Overview

This manuscript describes an approach for strategically using derating of individual turbines to achieve lifetime extension and NPV uplift for wind farms. The approach herein uses a time-monolithic nonlinear program-based optimization over the sector average-driven estimates of the AEP performance and lifetime load accumulation using both global and local (in the turbine/farm sense) strategies. The finding is that NPV increases are available by extending the operational lifetime of turbines while decreasing AEP in all present studies except for when limiting blade edgewise loads. Overall this paper is a strong contribution to the literature, and begins to elucidate the relationship between derating strategies and overall economic performance, though many outstanding questions remain, as the authors note, about how to act on the findings.

Major comments

• Overall, I would consider simplifying the notation a little bit and adding some more straightforward and immediate clarification for the definitions- it's taking me a few passes through a few sections to fully understand the material as it's presently presented.
• A key aspect I would like to see foregrounded a little more is what to do with the results here in terms of operational strategy.
  • If I were operating a farm in real time, I'm not necessarily compelled to work in this time-monolithic way, but rather projecting out the cost/value of a) what already happened to my farm/what I can detect about it and b) what I see, say 1 decade into operation, as the future circumstances: I might be in a totally different interest rate regime, or cost regime, or whatever, and I'm going to think of fatigue optimal choices w.r.t. this and my projections of the next 10-, 15-, 20-year period.
  • There are likely future moments where repowering is the optimal choice given the trajectory I've taken to a given point, and futures in which it absolutely does not make sense to repower at all- and the perspective from which that question is relevant is not a time-monolithic one at all.
• I would like to see a little more explication of the modeling approaches for 1) the surrogate modeling as well as 2) the development of the per-turbine optimization approach.
  • The latter is likely to include some impact from ordering to capture non-linear effects, such as in Algo. 1, L4, where optimal choices will propogate to neighboring turbines.
•  

Minor comments

• L17: "For energy, the objective is clear: ..."
  • I'm not sure the objective is totally clear especially in the context of a fatigue life study
  • rate of return, capacity factor/capacity utilization factor, PPA and debt service periods vs. post-contract pure-profit operations, etc. all should factor into what this objective actually might be
    • highly variable depending on actor, regulatory climate, etc.
• L17, "For loads..." and subsequent L18 "In contrast, ..." lines are duplicative; suggest rewording
• L116: XambXamb isn't defined well enough?
  • lead with Table 1?
  • local conditions (xx): is this upstream of the turbine w/ wake effects from upstream turbines but no induction effects? some clarification, please!
• L126: what are the surrogate models? are there details in here somewhere?
• L288, typo: your "NWP" I think should be "NPV"
• Fig. 17: caption is unclear, appears to be incorrect or at least ambiguous
• on vizualizations with "turbine index", since you have a 9-turbine farm, perhaps use "N", "NE", "E", ..., "NW", and "C" (for center) rather than indices, to help the reader interpret which row/column is impacted by derating choices