the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 18 Jun 2026
Wind resource assessment in flow-distorting terrain: Techno-economic comparison of fixed-wing UAVs & profiling LiDAR
Abstract. This study conducts a head-to-head techno-economic comparison of vertical profiling Light Detection and Ranging (LiDAR) and fixed-wing Unmanned Aerial Vehicles (UAVs) deployments for wind resources assessment at a representative 120 MW onshore project in flow‑distorting terrain in Germany. A simplified energy yield model and a Monte Carlo framework propagate literature-based measurement and flow model uncertainties through different measurement scenarios to the exceedance levels (P50, P90) of Annual Energy Production (AEP), Net Present Value (NPV) and Levelized Cost of Electricity (LCOE). The analysis shows that both technologies achieve similar point measurement uncertainties (±0.4–0.5 m/s, ±7–10°). The dominant lever for reducing AEP uncertainty and improving P90 is increased direct spatial coverage, not marginal gains in instrument accuracy. Under consistent cost and financing assumptions, UAV-based deployments deliver the largest P90 uplift, higher NPV and lower LCOE than LiDAR configurations.
Competing interests: Danial Hassani is the founder of Alveo AB, which is commercially developing the UAV platform mentioned in this study, and provided the technical data used in designing four measurement scenarios. The remaining authors declare no competing interests.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.- Preprint
(1062 KB) - Metadata XML
-
Supplement
(400 KB) - BibTeX
- EndNote
Status: closed
- CC1: 'Comment on wes-2026-92', J. Gordon Leishman, 13 Jul 2026
-
RC1: 'Comment on wes-2026-92', Anonymous Referee #1, 07 Sep 2026
Review of the manuscript „Wind resource assessment in flow-distorting terrain: Techno-economic comparison of fixed-wing UAVs & profiling LiDAR“
The manuscript of Nguyen et al. presents a comparison of lidar and drone technologies for investigating wind resource in complex terrain. Although the topic is interesting for the wind energy community, the manuscript suffers from fundamental drawbacks:
- It should be made clear from the beginning if the analyses are based on real-world measurement data, or on theoretical considerations. The lidar data set of Rausch et al. (2022) is mentioned several times, but it seems not to be used for any analyses, besides showing a wind rose and Weibull distribution of the measurement site
- The different measures for resource assessment should be introduced at least quickly to be able to understand the calculations
- Acronyms should be introduced, e.G. WAsP
- It remains totally unclear how the drone is used and what kind of drone is applied. What is the size and weight? Instrumentation? Flight patterns? It is stated that the drone flies at hub height over “larger areas”. How large is the area? What about flight permissions? Can the drone fly beyond visual line of sight? During day and night? What about bias due to precipitation, clouds, high wind speed?
- Drones are supposed to retrieve information on spatial variability of the wind field. How can this be distinguished from the temporal variability, e.g. low-level jets changing altitude, modifications of the atmospheric boundary layer during the morning transition?
- The literature review in Sect. 2.1 remains unclear. How was the information derived? Based on which data base – Web of Science or Scopus? How many entries were provided? How were the results of the search handled? Were they checked manually? The search string should also contain “UAS”, “RPAS”, and the full names of the acronyms
- 3 should provide information on the colours. Is it terrain?
- Please provide definitions of P50, P90, etc.
- Please explain price estimates for drone operations, and what is covered
- Describe how the drone data are combined with the simulations. Data assimilation?
- It is totally unclear how the NPV and LCOE are calculated, and what is the base for estimating the cost of the drone measurements.
In my opinion, drone and lidar technology are complementary methods that should ideally be combined – the strength of the lidar is continuous time series and obtaining the wind speed and direction simultaneously at different altitudes. Drones provide in-situ measurements. Their strength is that they can also measure temperature and provide information on atmospheric stability. They are normally not operated 24/7, but for intensive observation periods, with limitations concerning precipitation, clouds, strong wind speed.
In summary, I recommend rejecting the article in the current form.
