Models are approximations of reality. Groundwater flow models in particular are abstractions of an unseen subsurface. While efforts are underway to reduce uncertainty by imaging the subsurface in portions of Illinois, we will never have a perfect understanding of how groundwater moves beneath our feet.
Fortunately, for most applications, we do not need a perfect realization or a perfect model. Rather, ISWS scientists can run hundreds or thousands of models to evaluate uncertainty in model parameter values and conceptualization of a system. The following questions will guide you through the initialization of your own evaluation model.
Are you modeling in an area with unreported withdrawals?
Start the model with your best guess of pumping and allow that to be varied within a realistic range.
Do you have a model with open boreholes creating a series of holes in an aquitard?
Allow the vertical hydraulic conductivity (kv) in that aquitard to vary over a series of simulations.
Are you uncertain of the impact of a fault system on groundwater flow dynamics?
Create two conceptualizations, one that promotes flow and the other that restricts it and run a series of models with different parameterizations of those simulations.
PEST++ Iterative Ensemble Smoother
PEST is the most commonly used uncertainty evaluation and parameter estimation tool applied to groundwater flow models (and several other models as well). Combining modern PESTPP-IES infrastructure with high performance computing, we can simulate thousands of simulations within a reasonable range of parameters. These automated approaches allow ISWS scientists to explore both “best fit” and “reasonable” solutions to provide better evaluations of groundwater conditions, and lay the potential to create formalized risk evaluations with simulated probabilities of risk occurring.
Genetic Algorithm
The ISWS has customized machine learning, and specifically genetic algorithm, approaches to also evaluate uncertainty in multiple applications, including contaminant transport and evaluation of soil moisture. The flexibility of this code allowed ISWS scientists to chain models together, allowing for the testing of seasonal variability in parameters that are typically assumed to be constant. More details on this study can be found here.
Model simulations using (a) constant and (b) seasonal hydraulic parameter (SHP) architectures.
Currently, scientists are able to use the flexible genetic algorithm to test the “best fit” statistic, which can greatly influence the parameter estimation and uncertainty evaluation results.