Wind Turbine Sound in Cold Climates

(1)

UNIVERSITATISACTA UPSALIENSIS

UPPSALA

Digital Comprehensive Summaries of Uppsala Dissertations from the Faculty of Science and Technology 1784

Wind Turbine Sound in Cold Climates

KRISTINA CONRADY

ISSN 1651-6214 ISBN 978-91-513-0601-8

(2)

Dissertation presented at Uppsala University to be publicly examined in Axel Hamberg salen, Geocentrum, Villavägen 16, Uppsala, Monday, 13 May 2019 at 10:00 for the degree of Doctor of Philosophy. The examination will be conducted in English. Faculty examiner: Associate dean, senior lecturer Sabine von Hünerbein (University of Salford).

Abstract

Conrady, K. 2019. Wind Turbine Sound in Cold Climates. Digital Comprehensive Summaries of Uppsala Dissertations from the Faculty of Science and Technology 1784. 47 pp. Uppsala:

Acta Universitatis Upsaliensis. ISBN 978-91-513-0601-8.

The increase in the number of wind turbines (WTs) in populated areas in cold climates increases the number of people potentially being affected by WT sound. Outdoor sound propagation is strongly dependent on meteorological conditions, however, limitations in the knowledge exist regarding the implications of meteorological conditions in cold climates. Long-term acoustic and meteorological measurements were conducted in the vicinity of two wind farms in northern Sweden, to investigate the effect of snow and low-level wind maxima on WT sound, to analyse the occurrence of amplitude modulation and to evaluate selection methods for WT sound measurements. Different selection methodologies were applied to the acoustical data. The simplest method only includes a minimum rotational frequency of the WTs, while the most comprehensive method additionally includes criteria based on spectral resemblance, temporal variation of the sound level, amplitude modulation and wind speed. The effect of snow on WT sound depends on the snow quality. Snow on trees lowers the sound level by ca. 2 dBA.

Low-level wind maxima below hub height reduce the sound level near the surface. Since this effect is increasing with increasing strength of the low-level wind maximum, the WT sound is assumed to be partly trapped above the low-level wind maximum. Amplitude modulation was shown to be dependent on atmospheric stability and was most common for very stable conditions. Moreover, a clear difference between the occurrences of amplitude modulation for the two crosswind sectors was observed. The choice of selection method needs to be taken into account when comparing different studies since it affects the results and conclusions. The studies emphasise to include the effects of individual meteorological conditions of a site in the formulation of guidelines on WT sound.

Keywords: atmospheric acoustics, wind turbine sound, outdoor sound propagation, cold climates

Kristina Conrady, Department of Earth Sciences, LUVAL, Villav. 16, Uppsala University, SE-75236 Uppsala, Sweden.

urn:nbn:se:uu:diva-379554 (http://urn.kb.se/resolve?urn=urn:nbn:se:uu:diva-379554)

(3)

”Never be limited by other people’s limited imaginations.”

Mae Jemison

(4)

(5)

List of papers

This thesis is based on the following papers, which are referred to in the text by their Roman numerals.

I Conrady K, Sjöblom A, Larsson C. 2018. Impact of snow on sound propagating from wind turbines. Wind Energy, 21:1282–1295.

II Conrady K, Bolin K, Sjöblom A, Rutgersson A. 2019. Amplitude modulation of wind turbine sound in cold climates. Under review.

III Conrady K, Bolin K, Sjöblom A, Rutgersson A. 2019. Selection criteria for ﬁltering wind turbine sound measurements. Manuscript.

IV Conrady K, Bolin K, Sjöblom A, Rutgersson A. 2019. Impact of low-level wind maxima on wind turbine sound propagation. Submitted.

Reprints were made with permission from the publishers.

The author had the main responsibility for the data analysis, structuring, as well as writing the manuscripts of Paper I, II, III, and IV. The author was partly responsible for setting up instrumentation for the measurements used in Paper II, III, and IV. The ideas for Paper I and III were developed in collab- oration with the co-authors and the ideas for Paper II and IV mainly by the author.

(6)

(7)

1. Introduction

Wind energy has been expanding during the last years and a considerable part of the wind turbines (WTs) have been installed in cold climates (Wallenius and Lehtomäki, 2016). In the course of this expansion, the number of WTs and their heights have been increased. Residents in the vicinity of wind farms can be affected by WT sound (Michaud et al., 2016; Pedersen et al., 2009).

In several countries guidelines or recommendations exist, to limit the risk of humans being affected by WT sound. However, these guidelines are usually based on standardised meteorological conditions, which are often not applicable in cold climates. Usually, the standardised conditions do not represent the atmospheric boundary layer (ABL) if it is stably stratiﬁed, which is common in cold climates and in temperate climates during clear nights.

