Wind Turbine Clutter (WTC) mitigation is a challenging issue that worldwide weather services are facing at the moment. The use of wind farms to generate electricity is growing increasingly due to the importance of being a renewable energy source. Wind farm installations relatively near to weather radars may block the normal propagation of the radar signals and cause reflectivity clutter returns.
Due to the motion of blades, with tip velocities near 90 m s-1, the standard clutter mitigation techniques fail to mitigate the effects of wind farms on radar products. Thus, this clutter and interference causes false estimates of rain accumulation and rain Doppler velocity.
Several studies addressing the problem in C-band and S-band Doppler weather radars have been published to date, with encouraging results, but most of them are oriented to WTC lessening in spotlight operation mode; that is, with the antenna dish stationary.
In operational radars these techniques are useless, as the radar antenna is constantly rotating in a surveillance operation mode. Some studies propose interpolation in order to remove WTC from reflectivity and velocity plan position indicators (PPIs), but a previous knowledge about the location of wind turbines is required.
A fixed wind turbine clutter map has two main disadvantages. Firstly, wind farms can occupy distances of more than 50 km, where inter-turbine distances can be smaller or not than the radar range resolution. This can cause an important loss of valuable weather information. Secondly, wind turbines are not always running. The blades only move when the velocity of the wind is appropriate for energy production. Then, on-the-go detection is significantly more efficient for weather data accuracy.
In this work, a holistic approach to wind turbine clutter detection is performed. The detection is based on feature extraction. Although the technique is employed in surveillance operation mode, the experience with spotlight data will be used, as it provides the information about the behaviour of WTC in time. Then, several parameters will be used as indicators of WTC presence in radar data. Both simulated and recorded radar data from clutter and weather will be used for this purpose.
Wind Turbine Clutter effects
A typical wind turbine is made up of three components: the tower, the nacelle and the rotor. The tower means a constant zero velocity return that can be easily minimised by means of an appropriate clutter cancellation. The blades can have radial velocities up to 30 or even 50 rpm (Freris, 1992), and these returns can’t be mitigated with conventional clutter filters. To sum up, the main effects wind turbines have on radars are the following (Perry & Biss, 2007):
• The magnitude of the reflections from the wind turbine can cause radar receivers to be driven to saturation. In fact, typical tower heights reach 100m, with blades from 30 to 50m long. See Table 1 (Gamesa, 2009)
• The rotation movement of blades cause Doppler shifts. The velocity of a blade depends on the distance from the centre, therefore, there is an increasing Doppler shift from the centre to the tip of the blade. This spectrum can fall within the limits of some radars or exceed them
In weather radars (Vogt et al, 2008), the clutter, signal blockage and interference may cause the misidentification of thunderstorm features and meteorological algorithm errors such us false radar estimates of precipitation accumulation, false tornadic vortex and mesocyclone signatures, and incorrect storm cell identification and tracking.
Wind Turbine Clutter mitigation efforts
Wind Turbine Clutter is unpredictable. It can fluctuate from one scan to the following. The blades rotate at such a rate to produce Doppler shifts of 70 or even 90 m s-1. These values can exceed the maximum non-ambiguous Doppler velocity of some radars and then make WTC detection and mitigation more difficult.
Apart from processing techniques, stealth solutions are also being studied to reduce the problem (Matthews et al, 2007). These techniques try to develop radar absorbing materials as well as to design new wind turbines with reduced radar cross section, preserving the efficiency of turbines in terms of electricity production and construction costs. The main inconvenience of these solutions is that the materials employed might only be efficient for very specific radar frequency bands.
Some of the mitigation efforts are focused on the prevention of this clutter (Donaldson et al, 2008). The assessment for new wind farms should be planned taking into account nearby weather radars by using line of sight calculations, standard 4/3 radio propagation model and echo statistics. Already built wind farms, however, are still distorting weather radars, and in such circumstances specific processing is needed.
Spotlight operation mode offers more information about the behaviour of WTC so that easier detection and mitigation algorithms can be used such us median filtering (Isom et al, 2007) or imaging techniques (Gallardo- Hernando et al, 2008). In normal operation modes, however, with the antenna rotating at a minimum of 2rpm, the dwell time is not long enough to let WTC specific features be observed. The Doppler resolution is very poor and more sophisticated algorithms are needed. Adaptive arrays (Palmer et al, 2008) and new signal processing techniques such as the use of polarimetric signatures or super-resolution (Gallardo-Hernando et al, 2010a) have to be studied for this purpose.
In this work we propose a technique to detect WTC in scanning radar tasks. The most significant WTC features are used: high zero Doppler returns and broad Doppler shifts.
Wind Turbine Clutter characterisation
In this work we used experimental data from a C-band weather radar near Palencia, Spain. See Table 2 for detailed radar parameters. Up to three different wind farms can be seen in a narrow sector between 30 and 45km away from the radar. The main wind farm is composed of 54 wind turbines model G-58 (Gamesa, 2009), which provide an average power of 49300 kW. In Figure 3, the data gathered on a clear day (a) and a rainy day (b) are represented. The three wind farms can be easily distinguished, but not the individual wind turbines in every case. Notice that the raw matrices are plotted and how the wind turbines, as point targets, are expanded along pulses in slow-time.
