Methodological and algorithmic aspects of combining mathematical models with the results of observations and experiments in studying real processes are discussed. The theoretical grounds for such constructions are the variation principles and the methods of optimization as applied to joint models of atmospheric hydrothermodynamics and the models of transfer and transformation of pollutants. Relations of the theory of model sensitivity and algorithms for their implementation are constructed for these combinations. The algorithms allow simultaneous estimation of relative contributions from each factor to variations of the characteristics under study, as well as evaluation of the tendencies of their influence.