A new hybrid framework based on integration of optimization algorithms and numerical method for estimating monthly groundwater level

Abstract

Prediction of groundwater level (GWL) is an important issue for optimal planning and management of groundwater resources. MODFLOW, which is a modular, 3D, finite-difference model, is widely used to simulate GWL. Although MODFLOW is a powerful model for estimating GWLs, it has some unknown parameters, such as specific yield (Sy) and hydraulic conductivity (K). The Dezful-Andimeshk plain, located in the southwest of Iran, was considered as a case study, and its monthly GWLs were simulated. Aquifer boundaries (inflow, outflow, and no-flow), piezometers, operational wells, recharge, drainage, rivers, and evaporation were deemed as the inputs for MODFLOW. Three different algorithms (i.e., shark smell optimization (SSO), particle swarm optimization (PSO), and firefly (FF)) were integrated with MODFLOW (called MODFLOW-SSO, MODFLOW-PSO, and MODFLOW-FF) to find the most accurate values of K, Sy, and GWL of the aquifer. Results revealed that MODFLOW-SSO decreased the root mean square error (RMSE) by 70.2%, 74.9%, and 84.7% at calibration stage and by 68.2%, 73.4%, and 84.2% at validation stage, compared to MODFLOW-PSO, MODFLOW-FF, and MODFLOW, respectively. Other statistical indexes of MODFLOW-SSO (e.g., coefficient of determination (R2) and RMSE) were satisfactory as well. The generalized likelihood uncertainty estimation (GLUE) was applied to calculate the uncertainty of the models with respect to Sy and K. Results indicated that MODFLOW-SSO had the least uncertainty compared to all other models. In general, performance of the MODFLOW was greatly improved by integrating with the SSO algorithm.