HGS RESEARCH HIGHLIGHT – An adaptive zone-based refinement method for characterizing a highly complex aquifer system model
Hwang, H.-T., Jeen, S.-W., Lee, S.-S., Ha, S.-W., Berg, S. J., Miller, K. L., Frey, S. K., Gharedaghloo, B., Merrick, D., Sudicky, E. A., & Lee, K.-K. (2021). An adaptive zone-based refinement method for characterizing a highly complex aquifer system model. In Journal of Hydrology (Vol. 603, p. 126961). Elsevier BV. https://doi.org/10.1016/j.jhydrol.2021.126961
CLICK HERE TO READ THE ARTICLE.
This new paper by Aquanty senior scientist Hyoun-Tae Hwang introduces an innovative new method to iteratively refine model meshes based on model sensitivity and uncertainty, as calculated by PEST. The paper presents an initial proof-of-concept for this new method, based on the K-COSEM test site located in Eumseong-gun, South Korea. This is a highly studied and data-rich site, making it perfect for the application of this new adaptive zone-based mesh refinement method.
In simple terms, this new method involves the construction of a parameter-zone based HydroGeoSphere model. The model is fed into PEST to 1) calibrate the model and 2) obtain parameter sensitivities. Based on the resulting parameter sensitivities a refinement indicator is calculated for each parameter zone to determine whether should be partitioned (i.e. refined). Through iterative zone refinement, the model’s ability to reproduce field observations is improved with the targeted refinement of zones with the highest sensitivities.
The overall performance/parameterization of the model (and associated uncertainty) were compared to the results of cross-well hydraulic tests, electric resistivity tomography and a full model calibration approach (where all parameter zones were refined). The resulting improvement in model calibration and reduction in parameter sensitivities were positive in comparison to these other methods, indicating that this adaptive zone-based refinement method is a useful approach for the construction of highly complex real-world models.