Testing the Performance of a Ground-based Wind LiDAR System One Year Intercomparison at the Offshore Platform FINO1

Testing the Performance of a Ground-based Wind LiDAR System One Year Intercomparison at the Offshore Platform FINO1 B. Cañadillas, A. Westerhellweg, T. Neumann; DEWI GmbH, Wilhelmshaven B. Canadillas English Introduction Traditional meteorological masts that are commonly used in wind energy applications are limited and very expensive to install in offshore locations. Therefore, it is necessary to search for an alternative method to replace the standard wind measurements (cup/sonic anemometers, vanes) used on a mast. The LiDAR (Light Detection And Ranging) technique has become a reasonable alternative in the last years with the advantage that higher vertical resolution measurements over the whole rotor diameter of a wind turbine can be performed [1], [2]. However, before taking lidar systems as an accepted wind measurement alternative, we have to validate as well as understand how this technique computes both mean and turbulent wind parameters. One important issue is the time resolution at which the wind measurements have to be sampled which will depend mostly on the application for which the wind measurements will be used. For instance, for wind resource and power curve assessments, a 10-min interval is recommended by the standards [3], [4]; however several investigations pointed out that a higher resolution (1 Hz or even less) is needed for a better definition of the power curve [5]. In addition, monitoring the rapid changes in the wind conditions is important for optimal operation of wind farms. In the last years, several field experiment intercomparisons with in-situ sensors located on meteorological masts have proven the lidar system to be a reliable instrument for obtaining mean values (10-min). In particular, similar lidar systems as the one used in this study (Leosphere WindCube) have been tested over homogeneous terrain (e.g., [6], [7]) with excellent results. However such comparisons have only relied on 10-min averaged data. Within the joint research project RAVE (Research at AlphaVEntus), the RAVE-LIDAR work package deals with remote sensing measurements using lidar systems at the German Offshore test site [8]. Part of the work carried out by DEWI is presented in this article. This study aims to examine the performance of a wind lidar measurement located at FINO1 site, using the mast to provide reference measurements. The present study is split into two major parts. In the first part, an intercomparison between lidar and conventional mast-based instruments, such as cup/sonic and vanes, is performed based on 10-min data. Several wind parameters are considered, namely wind speed, wind direction and turbulence intensity. In the second part, we complement the 10-min data intercomparison with a spectral analysis in order to get a more complete picture of the flow in the ranges between one hour and about one second. 58 DEWI MAGAZIN NO. 38, FEBRUARY 2011

The Field Measurement Campaign A ground-based pulsed lidar system, the so-called WindCube, developed and manufactured by the French company Leosphere has been used in this study. The lidar system was positioned on a container roof at approximately 10 m distance to the northwest of the offshore research mast FINO1 (Fig. 1) and performs continuous measurements since July 2009. The FINO1 platform (lat. 54 0.86 N, long. 6 35.26 E), equipped with a 100 m lattice tower, is operating since September 2003 in the German Bight approx. 45 km to the northwest of the island of Borkum. In this study, the FINO1 mast is used as ground truth to provide a reference to which the lidar data shall be compared. The mast instrumentation used here (Tab. 1) are cup anemometers at 101.5 m (top-mounted) and at 90 m, 80 m and 70 m located on the South-East side of the mast, ultrasonic sonic anemometer at 80 m and wind vanes at 90 m and 70 m located on the northwest side of the mast. Slow-response profile instrumentation is sampled once per second (1 Hz) and stored as 10 min averages. Fastresponse instrumentation is sampled with a frequency of 10 Hz. Moreover, east of FINO1 the Alpha Ventus offshore wind farm is located at a distance of 405 m to the nearest wind turbine. The farm consists of 6 REpower 5M wind turbines with a hub height of about 92 m LAT and a rotor