Fourteen guidelines for effective model calibration
More sophisticated methods of model calibration and sensitivity analysis are
being developed, but they can be confusing and it is not always clear how to
use them most effectively. While the best approach is problem dependent,
many considerations are common to all problems. The
Methods and Guidelines
Report and, more recently, the textbook by Hill and Tiedeman (2007) suggests fourteen guidelines that are likely to be useful when
calibrating models for groundwater or other problems. Many of the ideas
will be familiar to experiences modelers; indeed, the guidelines can be
thought of as organized common sense with some new statistical twists. Brief
summaries of the guidelines are listed in the following table.
Table 1: Guidelines for effective model calibration.

Apply the principle of parsimony (start very simple; build complexity slowly)
 Use a broad range of information to constrain the problem
 Maintain a wellposed, comprehensive regression problem
 Include many kinds of data as observations in the regression
 Use prior information carefully
 Assign weights which reflect measurement errors
 Encourage convergence by making the model more accurate
 Evaluate model fit
 Evaluate optimized parameters
 Test alternative models
 Evaluate potential new data
 Evaluate the potential for additional estimated parameters
 Use confidence and prediction intervals to indicate parameter and prediction uncertainty
 Formally reconsider model calibration from the perspective of the desired predictions
Methods and Guidelines Report