Refraction is one of the most significant meteorological effects on sound propagation (Larsson and Israelsson, 1991) and can be described as bending of the path of a sound ray due to gradients of the effective sound speed. Since the effective sound speed mainly depends on wind speed and temperature, refraction is caused by wind speed and temperature gradients along the ray path. A simplified concept has commonly been assumed and is based on a continuously increasing wind speed with height. Thus, downward refraction occurs downwind of a sound source and upward refraction occurs upwind of a sound source (e.g., Larsson and Israelsson, 1991; Salomons, 2001; Boué, 2007; Lamancusa, 2008). As a result, sound levels are higher in the downwind direction and lower in the upwind direction. Especially in cold climates, this simplified concept is often not applicable, due to commonly occurring low-level wind maxima (LLWM). Typical processes causing LLWM are e.g., inertial oscillations (e.g., Andreas et al., 2000), baroclinicity due to sloping terrain (e.g., Tuononen et al., 2015) and katabatic winds (e.g., Renfrew and Anderson, 2006). LLWM alter the commonly applied, simplified refraction pattern. If a LLWM above hub height is present, the downward refraction is enhanced. The LLWM traps the WT sound and the sound level close to the surface is increased in a certain distance from the WT (Boué, 2007). The impact of LLWM on sound propagating around offshore WTs has been investigated by Johansson (2003) and Törnblom (2006). A channelling effect in downwind direction was observed as well as increased sound levels near the surface. Consequently, LLWM below hub height lead to trapping of the sound waves above the LLWM and therefore, to lower sound levels close to the surface.

(10)

Another common feature in cold climates is snow. Snow is commonly assumed to attenuate sound due to its low acoustic impedance (e.g., Embleton et al., 1983). However, the existence of different snow types with varying porosity complexiﬁes the idea of the snow. Moreover, under speciﬁc circum- stances, snow remains on the vegetation and constructions, which is called upplega. The surface additionally covered by snow might impact the sound propagating from WTs. A damping effect of snow on sound propagation has been shown in several studies (e.g., Albert and Orcutt, 1990; Öhlund and Lars- son, 2015). However, snow metamorphism was not considered, and snow has been treated as one substance.

One characteristic of WT sound is amplitude modulation (AM). Several studies have analysed the impact of meteorological parameters on the occurrence of AM (e.g., van den Berg, 2004; Oerlemans and Schepers, 2009; di Napoli, 2011; Larsson and Öhlund, 2014; Paulraj and Välisuo, 2017). However, nei- ther the cause of AM nor which meteorological parameters impact AM is fully understood. AM is not particularly conﬁned to cold climates but it is a common feature.

The base of experimental studies on WT sound are long-term measurements.

However, these measurements are disturbed by sound from ambient sources, as for instance, wind-induced sound in the vegetation, at constructions or in the microphone, wildlife especially birdsong, human activity, etc. To avoid treating the ambient sound as WT sound, it has to be identiﬁed and excluded from the measurements. In previous studies, several selection methods were used. The selection methods were partly based on one criterion (van den Berg, 2004) or a combination of several criteria (Furuholm and Hultberg, 2013; Öh- lund and Larsson, 2015; da Silva, 2017). An evaluation of these methods is necessary to assess the quality of the selected data and the validity of results and conclusions.

The aim of the thesis is to gain knowledge on the impact of meteorological conditions in cold climates on WT sound (Paper I, Paper IV), on its propagation and on its character, namely AM (Paper II). Moreover, a selection method is introduced to exclude ambient sounds from long term WT sound measurements (Paper III).

(11)

2. Theoretical background

Larsson (1997) states that atmospheric absorption, scattering by turbulence, and refraction are the most signiﬁcant meteorological effects for sound propagation. These effects as well as the effect of the ground are described in this chapter.

2.1 Refraction

The term refraction describes the process of wave bending. The bending is caused by a speed gradient along the wave front (e.g., Gabrielson, 2006). The wave paths bend towards regions of lower sound speed while propagating.

Several factors cause bending of sound waves within the ABL. Below, refraction due to temperature gradients and wind speed gradients are described.

Moreover, LLWM are addressed because they alter the standardised conditions and are common in cold climates.

2.1.1 Refraction due to temperature gradients

In case of refraction of sound waves due to temperature gradients, the speed gradient along the wave front occurs due to the dependency of the sound speed on the ambient temperature, T . The adiabatic sound speed as a function of height, z, is then deﬁned as

c(z) =

γ RdT(z), (2.1)

where γ = 1.4 is the ratio of the speciﬁc heat at a constant pressure and the speciﬁc heat at a constant volume for air and R_d= 287.1Jkg⁻¹K⁻¹is the uni- versal gas constant for dry air (Rossing, 2007).

Refraction of sound waves due to temperature gradients can lead to two situ- ations. Firstly, since the speed of sound is proportional to the temperature, a decrease in temperature leads to a decrease in the speed of sound (Fig. 2.1a).

Therefore, the resulting effect is an upward bending of the paths of the sound waves. This leads to regions that are not directly reached by sound waves – the so-called shadow zones. Shadow zones are characterised by low sound pressure levels. However, these zones can be penetrated by sound waves due to diffraction and scattering caused by turbulence (Wiener and Kneast, 1959;

(12)

Salomons, 2001). Secondly, a positive temperature gradient (Fig. 2.1b) leads to an increase in sound speed with height (Heimann and Salomons, 2004).

Therefore, downward bending of the sound waves occurs and causes higher sound levels compared to the upward-bending case close to the surface.

(a) (b)

Figure 2.1. Schematic refraction due to a) a negative and b) a positive temperature gradient, where z is the height and T is the temperature (e.g., Lamancusa, 2008, mod- iﬁed). Bold arrows show the direct paths of the sound waves and the dashed arrows the reﬂected paths. Reprints of Figure 1b and 1c in Paper I.

In acoustic literature, there is a tendency to distinguish between day time – upward bending of sound wave paths due to a temperature increase with height – and night time – downward bending due to a temperature decrease with height.

