COMPARING INVERSION AND TRIAL-AND-ERROR METHODS TO DETERMINE OPTIMUM YIELD AT A RIVERBANK FILTRATION SITE, KOREA

Traditionally trial and error has been used for model calibration and determination of optimum groundwater development. However, it is often inconvenient and time-consuming when matching calculated heads to observed heads to obtain reasonable results because hydraulic head is often a nonlinear function to many parameters. As the demands for accuracy become greater, the trial and error method becomes more arduous. More recently, several methods have been developed for automated calibration to allow for efficiency when performing model inversion. In this study, the inversion capability of MODFLOW-2000 is applied to calibrate a groundwater model and determine the optimum development of filtrate at a riverbank filtration site, Daesan-Myeon, Korea. Sensitivity analysis is then used to examine the effect of hydraulic conductivity on optimal decisions for a 5 layer model. The highest composite-scaled sensitivity shows a value of 0.4593 at layer 4 (sand/gravel layer), the main aquifer. A sensitivity of 0.3363 appears at layer 3 (fine sand layer) and a value of 0.3184 at layer 2 (medium sand layer). The results indicate that the relative change of hydraulic conductivity greatly affects the main aquifer whereas the sensitivities of layer 1 (fine sand layer) and layer 5 (weathered zone) are very low. The residual sum of squares by inversion and the trial-and-error method are compared as 0.1272 and 0.1335, respectively. Similar to the head distributions based on the observations, the hydraulic heads by inversion are higher near the Nakdong River and lower farther from the river than those by the trial-and-error method. This indicates that inversion is a better method than trial-and-error for model calibration. Optimal yield using the estimates of hydraulic conductivity determined by inverse modeling is different from those determined by the trial-and-error method. Thus, precise estimates of hydraulic conductivity are very important to find the optimal yield. The authors wish to acknowledge the financial support of the Sustainable Water Resources Research Center under the program of the 21st Century Frontier R&D Program (project no: 3-4-1).