A stochastic approach to model validation

Abstract This paper describes a stochastic approach for assessing the validity of environmental models. In order to illustrate basic concepts we focus on the problem of modeling moisture movement through an unsaturated porous medium. We assume that the modeling objective is to predict the mean distribution of moisture content over time and space. The mean moisture content describes the large-scale flow behavior of most interest in many practical applications. The model validation process attempts to determine whether the model's predictions are acceptably close to the mean. This can be accomplished by comparing small-scale measurements of moisture content to the model's predictions. Differences between these two quantities can be attributed to three distinct ‘error sources’: (1) measurement error, (2) spatial heterogeneity, and (3) model error. If we adopt appropriate stochastic descriptions for the first two sources of error we can view model validation as a hypothesis testing problem where the null hypothesis states that model error is negligible. We illustrate this concept by comparing the predictions of a simple two-dimensional deterministic model to measurements collected during a field experiment carried out near Las Cruces, New Mexico. Preliminary results from this field test indicate that a stochastic approach to validation can identify model deficiencies and provide objective standards for model performance.