In this study, the author investigates the preview control of wind turbines under varying incoming wind directions. Overall, this reviewer finds that this manuscript is of good quality and can be accepted with some minor revisions:

• In line 193, the author should cite the original literature of ADMR (Wu and Porté-Agel, 2011), which was also cited in Dörenkämper et al. (2015).
• In line 195, the author mentioned that “As a result, a transient wind direction change propagates through the domain with a time-varying wind speed and direction”. To me, it seems to indicate that the spatial variation of wind direction in the simulation domain is associated with the incoming wind speed. Considering the fact that the authors adopt a Taylor frozen hypothesis in their control model, this reviewer would like to ask the authors to add a short discussion on the possible influence of different incoming wind speeds on the preview control.
• In Section 4.3, the authors present a very interesting comparison of the results from LES and the computational cheap engineering model. This reviewer would appreciate a brief analysis of the possible directions for the improvement of the engineering model, which shows a high correlation but also a systematic bias to the LES results. 

This paper describes quite clearly the results of some rather thorough investigations into the value of preview information for yaw control and wake steering control. My main criticism is in the balance between the amount of careful and dedicated work which was done compared to the value of the results. The main conclusion is that there is value in using preview information; but it would be rather surprising if that were not the case. Using a number of more and less detailed simulations, the paper attempts to quantify the benefits in terms of energy production gain. The accompanying changes in the amount of yawing action are only mentioned qualitatively, without trying to quantify this. Turbine loading implications are not covered - that would be well beyond the scope of the paper, but it should perhaps be mentioned as an issue.

The simulations are certainly useful as illustrations, and as prompts for thinking about some of the issues involved. The authors rightly identify many of the shortcomings in the modelling, but with all the uncertainties, I think it would be difficult to generalise the quantitative results in any meaningful way.

To mention some specific issues:

The preview is modelled as coming from a notional met mast upstream of the turbines, providing a perfect preview which advects to the turbines as frozen turbulence. The authors acknowledge that this is not practically realistic, and mention some other possibilities, such as using turbine data or multiple scanning LiDARs. Both of these are so different from what is assumed that they should be modelled explicitly. The real question is, what additional information is it practical to obtain, and what performance gains are practically realisable from such information, given its inevitable imperfections, and is the benefit sufficient to make it worthwhile to obtain this information? Otherwise, the conclusion that perfect preview information is useful is self-evident.

The paper focuses on a wind farm of two turbines. This makes the quantitative results even less valuable. Repeating all these LES simulations for a large wind farm would be hugely onerous, but it should be feasible with the engineering model.

The conclusion that optimum preview time for wake steering doesn't change much is probably because it depends mainly on wind speed and the control algorithm's timestep and dynamic response, all of which were fixed in this paper, so it's not surprising.

The conclusion that preview is most beneficial when wind direction changes a lot is surely also self-evident.

The paper compares against conventional or 'greedy' yaw control, but this is with fixed parameters which have probably never been optimised, and certainly not for different conditions, let alone for the case where preview information is available.

In a stochastic world the effects of preview can only be determined by running quite long simulations with realistic low-frequency changes in direction. The short simulations with linear direction change ramp don't seem to be worthwhile, even with the addition of high-frequency turbulence, as the conclusions drawn from these are quite self-evident.

Some detailed comments:

Equation 3: different authors use different definitions of thrust coefficient for yawed rotors. Which definition is used here, i.e. which thrust force component in the numerator and which wind speed component in the denominator?

Section 3.2.3: the wind vane signal "was chosen ... to mimic a nacelle wind vane not disturbed by the rotor. Surely in reality the wind vane is very much disturbed by the rotor. Offset bias can be compensated for by calibration, but the yaw control dynamics could be strongly affected, which would be relevant here.

Lines 203-7: is this physically realistic, given that cyclic boundary conditions are a modelling necessity, not reflecting reality? This is acknowledged later, but how much uncertainty does it add to the conclusions?

The wording in section 4.3 seems to make an explicit assumption that the LES results are the 'correct' ones. The wording should be changed to be consistent with the fact that neither model is 'correct'. There may be justification for assuming that the LES model is likely to be more nearly correct, at least when the results are averaged over many specific runs, because of the simplicity and empiricism of the engineering model.

The word 'exemplary' is used many times in the paper, but I think this should be 'example'. The meaning is rather different.

Line 207 typo: 'alters'.