Modeling the volatility changes in Lake Urmia water level time series

Abstract

The decline in Lake Urmia (LU) water level during the past two decades has been addressed by several studies. However, the conducted studies could not come across a practical solution by considering the sample mean in the lake water level time series. For this, the present study suggests a fresh look to the lake water level decline in LU by addressing the volatility changes instead. The Bayesian change-point detection method was used to define the major and critical change points during the study period from January 1966 to December 2016 on a daily scale. Results indicated that major changes occurred in early 2000, and the time series can be studied under the pre- and post-change point events. Afterward, several methods namely shift-track and mono- and multiple-trend line analyses were used to remove the trends associated with the lake water level time series. The de-trending approaches later were applied separately for the entire study period, before 2000 (i.e., 1966–1999) and afterward (i.e., 2000–2016). Then, the de-trended time series were used, and a generalized autoregressive conditional heteroscedasticity (GARCH) model was fitted to the de-trended time series to predict the volatility changes in the data run. Results indicated to descending and ascending changes, respectively, in short- and long-term persistence after 2000. The GARCH(1,1) model was found to be satisfactory to interpret the pre- and post-turn point events, while the changes in short- and the long-term persistence were calculated as 0.53 to 0.75 and 0.46 to 0.24, respectively. In addition, by considering the lake water level anomaly and coefficient of variation in LU and two neighboring cases of Lake Sevan and Lake Van, it is concluded that the changes are exclusive to LU, and the rate of changes was accelerated after 2006.