| 30 Oct 2025
Spatio-temporal heterogeneity of the wind resources over a deciduous forest in the U.S. Southeast
Abstract. The Southeastern United States is predominantly characterized by moderately tall forests (≈ 20 m) which, by absorbing a portion of flow momentum from the Atmospheric Boundary Layer, reduce wind speeds within the rotor layer of modern wind turbines. Taller wind turbines (with hub heights greater than 100 m) are likely to experience higher wind speeds, assuming that wind resources located farther away from the ground are less affected by the forest layer. However, forest canopy drag and heterogeneity effects on wind resources at high altitude above ground have not been fully investigated in the U.S. . In this work, three sites located in a deciduous forest in the Appalachian mountains of the U.S. Southeast are investigated from September 2024 through June 2025 as part of the Lidar Experiments for Assessing Flow over Forests (LEAFF) campaign. Wind statistics are resolved both within the canopy (by a meteorological tower) and above it or in clearings (via four Doppler Lidar). A reference tall wind turbine (with a hub height of 110 m and a rotor layer spanning 45 m to 175 m) is assumed for each site to estimate the available power resource. The wind statistics considered here are the mean wind speed (U), the turbulence intensity (TI) and the cube of the mean wind speed U3, assumed as a proxy for the power in region II of a turbine power curve. The two dominant physical features affecting the wind, i.e. the momentum absorption at the canopy interface (quantified by the drag coefficient, Cd) and the momentum entrainment from the free atmosphere, are quantified as well based on Doppler Lidar data. The present analysis aims to: 1) quantify the monthly variability of wind resources induced by the annual cycle of leaf coverage and changes in the synoptic wind conditions; 2) quantify the correlation of canopy drag and free-atmosphere wind speed with rotor-layer wind statistics; and 3) quantify the wind resource heterogeneity between canopy and nearby forest clearing sites. The present analysis reveals that site inhomogeneities in the wind resources are still found within the bottom half of the rotor layer (i.e., up to the hub height of 110 m) of a tall wind turbine. Additionally, the examined wind resources are more correlated with the wind speed in the free atmosphere than the Cd within the rotor layer, with the only exception of the TI which shows equal correlation with these two quantities. Finally, the largest vertical extent featuring site heterogeneity is found between November and January, which corresponds to the period of minimal leaf coverage (i.e., minimum leaf area index). Overall, the present study shows that, even for tall wind turbines, the wind resources within the rotor area are affected by spatial heterogeneity in surface drag and by the seasonal transition of the canopy leaf coverage. These results have implications for the siting and operation of wind turbines in forested regions, as well as for the siting of Lidar instruments during future observational campaigns.
The manuscript presents an analysis into characteristics of the wind resource over a forested area in the U.S Southeast. The topic is of interest due to limited number of previous studies in that particular area and due to addressing issues that are general for wind power in other forested areas as well. The presentation of the material is clear and the structure is easy to follow. The figures are pleasant in appearance. I do see several major issues with the conclusions that the authors have drawn and the way that the analysis is performed. A major weakness is the definition of the drag coefficient used in the study. The way that it is calculated limits the conclusions that can be drawn due to the omission of atmospheric stratification and/or turbulence length scale from the analysis, the use of single height momentum flux and wind speed in the calculation and the lack of direct quantification of plant area density in the expression. Furthermore, the usefulness of some of the statistics presented in the manuscript, such comparison between correlation coefficients is not clearly motivated and does not seem self-evident. I suggest that the authors revisit the way that the drag coefficient is determined from the measurements, use the tower 2D anemometers together with estimations of plant area densities to calculate a reference drag coefficient. This, together with studying the effect of atmospheric stratification on the drag coefficient defined in Eq. 10 of the manuscript, either by use of the heatflux from the 3D sonic or by studying diurnal variations, would permit better founded conclusions (that may still not be as far reaching as the present conclusions, unfortunately). I furthermore suggest that the authors relate their work on heterogeneity to the body of work that already exists on the blending height and footprint of flow over forested areas. Finally, I am slightly surprised that a manuscript with 8 authors dedicated to the role of “providing feedback and supervising the work” is not more grounded in the previous literature and methodology of the field.