diameter of 126 m and 6 Areva Wind M5000 wind turbines with a hub height of about 91 m LAT and a rotor diameter of 116 m. The first wind turbine was installed on 2009-07- 15 and went into operation on 2009-08-12. However, for the time period used in this study not all wind turbines were in operation at the same time, so that wind coming from the east may be disturbed by those wind turbines for some periods. The WindCube Lidar System A Doppler wind lidar used for atmospheric purposes, measures the frequency shift (Doppler shift) of narrow bandwidth laser light backscattered from microscopic particles or aerosols in the air being transported by the wind. The frequency shift is proportional to the velocity of the scattered target along the propagation path of the laser beam or Lines-Of-Sight (LOS) which is estimated with a specific signal processing algorithm implemented in the device. The range to the target is determined by the time of traveling back and forward. The WindCube Doppler lidar system used in this study emits 200 ns pulses of 10 μj energy at a wave length of 1.54 μm [9]. There is a direct interface to the FINO-network, which allows the data transfer, as well as the control of the device by using remote control software. The measurements were set-up at heights from 70 to 101 m every 10m, from 100 to 200 m every 20 m and at 250 m. The lowest height is 70 m because the height of the lidar device is already 23 m msl and the Windcube has a minimum measurement height of 40 m. From lidar measurements, the estimation of the wind velocity vector (u, v, w) requires retrieval of the LOS velocity (also called as radial velocity v r ). The Windcube, located at the FINO1 platform, uses the Velocity- Azimuth-Display (VAD) technique of lidar scanning (conical scan lidar beam at a fixed elevation angle) to derive the 3D components Power Quality DEWI carries out measurements and evaluations to determine the electrical characteristics of single wind turbines and of wind farms according to the currently applicable standards (e.g. IEC 61400-21) and is accredited by German Accreditation Council in line with EN ISO/IEC 17025:2005 and by MEASNET. As one of the leading international consultants in the field of wind energy, DEWI offers all kinds of wind energy related measurement services, energy analyses and studies, further education, technological, economical and political consultancy for industry, wind farm developers, banks, governments and public administrations. DEWI is accredited to EN ISO/IEC 17025 and MEASNET for certain measurements and is recognised as an independent institution in various measurement and expertise fields. DEWI MAGAZIN NO. 38, FEBRUARY 2011 59 www.dewi.de

Fig. 1: Lidar position at the offshore research platform FINO1 Sensor Type Height msl [m] cup, vane (Vector A100 LK, Thies Wind Vane Classis) 70 m cup, sonic (Vector A100 LK, Solent 1210R3-50) 80 m Tab. 1: Overview of the FINO1 measurements involved in the intercomparison cup, vane (Vector A100 LK, Thies Wind Vane Classis) Cup (Vector A100 LK) 90 m 101.5 m Z w Vr90 v Vr0 Vr180 u Vr270 Y X Fig. 2: Sketch of the scan method performed by the WindCube lidar Fig. 3: Data availability at FINO1 for the period from 2009-08-01 to 2010-07-31 Fig. 4 Wind speed ratio lidar/cup (curves shifted). Sector between black vertical lines is the sector used for the intercomparison. Red circular areas show approx. the mast shadow on the lidar measurements 60 DEWI MAGAZIN NO. 38, FEBRUARY 2011

Height msl [m] Tab. 2: (y = m x) (y = a x +b) Bias (v lidar -v cup ) m [-] a [-] b [m/s] R 2 [m/s] C S Counts [-] Remarks 101.5 0.998 0.997 0.010 0.998-0.02 0.13 0.13 6770 dir sector [190;240] 90 0.988 0.991-0.030 0.997-0.13 0.22 0.18 6831 dir sector [190;240] 80 0.992 0.997-0.051 0.997-0.08 0.19 0.17 6928 dir sector [190;240] 80-USA 1.003 1.003 0.007 0.996 0.03 0.19 0.19 6704 dir sector [190;240] 70 0.997 1.000-0.036 0.997-0.03 0.16 0.16 6981 dir sector [190;240] Statistics of the filtered horizontal wind speed. Fig. 5: Scatter plot of horizontal wind speed of lidar against sonic measurements at 80 m Fig. 6: Scatter plot of lidar wind direction against sonic wind direction at 80 m Height [m] Tab. 3: (y = a x +b) Bias (dir lidar -dir cup ) a [-] b [ ] R 2 [ ] C S Counts Remarks 90 1.013-0.106 0.995 2.7 2.87 0.18 6831 dir sector [190;240] 80-USA 1.019-2.140 0.996 1.9 2.12 0.19 