However, even if this may hold for lower latitudes in general, this sharp case differentiation does not match the reality. The use of stratification to distinguish between upward and downward bending is much more appropriate and does not necessarily coincide with day and night, respectively. Furthermore, since cold climates are the focal point of this thesis, such an oversimplifica- tion is not applicable here. For higher latitudes, the alternation of polar night and polar day impact the climate. Stable conditions can be highly persistent in cold climates, especially during the coldest months. That in turn prevents a classical diurnal cycle of the vertical temperature gradient and the change from stable and unstable stratification.

2.1.2 Refractions due to wind speed gradients

The speed of sound is inﬂuenced by the wind speed and wind direction. If sound waves are traveling in the same direction as the wind, i.e. the downwind case, the wind speed enhances the speed of sound, while a reduction in sound speed is valid for the opposite situation, i.e. the upwind case. The re- sulting speed is the effective sound speed, c_{ef f}(z) (ms⁻¹), at the height z (m) (Salomons, 2001; Öhlund and Larsson, 2015):

cef f(z) =

γ RdT(z)(1 + 0.16q(z)) + ucomp(z), (2.2) where q is the speciﬁc humidity (kg kg⁻¹). The ﬁrst term on the right hand side represents the sound speed in a non-moving atmosphere. The second

(13)

term, u_comp(z) represents the horizontal component of the wind speed in a speciﬁc sound propagation direction (m s⁻¹) and is deﬁned as

ucomp(z) = −|U(z)|cos(wd(z) − dir), (2.3) where wd is the wind direction (^◦), U is the wind speed (m s⁻¹) and dir is the sound propagation direction (^◦), i.e. dir= 0^◦ indicates northward propaga- tion, whereas dir= 90^◦ indicates eastward propagation. Note that the sound propagation direction is deﬁned opposite to the wind direction.

The vertical gradient of c_{ef f}, Δcef f is a measure for the refraction of sound waves and is deﬁned as

Δcef f =c_{ef f}(z2) − cef f(z1)

z2− z1 , (2.4)

where z1and z2are two different heights. Δcef f was used for analyses in Pa- per I.

Assuming a logarithmic wind profile, sound waves are bent downward downwind of the source and bent upward in the upwind area (Fig. 2.2). Upward bending leads to a formation of shadow zones. However, logarithmic wind profiles are a feature of neutrally stratified boundary layers. During stable conditions, LLWM (or low-level jets) are a common phenomenon (e.g., Kilpeläi- nen et al., 2012). This makes the vertical sound speed profile much more complex and up- and downwind zones cannot easily be separated into two zones of upward or downward bending.

Figure 2.2. Schematic refraction due to wind speed gradients, where z is the height and u is the wind speed. Bold arrows show the direct paths of the sound waves and the dashed arrows the reﬂected paths. Reprint of Figure 1a in Paper I.

(14)

2.1.3 Low-level wind maxima

Low-level inversions and LLWM can cause a notable feature called stratified spreading. Large gradients of vertical temperature or wind speed can cause an extreme downward bending of the sound waves, preventing them to reach other heights than those close to the ground. In other words, the large gradients limit geometrical spreading by creating an upper boundary, which leads to a shift from spherical spreading to cylindrical spreading (Boué, 2007). The decrease of sound pressure by doubling the distance between source and receiver is therefore 3 dB instead of 6 dB as for spherical spreading. That, in turn, leads to higher sound pressure levels at the immission point. Additionally, stratified spreading leads to a higher relevance of the ground effect due to an increased possibility of multiple reflections. The impact of LLWM on WT sound was analysed in Paper IV.

2.2 Ground effect

When sound waves propagate inside the atmospheric boundary layer, they potentially hit the surface. As with every interface between two different media, the waves are altered in several ways by interacting with the new medium. If a ground-reﬂected wave meets a direct wave – which, so far, has not interacted with the ground – interference takes place, and results into the ground effect (Attenborough, 2002). Due to the difference in the distances they travelled and the resulting delay, the phases of the reﬂected and the direct waves are different. Dependent on the phase difference, constructive or destructive interference occur, which cause higher or lower sound pressure levels, respectively.

Ground conditions impact this effect. Following Attenborough (2002) from an acoustic point of view, the most important parameter characterising the ground is its ﬂow resistivity. High values mean that air cannot easily penetrate into or exit the ground, which is typically linked to a low porosity. The more porous the uppermost layer of the ground is, the more important its thickness becomes as well as potential sublayers. Low porosities lead to an enhance- ment in sound pressure. However, porous surfaces can also increase the sound pressure, which at least, holds for waves with low frequencies. The longer the wave length, the less able the wave is to enter the ground through the pores (Attenborough, 2002).

The acoustical ground characteristics of the area close to the sound source and the receiver are mainly effecting the sound propagation. To which extent these two regions and the region between them are contributing to the total effect is dependent on the height of the source and the height of the receiver (International Standards Organisation, 1996). Since the ground properties of one region are likely to be non-homogeneous, the determination of the ground

(15)

effect is complex and an implementation into a model is not trivial. The fact that a certain ground may change its acoustical properties with time, makes it even more complex. The ﬂow resistivity may vary due to different kinds of precipitation, the growth of vegetation, freezing of the ground and/or human construction activities.

In Paper I, snow quality was classified as dry, damp, wet, or frozen, which is based on the classification used by Lundberg and Halldin (2001), who distin- guished only between dry and wet snow. The classification was complemented with damp and frozen snow. The motivation behind that distinction is the hy- pothesis that different snow qualities have different surface characteristics and therefore diverse acoustic effects.