Detailed comments follow below. After section 4.1 I have not provided specific comments, reflecting my opinion that those sections needs to be rewritten.
line 9: Lidar -> Lidars
line 19. Unclear at this Point in the read what "correlated" refers to. Cd is normally a constant, but correlation assumes variation.
line 20. Why correlate TI and not sigma_u? U is already examined and there is a lot of covariation between sigma_u and U. In fact, omitting effects related to outer scales such as the PBLH, TI should be constant for a given atmospheric stratification. That means when you are correlating changes in TI, you are actually correlating changes in stratification.
line 20. Instead of using "site heterogeneity" I suggest you use "blending height" unless there is a specific reason you have opted not to.
line 58. Nowadays there is the standard to measure with met-towers much higher than 100 m in many parts of the world. Please specify regional restrictions to clarify to the reader if this is the case here, otherwise revise the sentence.
line 65-66. That the lidar showed promising ability to reproduce turbulence up to 4:th order statistics seems like a rather serious misrepresentation to me. If I recall correctly the tower was only 60 m high and was placed 5 km away, limiting the ability to draw any conclusions at all regarding the ability of using the lidar for turbulence at rotor relevant heights. Also, I seem to recall a rather serious underestimation of the wind by the lidar at the tower comparison. I think the reference must be updated to more accurately represent the findings in the study.
line 67. Please clarify what distances you are referring to.
line 89. “conclusions are discussed” is a bit ambiguous. Clarify if it is discussion and conclusions or only conclusion.
line 99. Please clarify if you mean LAI (Leaf Area Density the area of the leaves) or PAI (Plant Area Density the area of all forest biomass). The wind energy community have an unfortunate tendency to use the label LAI when actually using data and model assumptions that reflect PAI. Since this study refers to differences specifically related to the presence of leaves, it is especially important to get this right.
line 107. Small comment, but all these variables are not measured at 20 Hz. The wind vector and virtual temperature is. The rest is post-processed to 30 minute blocks, I assume.
line 124. While a good hypothesis, I think that it is premature at this point in the study to attribute the variation of wind speed just above the canopy to a general variation in momentum absorption. The displacement height is also likely to change and while the wind speed may be lower just above the forest, changes in turbulence length scales may mean that the situation is different higher up. It is also unclear how effects of atmospheric stratification, which also have a seasonal cycle, plays into this.
line 125. Drag is not just PAI, it is the vertical integral of PAD*U^2. So while the plant area density may be higher, the wind speed is lower so one cannot tell on the PAI alone if the drag is actually higher or lower unless a quantified analysis including wind speed is presented.
line 153. Filtering on CNR means there may be an implicit filiter on wind speed (see work by Gryning). Perhaps this could be worth mentioning in the discussion.
line 158. I suspect there was a particular reason for the chosen width (2.5 sigma instead of 3.5), I think it would be appropriate with a motivation of the reason for the stricter filter.
line 163. Given the filter for spike removal (\pm 2.5 sigma) one would expect only a few percent of the data being rejected if the radial wind is normally distributed. The numbers in figure 4 report much larger rejection rates which point to very long tail(s) of the radial wind distribution. It would be good to know for the reader if the rejection rates include only the data rejected by the \pm 2.5 sigma-filter or if data rejected by any internal filters of the lidars are also included. As far as I know the ZX does not use CNR, so I guess there must be difference between the scanning lidars and the ZX lidars in this respect?
Eq. 6 and 7. Please clarify if the wind direction relative to the rotor plane has been taken into account or not.
Line 260. The momentum loss of importance is that through the entire canopy and the use of "interface" is somewhat ambiguous in that sense.