6667 dir sector [190;240] 70 1.010-1.455 0.995 0.8 1.30 0.16 6928 dir sector [190;240] Statistics of the filtered wind direction of the wind. This technique is based on the following radial velocity equation: v r =u cos(θ) sin(φ) + v sin(θ) sin(φ) + w cos(φ) where u, v and w are the wind vector components and θ and φ are the azimuthal and zenithal angles of the wind vector. The WindCube lidar performs four successively radial velocity measurements at azimuth-angle intervals of 90 around the circle formed by a conical scanning, i.e. at θ=0, θ=90, θ=180 and θ=270 and at a fixed elevation angle φ, as illustrated in Fig. 2. The limitations of this approach include the assumption of horizontal homogeneity of the wind field over the sensed height. Its temporal resolution for acquiring a full 3-D wind vector is about 4.6 sec. (0.22 Hz), i.e. each revolution takes about 4.6 seconds. However, the last three radial velocity values plus a new measured value are used to derive wind speed and direction profiles. Experimental Results Intercomparison Based on 10-min Averaged Values In this section, an intercomparison study based on 10-min values of the measured quantities is performed. The variables used to describe and compare the sensors are wind speed, wind direction and standard deviation of the horizontal wind speed. In order to determine how well both data groups (lidar and FINO1) are statistically related to each other, a correlation analysis is carried out. Moreover, the criteria used to statistically describe the accuracy and precision of the lidar in comparison to a reference sensor are: Sample bias (B), Comparability (C), Precision or standard deviation of the differences (S). Data Availability The evaluated data cover a one year measurement period from 01 August, 2009 to 31 July, 2010. The availability of the different measuring instruments at the different heights is shown in Fig. 3. For the cup anemometers it is close to 100 % at all heights. The sonic anemometer shows a lower availability with approx. 97.5 %. For the lidar two different curves are displayed. The overall availability of the lidar is given by the blue line, i.e. here, all 10-min mean values were included in which at least one single value exists. The black graph shows the availability considering only 10-min data DEWI MAGAZIN NO. 38, FEBRUARY 2011 61

Fig. 7: Scatter plot of lidar wind standard deviation against sonic (left) and turbulence intensity (right) against mean wind speed at 80 m with 100% availability. At heights comparable with the FINO1 data the lidar availability amounts to 98 %. The availability decreases with increasing height to 91 % at a height of 200 m, respectively 83 % at 250 m. This is due to the decreasing of the signal to noise ratio with height. Mast Shadow on Measurements In Fig. 4 the ratio of wind speed measured by the lidar and cup as a function of the wind direction at 90 m is depicted for all compared heights. The curves are shifted vertically artificially for better visualization. As can be seen, the mast shadow effects on the anemometers and laser beams are clearly noticeable at certain wind directions. The plot shows mast shadow effects on the lidar beams at approximately 40, 110, 180 and 270. These are approximately the directions where one of the four lidar beams is on mast shadow. These directions are slightly shifted for different sensing heights. Moreover the graphs show the lightning cage (90, 180, 270 and 360 ) for the height at 101.5m and the upwind and downwind flow retardation (280-340 ) for heights at 90 m, 80 m and 70 m. The shadow effect from the neighbouring wind turbines from Alpha Ventus wind farm is also appreciable. Data Filtering In order to ensure the data comparability between the lidar and reference instruments, the following filtering criteria are applied before the analysis: Rain: Events with rain periods are discarded. Lidar availability: Only lidar data with an availability of 100% for each 10 min value are used. Signal to noise (CNR) threshold: Values with CNR < -22 db are discarded. Wind speed: Only mean wind speeds in the range of 4-16 ms -1 are considered in the comparison. This corresponds to the range where the cup/sonic anemometers were calibrated. Wind direction: For the wind speed comparison and for the heights of the boom mounted anemometers only wind directions within the