集成奇异谱分析和长短期记忆网络的区域海平面变化预测

Regional Sea Level Change Prediction with Integration of Singular Spectrum Analysis and Long-short-term Memory Network

摘要:

基于我国首套高精度全球海洋气候数据集（CDRs），选取季节变化较为明显的黄海作为研究区域，利用奇异谱分析（SSA）对黄海海域海平面异常（SLAs）数据进行了时间序列与时空序列的分解去噪，并结合长短期记忆（LSTM）网络构建了SSA-LSTM组合模型，对黄海海域海平面变化趋势进行预测。结果表明：SSA-LSTM组合模型对时间序列的预测精度显著提高，预测长度5年的均方根误差最小为35.04 mm；在对时空序列的预测中，预测第1年的均方根误差最小为19.68 mm。同时，利用空间模态进行了海平面变化时空分异规律研究，发现黄海海域海平面变化趋势具有高度一致性，并且与季节、纬度显著相关。预计2016年―2025年黄海海域海平面将以每年3.65±0.79 mm的速率持续上升。

关键词:海平面异常（SLA）;时空序列;奇异谱分析（SSA）;长短期记忆（LSTM）网络;黄海海域

Abstract:

In this paper， the China’s first global ocean climate data records （CDRs） are used to analyze and predict the sea level changes in the Yellow Sea with obvious seasonal changes. Based on the singular spectrum analysis （SSA）， the time and spatio-temporal series of sea level anomalies （SLAs） in the Yellow Sea are decomposed and de-noised. Then the SSA-long short-term memory （LSTM） network （SSA-LSTM combined model） is established to predict the sea level trends of the Yellow Sea. Compared with the traditional methods， the prediction accuracy of the SSA-LSTM combined model is significantly improved with 35.04 mm of the minimum root-mean-square error for the SLAs time series prediction length of 5 years. For the first-year prediction of spatial-temporal series of SLAs， the minimum root-mean-square error is only 19.68 mm. The law of spatial-temporal differentiation of the sea level change in the Yellow Sea is also analyzed by the spatial modes. It is found that the sea level trend of the Yellow Sea is highly consistent and significantly related to the season and latitude. According to the SSA-LSTM combined model， the sea level rise rate of the Yellow Sea will remain at 3.65±0.79 mm per year from 2016 to 2025.

Key words:sea level anomaly (SLA);spatio-temporal series;singular spectrum analysis (SSA);long short-term memory (LSTM) network;the Yellow Sea