Predicting Caspian Sea Surface Water Level by ANN and ARIMA Models

Abstract

Fluctuations of the Caspian Sea's mean monthly surface water level for the period of January 1986 to December 1993 were studied. The time series data showed an increasing trend and seasonal variations. Artificial neural network (ANN) and multiplicative autoregressive integrated moving average (ARIMA) modeling were used to predict the time series data. The ANN's input and output consisted of the last 12 months and the current month surface water levels, respectively. The selected ARIMA model required one-month regular differencing, 12-month seasonal differencing, and had a moving average component of lag 12. The ANN and ARIMA predictions for the period of January to December 1993 were very reasonable when compared with the recorded levels. On average, the ANN model underestimated the sea level by three cm, whereas the ARIMA model overestimated it by three cm. The monthly predictions for January to December 1994 presented a continuation of the Caspian Sea water surface level rise that would have various adverse effects or its neighboring countries.