range interval 190-240 are used which avoids the wakes from the wind turbines, and the direct mast shade on the anemometers and lidar beams (see Fig. 4). After filtering the 10-min dataset is reduced to 6981 values (13%, direction sector 190-240, 70 m msl). Wind Speed Values of correlation coefficient, slope, offset, bias, comparability, precision and number of observations for all heights are tabulated in Tab. 2. At each of the compared heights a scatter plot of 10-min averaged horizontal wind speed is performed, however, as an example, only the one at 80 m is displayed in Fig. 5 (left). Two versions of the linear regression analysis (with (y=ax+b) and without (y=mx) offset) are considered. Correlation analysis suggests that the wind speeds from the lidar system agree well with the tower measurements with values of R² higher than 0.99 for all heights. The estimates of the biases show that at almost all heights (except for the sonic) the lidar wind speed tends to be smaller than the tower wind speed (negative bias). Values of C and S are similar and fall between 0.19 and 0.13 ms -1. Considering the different sampling and averaging strategies of the two measurement systems, very good agreement can be noticed. Wind Direction Tab. 3 summarizes the results of the wind direction comparison. Biases for all heights are positive, which could be attributed to an inaccuracy in the north alignment during the lidar setup. The comparison of the wind direction measurements shows a good correlation between lidar and vanes with a coefficient of correlation higher than R² = 0.99. A Scatter plot at 80 m, using the sonic anemometer to measure the wind speed direction, is depicted in Fig. 6. Wind Standard Deviation Before the computation of the turbulence intensity (TI=σ U /U), comparisons of the 10-min averages of the standard deviation of the horizontal wind speed (σ U ) at all investigated 62 DEWI MAGAZIN NO. 38, FEBRUARY 2011

Run-mean Run-mean filter Fig. 8: Comparison (sonic (blue) versus lidar (red)) of horizontal wind speed (left) and radial velocity (right) spectra at 80m. In this example (right), the sonic data were projected onto a vector aligned into the 90 beam direction heights are carried out. A Scatter plot for the sonic at 80 m is depicted in Fig. 7 (left). Unlike the wind speed comparison, the scatter of the standard deviation is larger. The WindCube Lidar system measures higher values of σ U than the cup anemometers at all observing heights (positive bias). In contrast, in [11] for instance, it was shown that using another lidar system (ZephIR) smaller values of σ U (approximately 20%) are retrieved as a consequence of the spatial and temporal averages applied to the lidar measurements. The exact nature of this discrepancy found in this intercomparison has not yet been determined. Some possible reasons for the discrepancy include data processing technique. The curves represented in Fig. 7 (right) show the turbulence intensity as a function of the wind speed at 80 m height. The highest values for the turbulence intensity arise with very low wind velocities. There is a constant decrease up to the range of approx. 12 m/s. For higher wind speeds the turbulence intensity increases because the roughness of the sea increases. For all wind speeds however, the lidar measures turbulence intensity higher than the sonic anemometer. A quite similar picture results for all heights investigated. Spectral Analysis The 10-min comparison addressed in the previous section provides a measure of the mean characteristics of the turbulent flow where higher frequencies of the wind are filtered out, although there is considerable information about the higher frequency data with relevance to wind energy application including, among others, wind load studies and wind turbine design. Here the Power Spectrum Density (PSD) of wind velocity fluctuations is evaluated. We are trying to understand from the turbulence point of view how both measurements correlate to each other and to explain why turbulence measured with this LiDAR shows higher values with respect to insitu measurements. For this comparison study measurements from the sonic anemometers at 80 m are used. To perform the spectral analysis, the Welch algorithm has been used. For further information, please refer to [12]. The power spectrum from the lidar is compared to the power