Citation: https://doi.org/10.5194/wes-2026-92-RC1 -
RC2: 'Referee report wes-2026-92', Anonymous Referee #2, 02 Oct 2026
Recommendation: reject. There is no clear research question being posed nor answered. A wind resource assessment is not performed as the title implies. The paper's central claim — that a fixed-wing UAV deployment give the largest P90 uplift, higher NPV and lower LCOE than profiling LiDAR — does not follow from evidence. It follows from how the uncertainty and financial models were set up.
The manuscript runs a hypothetical 20-turbine, 120 MW farm at Clausthal through a WAsP-type yield model and a Monte Carlo propagation, then a project-finance model. The original dataset had only one vertical profiling lidar (Vaisala Leosphere WLS8-8). No UAV or second LiDAR was ever deployed at that site. Hence, each measurement technology is reduced to one Gaussian σ for speed and one for direction. Nothing that actually distinguishes the technologies as measurement systems is modelled and, precisely here, lies the manuscript's single greatest flaw and, at the same time, the interesting research challenge.
Once the measurement technologies are not actually modeled, the instrument error is applied as an independent noise to the reference lidar measurements. Systematic biases, UAV sampling representativeness, weather-limited availability, vertical coverage, operational issues, regulation and bankability to get a certifiable P90 are all absent. The authors do concede that only random components are modelled, yet the abstract and conclusion draw conclusions about instrument accuracy, which can't be verified.
The paper is a comparison of two measurement technologies, but never models how either one measures. Its conclusion that "both technologies achieve similar point measurement uncertainties" and that measurement accuracy is a minor lever is a consequence of the error model, not a finding. What actually drives measurement uncertainty is systematic and is missing. In energy-yield practice, instrument uncertainty is correlated over the whole campaign and does not average down.
Once systematic uncertainty is split from random noise, the scenario ranking depends on the relative systematic uncertainty of each technology — which the paper never estimates. For an unvalidated UAV platform that term is the largest one in the problem. (A community comment by J. G. Leishman, CC1, already raised the random-cancellation point in the public discussion.)
Another fatal flaw comes from a flow model uncertainty that is just assigned by the authors. Assigning 0% for UAV_4 just guarantees it wins. Plus, the model contains no real spatial information that the UAV could add, as the turbine "truth" is itself the flow-model output.
An important reference for the argument's presented in the paper, Boventer et al. (2024), is misreported. The paper cites it for "~0.1 m/s" 10-min UAV uncertainty and as a comparison "at a mountainous site" with 30-min LiDAR averaging (l. 68–70, 213–214). However, the reference study is over flat farmland at Falkenberg, hence much more "gentler" terrain. No derivations on the impact of terrain slope on measurement uncertainty (e.g. roughness index) are mentioned or applied.
Consider the P90 farm standard deviation of 1.90 % against a turbine standard deviation (flow model) of ~8 %, which matches 8 %/√20 = 1.79 %, i.e. errors are independent across turbines. This means that there is zero spatial correlation in the flow model, which is not true. Errors extrapolated from one mast are strongly correlated in reality. This alone changes the P90 by a factor of 3-4. The realistic base-case P90/P50 is ~0.85–0.92 in industry practice. The paper's 0.975–0.9997 is not credible for bankable financing.
As an atmospheric scientist and wind resource expert, I will restrain myself to comment on the major flaws in the inputs to the financial model, and not on the financial model itself, which in my view are already enough to reject the paper.
As a final comment, I do encourage the complementarity of UAVs and lidars for atmospheric studies, particularly for wake measurements and high-frequency sampling along a short and focused flight paths/study areas for an intensive measurement period. More studies on UAVs measurement uncertainty, spatial representativeness, year-round operability and weather-related biases will definitely be appreciated by the community.
Citation: https://doi.org/10.5194/wes-2026-92-RC2
Status: closed
-
CC1: 'Comment on wes-2026-92', J. Gordon Leishman, 13 Jul 2026
The manuscript compares four wind-resource-assessment deployment scenarios by propagating assumed measurement and flow-model uncertainties through an AEP and financial model. Several parts of the methodology require clarification before the numerical differences among the scenarios can be interpreted.
