A QUANTITATIVE MEASURE OF MODEL VALIDATION AND ITS POTENTIAL USE FOR REGULATORY PURPOSES

The accuracy of computer model calculations is usually examined via a validation study. The results of such studies are commonly presented by plotting model calculations and measured values on the same axes, and the match between them is usually evaluated by inspection. This is a subjective approach, which is not an appropriate basis for making regulatory decisions concerning public health and safety. It is suggested that linear regression analysis of calculated results and measured data can be used to evaluate the results of a validation in a more objective and quantitative manner. The coefficients of the resulting regression line are interpreted as an empirical indicator of model bias. The confidence interval based on the standard error of the regression is interpreted as a measure of the uncertainty in the model predictions. The merits and limitations of the proposed evaluation scheme are discussed from the point of view of having to assess the reliability of calculations performed by someone else. The simplicity of this approach is both its largest advantage and its greatest limitation. Linear regression of calculated and measured data provides no insights into the cause of model inaccuracies, but the ease and simplicity of obtaining an empirical, quantitative measure of the accuracy of model predictions compensates for the potential for misinterpretation of the results. The evaluation method is demonstrated by a validation of groundwater flow and contaminant transport calculations for a uranium tailings site.