Hydraulic conductivity estimation via the AI-based numerical model optimization using the harmony search algorithm

Abstract

Groundwater as a vital resource for humankind is being debilitated by enormous over-extraction and intensifying contamination. Insightful advancement and protection of this significant resource needs a careful understanding of aquifer parameters. In the present study, the groundwater level was predicted at first, using a hybrid wavelet artificial neural networks and genetic programming (wavelet-ANN-GP) model. The hybrid model results were then evaluated using the performance evaluation criteria including R square, root mean square error (RMSE), mean absolute error and Nash–Sutcliffe efficiency, respectively ranged from 0.81 to 0.97, 0.070 to 4.45, 0.016 to 3.036 and 0.74 to 0.96, which revealed the high applicability of the hybrid model. The groundwater levels were predicted using wavelet-ANN-GP and then entered into the numerical model. Harmony search (HS) was used for the optimization of the numerical model. Hydraulic conductivity (HC) was estimated during the optimization process. Then, the estimated HC was extended throughout the aquifer domain by the empirical Bayesian kriging (EBK) method. Eventually, estimated hydraulic conductivity was compared by defined hydraulic conductivity through the pumping test. The plotted map of the estimated hydraulic conductivity showed about 87.5% conformity to points with distinct hydraulic conductivities obtained from the pumping test. The results proved the applicability of AI-based meta-heuristic optimization models in water resource studies.