Regressive-stochastic models for predicting water level in Lake Urmia

Abstract

This study develops a set of models to investigate the water budget of Lake Urmia in Iran, a permanent hypersaline lake that has suffered a declining water level since the late 1990s. The models are of the regressive-stochastic type, a combination of multilinear regression and autoregressive integrated moving average stochastic models. The multilinear regression models were used to construct the core of the relationship of lake water level to streamflow, precipitation, evaporation and groundwater depth. Afterward, stochastic models were used to generate data for each independent variable to estimate the oscillation in the lake water depth. Several criteria were used to compare the performance of the models in the aggregated and disaggregated cases with which the pre- and post-encroachment periods are considered, respectively. The regressive-stochastic models are found to be competitive with the existing models developed so far for Lake Urmia water level.