Eq 10. The Yi 2008 paper presents a rather elaborate investigation into the Cd dependency on height within the canopy. Crucially, the definition of Cd also assumes that there is a local (to each height) balance of the shear stress divergence (see eq. 8 or 10 of Yi 2008 for instance). The rest of the paper studies the vertical variation of Cd, which is a different thing from the vertical variation of the canopy density. In your Eq. 10 you have only a bulk relation between the shear stress and the local velocity and this is very different. I think the citation of Yi is very misleading. Bulk relations, such as your Eq. 10 are normally used over surfaces with very small displacement heights, to represent the skin (or surface) friction but I would argue that they are unsuitable when the displacement height is a factor, in other words, when the drag is distributed vertically. In your case, the density of the vegetation will vary across season, something that will be implicit in your Cd value and that makes it very different from the Cd value of Yi 2008 and most of the Cd values used in the wind energy community. Furthermore, given a fixed height in the atmosphere, the RHS of your eq.10 will be a function of the atmospheric stratification (it is the reciprocal of the square of U/u_*) which makes it unsuitable to evaluate on seasonal basis (given there is a seasonal variation in stratification). Finally, the equation should contain the total shear stress, not just the longitudinal component.
Line 278. I wonder if the non-zero intercept is rather that you are fitting a linear curve to a nonlinear relationship. The volume filter of the lidar is fixed with time, but the length scale of turbulence is very different in stable and unstable conditions, hence the lidar measurement should agree better for larger values of the shear stress (since they tend to happen in unstable conditions when the length scale is larger) and vice versa.
Line 285, Figure 7. I wonder how much of the distribution is actually just due to the atmospheric stratification. I made a quick using measured values of z/L and plotted "C_d" from the square of the reciprocal of the MOST-wind profile. While the value needs to be adjusted for the "missing" plant area to match your value, there is considerable variation even though the "plant area" is not changed. The variation only comes from the variation in the length scale of the turbulence in relation to the height z-d.
Lines 300-308. It is important to differentiate between the effects of displacement height and roughness length. First you state that the roughness is larger, then that it is the displacement height. A denser natural canopy would normally mean a larger displacement, but a lower roughness. In the same way it is by no means necessary that a denser canopy has more drag (it is blocking the wind and the drag is proportional to the plant density times the wind speed squared). Figure 6 in Jackson 1981 gives a good illustration of this.
Line 316. See above comment.
Line 339. As the authors themselves argue in the above section, there is considerable variation due to the diurnal cycle and the same should then apply to the annual cycle. Whereas the annual cycle also has vegetation variations, the diurnal does not. Perhaps a good test is to redo figure 10, but for an average diurnal cycle, to verify that what you see is not simply the cycle of PBL stratification.
Line 348. Again, you have not actually shown that the momentum absorption in the canopy is lower.
Line 355 and above paragraph. I think it is important that you consider and demonstrate that what you see in these figures is not the atmospheric stratification. The shear exponent is a very strong function of the atmospheric stratification and the variation shown in Fig. 11 seems consistent with the changes you would expect from the seasonal cycle of stratification.
Line 360. I don't think you have shown this. It is a strong statement that requires careful investigation. See above comments.
Section 4.2 This whole section needs to be reconsidered. I don't really understand the motivation behind the analysis. Cd is a nondimensional property, so it is clear that \tilde(U) will correlate better with U than Cd will. How should the results be used? For me it is difficult to understand from the plots what is spurious correlation and what is a physical connection between the variables. In Figure 14: How come Cd and TI have so low correlation close to 30 m? It is almost the same quantity. Same goes for Figure A1, it is surprising to me that the correlation is zero between Cd (essentially TI squared) and hub-height TI.
Section 4.3, discussion around figure 15. I don't think that this analysis actually shows how heterogeneity of the surface is reflected on the wind resource. The correlation could come from non-local features such as large scale variations and I suspect it is highly sensitive to the averaging time used to determine the statistics. Consider a situation with an offshore site, it is unclear to me what the differences would be to the plots you present. One would still expect to see less correlation on lower height simply due to the random-error effect of turbulence on the statistics. This is also consistent with the observation in lines 424-425 since stronger wind resource likely means larger variations.
Section 5. I believe the whole section needs to be rewritten following further analysis into the role of atmospheric stratification on the results. I would have liked to see that the heterogeneity section relates to previous work on blending heights above forests and potentially also on footprint. I am skeptical of the method and value of comparing correlation coefficients of Cd (as per Eq. 10 in the manuscript) and U_1km with the wind speed at the rotor heights. Firstly, a direct comparison is of limited value since dimensions of the compared quantities are different, secondly, the 1 km wind speed is very rarely available in wind resource estimation, so I would like to see a better motivation of how the results are useful.