Ensemble data assimilation algorithm, based on a maximum likelihood approach, has been developed for application to NASA’s GEOS column precipitation model. The algorithm produces optimal initial conditions and estimates the model error, employing the state augmentation approach. In each data assimilation cycle, the analysis and forecast error covariance matrices are updated, allowing for cross-correlations between the model state and the model error components. Preliminary data assimilation experiments indicate a good algorithm performance, as measured by standard innovation statistics scores (e. g., chi-square test, probability distribution of the innovations, RMS errors). Experimental results examining the impact of the model error estimation will be presented and discussed in more detail.