the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 05 Feb 2024
Characterization of Local Wind Profiles: A Random Forest Approach for Enhanced Wind Profile Extrapolation
Abstract. Accurate wind speed determination at the height of the rotor swept area is critical for resource assessments. ERA5 data combined with short-term measurements through the "Measure, Correlate, Predict" (MCP) method is commonly used for offshore applications in this context. However, ERA5 poses limitations in capturing site-specific wind speed variability due to its low resolution. To address this, we developed random forest models extending near-surface wind speed up to 200 m, focusing on the Dutch part of the North Sea. Our results show that the random forest model trained on site-specific wind profiles outperforms the MCP-corrected ERA5 wind profiles in accuracy, bias, and correlation. In absence of rotor-height measurements, a model trained within a 200 km region handles vertical extension effectively, albeit with increased bias. Our regionally trained random forest model exhibits superior accuracy in capturing wind speed variations and local effects, with an average deviation below 5 % compared to corrected ERA5 with a 20 % deviation from measurements. The random forest model adeptly captures the inertial subrange of the power spectrum where ERA5 shows degradation. Our study highlights the potential enhancement in wind resource assessment by means of machine learning methods, specifically random forest. Future research may explore extending the random forest methodology for higher heights, benefiting new generation of offshore wind turbines, and investigating cluster wakes in the North Sea through a multinational network of floating lidars, contingent on data availability.
- Preprint
(2097 KB) - Metadata XML
- BibTeX
- EndNote
Status: closed
-
RC1: 'Comment on wes-2023-178', Anonymous Referee #1, 26 Feb 2024
Anonymous Comments on "Characterization of Local Wind Profiles: A Random Forest Approach for Enhanced Wind Profile Extrapolation"
This manuscript developed a random forest model in attempt to efficiently extend near-surface wind speed up to 200 m. Through comprehensive analysis and validation, the study demonstrates that the random forest model exhibits superior accuracy in capturing wind speed variations and local effects. In general, this paper is well written. I recommend the publication of this manuscript, while I have some comments below for the authors to address.
- In the abstract (Line 9), there is no limit on the collection time and sample size of comparative data, only indicating that the comparative results are meaningless. In addition, the analysis of wind speed variability, spectral analysis, and atmospheric stability in the results section was not mentioned in the abstract.
- In section 2.1, as the author stated, “A reliable dataset is key to train and validate a data driven model”. Here should be an explanation of the accuracy of the LiDAR wind profile product.
- In section 2.3, only 15% of the sample size was used for training and validation. Lack of explanation on the total number of training samples. Additionally, in subsequent inversion and comparison, did the sample point use hourly average wind speed data or minute level data?
- In section 2.6, atmospheric state partitioning is based on the Richardson number for subsequent analysis. Here, it should be clarified how many samples are stable conditions and how many samples are unstable conditions for each site.
- There are no sub image numbers such as (a), (b), and (c) in all the images, making it difficult to understand what results each sub image corresponds to. Taking Figure 10 as an example, do the four subgraphs correspond to four sites?
- Lines 39 and 57: The reference format is incorrect. “Bodini and Optis (2020a)” is supposed to appear first. In addition, there are some pioneering papers that are highly relevant to this manuscript, which should not be ignored. I just list several references as follows:
https://doi.org/10.1016/j.rser.2022.112897;
https://doi.org/10.1073/pnas.2119369119;
https://doi.org/10.5194/acp-23-3181-2023;
https://doi.org/10.5194/egusphere-2023-2727.
Citation: https://doi.org/10.5194/wes-2023-178-RC1 -
RC2: 'Comment on wes-2023-178', Anonymous Referee #2, 05 Mar 2024
The submitted manuscript details a methodology to extrapolate wind profile from surface measurements to desired heights. In resource assessment this is an important technical step and a vital input. In the industrial world, MCP method using ERA5 data has been used for a long time and is considered as the go-to method, although it has low resolution in both space and time. This work proposes an alternative method by means of ML algorithms, which shows much better accuracy, especially for wind speed variability.
While I recommend the publication of this work, I do have several questions/thoughts as follows:
1. How does the proposed method compare to ERA5-corrected method in terms of efficiency? ML methods normally take a considerable amount of time to train, but a typical MCP process using ERA5 data can be very fast.
2. The hyperparameters listed in Table 3 for HKW, HKN and HKZ are different. To my limited understanding towards ML methods, hyperparameters act as choice of models/methods in numerical simulations. For example, both RANS equations and LES can be solved to numerically simulate a fluid flow, the choice between them depends on flow physics for a specific problem. Towards this end, the difference in hyperparameters may indicate different wind conditions among these sites. Would the authors make some comments on this?
3. In Figure 4, the left plot shows the wind profile predicted using ERA5-corrented data, but in Figure 5 ERA5 data is used for comparison. What is the difference between "ERA5-corrected" data and "ERA5" data? Also, do both cover the same period shown in Table 3?
4. In Figure 9 "ERA5" data is used, but as is stated in the manuscript, it is expected that ERA5 data is not able to resolve the inertial subrange of the power spectrum. It would be much more interesting to show how ERA5-corrected data perform in this scenario.
5. In Figure 4 and 5, shaded area (denotes standard error of the mean) and line (denotes the mean) with same color are used to represent results from one specific model, but their colors are not easy to distinguish from one and another.