2.3 Amplitude modulation

The regular variation in sound level due to the rotation of the WT blades is called amplitude modulation (AM). According to its source mechanisms, AM can be classified into two categories. Normal AM (NAM) refers to the aero- dynamically induced WT sound, which is amplitude modulated by the source itself (Cand et al., 2012). The directivity of the trailing edge noise, created at the downward moving blade, in combination with convective amplification causes NAM (Oerlemans, 2011; Cand et al., 2012). The perceived sound is often described as swishing. NAM reduces with distance and at distances more than three rotor diameters away from the WT, it is only detectable in the crosswind direction (Oerlemans, 2011). The second category is other AM (OAM), also known as enhanced AM (EAM) (Oerlemans, 2011). The sound character of OAM is described as thumping. It is more impulsive and/or is shifted to lower frequencies compared to NAM. Local detached flow and enhanced in- flow turbulence are assumed to effect OAM (Oerlemans, 2011), however, the source mechanisms behind OAM are not completely understood. The variation in sound level is increased compared to NAM. Unlike NAM, OAM can be detected at distances of several rotor diameters up- and downwind of a WT.

AM in cold climates is addressed in Paper II. Moreover, AM is used as a criterion in a selection method in Paper III since it is an indicator for WT sound.

2.4 Atmospheric absorption

The term atmospheric absorption describes the loss of energy of propagating sound due to friction. This process is separated into classical absorption and molecular absorption. The ﬁrst describes the transformation of acoustical

(16)

energy into heat. This is assumed not to contribute much to the total atmospheric absorption. The latter phenomenon, also called molecular relaxation losses (Putnam, 1975), represents the transformation of acoustical energy into internal energy of the air molecules. This forms the largest part of the whole atmospheric absorption (Putnam, 1975). Atmospheric absorption is dependent on the frequency of the sound wave, relative humidity, temperature of the air, the atmospheric pressure, and the distance between sound source and receiver (International Standards Organisation, 1996). Atmospheric absorption has not been examined in greater detail. However, it is included in the calculations in Paper I and Paper III.

(17)

3. Measurements

3.1 Measuring Site A

The anonymised Site A (Fig. 3.1) is located in an undulating landscape in northern Sweden. The area is predominantly covered by forests and swamps, and is sparsely populated. The closest settlement consists of less than ten households, accessible by a road with little trafﬁc. Approximately 1 km from the road and several hundred meters from the closest occupied building, the acoustic station was placed. On a hill approximately 1 km southwest of the acoustic station, twelve Enercon E82 2 MW WTs with a hub height of 108 to 138 m are located. A meteorological mast was erected between the acoustic station and the settlement.

Figure 3.1. Overview of Site A, with twelve wind turbines (black dots), acoustic station (+) and an 18-m mast (x), the numbers indicate the height above sea level, large roads (bold lines), small roads (dotted lines) and railway (dashed line). Modiﬁed reprint of Figure 2b in Paper I.

Meteorological measurements

A 18-m mast is located ca. 60 m downslope of the WTs, in close vicinity of a forest. The forest is in the southwest direction of the 18-m mast and only the highest level of the mast is above the tree crowns. Temperature, wind speed and wind direction measurements were conducted at 0.5, 1.5, 5 and 18 m, while the atmospheric pressure and relative humidity were measured at 1.5m. Close to the 18-m mast, measurements of snow depth and observations of the snow quality were taken once a day by local residents. Moreover, the

(18)

local residents collected information on upplega. Additionally, meteorological measurements were conducted on a tower 10 km southeast of the 18-m mast.

This tower is situated on a hill, which has approximately the same height as the hill on which the WTs are located (see Fig. 2 in Paper I).

Acoustic measurements

The acoustic station was placed on a slope in the forest outside which the 18-m mast was erected. The acoustic instrumentation consists of a Norsonic NOR140 sound level meter and a Nor1214 outdoor microphone equipped with a rain hood and a dust mesh. The microphone is placed 1.5 m above the bare ground. Measurements of 1/3-octave bands with midband frequencies between 6.3 to 2,000 Hz were conducted. The collected parameters were:

the equivalent sound pressure level, L_Aeq, and the 5th and the 95th percentiles (sound pressure levels which are exceeded 5% and 95% of the time within each 10-min interval), as well as 10-min averages of the sound spectrum.

The location of the acoustic station is approximately 40 m lower than the location of the WTs and is not shielded from the WTs by any topography. How- ever, measurements could have been conducted in a shadow zone depending on the refraction conditions.

Operational data

Measurements of the rotational frequency of each WT was provided by the wind farm operator.

All measurements were averaged over the same 10-minute intervals, except the snow observations. The measuring period covers the snow season between 05.11.2013 and 30.04.2014. These measurements were used for the study presented in Paper I.

3.2 The Malå site

The Malå site (Fig. 3.2) is located in close vicinity to the small settlement Nåda, within the Malå municipality in northern Sweden (65.09^◦N, 18.87^◦E).

The sparsely populated landscape is characterised by hills and ridges and is predominantly covered by forests. The acoustic station is about 200 m from the closest houses and approximately 500 m from a road with little trafﬁc. The closest of 22 Vestas V90 (2 MW) WTs is located ca. 1 km northwest of the acoustic station. The WTs have a hub height of 105 m and are situated on the Nådagubbliden, which is a ridge with elevations up to 510 m above sea level. Meteorological measurements were conducted at the edge of Nåda and approximately 100 m southeast of the acoustic station.

