Application of multi-model ensemble averaging techniques for groundwater simulation: synthetic and real-world case studies

Growing demands in arid regions have increased groundwater vulnerabilities necessitating appropriate modeling and management strategies to understand and sustain aquifer system behaviors. Sustainable management of aquifer systems, however, requires a proper understanding of groundwater dynamics and accurate estimates of recharge rates which often cause error and uncertainty in simulation. This study aims to quantify the uncertainty and error associated with groundwater simulation using various multi-model ensemble averaging (MEA) techniques such as simple model averaging, weighted averaging model, multi-model super ensemble, and modified MMSE. Two numerical solutions, i.e., finite difference and finite element (FE), were first coupled under three schemes such as explicit scheme (ES), implicit scheme, and Crank-Nicolson Scheme to numerically solve groundwater simulation problems across two case studies, synthetic and real-world (the Birjand aquifer in Iran) case studies. The MEA approach was considerably successful in calibrating a complex arid aquifer in a way that honors complex geological heterogeneity and stress configurations. Specifically, the MEA techniques skillfully reduced the error and uncertainties in simulation, particularly those errors associated with water table variability and fluctuation. Furthermore, a coupled FE-ES method outperformed other approaches and generated the best groundwater-level simulation for the synthetic case study, while stand-alone FE was particularly successful for the Birjand aquifer simulation as a real-world case study.

The simulation of groundwater spatiotemporal changes under the uncertainty of hydrogeological conditions

<p>Groundwater is a reliable freshwater resource in many areas, and it is also an important source of backup water during the drought. Therefore, understanding the characteristics of groundwater resources is crucial and can be explored by building correct hydrogeological models for simulation. To build a perfect hydrogeological model, it is necessary to grasp the correct geological conditions and hydrogeological parameters to establish an effective numerical simulation of groundwater flow. However, geological conditions always contain some uncertainties, which may cause a certain impact on the spatiotemporal changes of groundwater.</p><p>Therefore, this study uses the groundwater flow numerical model, MODFLOW, to build the groundwater simulation model. The ideal case and real case at Touqiao Minshiung Industrial Zone in Chiayi is built from 2009 to 2013. The results show that under different hydrogeological parameters, geology, and other conditions, groundwater will have different patterns of variation. The Empirical Orthogonal Function (EOF) method is also used to compare the dominated patterns. The simulation results show the R<sup>2</sup> can all reach 0.9 compare with the groundwater real observation data. This study can further explore the drought-resistant availability of groundwater in various regions under different geological conditions, it will help relevant agencies and local governments to better manage groundwater resources.</p><p>Keywords: groundwater simulation, MODFLOW, uncertainty, hydrogeology, EOF</p><p> </p><p>__________________</p><p>*Department of Civil Engineering, National Central University</p>

A surrogate-based simulation–optimization approach for coastal aquifer management

This study proposed a pumping-injection (P-I) groundwater management strategy based on a simulation–optimization (S-O) framework to mitigate seawater intrusion (SI). The methodology was applied to a real case in Longkou, China. A three-dimensional variable-density groundwater simulation model was established to simulate and predict the SI process. In the S-O framework, while solving the optimization model, it is required to call the simulation model thousands of times, which leads to enormous computational load. In this case, the Kriging and support vector regression (SVR) surrogate models were established for the simulation model respectively. Furthermore, the ensemble surrogate modeling technique was applied to construct the Kriging-SVR ensemble surrogate model. The most accurate surrogate model was selected as the substitute for the simulation model, saving considerable computing costs. The results show that the ensemble surrogate model performs better than the stand-alone surrogate models in accuracy, indicating that combining stand-alone surrogate models is a potential modeling method for the surrogate model of the variable-density groundwater simulation model. By solving the optimization model, the optimal pumping and injection schemes under different scenarios were obtained. The optimization results demonstrate that the proposed methodology is effective and stable in coastal groundwater management.

Automatic calibration of the groundwater simulation model with high parameter dimensionality using sequential uncertainty fitting approach

Calibration of the groundwater simulation model is one of the main challenges in the modeling process. In addition, hydrogeological complexities and the lack of field data in terms of time and space lead to uncertainty in the model. Therefore, the present study linked the groundwater simulation model (MODFLOW) and sequential uncertainty fitting approach (SUFI-2) to the uncertainty-based automatic calibration of the Ardabil groundwater model located in northwestern Iran. Hydraulic conductivity, specific yield, recharge rate, the hydraulic conductivity of the riverbed material, and the boundary conductance of the aquifer were considered as the uncertain parameters. Furthermore, the Newton solution method for the unconfined aquifer was used for solving the groundwater flow equation. A Normalized Total Uncertainty Index was defined to evaluate the performance of the SUFI-2 algorithm. According to the MODFLOW-SUFI-2 calibration findings, 60% of observational data was bracketed by a 95% confidence interval, on average. The Ardabil groundwater model was also calibrated with the PSO algorithm. In comparison with SUFI-2, although this method resulted in good coverage of the solution, it obtained irrational values for most parameters since they only aimed to match observational and computational values. Eventually, SUFI-2 showed a small number of simulation runs compared with the PSO algorithm.