By calculating the Doppler spectrum, defined as the power weighted distribution of radial velocities within the resolution volume of the radar (Doviak & Zrnic, 1984), for each azimuth angle, for a particular range gate, the spectral content versus the azimuth angle can be studied.
I-Q radar data were gathered with the slowest antenna velocity, the lowest elevation angle (the most affected by the presence of wind farms) and the highest pulse repetition frequency (PRF). The spectral content of several range bins has been studied using a Short Time Fourier Transform (STFT) of partially overlapped time sectors to build a spectrogram.
A Hamming window was used in order to diminish windowing effects. There are a limited number of coherent pulses that can be used for this purpose; that is, an n-point window has to be used, ‘n’ being the number of points a punctual reflector would expand when being illuminated by the radar, n = TI ·PRF = θB ·PRF/Vantenna, where θB is the azimuth beam width and Vantenna is the rotation velocity of the radar antenna.
An example has been represented in Figure 2(a). There was an isolated windmill in the selected range gate, so, the spectrum is located at a very specific azimuth angle. This spectrum is extremely wide, as some of its components seem to be overlapped. Figure 2(b) shows the same range bin in rain. Both spectra are clearly different.
In order to understand the physical behaviour of WTC and also to have more information about WTC features to use in a detection algorithm, WTC time series were simulated (Gallardo-Hernando et al, 2010b). These simulations were generated to reproduce the real acquisition scenario as accurately as possible. Hence, the real parameters from wind turbines (Table 1) and radar (Table 2) were used. Figure 3 shows the 3D acquisition scenario. A rotating radar (equipped with a directive antenna in azimuth and elevation) illuminates a single wind turbine, composed of a tower and three rotor blades.
An individual wind turbine is considered to be composed of three lineal elements representing the tower and the blades. Each element is simulated by a certain number of scatterers situated along a line (the higher the number the more accurate the simulation is). Then, considering the transmitted radar signal and the geometry in Figure 3, the radar return can be computed for each scatterer.
Since both the wind turbine and the radar are moving, these computations are made under the stop-and-go supposition; that is, for every computation, the radar and the turbine are static, but for the following, their positions have changed according to their individual movements.
Figure 4(a) shows an example of a simulation. In this case, we tried to reproduce the real case plotted in Figure 4(b). No noise was added because the goal was to understand the nature of this clutter as well as to observe its main features: large zero Doppler return and broad spectrum width.
Weather data were also simulated for this purpose. We used the method described in Zrnic, 1975 to generate weather-like spectra. The simulated data were added to the real WTC data. This allows us to have an estimation of how badly WTC affects rain estimates such as reflectivity and velocity. An example is plotted in Figure 5 (Gallardo- Hernando et al, 2009). The WTC dataset is the same from section 2, whereas the rain is simulated. Figure 5 shows the estimates for reflectivity and velocity for rain only and for the addition of real WTC and simulated rain for a single azimuth angle in range.
Detection of WTC
Any mitigation scheme can be applied for an a priori known WTC map, but automatic detection patterns are also a need. A fixed wind turbine clutter map has two main disadvantages. Firstly, wind farms can occupy more than 50km long where inter-turbine distances can be smaller or not than the radar range resolution. This can cause an important lost of valuable weather information. Secondly, wind turbines are not always running. The blades only move when the velocity of the wind is the appropriate for energy production. Then, on-the-go detection is much more efficient for weather data accuracy.
In this section, a novel detection algorithm in the frequency domain is explained. The parameters of the radar data gathering are detailed in Table 2, and the section under study is represented in Figure 1.
This detection algorithm is based in the most visible features of WTC: large radar cross-section and large Doppler shifts. These features have been extensively studied in real and simulated data, as we have seen in the previous section. The detection is performed in the Doppler frequency domain, zero-velocity components are studied first, and then, information about the spectrum width is taken into account.
1. For each range bin of the raw data radar matrix (I-Q data): analyse the spectral content (STFT). Use an n-point running window, ‘n’ being the maximum of coherent pulses available. Using a running window simplifies the detection of spectrum peaks.
2. Build a zero-Doppler range-azimuth matrix. The zero-Doppler power in a WTC range-azimuth bin will always be very high, as we explained earlier.
3. Apply a threshold to obtain the positions of possible WTC affected bins. This threshold has two components: power (bins under the threshold are rejected), and length (bins with zero-Doppler components in a wide azimuth area are expected to be ground, but not wind turbines). The threshold was set with the aid of the simulated data. A point target should expand in an a priori known number of pulses, but also adjacent wind turbines are considered, and they would expand in a higher number of pulses. Applying step 4 solves this problem.
