**1. Introduction**

Wind energy is expanding rapidly in France, as it is everywhere else in the world. Specific rules govern the design and implementation of wind farms in order to limit the noise they produce when in operation. However, the population living near these installations is often concerned about the health impacts of sound levels emitted by wind turbines (WT) and there is a lack of available scientific data on this topic.

Most studies have found a significant positive association between WT sound levels and the percentage of highly annoyed people [1,2]. Very few studies have investigated the effects of WT sound on sleep disturbance, cardiovascular disease, effects on metabolic or endocrine systems, or on cognition or mental health. Therefore, the WHO guidelines on environmental noise published in October 2018 pointed out that the evidence on the health effects of wind turbines noise is either non-existent or of low quality [3]. The WHO and Anses (French Agency for Food, Environmental and Occupational Health & Safety) in France therefore recommended implementing epidemiological studies [3–5]. However, a number of issues remained to be overcome before such a study can be carried out in France.

The first issue concerned the estimation of exposure to WT sound. Indeed, the quality of epidemiological studies evaluating the risks related to environmental exposures depends in part on the quality of the estimation or measurement of the participants' exposure. However, there was no real consensus on a WT sound prediction model and it was necessary to provide validation criteria to identify the most relevant model.

The second issue concerned the count of the number of people exposed to WT sound. Unlike other sources of noise pollution (e.g., transportation noise), wind farms are generally built in sparsely populated areas, and consequently the number of people potentially

**Citation:** Ecotière, D.; Demizieux, P.; Guillaume, G.; Giorgis-Allemand, L.; Evrard, A.-S. Quantification of Sound Exposure from Wind Turbines in France. *Int. J. Environ. Res. Public Health* **2022**, *19*, 23. https://doi.org/ 10.3390/ijerph19010023

Academic Editors: Roberto Alonso González Lezcano, Francesco Nocera and Rosa Giuseppina Caponetto

Received: 9 November 2021 Accepted: 17 December 2021 Published: 21 December 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

exposed to WT sound seemed a priori to be much smaller than the number of residents exposed to other sources of anthropogenic noise, such as transportation noise, for example. In order to conduct an epidemiological study, it would be necessary to recruit a sufficient number of individuals exposed to different and relatively contrasting WT sound levels. Thus, it was necessary to be able to estimate the population exposed to various WT sound levels at the scale of an entire country. This had never been done before in France.

In this context, a feasibility study for an epidemiological study called Cibélius (Knowing the impact of wind turbine noise on health) was conducted in France between 2017 and 2019. The objective was to propose a methodology for calculating WT sound levels and to identify the number of French residents exposed to different sound levels of wind turbine.

The aim of the research presented in this paper was to quantify the number of windfarms' residents in France exposed to audible WT sound. For this purpose, a WT sound prediction model was selected and validated for the calculation of sound exposure. Then a methodology was suggested for estimating the number of people exposed to WT sound at the scale of all metropolitan France. A brief comparison with transportation noise exposure was also investigated.

#### **2. Material and Methods**

#### *2.1. Overview of the Methodology*

As it would not be feasible to make sound levels measurements in all the dwellings of people exposed to WT sound at the scale of the whole French national territory, it was necessary to estimate sound exposure by using an appropriate numerical modeling of WT sound emission and propagation. We first presented the numerical model used, and its performances for WT sound prediction by quantifying its uncertainties. Then, we detailed the method of calculation of sound exposure at the scale of the metropolitan French territory. Finally, the method for evaluating the count of people exposed to WT sound was shown.

### *2.2. Selection and Validation of a Numerical Wind Turbine Sound Model*

The sound levels prediction model was selected from a literature review based on the following essential criteria in the context of WT sound: ability to account for a high noise source (hub height above 60 m), topography properties, meteorological (vertical profiles of wind speed, wind direction and temperature) and ground effects on sound propagation. The model must also be able to parameterize the sound power of the source as a function of wind speed. Although some research propagation models could meet these criteria for WT sound [6–9], they were not suitable for sound levels modeling on a scale as large as a country's territory, because of their high computational time.

