Sea level prediction using time series obtained from satellite altimetry observations

Abstract

In coastal management and ship navigation activities, there is an increasing demand for accurately predicting sea level fluctuations. In order to achieve this goal, accessible high-quality data and proper modeling process are critically required. The main purpose of the study focuses on developing and validating different modelling approach for analysis and forecasting of Caspian Sea level anomalies based on Topex/Poseidon and Jason-1 altimetry data generally covering 1993-2008-2013, which are originally developed and optimized for open oceans but have the considerable capability to monitor inland water level changes. Since these altimetric measurements comprise of a large datasets and then are complicated to be used for our purposes, for the first stage of the study, Principal Component Analysis (PCA) is adopted to reduce the complexity of large time series data analysis. Furthermore, Autoregressive Integrated Moving Average (ARIMA) model is applied for further analyzing and forecasting the time series. According to our analysis, ARIMA (1,1,0)(0,1,1) model has been found as an optimal representative model capable of predicting pattern of Caspian Sea level anomalies reasonably. Due to presence of temporal and spatial data gaps, least squares polynomial interpolation is thus performed to fill the gaps of along-track sea surface heights used for the next stage of study. The data were then adapted to Holt-Winters exponential smoothing (HWES) for investigating the capability of another linear approach for predicting the Caspian Sea level behavior. Although the modeling results agree well with the observed time series, but due to stochastic and nonlinearity nature of most water resources time series, these methods may not always perform well when applied in modeling hydrological time series. Therefore, in order to provide more applicable modelling approach, different artificial intelligent techniques were used for the short term Caspian Sea level forecasting. The forecast is performed by Multi-layer Perceptron network (MLP), Radial Basis Function (RBF), and Generalized Regression Neural Networks (GRNN), Support Vector Machine (SVM), and Gene Expression Programming (GEP). The overall results show that comparing with a routine Autoregressive Moving Average (ARMA) model, different neural network methodologies perform satisfactorily as a powerful tool in providing reliable results for predicting the short term Caspian Sea level anomalies. While all artificial intelligent approaches showed superior performance compare with conventional linear methods, the inter-comparison analysis verified that SVM has the best performance in predicting Caspian Sea-level anomalies, given the minimum Root Mean Square Error (RMSE=0.035) and maximum coefficient of determination (R2=0.96). The results of the study may lead to a better understanding of applicable tools in forecasting stochastic time series and giving an effective insight for more precise prediction-based decision making in water management scenarios.