Sea Level Prediction in the North-Western Black Sea Using Autoregressive Integrated Moving Average and Machine Learning Models

Abstract

Data prediction models are essential for estimating extreme environmental changes and predicting anomalies by learning when the actual data is outside previously accepted values. This paper focuses on predicting two years of sea level in the North-Western Black Sea region. Data from the UNESCO/ IOC tsunami observation and Permanent Service for Mean Sea Level archives were analysed using Auto Regression - and Seasonal-Regression Integrated Moving Average models. This work proposes one such model obtained by using modern Machine Learning algorithms, and the results are compared with standard models such as ARIMA obtained for the same data. Using Machine Learning can produce software models ready to run with hardware using much lower specifications than those used for model training which is not the case for standard statistical models. The merged dataset in the analysed period (2006-2016) from the tide gauges along the Romanian Black Sea Coast is consistent and satisfactorily used to develop and validate a Seasonal Regression Integrated Moving Average and Machine Learning model for sea-level forecasts. The data show that the sea level evolution in cyclical changes of the other parameters that influence it. Furthermore, slight demarcation of the two models was observed between the comparison of observed and predicted values