Model CalibrationAgency: Environment, Great Lakes, and Energy
Model calibration consists of changing values of model input parameters in an attempt to match field conditions within some acceptable criteria. This requires that field conditions at a site be properly characterized. Lack of proper site characterization may result in a model that is calibrated to a set of conditions which are not representative of actual field conditions. The calibration process typically involves calibrating to steady-state and transient conditions. With steady-state simulations, there are no observed changes in hydraulic head or contaminant concentration with time for the field conditions being modeled. Transient simulations involve the change in hydraulic head or contaminant concentration with time (e.g. aquifer test, an aquifer stressed by a well-field, or a migrating contaminant plume). These simulations are needed to narrow the range of variability in model input data since there are numerous choices of model input data values which may result in similar steady-state simulations. Models may be calibrated without simulating steady-state flow conditions, but not without some difficulty.
Comparison to Field Conditions
Where appropriate, model calibration should include comparisons between model-simulated conditions and field conditions for the following data:
- Hydraulic head data,
- Groundwater-flow direction,
- Hydraulic-head gradient,
- Water mass balance,
- Contaminant concentrations,
- Contaminant migration rates,
- Migration directions, or
- Degradation rates.
These comparisons should be presented in maps, tables, or graphs. Each modeler and model reviewer will need to use their professional judgment in evaluating the calibration results. There are no universally accepted "goodness-of-fit" criteria that apply in all cases. However, it is important that the modeler make every attempt to minimize the difference between model simulations and measured field conditions. Typically, the difference between simulated and actual field conditions (residual) should be less than 10 percent of the variability in the field data across the model domain.
Should all models be calibrated?
There should always be an attempt to calibrate a model, whether it's a numerical or analytical model. Whether a model is used for screening-level purposes or to make decisions concerning possible remedial actions or environmental compliance, there needs to be a demonstration that the model and its parameter values are reasonably representative of site conditions. This is done through the process of model calibration and, if possible, history matching. Without model calibration it is not possible to assess whether predictions made with the model are reasonable.
Of what use is a non-calibrated model?
For initial assessments, it is possible to obtain useful results from models that are not calibrated. The application of uncalibrated models can be very useful in guiding data collection activities for hydrogeological investigations or as a screening tool in evaluating the relative effectiveness of remedial action alternatives.