the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 15 Jul 2026
Influence of environmental characteristics on power production in an offshore wind turbine in the Belgian North Sea
Abstract. Quantifying the influence of inflow atmospheric characteristics on wind turbine's operation is fundamental for performance assessment, operational diagnosis, and the estimation of power production. While it is well-established that wind speed is the primary driver of power output, additional environmental factors such as turbulence intensity, atmospheric stability, shear and veer are also known to influence the rotor captured inflow and the turbine's response. However, isolating and interpreting the contribution of each of these factors remain a continuing effort. This study presents a framework using operational data, as nacelle-mounted lidar, SCADA and environmental data, to analyse the impact of these factors in a single wind turbine in a wind farm on the Belgian offshore zone. A machine-learning model is trained to represent the power production deviation behaviour, and SHapley Additive exPlanations (SHAP) are applied to quantify the contribution of each environmental variable to predicted power output. Results show that, excluding wind speed, turbulence intensity and air density are the biggest predictors in the transition of torque-control to pitch-control of the wind turbine operation, with respective contributions that influence the power deviation prediction of 16.8 % and 14.5 %. As for higher wind speeds, air density appears as the main influential factors, contributing up to 32.8 % for the above-rated wind speed of the power curve.
Competing interests: At least one of the (co-)authors is a member of the editorial board of Wind Energy Science.
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
(1558 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on wes-2026-117', Anonymous Referee #1, 27 Jul 2026
-
RC2: 'Comment on wes-2026-117', Anonymous Referee #2, 21 Sep 2026
General Impression:
This paper aims to calculate the effect of environmental factors other than wind speed on wind turbine power performance. The authors collect a highly thorough list of variables and explore their potential impact on power production through a correlation analysis and subsequently a gradient boosting algorithm and associated SHAP analysis. I find the very thorough list and descriptions of potentially influencing variables and their subsequent reduction very valuable, both for improving one’s understanding of wind turbine power production, as well as for the stated aim of producing more accurate but crucially more interpretable power forecasts. I think the methodology is generally very detailed and have included a few areas where I would appreciate a little more detail on the steps followed or the rationale behind them. I believe the framework developed could provide helpful guideline for other researches or professionals in industry forecasting wind power, where current standard practices are lagging far behind. I feel the value and scientific rigour of the paper would be further improved by increasing clarity on whether the SHAP values are interpreted as contribution factors for the machine-learning model or for our understanding of real world phenomena. I would also appreciate suggestions from the authors for how people adopting this framework could further improve its rigour.
Minor Comments
- It is claimed that the SCADA wind speed data comes entirely from anemometers (sonic or cup), though in some wind turbines the wind speed recording may sometimes be calculated partly via a transfer function relating turbine power to wind speed - has this been ruled out in the SCADA data used for this project?
- Please provide some brief information on whether the SHAP analysis was undertaken using a specific software library and which methods/functions were used and why (e.g. TreeSHAP vs KernelSHAP or others)
- The paper aims to develop a framework for analyzing the effect of certain variables on real world turbine power production. My understanding is that SHAP analysis was originally developed to understand the contribution of different features in a machine learning model. At the same time, there are other examples in literature of its values being interpreted as a representation of real-world relationships and causality. I think this may be more robust when the model accuracy is very high and different machine learning models have been analyzed. I believe it would be helpful, for the scientific community, to be more discerning between the model and the real world in the discussion. In particular, I would value more consideration of whether and why model feature contributions (as calculated by SHAP) can be considered to represent real-world relationships. Line 574 of the conclusion may be an additional good location to make a brief comment on this (interestingly in the previous sentence the authors instead claim to be focused on the influence in the model). In general, it may be helpful to be a little more consistent with stating the aim and outcome of the research in terms of the features contributions to the model vs the real-world.
- line 67 – is it possible to provide a description of how the “automated profile classification” is performed
- line 100 describes the measurements being 10m above sea level – is the discrepancy to turbine height considered somehow?
- In lines 116 to 119, it wasn’t clear to me how with the 50Hz sampling frequency and one revolution per second second per scanning plane “produce” 64 azimuthally averaged positions per LiDAR scan. I feel the sentence is worded such that it should be an explanation of this – either provide more description of the procedure OR replace the word “producing” which implies a causality between the numbers. It is also not clear how many planes are scanned.
- in section 2.1, data losses from the separate data sources (e.g. LiDAR, SCADA, MVB (though no losses mentioned) are discussed – is it possible to summarize how much of the overall data remained if e.g. it was required to have SCADA, LiDAR and MVB that was chronologically aligned (SCADA and LiDAR from the same periods in time)?
- Section 3.4 could also mention how atmospheric stability and turbulence intensity are often related
- Lines 245 & 246 – is it possible to make a comment on the likely validity of the assumption (lidar aligned with turbine yaw)?
- Lines 303-305 – is it possible to describe why the chosen metrics were chosen?
- I generally find that the authors have completed some very detailed and interesting work, and I wonder if it is possible to share ideas for additional improvements to the framework in the conclusion? Could more correlation analyses that consider more complex interrelationships be suggested or methods to further test the validity of extending the model based SHAP values to our understanding of the real world?
- In section 5.2.1 I was unsure whether this fitting was a different fitting to the MLR in table 3, as the R squared values seemed different
- In section 5.3, I was confused by the use of the word “baseline” as I understood SHAP analysis to be calculated on a single model, but in the previous section (5.2.1) the MLR was described as the “baseline” model – is this just a terminology clash? Could a different word be used for one of these sections or some clarity be added here?
Technical Corrections and Typos:
- lines 30-32 – is this statement based on the authors experience (or that of colleagues), or from any literature?
- line 41 – abbreviation MRL should be MLR
- line 72 “quantifies the impact of each environmental variable has” – grammar
- line 73 “becoming possible the choice of which variable to be included in further prediction methodologies.” I find this wording awkward and would recommend the word "facilitates" (the choice of...)
- line 133 – “data points with deviations….are removed” not “it is removed data points…”
- line 169 - "allows to assess” is for me strange wording – “allow us to assess” or “makes it possible to assess” would sound more natural
-line 198 – note the different measurement devices may have quite different suitability for measuring turbulence intensity
- lines 337 to 338 – note that this extra region of the power curve is sometimes referred to as zone 2.5 in other literature
-line 362 – R squared is shown not showed
Citation: https://doi.org/10.5194/wes-2026-117-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 392 | 150 | 76 | 618 | 92 | 88 |
- HTML: 392
- PDF: 150
- XML: 76
- Total: 618
- BibTeX: 92
- EndNote: 88
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Please see the attached document for a detailed overview.