Multi-scale prediction of regional sea level variations based on EEMD-BP combined model

Sea level variability has the characteristics of non-linear, non-stationary and multi-time scale, and the traditional prediction method based on statistical model has some limitations, thence the prediction results are not ideal. Based on the ensemble empirical mode decomposition (EEMD) and back propagation (BP) neural network, the paper proposes an improved sea level multi-scale prediction method, namely, EEMD-BP combined model. Firstly, the multi-scale frequency oscillatory modes (intrinsic mode functions, IMFs) representing different oceanic processes are extracted by EEMD from the highest frequency to the lowest frequency oscillating mode. The remaining non-oscillating mode is the residual, or the sea level trend. Secondly, BP neural network is used to establish prediction models for different scale IMF to analyze their future trends, and each IMF is used as an input factor of the BP neural network separately. Finally, the prediction results of each IMF with BP neural network are reconstructed to obtain the final sea level prediction results. The results showed that EEMD is particularly suitable for analyzing non-linear and non-stationary time series, and BP neural network is applicable for regional sea level prediction at different scales. Comparing with the prediction by BP neural network directly (R = 0.74, RMSE = 37.51 mm, MAE = 48.02 mm), the EEMD-BP combined method improved prediction accuracy significantly (R = 0.88, RMSE = 29.23 mm, MAE = 37.50 mm). The results showed that EEMD-BP combined model offers a new method for regional sea level change prediction and is especially suitable for the prediction of non-linear time series. © 2019, Editorial Board of Journal of Systems Engineering Society of China. All right reserved.