Characterisation and quantification of wind farm noise

Author: Duc Phuc Nguyen

Nguyen, Duc Phuc, 2022 Characterisation and quantification of wind farm noise, Flinders University, College of Science and Engineering

Terms of Use: This electronic version is (or will be) made publicly available by Flinders University in accordance with its open access policy for student theses. Copyright in this thesis remains with the author. You may use this material for uses permitted under the Copyright Act 1968. If you are the owner of any included third party copyright material and/or you believe that any material has been made available without permission of the copyright owner please contact with the details.


This thesis presents a study on the characterisation and quantification of wind farm noise (WFN) at long-range locations. The primary research goals in this thesis were to develop a machine learning model to detect amplitude modulation (AM) of WFN. This model was used to automate the AM detection process, and to quantify and characterise AM in long-term measured data sets. A further aim was to investigate the audibility of unique characteristics of WFN such as infrasound and amplitude modulated tones. The thesis concludes with an exploration of deep learning techniques, which were used to examine and automate the characterisation of WFN. Chapter 1 provides an overview of recent advances in WFN research, followed by a description of field work in Chapter 2 and then four major results chapters.

Chapter 3 presents an approach to detect and characterise AM in a comprehensive and year-long wind farm noise data set evaluated using human scoring. Benchmark AM characteristics were established towards further validation and calibration of results obtained using automated methods. Using these data, an advanced AM detection method was then developed, with predictive power close to the practical limit set by human scoring. However, given that noise impacts on humans remain of primary interest, human-based approaches should be considered as a benchmark method for characterising and detecting unique noise features most relevant to human WFN perception and impacts.

Chapter 4 quantifies and characterises AM over 1 year using acoustical and meteorological data measured at three locations near 3 wind farms. Substantial diurnal variation of outdoor AM prevalence was found, with nighttime prevalence approximately 2 to 5 times higher than daytime prevalence. On average, indoor AM occurred during the nighttime from 1.1 to 1.7 times less often than outdoor AM. However, the indoor AM depth was higher than that measured outdoors. An association between AM prevalence and sunset and sunrise was also observed. These data showed that AM occurs more often during downwind and crosswind compared to upwind conditions. These findings provide important insights into long term WFN characteristics needed to help better inform future WFN assessment guidelines.

Chapter 5 used a computational approach to assess the audibility of infrasound and amplitude modulated tones (AM tones) at long-range locations, which also considered the uncertainty associated with WFN measurements and human hearing variability. It was demonstrated that infrasound is rarely audible to residents with normal hearing who live at distances greater than 1 km from a wind farm, but that AM tones occurring at a low frequency are readily audible at distances up to 9 km. These results suggest that AM tones could be the main source of WFN complaints at long-range locations, and thus clearly warrant further attention towards ensuring that wind farms have minimal impacts on nearby residents located within 9 km of the nearest wind turbine.

Chapter 6 explored an approach for the characterisation and assessment of WFN. This was based on extraction of acoustic features from a pretrained deep learning model (referred to as deep acoustic features). Using data measured at a variety of locations, deep acoustic features were shown to contain meaningful information about noise characteristics. Deep acoustic features were also shown to reveal an improved spatial and temporal representation of WFN compared to traditional spectral analysis and statistical noise descriptors. These very promising novel findings provide a clear framework for improved WFN assessment in the future.

Taken together, this thesis work provides a major and important new contribution towards the understanding of some of the most prominent WFN features audible to humans. These new methods provide an important framework towards improved noise assessments and wind turbine designs better able to minimise impacts on surrounding communities. Ultimately, this approach, along with future improvements in wind farm planning, design, noise assessment and abatement strategies will all likely be needed to help ensure that wind energy is acceptable to surrounding communities.

Keywords: Wind farm noise, Amplitude modulation, Infrasound, Deep acoustic features

Subject: Environmental Science thesis

Thesis type: Doctor of Philosophy
Completed: 2022
School: College of Science and Engineering
Supervisor: Dr Kristy Hansen