The largest issue is the treatment of uncertainty. Measurement errors are introduced as zero-mean Gaussian perturbations to 10-minute wind-speed and wind-direction data. When these perturbations are applied over a full year and across many turbine locations, much of the random error cancels. This appears to contribute to the very small AEP standard deviation reported for UAV_4, only 0.02% of P50. At the same time, long-term correction, interannual variability, wake uncertainty, power-curve uncertainty, calibration bias, and correlated spatial or temporal errors are excluded. The manuscript should distinguish between random sampling error and persistent or correlated uncertainty and should explain exactly how each uncertainty term is sampled in the Monte Carlo analysis.
The assumed flow-model uncertainties also control much of the result. The scenarios use 8%, 5%, 5%, and 0% flow-model uncertainty, respectively. The reduction from 8% to 5% is based on literature values, while the UAV_4 case removes flow-model uncertainty entirely because each turbine location is directly sampled. Direct sampling at a turbine location does not remove all uncertainty associated with representativeness, vertical extrapolation, temporal coverage, long-term correction, or spatial variability between flight passes. The basis for assigning zero residual flow-model uncertainty should therefore be reconsidered or more narrowly defined.
The increase in P50 from 468,893 MWh in LiDAR_1 to 472,423 MWh in UAV_4 also requires explanation. If the same underlying wind field is used and only zero-mean uncertainty is reduced, the main expected effect should be a narrowing of the AEP distribution rather than a systematic increase in its median. A deterministic no-uncertainty result should be included so that the source of the P50 shift can be identified.
The financial model requires additional detail. Equation (2) is described as an equity NPV calculation, but it discounts after-tax EBIT using WACC and subtracts total CAPEX. This is closer to a project-level discounted cash-flow formulation than an equity cash-flow formulation. The debt-sizing procedure, debt-service schedule, treatment of interest and principal, and derivation of the reported debt shares should be shown explicitly. The use of P90 energy in the LCOE denominator should also be separated from a conventional P50-based LCOE, since using P90 makes reduced uncertainty appear as a reduction in generation cost even when expected energy changes only slightly.
Several smaller inconsistencies should also be corrected. Twenty V162-6.2 MW turbines give 124 MW rather than 120 MW. The manuscript alternates between describing the site as low-complexity terrain and using it to support conclusions for moderate-to-high-complexity terrain. The notation “±0.4 m/s” is also used for what appears to be a Gaussian standard deviation, which should be stated directly as σ = 0.4 m/s rather than as an uncertainty bound. Clarifying these points would make the comparison more reproducible and would separate results produced by the data from those produced by the scenario assumptions.
Disclaimer: this community comment is written by an individual and does not necessarily reflect the opinion of their employer.Citation: https://doi.org/10.5194/wes-2026-92-CC1 -
RC1: 'Comment on wes-2026-92', Anonymous Referee #1, 07 Sep 2026
Review of the manuscript „Wind resource assessment in flow-distorting terrain: Techno-economic comparison of fixed-wing UAVs & profiling LiDAR“
The manuscript of Nguyen et al. presents a comparison of lidar and drone technologies for investigating wind resource in complex terrain. Although the topic is interesting for the wind energy community, the manuscript suffers from fundamental drawbacks:
- It should be made clear from the beginning if the analyses are based on real-world measurement data, or on theoretical considerations. The lidar data set of Rausch et al. (2022) is mentioned several times, but it seems not to be used for any analyses, besides showing a wind rose and Weibull distribution of the measurement site
- The different measures for resource assessment should be introduced at least quickly to be able to understand the calculations
- Acronyms should be introduced, e.G. WAsP
- It remains totally unclear how the drone is used and what kind of drone is applied. What is the size and weight? Instrumentation? Flight patterns? It is stated that the drone flies at hub height over “larger areas”. How large is the area? What about flight permissions? Can the drone fly beyond visual line of sight? During day and night? What about bias due to precipitation, clouds, high wind speed?
