253. doi: 10.1080/07055900.2017.1406331
[30]	XIAO CHANGJIANG, CHEN NENGCHENG, HU CHULI, et al, 2019a. A spatiotemporal deep learning model for sea surface temperature field prediction using time-series satellite data[J]. Environmental Modelling & Software, 120: 104502.
[31]	XIAO CHANGJIANG, CHEN NENGCHENG, HU CHULI, et al, 2019b. Short and mid-term sea surface temperature prediction using time-series satellite data and LSTM-AdaBoost combination approach[J]. Remote Sensing of Environment, 233: 111358. doi: 10.1016/j.rse.2019.111358
[32]	ZENG XIANMING, LI YIZHEN, HE RUOYING, 2015. Predictability of the Loop Current Variation and Eddy Shedding Process in the Gulf of Mexico using an Artificial Neural Network approach[J]. Journal of Atmospheric and Oceanic Technology, 32(5): 1098-1111. doi: 10.1175/JTECH-D-14-00176.1
[33]	ZHANG QIN, WANG HUI, DONG JUNYU, et al, 2017. Prediction of sea surface temperature using long short-term memory[J]. IEEE Geoscience and Remote Sensing Letters, 14(10): 1745-1749. doi: 10.1109/LGRS.2017.2733548
[34]	ZHANG ZHENGGUANG, QIU BO, 2020. Surface chlorophyll enhancement in mesoscale eddies by submesoscale spiral bands[J]. Geophysical Research Letters, 47(14): e2020GL088820.
[35]	ZHANG ZHENGGUANG, WANG WEI, QIU BO, 2014. Oceanic mass transport by mesoscale eddies[J]. Science, 345(6194): 322-324. doi: 10.1126/science.1252418 pmid: 25035491

模型名称	1周预报误差 (经向/纬向预报误差)/km	2周预报误差 (经向/纬向预报误差)/km	3周预报误差 (经向/纬向预报误差)/km	4周预报误差 (经向/纬向预报误差)/km
回归模型 (Li et al, 2019)	— (32.7/29.5)	— (55.1 /47.3)	— (72.5 /61.4)	— (89.2 /73.5)
LSTM	36.5 (23.4/28.0)	51.8 (34.8/38.4)	68.7 (47.3/49.8)	84.8 (58.5/61.4)
Gradient boosting	31.2 (22.5/21.6)	59.1 (41.6/41.7)	83.6 (57.8/60.1)	104.3 (72.2/74.9)
随机森林	51.3 (40.2/31.6)	68.1 (51.8/44.0)	86.9 (64.4/58.1)	105.1 (76.6/71.5)
Extra tree	47.2 (38.0/27.8)	69.1 (53.3/43.6)	89.9 (67.0/59.6)	108.8 (79.8/73.4)
Extra tree (Wang et al, 2020)	— (28.8/23.8)	— (36.9/30.6)	— (41.9/34.1)	— (47.2/37.2)
PredRNN++	28.1 (19.7/20.0)	36.4 (25.6/25.8)	40.8 (28.8/28.7)	42.1 (29.3/30.0)

预报时效	观测涡旋轨迹追踪到个数	涡旋轨迹第一天被追踪到的涡旋轨迹个数	涡旋轨迹最后一天被追踪到的涡旋轨迹个数
1	549	297	381
2	549	239	388
3	549	211	367
4	549	181	357
5	549	165	338
6	548	153	316
7	548	129	319
14	528	71	192
21	482	66	110
28	426	47	82