(19)

Figure 3.2. The Malå site with the 14 closest wind turbines (dots), meteorologi- cal tower (rhombus), acoustic station (+), meteorological mast (x), large roads (bold lines), small roads (dotted lines) and houses (black squares). The distance between the isohypses is 20 m; the numbers indicate the height above sea level. Reprint of Figure 2 in Paper IV.

Meteorological measurements

A meteorological mast was erected in the centre of a grass-covered area, about 50 m from the closest trees, bushes and buildings, to ensure measurement conditions as undisturbed as possible. Measurements of temperature, wind speed and direction were conducted at three heights with the uppermost level at 4.6 m. Relative humidity and atmospheric pressure were measured at one height. Additionally, wind measurements (speed and direction) at 60 m, 82.5m and 105 m on a meteorological tower were provided by the operator of the wind farm and were used in Paper IV.

Acoustic measurements

Similar to the acoustic station at Site A, the acoustic station at the Malå site was placed within the forest and consisted of a Norsonic NOR140 sound level meter and a Nor1214 outdoor microphone equipped with a rain hood and a dust mesh. The microphone was also placed 1.5 m above the bare ground.

The acoustic parameters collected were basically the same as for the Site A:

the equivalent sound pressure level, L_Aeq, and the 5th and the 95th percentiles (sound pressure levels which are exceeded 5% and 95% of the time within each 10-min interval), as well as 10-min averages of the sound spectrum of 1/3-octave bands with midband frequencies between 6.3 to 2,000 Hz. Further- more, 8-Hz measurements of the instantaneous sound pressure level were used in Paper II for the detection of AM.

(20)

Operational data

The operational data consists of 10-min averages of wind direction and wind speed measurements at hub height, and the rotational frequency of each WT.

In Paper III, information about the electrical power generated by each WT was additionally used. The operational data was provided by the operator of the wind farm.

As for Site A, all measurements were averaged over the same 10-minute intervals. The measuring period was from 3.10.2016 to 24.6.2017. If not otherwise stated, these measurements were used for the studies presented in Paper II, Paper III and Paper IV.

(21)

4. Data selection

Even though low background levels were observed at both measuring sites, the sound measurement also contain sound from other sources than WTs. To prevent ambient sound sources from affecting the results, the application of selection methods suitable for the speciﬁc study is necessary. The selection methods vary between the papers since the studies address different objec- tives. The selection criteria described below are mainly from Paper III but references to the other papers are made when applicable.

4.1 Selection criteria

Two factors determine whether WT sound is the dominant source in a sample:

the level of the sound generated at the WT, and the level of the ambient sound.

In consideration of atmospheric absorption, geometrical spreading and ground reﬂection, the level of the generated sound must be high enough compared to the background sound and dynamics of the measurement system. Ambient sound sources must not conceal WT sound at the immission point. Potential ambient sound sources can, for example, be birdsong, human activity, wind- induced sound in the vegetation and in the microphone. To examine these two factors, the most inﬂuential parameters are described below:

4.1.1 Rotational frequency

In order to estimate the potential of WT sound to be detected at the immission point, the median of the rotational frequency of the WTs, TRF, is used as an indicator. Considering the attenuation factors that impact sound propagation, WT sound is possible, but not necessarily detectable if T_RF≥ 10rpm. If TRF<

10 r pm, WT sound is unlikely, but not impossible. The same criterion is used in Paper II. In Paper IV, the threshold for T_RFis raised to 14 r pm. There, T_RF is the only indicator applied to identify WT sound, and therefore, more strict.

4.1.2 Spectral resemblance

Spectral resemblance is a method to identify ambient sounds in a measured spectrum, L_mea( f ), by comparing it to the estimated immission spectrum, Lim( f ), where f is the mid-band frequency of the 1/3-octave frequency bands

(22)

(Hz). A measurement is considered to be WT sound if the shape of L_mea( f ) is similar to the shape of Lim( f ). The range of 1/3-octave bands between 400 Hz and 1.6 kHz was chosen. Frequencies above 1.6 kHz are not analysed since the impact of atmospheric absorption is increased for these frequency bands, while the signal of vegetation from coniferous trees is increased (Fégeant, 1999b,c,a; Bolin, 2009). Source levels of WT sound are usually low for frequencies below 400 Hz and hence not analysed.

The steps of the algorithm are:

1. Calculation of the free ﬁeld immission spectrum, Lim( f ):

L_im( f ) = LW( f ) − Lattenuation( f ), (4.1) where LW( f ) is the emission spectrum, Lattenuation( f ) is atmospheric atten- uation and f is between 400 Hz and 1.6 kHz.

2. Normalisation of the 1/3-octave data:

L_im_,norm( f ) = Lim( f ) − max(Lim( f )), (4.2a) L_mea_,norm( f ) = Lmea( f ) − max(Lmea( f )), (4.2b) where max(Lim( f )) and max(Lmea( f )) are the maximal levels.

3. Calculation of the average difference:

TSR= mean(abs(Lim,norm( f ) − Lmea,norm( f ))). (4.3) The applied ranges are T_SR_,1≤ 1dBA, TSR,2≤ 1.5dBA and TSR,3≤ 2dBA.

