A multivariate variational data assimilation scheme for the Multiple-scale Atmospheric Transport and CHemistry (MATCH) model is presented and tested. A spectral, non-separable approach is chosen for modelling the background error constraints. Three different methods are employed for estimating background error covariances, and their analysis performances are compared. Observation operators for aerosol optical parameters are presented for externally mixed particles. The assimilation algorithm is tested in conjunction with different background error covariance matrices by analysing lidar observations of aerosol backscattering coefficient. The assimilation algorithm is shown to produce analysis increments that are consistent with the applied background error statistics. Secondary aerosol species show no signs of chemical relaxation processes in sequential assimilation of lidar observations, thus indicating that the data analysis result is well balanced. However, both primary and secondary aerosol species display emission- and advection-induced relaxations.