Please kindly find my comments on the manuscript in the supplementary pdf file.

Control simulation experiment (CSE) is a method recently introduced by Miyoshi and collaborators with the aim of controlling chaotic dynamical systems. The ultimate goal is to apply this approach to the weather system, but so far only preliminary investigations have been performed on the Lorentz-63 and Lorentz-96 models. In the original paper of Miyoshi and Sun on the Lorentz-63 model, the magnitude D of the perturbations is kept constant during each experiment, and the efficiency of the control is tested for various values for D. The current manuscript discusses the use of the bred vector (BV) and of the singular vector (SV) for reducing the time of control and the total magnitude of the perturbations. More precisely, after a thorough study of the behavior of the BV and of the SV for the system under control, the authors propose to update the magnitude of the perturbations based on the maximum growth rate of the SV. On average, this implementation leads to a reduction of about 40% of the time of control, and of about 20% of the magnitude of total perturbation applied to the system.

The manuscript is clear and easy to read. In addition, since reducing the control time and the magnitude of the perturbations are two of the key factors for a successful implementation of a CSE to more complex systems, discussing any method for achieving this goal is valuable. As a consequence, the referee recommends this manuscript for publication in Nonlinear Processes in Geophysics, once the comments mentioned below are taken into account.

The first part of the investigation consists in studying the evolution of the BV and of the SV for the system under control. Indeed, it is known that these vectors can be used to predict the regime change for the Lorentz-63 model, and the aim of the SCE is precisely to keep the system always in the same regime. Quite surprisingly, Figure 4 shows a very different behavior of the BV and SV for the system under control, with large values taken by the BV despite the absence of change of regime. A qualitative difference is already visible in Figure 3. How can one understand the behavior of the BV ? Is it due to an inadequate definition of the BV for the controlled system ? Unfortunately, the paragraph describing the computation of the BV is not clear (lines 95 to 99). It would be useful to clarify this paragraph, and to interpret the unexpected behavior of the BV for the system under control. Along the same line, would it be possible to discuss the observation reported at the end of Section 3.2: "The directions and magnitudes of SV are similar for thos without and under control", could this be expected ?

The BV and the SV can be computed for any dynamical system, but they turned out to be useful for predicting the change of regime for the Lorentz-63. This property is then used for rescaling the perturbation amplitude based on the knowledge of the SV. If other chaotic systems are considered, without such a clear relation between the property under control and the BV or the SV, can one still expect any use of these vectors for reducing the time of control or the magnitude of the perturbations ? In other words and even if the question is very vague, is there an abstract applicability of your approach, or does it intrinsically depend on the existing link between the growth rates of the BV or SV and the change of regime ?

In Section 2.4, if To is 1 time step, does it mean that the product of matrices in (8) reduces to only 1 matrix ? Similarly, if n=3, is it necessary to mention the general case for S as a n times n diagonal matrix ? This short section can probably be simplified a little bit.

On line 168, is it really 2.96 or 0.0296 ?

On line 203 – 204, the sentence starting with Nevertheless… should be improved.