Groundwater estimation of Ghayen plain with regression‑based and hybrid time series models

چکیده مقاله

Groundwater estimation to be aware of groundwater level and its decline, as well as the well discharge rate in different time periods, is one of the important and practical issues in the field of groundwater level management and abstraction management. Quantitative information on groundwater resources is essential to manage those best and to propose solutions and constraints on aquifer abstraction. Accurate modeling and prediction of this information also helps avoid financial losses and contributes to the protection of natural resources. Regarding the importance of knowing the groundwater level in the region in future periods, in the present study, the time series of groundwater levels in seven wells in the Ghayen plain on a monthly scale in the statistical period 1997–2018 was used. Contemporaneous Autoregressive Moving Average (CARMA) multivariate and time series models, integrated time series model with isotropy (CARMA-ARCH), multivariate regression and support vector regression were used aiming for the aforementioned. Ant colony optimization algorithms, ant-trap optimization strategies, salp swarm methods, dragonfly algorithms and multiverse algorithms were investigated in order to optimize the parameters of the support vector regression model. The results of simulations showed that in terms of the involvement of different optimization algorithms in SVR-based simulations, the accuracy of groundwater estimation in all wells is appropriate and acceptable. The lowest error rate in groundwater estimation of Shir Morgh Road, Pahnai, Firoozabad Road and Sineh-Kooh-Aboozar wells in terms of support vector regression method optimized with dragonfly algorithm–support vector regression, ant colony optimization–support vector regression, antlion optimizer–support vector regression, antlion optimizer–support vector regression algorithms was estimated at 0.192, 0.46, 0.091 and 0.63 m, respectively. Moreover, multivariate regression method in Birjand Ghayen Road, Northern Kalateh Khan and Northern Esfashad wells obtained the lowest error of 0.412, 0.163 and 0.238, respectively. The best efficiency extracted from the Nash–Sutcliffe coefficient in most wells is also related to CARMA method, whose efficiency improvement was over 97%. Since this study aims to provide an optimal method in groundwater quantity studies, the results indicate the proper performance of optimized and hybrid models that include different family models.