- Drones are supposed to retrieve information on spatial variability of the wind field. How can this be distinguished from the temporal variability, e.g. low-level jets changing altitude, modifications of the atmospheric boundary layer during the morning transition?
- The literature review in Sect. 2.1 remains unclear. How was the information derived? Based on which data base – Web of Science or Scopus? How many entries were provided? How were the results of the search handled? Were they checked manually? The search string should also contain “UAS”, “RPAS”, and the full names of the acronyms
- 3 should provide information on the colours. Is it terrain?
- Please provide definitions of P50, P90, etc.
- Please explain price estimates for drone operations, and what is covered
- Describe how the drone data are combined with the simulations. Data assimilation?
- It is totally unclear how the NPV and LCOE are calculated, and what is the base for estimating the cost of the drone measurements.
In my opinion, drone and lidar technology are complementary methods that should ideally be combined – the strength of the lidar is continuous time series and obtaining the wind speed and direction simultaneously at different altitudes. Drones provide in-situ measurements. Their strength is that they can also measure temperature and provide information on atmospheric stability. They are normally not operated 24/7, but for intensive observation periods, with limitations concerning precipitation, clouds, strong wind speed.
In summary, I recommend rejecting the article in the current form.
Citation: https://doi.org/10.5194/wes-2026-92-RC1 -
RC2: 'Referee report wes-2026-92', Anonymous Referee #2, 02 Oct 2026
Recommendation: reject. There is no clear research question being posed nor answered. A wind resource assessment is not performed as the title implies. The paper's central claim — that a fixed-wing UAV deployment give the largest P90 uplift, higher NPV and lower LCOE than profiling LiDAR — does not follow from evidence. It follows from how the uncertainty and financial models were set up.
The manuscript runs a hypothetical 20-turbine, 120 MW farm at Clausthal through a WAsP-type yield model and a Monte Carlo propagation, then a project-finance model. The original dataset had only one vertical profiling lidar (Vaisala Leosphere WLS8-8). No UAV or second LiDAR was ever deployed at that site. Hence, each measurement technology is reduced to one Gaussian σ for speed and one for direction. Nothing that actually distinguishes the technologies as measurement systems is modelled and, precisely here, lies the manuscript's single greatest flaw and, at the same time, the interesting research challenge.
Once the measurement technologies are not actually modeled, the instrument error is applied as an independent noise to the reference lidar measurements. Systematic biases, UAV sampling representativeness, weather-limited availability, vertical coverage, operational issues, regulation and bankability to get a certifiable P90 are all absent. The authors do concede that only random components are modelled, yet the abstract and conclusion draw conclusions about instrument accuracy, which can't be verified.
The paper is a comparison of two measurement technologies, but never models how either one measures. Its conclusion that "both technologies achieve similar point measurement uncertainties" and that measurement accuracy is a minor lever is a consequence of the error model, not a finding. What actually drives measurement uncertainty is systematic and is missing. In energy-yield practice, instrument uncertainty is correlated over the whole campaign and does not average down.
Once systematic uncertainty is split from random noise, the scenario ranking depends on the relative systematic uncertainty of each technology — which the paper never estimates. For an unvalidated UAV platform that term is the largest one in the problem. (A community comment by J. G. Leishman, CC1, already raised the random-cancellation point in the public discussion.)
Another fatal flaw comes from a flow model uncertainty that is just assigned by the authors. Assigning 0% for UAV_4 just guarantees it wins. Plus, the model contains no real spatial information that the UAV could add, as the turbine "truth" is itself the flow-model output.
An important reference for the argument's presented in the paper, Boventer et al. (2024), is misreported. The paper cites it for "~0.1 m/s" 10-min UAV uncertainty and as a comparison "at a mountainous site" with 30-min LiDAR averaging (l. 68–70, 213–214). However, the reference study is over flat farmland at Falkenberg, hence much more "gentler" terrain. No derivations on the impact of terrain slope on measurement uncertainty (e.g. roughness index) are mentioned or applied.
