This paper is a valiant original effort to address the design of state-of-the-art sequential data assimilation codes for large-scale problems. Its main objective is to present a package of various versions of the Kalman filter implementation, “whose code is independent of both the model and observations and is easy to use,” and, moreover, includes parallelization. It also includes Fortran 90 modules that implement reduced-rank square root versions of the Kalman filter.

For the ensemble Kalman filter (EnKF) implementation that does not offer covariance localization or covariance inflation, the tradeoff between simplicity of use and details is evident. The same is true for observation operators, which are not addressed. Therefore, the user must know his or her modeling system in detail, in order to deal with the above issues, and provide observation errors and covariance matrices.

Even so, this is a worthwhile original effort that presents many novel ideas, including examples of present package implementations for zero-, one-, two-, and three-dimensional cases, as well as a helpful user guide. The paper provides valuable guidelines, such as modularity and ease of use, for the data assimilation community. As such, Torres should be commended for his originality and scope.

I recommend this clear and easy-to-follow paper to data assimilation researchers and advanced graduate students.