4.1.3 Temporal variation

Due to a generally uniform character of WT sound, the temporal variation is usually low. To test the temporal variation of a sample, the difference between the 5th and the 95th percentile is calculated for a 10-min interval. The sample is identiﬁed as WT sound if the difference is within the range of Tvar,1≤ 4dBA (van den Berg, 2004; Öhlund and Larsson, 2015). Furthermore, T_var,2≤ 5dBA was applied.

4.1.4 Amplitude modulation

Even though WT sound generally has a uniform character, regular variations of the sound level can occur due to AM. Since these variations can exceed T_var_,1 and Tvar,2, measurements showing AM can falsely be rejected. Therefore, TAM

is applied in combination with T_var_,1or T_var_,2to prevent these false rejections.

A measurement is selected if Tvar and/or TAMare met. The method described by Larsson and Öhlund (2014) and based on the concept by Lundmark (2011) is used to detect AM:

(23)

1. Determination of the amplitude modulation spectrum (AMS): a Fast Fourier Transform (FFT) is performed for each 15-s interval of the instantaneous A-weighted sound pressure level (dBA) within a frequency range of 10 to 630 Hz and a time constant of 125 ms.

2. Calculation of the AM f actor: the maximum of the AMS is determined within a range of 0.6 Hz to 1.0 Hz, which is the typical blade passing frequency of the analysed WTs. A sample is identiﬁed as amplitude modulated if AM f actor≥ 0.4dBHz⁻¹, which is based on observations and the analysis of several AMS by Larsson and Öhlund (2014).

This method was also used to identify AM in Paper II.

4.1.5 Wind speed

To reduce the risk of masking by wind-induced sound in the microphone or vegetation, T_u is introduced. Measurements are selected if the wind speed meets the criterion Tu,1≤ 2ms⁻¹in the strict case or Tu,2≤ 3ms⁻¹in the less strict case. The wind speed used for this analysis is measured at the uppermost level of the meteorological mast. Even though the wind speed at the acoustic station is assumed to be lower than T_u, the wind speeds at the tree tops can be higher.

4.2 The three-criteria selection method

In Paper I, a selection method consisting of three criteria is applied. The criteria were previously used by Öhlund and Larsson (2015). A measurement is selected if

1. the temporal variation does not exceed 4 dBA, as described by Tvar,1, 2. the A-weighted sound level calculated for 1/3-octave bands between 800 Hz

and 20 kHz contributes less than 1.5 dBA to the total A-weighted sound level if the total sound level exceeds 25 dBA and

3. the total sound pressure level is greater or equal to 30 dBA at the immission point if free ﬁeld spreading from every WT is assumed.

The second criterion is applied because WT sound is usually in the lower frequency range. If however, sound of higher frequencies is emitted, the detected amount is expected to be low due to atmospheric absorption. The third criterion ensures that the WTs operate above a certain rate and therefore, generated high enough sound levels.

(24)

4.3 A more comprehensive selection method

A more comprehensive selection method applying the criteria explained in Section 4.1 is the objective of Paper III. The criteria are applied one after the other in each of the three steps beginning at the total amount of data (root) from the Malå site. Different thresholds (e.g., Tvar,1 and Tvar,2,) are applied in each step to reach different levels of strictness. The thresholds are indicated next to each branch. Each node shows: A) the percentage of selected data points after each step or after each sequence of steps, and B) the percentage of A when the median of the rotational frequencies of the turbines is equal to or greater 10 rpm, i.e., meeting the criterion TRF(see Section 4.1).

Order of steps in Sequence I (Fig. 4.1):

• Step 1: spectral resemblance, TSR,1, T_SR_,2and T_SR_,3,

• Step 2: temporal variation, Tvar,1and Tvar,2,

• Step 3: wind speed, Tu,1 and T_u,2. Order of steps in Sequence II (Fig. 4.2):

• Step 1: spectral resemblance, TSR,1, T_SR_,2and T_SR_,3,

• Step 2: temporal variation, Tvar,1 and Tvar,2, each in combination with am- plitude modulation, T_AM,

• Step 3: wind speed, Tu,1 and Tu,2.

Figure 4.1. Tree diagram to illustrate Sequence I. For reasons of simpliﬁcation, only one example for the procedure is featured (black), while the other possibilities are shown in grey. The root shows the total amount of data points, the numbers next to the branches state the different thresholds for each criteria and the nodes give under position A) the percentage of selected data points after each step or after each sequence of steps, and under position B) the percentage of A meeting the criterion TRF. Reprint of Figure 4 in Paper III.

(25)

Figure 4.2. Tree diagram to illustrate Sequence II. For reasons of simpliﬁcation, only one example for the procedure is featured (black), while the other possibilities are shown in grey. The root shows the total amount of data points, the numbers next to the branches state the different thresholds for each criteria and the nodes give under position A) the percentage of selected data points after each step or after each sequence of steps, and under position B) the percentage of A meeting the criterion TRF. Reprint of Figure 5 in Paper III.

4.3.1 Deﬁning the selection methods

Two options of Sequence I (Fig. 4.3) are chosen for further analyses. The selection method SM I_ashows in Step 1 that 9% of the total data fulﬁl T_SR_,1and 91% of the 9% meet the T_RF criterion. In Step 2, the selected data shrinks to 3% when Tvar,2is applied and the percentage of this data meeting the criterion T_RF increases to 95%. The remaining data is reduced to 2% by applying T_u_,1 in Step 3, while the percentage that meets TRF further increases to 96%. The second option, SM I_b shows in Step 1 that 30% of the total data fulﬁl T_SR_,3 of which 80% meet TRF. In Step 2, the selected data shrinks to 8% when Tvar,2 is applied and the percentage of this data meeting the criterion T_RF increases to 88%. The remaining data is reduced to 5% by applying T_u_,1 in Step 3, while the percentage that meets TRF remains at 88%.

