Abstract <p>The problem of data assimilation for nonstationary models of impurity transfer and transformation is considered as a sequence of coupled inverse problems of restoring the spatiotemporal structure of state functions, taking into account the measurement data received during modeling. Data assimilation is carried out together with the identification of an additional unknown source function, which we call the uncertainty function of the model. The purpose of this work is a brief historical overview and presentation of an up-to-date version of the data assimilation algorithms for atmospheric chemistry models based on sensitivity operators and ensembles of solutions to adjoint equations. A demonstration of the algorithm for a three-dimensional model with a nonlinear measurement operator is given in a modeling scenario for the Baikal region.</p>