spectrum from the sonic anemometer at 80 m, and the results are illustrated in Fig. 8 (left). To enhance the readability of the spectra, a log-log scale is used. Note that the spectra represent an average over all observed wind speeds for the period selected. Atmospheric stability for this period was mostly near-neutral. Winds were coming from the northwest sector. In Fig. 8 (left), the cut-off frequencies differ because of different time resolutions of sonic (10Hz) and lidar measurements (0.67 Hz). The black dashed line represents the theoretical slope of a Kolmogorov inertial sub-range (f -2/3 ). The green dashed vertical line corresponds to a frequency of 1/600 (10min). The loop at the end of the lidar spectrum (about f> 0.21Hz) is a consequence of the temporal averaging algorithm which acts as a run-mean filter with a time window of about 4.6 sec. (time which laser beam takes to scan a full circle). Note that this lidar system cannot predict frequencies higher than about 0.21 Hz and, therefore, above that value it is better not to derive any conclusions from the spectrum. As can be clearly seen, the lidar spectrum presents a signal increase at frequencies between 0.004 and 0.2 Hz. The cause of this apparent energy increase is unknown at present, but could be related to instrument noise, differences in sampling methods, or algorithm methods to construct the velocity vector. However, it is shown in the left hand of this figure that there is a good agreement between both wind spectra at low frequencies. It is well known that the shape of the turbulence spectra depends among others on the thermal stratification, the height above the ground surface, and the wind velocity. In [12] an analysis according to different ranges of wind speed, wind direction or atmospheric stability was performed and it showed that any of these parameters seem to have a direct influence on the lidar spectra shape. One of the averaging mechanisms applied during the WindCube lidar measurement is the probe length which leads to an averaging of the wind component along the laser beam within a sampling width of 20 m. In Fig. 8 (right), we focus on the radial velocity intercomparison in order to avoid the algorithm methods used to derivate the horizontal wind DEWI MAGAZIN NO. 38, FEBRUARY 2011 63

Height [m] (y = a x +b) Bias (σ lidar -σ cup ) a [-] b [m/s] R 2 [m/s] C S Counts [-] Remarks 101.5 1.069 0.030 0.834 0.06 0.13 0.12 6770 dir sector [190;240] 90 1.043 0.028 0.841 0.05 0.12 0.11 6831 dir sector [190;240] 80 1.034 0.028 0.851 0.05 0.12 0.11 6928 dir sector [190;240] 80-USA 0.973 0.053 0.789 0.04 0.14 0.13 6704 dir sector [190;240] 70 1.024 0.027 0.867 0.04 0.11 0.10 6981 dir sector [190;240] Tab. 4: Statistics of the filtered standard deviation of the horizontal wind speed. speed and thus to check whether the anomalous shape of the lidar spectra at certain frequencies can be a consequence of this velocity derivation algorithm or not. In order to compare the radial velocity along the laser beam, the 10Hz sonic data are projected onto a vector aligned with the same azimuth and elevation angle as the lidar beam (line-of-sight). Fig. 8 (right) suggests that there is a good agreement between both radial wind spectra which suggests that the probe-length averaging seems not to have an influence on the shape of the horizontal wind lidar spectrum. Conclusion A Lidar measurement has been performed over a period of a whole year with an availability of 98% for the heights involved in the intercomparison, proving that lidar can be a reliable wind measurement device even in harsh offshore weather conditions. Even for the highest measurement level of 250m and availability of 83% could be achieved, which proves that the lidar device is an adequate instrument to span the whole rotor area of modern offshore wind turbines. In general good agreements are found between both data sets when comparing 10 min averaged measurements. The comparison of the 10 min lidar wind speed data with mast data shows a high correlation with R 2 higher than 0.99 for all heights. One important result is that the mean wind deviations between lidar and cup measurements are smaller than the uncertainties of the mast measurements. In contrast to the onshore case, the offshore wind conditions are governed by small wind shear and small turbulence intensity, which are ideal