Consider the P90 farm standard deviation of 1.90 % against a turbine standard deviation (flow model) of ~8 %, which matches 8 %/√20 = 1.79 %, i.e. errors are independent across turbines. This means that there is zero spatial correlation in the flow model, which is not true. Errors extrapolated from one mast are strongly correlated in reality. This alone changes the P90 by a factor of 3-4. The realistic base-case P90/P50 is ~0.85–0.92 in industry practice. The paper's 0.975–0.9997 is not credible for bankable financing.
As an atmospheric scientist and wind resource expert, I will restrain myself to comment on the major flaws in the inputs to the financial model, and not on the financial model itself, which in my view are already enough to reject the paper.
As a final comment, I do encourage the complementarity of UAVs and lidars for atmospheric studies, particularly for wake measurements and high-frequency sampling along a short and focused flight paths/study areas for an intensive measurement period. More studies on UAVs measurement uncertainty, spatial representativeness, year-round operability and weather-related biases will definitely be appreciated by the community.
Citation: https://doi.org/10.5194/wes-2026-92-RC2
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 345 | 105 | 103 | 553 | 76 | 118 | 101 |
- HTML: 345
- PDF: 105
- XML: 103
- Total: 553
- Supplement: 76
- BibTeX: 118
- EndNote: 101
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
The manuscript compares four wind-resource-assessment deployment scenarios by propagating assumed measurement and flow-model uncertainties through an AEP and financial model. Several parts of the methodology require clarification before the numerical differences among the scenarios can be interpreted.
The largest issue is the treatment of uncertainty. Measurement errors are introduced as zero-mean Gaussian perturbations to 10-minute wind-speed and wind-direction data. When these perturbations are applied over a full year and across many turbine locations, much of the random error cancels. This appears to contribute to the very small AEP standard deviation reported for UAV_4, only 0.02% of P50. At the same time, long-term correction, interannual variability, wake uncertainty, power-curve uncertainty, calibration bias, and correlated spatial or temporal errors are excluded. The manuscript should distinguish between random sampling error and persistent or correlated uncertainty and should explain exactly how each uncertainty term is sampled in the Monte Carlo analysis.
The assumed flow-model uncertainties also control much of the result. The scenarios use 8%, 5%, 5%, and 0% flow-model uncertainty, respectively. The reduction from 8% to 5% is based on literature values, while the UAV_4 case removes flow-model uncertainty entirely because each turbine location is directly sampled. Direct sampling at a turbine location does not remove all uncertainty associated with representativeness, vertical extrapolation, temporal coverage, long-term correction, or spatial variability between flight passes. The basis for assigning zero residual flow-model uncertainty should therefore be reconsidered or more narrowly defined.
The increase in P50 from 468,893 MWh in LiDAR_1 to 472,423 MWh in UAV_4 also requires explanation. If the same underlying wind field is used and only zero-mean uncertainty is reduced, the main expected effect should be a narrowing of the AEP distribution rather than a systematic increase in its median. A deterministic no-uncertainty result should be included so that the source of the P50 shift can be identified.
The financial model requires additional detail. Equation (2) is described as an equity NPV calculation, but it discounts after-tax EBIT using WACC and subtracts total CAPEX. This is closer to a project-level discounted cash-flow formulation than an equity cash-flow formulation. The debt-sizing procedure, debt-service schedule, treatment of interest and principal, and derivation of the reported debt shares should be shown explicitly. The use of P90 energy in the LCOE denominator should also be separated from a conventional P50-based LCOE, since using P90 makes reduced uncertainty appear as a reduction in generation cost even when expected energy changes only slightly.
Several smaller inconsistencies should also be corrected. Twenty V162-6.2 MW turbines give 124 MW rather than 120 MW. The manuscript alternates between describing the site as low-complexity terrain and using it to support conclusions for moderate-to-high-complexity terrain. The notation “±0.4 m/s” is also used for what appears to be a Gaussian standard deviation, which should be stated directly as σ = 0.4 m/s rather than as an uncertainty bound. Clarifying these points would make the comparison more reproducible and would separate results produced by the data from those produced by the scenario assumptions.