4. Analyse the spectrum width for the bins of step 3. Apply a spectrum width threshold to confirm WTC. Again, this threshold was set using simulated data.
5. Now, once the contaminated bins have been detected, interpolation or other mitigation techniques can be used. For testing purposes this algorithm has been employed in both clear air and rainy conditions. In this paper the latter is exemplified.
Example of WTC detection
Figure 1(b) shows the matrix under study. Wind turbine clutter and rain are mixed in certain areas. No processing had been applied so far.
Each range bin is frequency processed separately with a running window. An example of a range bin Doppler spectrum is plotted in Figure 2(b).
Zero-Doppler frequency bins are separated and a range matrix is built. See Figure 6.
Now, these range bins are passed through a power-azimuth threshold. Notice that although rain spectrum can have zero-Doppler content, its shape will differ substantially from a wind turbine.
Once the preliminary WTC positions are plotted, see Figure 7(a), the spectrum width in each detection can be examined in order to confirm clutter presence.
Finally, the detections are plotted over a PPI representation of the data. In Figure 7(b) clutter cancellation in time had been previously used to remove ground clutter. It can be appreciated how the number of false alarms has decreased from Figure 7(a) to Figure 7(b).
We have presented a novel method to detect wind turbine clutter in scanning weather radar tasks. The algorithm is based on the most significant spectral features of wind turbine clutter, which allows its differentiation from other kinds of clutter and weather signals. We have also shown the performance of this algorithm in real data. The data contained both weather and wind turbine clutter.
This article was a paper originally presented by its authors to delegates of the Sixth European Conference on Radar in Meteorology and Hydrology (ERAD), held in Romania in September, 2010. Donaldson, N, Best, C & Paterson, B (2008). Development of Wind Turbine Assessments for Canadian Weather Radars. Proceedings of the Fifth European Conference on Radar in Meteorology and Hydrology, ERAD 2008, pp 1-4, Helsinki, Finland, July 2008. Doviak, RJ, & Zrnic, DS (1984). Doppler Radar and Weather Observations, Academic Press, Inc. Orlando, Florida. USA. Gallardo-Hernando, B, Pérez-Martínez, F & Aguado-Encabo, F (2008). Characterization Approach of Wind Turbine Clutter in the Spanish Weather Radar Network. Proceedings of the Fifth European Conference on Radar in Meteorology and Hydrology, ERAD 2008, pp 1-5, Helsinki, Finland, July 2008. Gallardo-Hernando, B, Pérez-Martínez, F and Aguado-Encabo, F (2009): Mitigation of Wind Turbine Clutter in C- band Weather Radars for Different Rainfall Rates. Proceedings of the 2009 International Conference on Radar, Bordeaux, France, October 2009. Gallardo-Hernando, B, Muñoz-Ferreras, JM, Pérez-Martínez, F and Aguado-Encabo, F (2010a): Super-resolution techniques in meteorological radars: the example of wind turbines. Proceedings of the 2010 International Conference on Radar, Washington, May 2010. Gallardo-Hernando, B, Muñoz-Ferreras, JM, and Pérez-Martínez, F: (2010b) Wind turbine clutter observations and theoretical validation for meteorological radar applications. IET Radio Sonar and Navigation. In press. Gamesa. G58-850 Wind turbine. http://www.gamesa.es/ Freris, L (1992). Inherit the Wind. IEEE Review, Vol 38, No. 2, April 1992, pp 155-159 Isom, BM, Palmer, RD, Secrest, GS, Rhoton, RD, Saxion, D, Winslow, JL, Reed, J, Crum, T, & Vogt, R (2007). Characterization and mitigation of wind turbine clutter on the WSR-88D network. Proceedings of the 33rd Conference on Radar Meteorology, August 2007, paper 8B.8. Matthews, J, Pinto, J & Sarno, C (2007). Stealth Solutions to solve the radar-wind farm interaction problem. Proceedings of the 2007 Antennas and Propagation Conference, pp 101-104, Loughborough, UK, April 2007, Palmer, RD, Le, K. D & Isom, B (2008). Wind Turbine Clutter Mitigation using Fully Adaptive Arrays. Proceedings of the 5th European Radar Conference, pp 356-359, Amsterdam, The Netherlands, October 2008. Perry, J & Biss, A (2007). Wind farm clutter mitigation in air surveillance radar. Aerospace and Electronic systems Magazine, IEEE, Vol. 22, No 7, July 2007, pp 35-40.. Vogt, R; Reed, J; Crum, T; Snow, J; Palmer, R; Isom, B & Burgess, D. (2008). Impacts of wind farms on WSR-88D operations and policy considerations. Preprints, 23rd Int. Conf. on Interactive Information Processing Systems (IIPS) for Meteorology, Oceanography, and Hydrology, San Antonio, TX, Amer. Meteor. Soc. Zrnic, DS, (1975). Simulation of weatherlike Doppler spectra and signals. J. Appl. Meteorol, 14, 619-620.
Published: 27th Nov 2013 in AWE International