Urmia Lake water-level change detection and modeling

Abstract

This study suggests a different view of the change in water level fluctuation of Urmia Lake (UL), Northwestern Iran, and its descriptive statistics in term of the seasonal variation of monthly average time series during 1966–2012. A significant change is demonstrated in the descriptive statistical characteristics, such as the mean and the variance after a change point in 1999 by nonparametric techniques. Therefore, two models namely, mono- and multiple-time trend (Mono- and Multiple-TT), are applied to remove trends observed in the UL water level. In case of Mono-TT model, the trend assessment of water level exhibit downward trend significantly. In case of Multiple-trend model, the time series is divided into before and after change point subseries and a linear time trend model is fitted to each subseries. The results showed a significant increasing and decreasing trend, respectively. Furthermore, this paper have developed hybrid time series models including Seasonal Auto Regressive Integrated Moving Average (SARIMA) and Generalized Autoregressive Conditional Heteroscedasticity (GARCH) fitted to detrended values by Mono- and Multiple-TT models. The criteria evaluation demonstrates the adequacy of SARIMA, Mono- and Multiple-TT-SARIMA models and their combination with GARCH approach for modeling monthly water level which have significant trends. The GARCH models indicate the existence of short-run persistency in the water level. The proposed SARIMA and Mono-TT-SARIMA and their combination with the GARCH models are found useful for modeling the water level with significant trends.