Dominant parameter selection in the marginally identifiable case Original Research Article

Often a rather limited set of experimental data is available for the identification of a dynamic model, which contains many parameters. This is, e.g. the usual case for crop growth models. In this situation, only some parameter values can be estimated. Based on an analysis of the Fisher information matrix, a method for a reasonable selection of parameters is suggested here. The method chooses the most sensitive parameters, i.e. those to which the model under the considered experimental conditions is most sensitive, and excludes both coupled parameters and those that exhibit multiplecorrelation. A comparison with different ridge regression methods is made. The methodology is illustrated with a simple lettuce growth model.