Also, lidar measurements at different heights are connected with lines, which indicates a certain relation between these measurements. However, rigorously speaking this relation is not known. This principle also applies to plotted lines in the left panel in Figure 4 and 5 (which shows random forest model results). Do wind speed for each height is predicted (e.g., with 1m interval) or only some specific heights are predicted (e.g., heights listed in Table 1)?
6. In Figure 7 only RF trained results are shown, what about QRF ones?
Citation: https://doi.org/10.5194/wes-2023-178-RC2 - AC1: 'Comment on wes-2023-178', Farkhondeh Rouholahnejad, 09 Aug 2024
Status: closed
-
RC1: 'Comment on wes-2023-178', Anonymous Referee #1, 26 Feb 2024
Anonymous Comments on "Characterization of Local Wind Profiles: A Random Forest Approach for Enhanced Wind Profile Extrapolation"
This manuscript developed a random forest model in attempt to efficiently extend near-surface wind speed up to 200 m. Through comprehensive analysis and validation, the study demonstrates that the random forest model exhibits superior accuracy in capturing wind speed variations and local effects. In general, this paper is well written. I recommend the publication of this manuscript, while I have some comments below for the authors to address.
- In the abstract (Line 9), there is no limit on the collection time and sample size of comparative data, only indicating that the comparative results are meaningless. In addition, the analysis of wind speed variability, spectral analysis, and atmospheric stability in the results section was not mentioned in the abstract.
- In section 2.1, as the author stated, “A reliable dataset is key to train and validate a data driven model”. Here should be an explanation of the accuracy of the LiDAR wind profile product.
- In section 2.3, only 15% of the sample size was used for training and validation. Lack of explanation on the total number of training samples. Additionally, in subsequent inversion and comparison, did the sample point use hourly average wind speed data or minute level data?
- In section 2.6, atmospheric state partitioning is based on the Richardson number for subsequent analysis. Here, it should be clarified how many samples are stable conditions and how many samples are unstable conditions for each site.
- There are no sub image numbers such as (a), (b), and (c) in all the images, making it difficult to understand what results each sub image corresponds to. Taking Figure 10 as an example, do the four subgraphs correspond to four sites?
- Lines 39 and 57: The reference format is incorrect. “Bodini and Optis (2020a)” is supposed to appear first. In addition, there are some pioneering papers that are highly relevant to this manuscript, which should not be ignored. I just list several references as follows:
https://doi.org/10.1016/j.rser.2022.112897;
https://doi.org/10.1073/pnas.2119369119;
https://doi.org/10.5194/acp-23-3181-2023;
https://doi.org/10.5194/egusphere-2023-2727.
Citation: https://doi.org/10.5194/wes-2023-178-RC1 -
RC2: 'Comment on wes-2023-178', Anonymous Referee #2, 05 Mar 2024
The submitted manuscript details a methodology to extrapolate wind profile from surface measurements to desired heights. In resource assessment this is an important technical step and a vital input. In the industrial world, MCP method using ERA5 data has been used for a long time and is considered as the go-to method, although it has low resolution in both space and time. This work proposes an alternative method by means of ML algorithms, which shows much better accuracy, especially for wind speed variability.
While I recommend the publication of this work, I do have several questions/thoughts as follows:
1. How does the proposed method compare to ERA5-corrected method in terms of efficiency? ML methods normally take a considerable amount of time to train, but a typical MCP process using ERA5 data can be very fast.
2. The hyperparameters listed in Table 3 for HKW, HKN and HKZ are different. To my limited understanding towards ML methods, hyperparameters act as choice of models/methods in numerical simulations. For example, both RANS equations and LES can be solved to numerically simulate a fluid flow, the choice between them depends on flow physics for a specific problem. Towards this end, the difference in hyperparameters may indicate different wind conditions among these sites. Would the authors make some comments on this?
3. In Figure 4, the left plot shows the wind profile predicted using ERA5-corrented data, but in Figure 5 ERA5 data is used for comparison. What is the difference between "ERA5-corrected" data and "ERA5" data? Also, do both cover the same period shown in Table 3?
4. In Figure 9 "ERA5" data is used, but as is stated in the manuscript, it is expected that ERA5 data is not able to resolve the inertial subrange of the power spectrum. It would be much more interesting to show how ERA5-corrected data perform in this scenario.
5. In Figure 4 and 5, shaded area (denotes standard error of the mean) and line (denotes the mean) with same color are used to represent results from one specific model, but their colors are not easy to distinguish from one and another.
Also, lidar measurements at different heights are connected with lines, which indicates a certain relation between these measurements. However, rigorously speaking this relation is not known. This principle also applies to plotted lines in the left panel in Figure 4 and 5 (which shows random forest model results). Do wind speed for each height is predicted (e.g., with 1m interval) or only some specific heights are predicted (e.g., heights listed in Table 1)?
6. In Figure 7 only RF trained results are shown, what about QRF ones?
Citation: https://doi.org/10.5194/wes-2023-178-RC2 - AC1: 'Comment on wes-2023-178', Farkhondeh Rouholahnejad, 09 Aug 2024
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 652 | 155 | 24 | 831 | 25 | 29 |
- HTML: 652
- PDF: 155
- XML: 24
- Total: 831
- BibTeX: 25
- EndNote: 29
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1