Although it was less frequently used [10,11] than other engineering models that can handle a large-scale territory (e.g., ISO 9613-2 [12], NMPB-2008 [13] or CNOSSOS-EU [14]), the Harmonoise model [15] was preferred here for WT sound prediction because it takes into account ground effects more accurately, and because it is able to model the effects of different wind speeds and directions, and different classes of atmospheric stability on sound propagation. These properties are essential for the modeling of WT sound propagation, and several authors have mentioned the capabilities of this model for WT sound predictions [16–18].

The uncertainties of the Harmonoise model were estimated by comparison between numerical modeling and in situ measurements. For this purpose, an experimental campaign was carried out near a wind farm whose characteristics were representative of the vast majority of French wind farms: flat site (terrain slope lower than 2% over 2000 m), quiet environment, good diversity of wind speeds and directions. The wind farm consisted of five wind turbines with a rated electrical power of 2 MW each. Each WT had a hub located at 100 m high, and three 46 m long blades. During the eight days of the measurement period, the wind farm operated in 1-h /1-h on/off cycles, in order to select only measurements with a satisfactory signal to noise ratio between WT sound and background noise of the site, and thus not to retain sound samples containing extraneous noise.

LAeq (10 min) sound levels were recorded using 15 sound level meters located on the two dominant wind directions of the site, and at distances from the wind farm ranging from 0 to 1500 m (Figure 1). This arrangement enabled sound levels to be measured for downwind and upwind situations, for which sound propagation differs [19]. The sound level meters (B&K 2250, ACOEM Solo and Cube, Rion NL62) were placed at a height of 1.5 ± 0.1 m above the ground and measured in the frequency bands [12.5 Hz; 20 kHz] or [1 Hz; 20 kHz] depending on the point. In addition to these sound level measurements, the sound level power of the wind turbines was derived according to the protocol described in the standard IEC 61400-11 [20], based on sound levels measurements made with a flushmounted microphone on a circular reflective plate of 1 m diameter, placed on the ground 143 m from a wind turbine (Figure 1).

**Figure 1.** General overview of the experimental set-up for the validation of the numerical model. Location of sound level measurements (black points), wind measurements (blue points), wind turbine sound power level measurements (red points), wind turbines (blue crosses).

Meteorological measurements were made in order to obtain the influence of wind speed and direction on both emission and propagation of WT sound. Wind speed and direction measured from a 3D ultrasonic anemometer (Campbell CSAT3) located at 3 m high were used to classify sound propagation conditions as required by the Harmonoise modeling [10]. Wind speed at hub height and wind turbine operation data (speed rotation of the rotor, electrical production) were obtained from the wind farm's Supervisory Control And Data Acquisition system (SCADA). The accuracy of the anemometers (of the Campbell device or the WT hub) was typically 0.1 m/s, which was satisfactory for not inducing a significant uncertainty in the wind speed classification required for the analysis.

The WT emission was modeled using the manufacturer's WT technical specifications, which give the sound power level as a function of wind speed. An additional 2.2 dBA was added to these values to match the actual sound power level measured at the IEC point. The uncertainties of the sound levels prediction model were estimated by calculating the distribution of the deviation between the predicted and measured sound levels. The bias was estimated with the mean of this distribution, and the standard uncertainty with its standard deviation [21]. The hypothesis adopted for the calculation process of WT sound assumed WT operating at full power (see Section 2.3.2). To be in line with this hypothesis, only periods when wind speed was above 6 m/s at 10 m high were kept for the calculation of the bias and the standard uncertainty. Similarly, in order to be consistent with French regulations that do not allow wind farms to be installed less than 500 m from local residents, only measurements for which the distance to the wind farm was greater than 500 m were kept. Bias was used to correct the sound levels predicted by the Harmonoise model, while the standard uncertainty was used to bound the estimate of the population exposed to WT sound and to account for the uncertainty in the Harmonoise

model's estimate of sound levels on the count of the exposed population. Indeed, the population count was performed by considering three scenarii of sound exposure: the first one (average scenario) used the sound levels predicted by the Harmonoise model, only corrected by the estimated bias, and the other two scenarii (upper and lower scenarii) used the sound levels predicted by the model, corrected by the bias, and increased or decreased respectively by a standard uncertainty in the sound level estimate. It should be noted that the range given by the uncertainty estimates based on the lower and upper scenarii did not correspond to confidence intervals associated with a level of reliability. They did, however, provide information on the best and worst-case values.

