Estimation of partitioned set of parameters

In the study of development of models for absorption of the infrared and millimeter waves due to various atmospheric constituents, it was found that the adjustable parameter set had to be decomposed into several subsets with varying level of association with data. In classical approaches, the subsets of parameters and obtained purely empilically one at a time. Therefore, the resulting parameter set had no optimality property. This paper proposes a method of simultaneously identifying all the parameters regardless of their association with data. First a generic model is defined in terms of generic parameters which represent individual subsets of real parameters. Then we utilize structured partition of the parameter set and a family of auxiliary variables to combine subsets of parameters and replace generic parameters. Thus our problem reduces to a standard parameter estimation problem. The developed approach was successfully applied to the generation of atmospheric absorption models.