A linear/non-linear hybrid time-series model to investigate the depletion of inland water bodies

摘要

Changing climate and human interference with natural phenomena are causing unprecedented changing patterns in hydro-climatic variables. These changes can manifest as dynamic changes of the stochastic properties of the datasets over time, which pose challenges for conventional time-series modeling. These datasets are dynamic in nature, even when trend and seasonality components are eliminated. Shrinking lakes are among the most notable examples of hydro-climatic-driven phenomena. This study demonstrates a framework that can capture the underlying dynamic and non-stationary structure of such environments using a case study of Maharlou Lake, Iran. To that end, a hybrid time-series model was developed to account for volatility in the data [i.e., SARIMA (1,1,2) x (1,1,2)(12)/GARCH(1,0)]. A series of statistical tests (i.e., augmented Dickey-Fuller test, the Ljung-Box test, the heteroskedasticity test, and the two-sample Kolmogorov-Smirnov test) were used to create, calibrate, and assess the model in the 95% confidence interval. The results indicate the decline and depletion of the lake. This reduction manifests as a general downward trend and a widening gap between the lake's intra-annual fluctuations. The changes could be an alarming signal, as this saline lake could be mimicking the tragic fate of similar inland water bodies like Lake Urmia or the Aral Sea.