As meteorological conditions (wind and temperature vertical profiles) could have significant effects on long-range propagation of outdoor sound, resulting in decreased or increased noise levels at dwellings, it was essential to separate daytime and nighttime periods where the meteorological influence on propagation often differs [19]. The uncertainties were thus calculated for two propagation conditions representative of daytime and night-time meteorological conditions.

#### *2.3. Estimation of Sound Exposure from Wind Turbines*

The selected sound levels prediction model was used to produce a noise map of all the wind farms on the metropolitan French territory. This mapping was built following the three steps detailed below.

2.3.1. Constitution of a Database with the Characteristics Required for the Prediction of Sound Levels from Wind Farms in Metropolitan FRANCE

The location of the wind farms (Figure 2), the hub height and the rated electrical power of the wind turbines were obtained from a 2017 database provided by www.thewindpower.net (accessed on 1 December 2021), which was the most complete database publicly available at the beginning of this research. This database listed existing wind farms as of 30 August 2017, of which only wind farms in operation in metropolitan France were considered here. The absolute position of each WT within each wind farm was obtained from the BDTOPO® topographic database [22] from the National Institute of Geographic and Forestry Information (IGN).

**Figure 2.** Localisation of metropolitan French wind farms (2017, www.thewindpower.net, accessed on 1 December 2021).

2.3.2. Calculation of Sound Contributions of Wind Turbines near the Dwellings of Wind Farm Residents

Sound levels calculations with the Harmonoise model presented earlier were performed with the environmental noise prediction software CadnaA from DataKustik [23]. In this software, each wind turbine was modeled as a point source located at the height of the wind turbine hub. Following the results of Botha et al. [24], Møller et al. [25] and Tachibana et al. [26], the sound emission of each wind turbine was estimated considering a linear sound emission spectrum (−4 dB/octave), whose overall sound power level was estimated from its rated electrical power [25], considering a wind speed of 7 m/s at 10 m height (nominal operation of the wind turbines in full operation).

The topographic data used in the propagation model came from the BDTOPO® database [22]. Sound levels were predicted for standard environmental conditions (temperature 10 ◦C, humidity 70%, partially sound absorbing ground) and for eight wind directions (from 0◦ to 315◦, in 45◦ steps). The sound level assigned to each building corresponds to the sound level in front of the most exposed façade, and to the maximum of the sound levels predicted for the eight wind directions. Two propagation conditions typical of the daytime and night-time periods were also considered. These conditions could be parametrized in the Harmonoise model by choosing classes of meteorological stability adapted to these periods (three classes for day, and two classes for night). For each time period, the class that favored sound propagation was chosen (classes S3 for daytime and S5 for night-time [27]). It is important to note that the assessment of daytime and night-time exposures was not related to the sound a resident would be exposed to during an entire daytime or night-time period (as an equivalent sound level indicator would quantify), but they were related to two different meteorological scenarii that influence the sound propagation differently during these two periods. Sound levels exposures below 30 dBA were excluded as too low to be significant. Finally, the calculations provided the sound level exposure at each building façade in BDTOPO database around all the wind farms. Figure 3 shows an example of the sound radiated from a wind farm (Figure 3a) and the exposure of nearby buildings (Figure 3b) estimated with these calculations.

**Figure 3.** Example of the result of the calculation of the radiated sound from a wind farm (**a**), and of the exposure of the buildings in the nearby village enclosed by the black block in (**a**,**b**).
