Assessment of parameter uncertainty of MODFLOW model using GLUE method (Case study: Birjand plain)

Background and objectives: Groundwater modeling often associates with uncertainties caused by incomplete knowledge of the underlying system or uncertainties due to natural variability in system processes and field conditions. Uncertainty in groundwater modeling has been evaluated by researchers in three main sources that can be classified as parameter uncertainty, conceptual uncertainty (model structure uncertainty) and input uncertainty (observation uncertainty). So far, there are few studies that they assess groundwater uncertainty in the country, and quantifying uncertainty has been limited to the statistical methods. Due to the importance of the water resources in the country and the necessity of estimating the uncertainty in order to achieve accurate and reliable results, in this study, parameter uncertainty of an arid region’s groundwater flow model was assessed by using a Monte Carlo-based simulation technique.
Materials and methods: First, conceptual groundwater model of Birjand plain, located in the southern province, was developed based on collecting all available data, including topography, observed and withdrawal wells information, recharge information, hydrodynamic properties of aquifer, surface elevation data. Then, the MATLAB-based MODFLOW model was used to simulate the groundwater flow. After initial calibration in steady state, for assessing parameter uncertainty in transient mode two scenarios were defined. In the first scenario uncertainty analysis was performed by assuming that the hydraulic conductivity is one of the major contributors to the model uncertainty. So the aquifer was divided into 17 homogeneous zones according to initial calibration of hydraulic conductity results, and the parameter uncertainty was assessed using Monte Carlo (MC) sampling technique, namely, the generalized likelihood uncertainty estimation (GLUE). In the second scenario, 9 recharge zones were additionally considered as the second parameters, and their influence on the hydraulic conductivity and the total uncertainty were estimated by the GLUE.
Results: Posterior parameter plots of hydraulic conductivity in the 17 homogeneous regions and recharge in the 9 inflow pathways and also, 95% confidence intervals for the simulated water table depth, were obtained as main results. The Indices, as criteria for the comparison, were used to quantify the goodness of uncertainty performance and the sensitive regions in the aquifer were specified by implementing global sensitivity analysis of the model.
Conclusion: Results indicate up to 86% of observed data bounded in the 95% confidence intervals that is emphatic the good performance of the GLUE and also the likelihood function, Weighted Root Mean Squared Error (WRMSE), in the assessment of parameter uncertainty in a groundwater simulation model.