The increase in the percentage that meets TRF from Step 1 to Step 2 shows that the combination of Step 1 and Step 2 delivers better results than the application of Step 1 alone. However, further increases or decreases in Step 3 are stochastical effects. The wind speed close to the ground is an environmental factor and not affected by the rotational frequency. Since SM I_a shows the highest percentage that meets TRF and includes the stricter wind speed range, T_u_,1, it is chosen for further analyses. SM I_bis chosen to compare the quality and quantity of data selected by applying TSR,1 and TSR,3.

The tree diagram in Figure 4.4 visualises Sequence II. Contrary to Sequence

(26)

I, the T_var criterion in Step 2 is complemented with the T_AM criterion. Step 1 and Step 3 remain the same. If Tvar,1 and TAMare applied jointly, Step 2 re- duces the selected data to 5%, which is clearly more than applying T_var alone as shown in Figure 4.3. The percentage of selected measurements meeting the TRF criterion in Step 2 is 96% and remains the same for Step 3. SM II is chosen for further analyses since it includes the strictest criteria.

Figure 4.3. Tree diagram of twelve options of Sequence I. The root node is the total amount of unﬁltered measurements, and the other nodes show the percentage of data selected (A) and the percentage of A after applying TRF(B). The different thresholds of the indicators are written on the branches. Paths chosen as selection methods (SMs) for further analyses are indicate by Iaand Ib. Reprint of Figure 6 in Paper III.

Note that the strictness of the chosen selection methods is suitable due to the large amount of data. For studies based on data of shorter measuring periods, an adjustment of the strictness might be appropriate. Otherwise, the amount of selected data might not be comprehensive enough for reliable results.

(27)

Figure 4.4. Tree diagram of four options of Sequence II. Similar to Figure 4.3 but T_var is complemented with TAM. Note that only the paths for TSR≤ 1dBA are shown. The path chosen as a selection method (SM) for further analyses is indicate by II. Reprint of Figure7 in Paper III.

4.3.2 Quality analysis

A confusion matrix can be used to visualise the performance of a classiﬁcation method. Usually, the classiﬁed data is compared with the true values. How- ever, true values do not exist here. T_RF is used to estimate the true values.

Based on this, the entries of the confusion matrix can be described as:

• true positives (TP): cases in which the SM predicted “yes” (WT sound was recorded), and WT sound was possible to detect based on T_RF,

• true negatives (TN): cases in which the SM predicted “no” (WT sound was not recorded), and WT sound was unlikely to detect based on T_RF,

• false positives (FP): cases in which the SM predicted “yes” (WT sound was recorded), and WT sound was unlikely to detect based on T_RF, and

• false negatives (FN): cases in which the SM predicted “no” (WT sound was not recorded), and WT sound was possible to detect based on TRF.

Table 4.1 shows the entries of the confusion matrices for SM I_a, SM I_b and SM II. The majority of data selected by the three SMs is classiﬁed as FN. That means, more than two-thirds of the data are rejected even though, it is possible that WT sound is the dominant source. With percentages between 28.1% and 28.7%, the second largest share is classiﬁed as TN. TP varies between 2.1%

(SM Ia) and 4.6% (SM I_b). The smallest percentages, namely 0.1 to 0.7%, are classiﬁed as FP. Thus, the probability for the presented SMs to select measurements, even if WT sound is unlikely to be the dominant source, is relatively low.

(28)

Table 4.1. Confusion matrix assessing the performance of SM I_a, SM I_band SM II based on T_RF. Amount of data points and probability are shown for true positives (TP), true negatives (TN), false positives (FP) and false negatives (FN).

selection method: “yes” selection method: “no”

SM I_a SM I_b SM II SM I_a SM I_b SM II

T_RF: “possible” TP FN

# 587 1290 1029 19349 18646 18907

percentage 2.1% 4.6% 3.7% 69.2% 66.6% 67.6%

T_RF: “unlikely” FP TN

# 23 182 38 8020 7861 8005

percentage 0.1% 0.7% 0.1% 28.7% 28.1% 28.6%

The confusion matrix (Tab. 4.1) reveals one major weakness of all three SMs:

the large amount of rejected measurements even though T_RFindicates that the detection of WT sound was possible (FN). To identify the reasons for the re- jection, several criteria are tested for measurements classiﬁed as FN (Tab. 4.2).

52 to 54% can be explained by assuming the independence of T_ufrom T_RF. In these cases, enough WT sound was generated but rejected due to the risk of masking by wind-induced sound. For additional 3 to 4% of the data, the rotational frequency of the closest WT is below 10 rpm. Thus, the sound generated by the other WTs could be too low to be detected at the immission point. For another 13 to 15% of the data, the wind conditions were other than downwind, which includes wind directions smaller than 258.5^◦ and larger than 348.5^◦. Thus, the immisson point could have been located in a shadow zone, caused by upward refraction. The remaining 28 to 30% of the measurements clas- siﬁed as FN cannot be explained by these criteria. Absorption due to snow, either on the ground or in the form of upplega, LLWM below hub height or other processes may cause this discrepancy.

