Model Calibration Based on Random Environmental Fluctuations

Abstract

A simple model‐calibration technique that takes into account the random fluctuations of field measurements and yields probability distribution of the model input parameters is tested. This technique is based on the following concept. A large number of model simulations is conducted using a wide range of model input parameter values. The specific values of each input parameter in a given simulation are randomly selected from specified probability distributions. Only a subset of all simulations will yield outputs that are satisfactorily close to the field observations. The input parameters used in this subset of simulations can then be analyzed and their mean and variance are computed. It is shown that this calibration procedure is a useful tool to determine the mean values of model input parameters that dominate the variance of the model's output. If the variance of the model's output is completely dominated by a single, uncorrelated parameter, the mean and variance of this parameter can be estimated correctly. If several parameters dominate the variance of the model's output or if those parameters are correlated, correct means can still be estimated but the variances obtained for these parameters are larger than their correct values.