Abstract
The covariance matrix of observation error is very important to estimate the credibility of the updated model. Besides the accuracy of the measurement devices, modeling error is also very importan as a factor of observation error. However, it is very difficult to consider every factor of the modeling error. In order to determine the level of the observation error from observation data statistically, the objective function for the inverse analysis is extended based on the maximum likelihood method. As an example, we discuss the identification problem of dynamic soil properties, in which the estimation of observation error and model selection by using information criteria such as AIC are demonstrated.