Table 4.2. Measurements classiﬁed as FN by SM I_a, SM I_band SM II tested for several criteria.

criteria amount of measurements [%]

SM I_a SM I_b SM II

Tu> 2ms⁻¹ 52% 54% 53%

TRF,WT22< 10rpm and Tu≤ 2ms⁻¹ 4% 3% 4%

TRF,WT22≥ 10rpm, Tu≤ 2ms⁻¹and not downwind

15% 13% 15%

else 29% 30% 28%

(29)

5. Meteorological effects on WT sound in cold climates

5.1 Impact of snow on WT sound propagation

Snow impacts WT sound as shown by Öhlund and Larsson (2015). However, the effect of snow is generalised and a distinction between different snow qualities is missing. Snow metamorphism changes the structure of the snow, which may alter the effect of the snow on sound. Therefore, a distinction was made between four snow qualities in order to investigate their individual effects.

Moreover, the effect of upplega was analysed. The analyses are based on measurements at Site A (see Paper I).

To compare the effects of different conditions, which are here either varying snow qualities or occasions with or without upplega, the measured sound levels relative to the emitted sound levels were compared. Therefore, the relative sound pressure level,ΔL, is determined by subtracting the sound pressure level calculated for free ﬁeld conditions from the measured sound pressure level at the immission point, LpA(dB), and is deﬁned as

ΔL = LpA− 10lg∑^k

i=1

10^(L^WAi^−10lg(4πR²ⁱ^)−αRⁱ^)/10, (5.1) where L_WA_i is the emitted sound power level from each WT i (dB), R_i is the distance between the receiver and each WT (m), and α is the atmospheric absorption coefficient (dB m⁻¹). In the calculation of the free field value, atmospheric absorption and spherical spreading are taken into account. The WTs are approximated as point sources. Positive ΔL indicate amplification, while negativeΔL indicate attenuation (e.g., van den Berg, 2004; Öhlund and Larsson, 2015).

5.1.1 Effect of snow qualities

To show the effects of the four different snow qualities during comparable meteorological conditions,ΔL as a function of Δcef f in Figure 5.1. Here,Δcef f is determined between the 1.5-m level of the 18-m mast and around the 120-m level of the tower. NegativeΔcef f were mostly observed for dry-snow conditions, thus, the comparison of the effects of different snow qualities for downward refraction (i.e. Δcef f > 0s⁻¹) is unnecessary.

(30)

The largest median ofΔL can be observed for damp snow combined with no or gentle downward refraction and decreases slightly with increasingΔcef f. The medians ofΔL for frozen snow are around 0 dBA for Δcef f ≈ 0.0s⁻¹and increase with increasingΔcef f.ΔL for dry snow increases with increasing Δcef f

similar to frozen-snow conditions but decreases for the largestΔcef f. The low- est medians of ΔL occurred for wet snow. Note that the sample size for wet snow is relatively small and the standard deviations are relatively large, especially for damp and dry snow. The described differences inΔL for the different snow qualities indicate that the impact of snow onΔL depends on its quality.

Furthermore, Figure 5.1 emphasises the impact ofΔcef f on sound propagation.

Figure 5.1. Relative sound pressure level,ΔL, as a function of effective sound speed gradient, Δcef f, calculated between 120 m and 1.5 m for dry (blue, dashed), damp (green, dotted), frozen (orange, dashed-dotted) and wet snow (purple). Circles, bars and numbers indicate medians, one standard deviation and sample size of each bin, respectively. Reprint of Figure 7A in Paper I.

5.1.2 Effect of upplega

Similar to Figure 5.1, the effect of upplega on ΔL in dependence of Δcef f

is shown in Figure 5.2. For occasions with upplega, the medians of ΔL are around 2 dBA lower than for occasions without upplega. For both,ΔL is lower for upward bending (i.e.,Δcef f < 0s⁻¹) and large for downward bending (i.e., Δcef f > 0s⁻¹). However, while for upplega, the transition from attenuation

(31)

to ampliﬁcation is also the transition from upward to downward bending, this transition is shifted towards slightly negativeΔcef f for occasions without upplega. An impact of upplega onΔL is clearly recognisable independent of the refraction.

Figure 5.2. Relative sound pressure level,ΔL, as a function of effective sound speed gradient,Δcef f, calculated between 120 m and 1.5 m for occasions with upplega (purple, bold) and no upplega (green, dashed). Circles, bars and numbers indicate medians, one standard deviation and sample size of each bin, respectively. Reprint of Figure 10 in Paper I.

5.2 Impact of LLWM below hub height on WT sound propagation

LLWM in altitudes below 100 m are a common feature in cold climates.

Their impact on WT sound propagation however, has not been investigated yet. Since the wind speed is not continuously increasing with height when LLWM are present, the simpliﬁed refraction pattern (e.g., Larsson and Israels- son, 1991; Salomons, 2001; Boué, 2007; Lamancusa, 2008) is not applicable.

To investigate the effect of LLWM on WT sound close to the surface downwind of a wind farm (i.e., 249^◦ to 359^◦), the occurrence of LLWM and their impact on L_Aeqwere analysed. The analyses are based on measurements at the Malå site (see Paper IV).

Wind Turbine Sound in Cold Climates

Wind Turbine Sound in Cold Climates

List of papers

Contents

1. Introduction

2. Theoretical background

3. Measurements

4. Data selection

5. Meteorological effects on WT sound in cold climates