Estimating groundwater recharge using the SMAR conceptual model calibrated by genetic algorithm

摘要

Proper groundwater management of a contaminated aquifer requires the accurate estimation of groundwater recharge, which carries the contaminated load. There are many direct and indirect methods and sophisticated models for estimating recharge. However, most of them require many field data or model parameters, which limit their actual field application. To overcome this limitation, the Soil Moisture Accounting and Routing (SMAR), a conceptual rainfall-runoff model, is employed. The SMAR model has the potential for estimating recharge using only rainfall, evaporation and groundwater level data. However, for an aquifer having prominent horizontal groundwater flow, this model cannot be used directly. For this reason a horizontal flow component is added to this model. Model parameters are calibrated by the Genetic Algorithm (GA) optimization technique. Sensitivity of calibrated parameters to model efficiency and estimated recharge, and parameter interdependence are investigated. This model is applied to 11 observation locations in four catchment areas of Miyakojima Island, Japan, where groundwater nitrate contamination is a threat. The effectiveness of the model is evaluated using the model efficiency (R-2), the mean of the sum of square errors (MSE), plots of observed versus estimated groundwater levels, scatter plots of observed versus estimated groundwater levels, measure of timing of the peaks, and the correlation between monthly rainfall and monthly estimated recharge. All show that this technique is very efficient for estimation of recharge. Model efficiency (R-2) up to 92%, minimum MSE 0.32 m(2)/day, average relative error of timing of the peaks 4.13%, and coefficient of determination (r(2)) up to 0.92 are obtained for the study area. The estimated recharge is 45% of the mean annual rainfall and agrees with other finding. It is thus concluded that the SMAR model could be a viable alternative since it can estimate dependable recharge with a minimum of input data. (c) 2004 Elsevier B.V. All rights reserved.