conditions for the lidar technique in VAD mode, which depends on homogeneity in the spatial and temporal scanning area. Hence the mean wind speed and wind direction can be measured with high quality; however the regarded lidar type (Leosphere Windcube) overestimates the turbulence intensity at the present stage. The lidar spectra present a signal increase at certain frequencies, which is independent on wind direction, wind speed or stability conditions. The cause of this apparent energy increase is unknown at present, but could be related to approximations in sampling or algorithm methods to construct the velocity vector. For the global turbulence this effect leads to a systematic overestimation. On the other hand, the radial wind spectra comparison suggests that the probe-length averaging seems not to have an influence on the shape of the horizontal wind lidar spectrum. In this study the processing techniques used by the WindCube lidar to derive the horizontal wind speed have not been analyzed. It is beyond the scope of this study to find out whether the lidar algorithms are responsible for the spectra behavior and/or can be optimized to compensate this error. Acknowledgements Within the joint research project RAVE (Research at AlphaVEntus), the RAVE-LIDAR project was set up to prepare the application of remote sensing measurements using lidar systems at the German Offshore test site [8]. Part of the work that is carried out by DEWI is presented in this report. We also like to thank Leosphere for their support. References: [1] Courtney, M., R. Wagner, und P. Lindelöw. Commercial lidar profilers for wind energy. A comparative guide. EWEC. 2008. [2] Emeis, S., M. Harris, und R. M. Banta. Boundary-layer anemometry by optical remote sensing for wind energy applications. Meteorologische Zeitschrift 16, Nr. 4 (2007): 337-347. [3] IEC61400-1. Wind turbines-part 1: Design requirements. Technical report, International Electrotechnical Commission (IEC). [4] IEC61400-12-1. Power performance measurements of electricity producing wind turbines. 2005. [5] Anahua, E., M. Lange, F. Boettcher, S. Barth, und J. Peinke. Characterization of the Wind Turbine Power Performance Curve by Stochastic Modeling. EWEC. Athens, 2006. [6] Albers, A., and A. Janssen. Windcube evaluation report. Technical report, Deutsche WindGuard Consulting GmbH, 2008. [7] Gottschall J., and M.l Courtney. Verification test for three WindCubeTM WLS7 LiDARs at the Høvsøre test site. Technical report, Risø-R-1732(EN), 2010. [8] Rettenmeier, A., et al. Development of LiDAR measurements for the German Offshore Test Site. 14th International Symposium for the Advancement of Boundary Layer Remote Sensing. IOP Conf. Series: Earth and Environmental Science, 2008. [9] Pauliac, Romain. WindCube. User s Manual. Leosphere, June 2008. [10] Mann, J, et al. Comparison of 3D turbulence measurements using three staring wind lidars and a sonic anemometer. Risø-R-1660(EN), 2008, 135 140. [11] Rozenn Wagner, Torben Mikkelsen, Michael Courtney. Investigation of turbulence measurements with a continuous wave, conically scanning LiDAR. EWEC. 2009. 10 [12] Canadillas, B., A. Bégué, T. Neumann. Comparison of turbulence spectra derived from LiDAR and sonic measurements at the offshore platform FINO1. DEWEK. 2010. Germany. 64 DEWI MAGAZIN NO. 38, FEBRUARY 2011

Werbung Kunde 1/1 s/w oder 4c 200m ultra-portable wind profiler Location 3 Period 3 Location 1 Period 1 Location 2 Period 2 SITE ASSESSMENT > PRE-EVALUATION > MICRO-SITING > POWER CURVE VERIFICATION Location 4 Period 4 200m 60m The original WINDCUBE is reducing uncertainty worldwide with units on every continent. You asked for the same comprehensive wind profiles, only from a small, lighter, lower-power unit. We listened. The new WINDCUBE v2 lidar remote sensor offers: Ultra-portable, easy to deploy, lowest cost of ownership Bankable data, to reduce your project financing costs Complement met mast data with variable measurement heights up to 200m Unattended, low-power, quiet operation in any weather or noise condition even off shore Satellite communications from anywhere in the world Download Free ROI Papers. www.lidarwindtechnologies.com/roi LidarWindTechnologies, S.A. Our global partnership provides sales, technical support and customer service worldwide, wherever your wind site location. Live demonstrations EWEC AWEA CanWEA