Abstract

Model calibration is usually an important part of the modelling process. A well structured approach of this activity, supported by general guidelines and techniques, will be especially beneficial of practical applications. Particularly the study of the sensitivity and the identifiability of the parameters (e.g. model coeffients, initial conditions) has to be a relevant part of model calibration. Such a study can reveal potential problems already during the early stages of the model calibration process, and can offer useful suggestions to prevent their occurrence. It will also be useful in post-calibration studies, e.g. when analysing the problems of unsuccessful calibration runs. Several simple methods are suggested for performing these analysis for a general class of calibration problems. The advantages and disadvantages of these methodes are briefly discussed. In particular attention is given to the problem of local versus global analyses in the parameter space.