Please see the attached document for more detailed comments

It was an interesting read and valuable analysis have been done. Please find my comments below:

Line 13: It is very rare to install sensors on all turbines of a farm. So, this claim that high-frequency data is impractical at fleet scale is not relevant.

Line 14 and 15: “This choice is convenient but tends to underestimate fatigue damage generated by cycles whose period exceeds the windows”. We already know that by a simple cyclecount on the residuals signal, this limit is solved. So this sentence is not 100% correct.

Having sub-minute windows has higher SCADA resolution compared to 10-minute windows, as usually the statistical values for SCADA are used to calculate bending moments, or to define the operational state of each window. I missed this discussion in your abstract.

Line 16: These parameters are not the only reasons behind low frequency cycles. It can also be wind direction, change in wave parameters in case of offshore turbines. It can even be the seasonal or daily temperature variation.

Line 19: Which cycles are discarded? We usually cycle count the high-frequency stress time-series within 10-min, so no cycles are discarded. The sentence needs clarification.

Line 44: “based on” is not a correct wording. It is usually based on strain data. I guess you mean data discretization in 10-minute windows follows the 10-minute SCADA format. The 10-minute windowing is not just because of SCADA, but also due to how simulations are done in the design.

Line 49: What do you mean by aggregation over 10-minute windows? Similar comment as above. If it is high-frequency strain measurement, the high-frequency cycles should be covered. Did you mean that by cutting data into 10-minute windows and using the statistical SCADA parameters, the high-frequency cycles might be smoothed?

Line 66: Can you explain here shortly why should someone need this separation?

Line 68: Can you discuss the limitations of reduced information strategies? For a general SHM campaign, can we have the FFT, PSD, strain calibration, virtual sensor signal? How accurate would be the 15s damage? Although longterm accumulated damage is important, many prediction models are based on 10min window. For example, in Figure 8, the green lines for 15s windows are very low, meaning that the 15s damage is not a good representative by itself. So if a machine learning model is trained based on the 15s data, it needs an extra information about low frequency cycles. While this problem is less significant for the 10minute data.

Line 144: 2.2 title should be High-frequency-cycle counting. High cycle is not defined in the manuscript.

In Figure 1, section € of figure, it should be sigma H f+h i. Also i missing in section g too. Usually the counted cycles are shown by small n. Capital N is for the maximum number of cycles that yields to failure.

In Figure 2, the link between merged full-cycle histogram and the gray area should be a bit higher. Now it gives the impression that pink and green boxes are linked. There is also another link between SHM raw data 50Hz and the pink box which might not be necessary.

Figure 3, title of the orange box should be min-max values per window? Caption is wrong ( elements d to g)

Table 1, N_raw, can you  use another symbol. N is for number of stresses to failure.

Line 247: 𝑖 ∈ {𝑙𝑜𝑤, ℎ𝑖𝑔ℎ} should be in line 246.

Line 247: “relative to the total damage 𝐷𝐿𝐹𝐹𝐷,𝑡𝑜𝑡𝑎𝑙” should not be here. Add the symbol  D LFFD,total to the end of line 246.

Line 272: If the yaw parameter was used for the conversion to FA/SS, the yaw mean was from a window of 15s and 10min SCADA for the 2 windowing of 15s and 10min, respectively?

Line 280: Can you define Run, Ready, STOP states shortly?

Figure 6: EC3 to be removed from the title of the subplots as they show both EC3 and DNV results.

Figure 7: There is no high 15-sec bar for (b)

Figure 9: Why the SS direction is more sensitive to lower sampling frequencies than the FA direction? What can be the physical explanation?

Table 3: Are the storage percentage compared to first row? If yes, first row should be 100 %? And second row should be 107% in total? The mentioned 1 histograms have negligible storage need?