LIBRARY

Michigan State

University

 

 

PLACE IN RETURN Box to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE

DATE DUE

 

mutt 9

SEP 0 2 2007

 

 

 

 

 

 

 

 

 

 

 

 

 

 

6/01 cJCIRC/DatoDue.p65-p.15

MEASUREMENT, QUANTIFICATION AND INTERPRETATION OF ACOUSTIC
SIGNALS WITHIN AN ECOLOGICAL CONTEXT

By

Brian Michael Napoletano

A THESIS
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
MASTER OF SCIENCE
DEPARTMENT OF ZOOLOGY

2004

ABSTRACT

MEASUREMENT, QUANTIFICATION AND INTERPRETATION OF ACOUSTIC
SIGNALS WITHIN AN ECOLOGICAL CONTEXT

By

Brian Michael Napoletano
Increasing awareness of the degree to which human activities are altering the state of the
Earth’s Biosphere has fostered an interest in ecological variables that clarify complex
relationships and represent the dynamic nature of living systems. The study of a location’s
acoustic signals (its soundscape) integrates an array of biophysical factors and changes with
the location’s characters. This study begins with the formation of an analytical framework to
quantify and interpret acoustic signals from the environment. This framework classifies three
primary constituent regions of the soundscape, the anthrophony (0.4 to 2 kilohertz), the
biophony (2.5 to 11 kHz) and the geophony (diffuse signal, full spectrum). Based on this
framework, the structure of the biophony is examined through a spectral analysis of a series
of avian, insect, and amphibian vocalizations. This analysis confirms the hypothesis that the
strongest concentration of biological activity is between 2 and 5 kHz. The analytical tools are
then applied to a series of acoustic observations gathered in the Muskegon River Watershed,
where the acoustic signatures of three different land cover types (urban, forested/outdoor
recreation, agriculture/ grassland) are examined in two sets. Analyses of Variance indicate
significant differences between the sites. Finally, indices of biological and anthropogenic
activity are compared to population density values. The results of these correlations indicate
that an accurate assessment of a region’s soundscape and corresponding biophysical
attributes requires an observation system of a sufficient spatiotemporal scale, thereby limiting
the value of acoustics as an ecological indicator. However, such an observation system would

enable the measurement and integration of an array of ecological variables.

Copyright by
BRIAN MICHAEL NAPOLETANO
2004

Dedicated to Champ

I’ll miss you, fierall

ACKNOWLEDGEMENTS

I would like to thank my advisor, Stuart Gage, for his generous support and
encouragement, as well as his outstanding patience. I would also like to thank Jan
Stevenson and Bryan Pijanowski, my other two committee members, for their helpful
contributions and positive input.

I would like to thank Manuel Colunga-Garcia for his help in both the writing of the thesis
and the research it describes. Chuck McKeown deserves much gratitude for his help on
planning, implementing and summarizing this research.

Additionally, I owe Michelle Siegrist a great deal of gratitude for her review and input.

I have received help and input from a number of people, but I would like to thank Sasha
Kravchenko for her help in the statistical analyses and William Hartmann for introducing
me to the physics of sound.

A special thank-you goes to Jay Harman for the many lively conversations and for
convincing me to pursue graduate school.

Much of the support for this work was provided by the Great Lakes Fisheries Trust.

Table of Contents

List of Tables ...................................................................................................................... x
List of Figures ................................................................................................................... xii
Introduction The Need for a Dynamic Ecological Variable ............................................... I
General Perspective ........................................................................................................ 1
Sound as a dynamic variable ........................................................................................... 2
Muskegon River Watershed ............................................................................................ 3
Goals and Objectives ...................................................................................................... 4
Methods and Outline ....................................................................................................... 4
Chapter One: Introduction to Ecological Acoustics ........................................................... 7
Introduction ..................................................................................................................... 7
Physical Characters of Sound ..................................................................................... 8
Human Perception of Sound ....................................................................................... 8
Sound in Ecology .......................................................................................................... 10
Chapter Two: Development of an Interpretive“ Framework to Assess Ecological Features
of Acoustic Signals ........................................................................................................... 14
Introduction ................................................................................................................... l 4
Definition of the Soundscape ........................................................................................ l4
Classification of Signals in the Soundscape ............................................................. 15
Quantification of Acoustic Signals ............................................................................... 17
Visual Representation of Acoustic Samples ............................................................. l8
Quantification of the Spectrogram ............................................................................ 19

vi

Index Value Assumptions ............................................................................................. 22

Chapter Three: Application of the interpretive framework to landscape-level acoustic data

........................................................................................................................................... 24
Introduction ................................................................................................................... 24
The Need for Automation ............................................................................................. 24
EAS as an interface layer .............................................................................................. 26
Stepwise Operation of EAS .......................................................................................... 28
Code Structure and Logic ............................................................................................. 35
Conclusion .................................................................................................................... 38

Chapter Four: Spectral Analysis of the Acoustic Signals of Three Classes of Organisms

in the Northeastern United States ..................................................................................... 40
Introduction ................................................................................................................... 40
Birds, Bugs and Bullfrogs ......................................................................................... 42

The Use of Acoustic Signals ..................................................................................... 42
Objectives ..................................................................................................................... 43
Methods ......................................................................................................................... 44
Sound Signal Analysis .............................................................................................. 44
Statistical Analysis .................................................................................................... 45
Results ........................................................................................................................... 46
Spectral Bandwidth Histograms ............................................................................... 46
Analysis of Variance ................................................................................................. 50
Discussion ..................................................................................................................... 52
Chapter Five: Acoustic Signatures of Different Locations ............................................... 54
Introduction ................................................................................................................... 54

vii

Objectives ..................................................................................................................... 56

Study One: 14 Days, 2 Locations ................................................................................. 56
Study Two: 3 Months, 2 Locations ............................................................................... 59
Discussion ..................................................................................................................... 63

Chapter Six: Correlation Analysis of Acoustic Signals and Ecological Features in the

Muskegon River Watershed .............................................................................................. 64
Introduction ................................................................................................................... 64
Objectives ..................................................................................................................... 66
Methods ......................................................................................................................... 67

Data Gathering .......................................................................................................... 67
Index Derivation ....................................................................................................... 68
Regression Analysis .................................................................................................. 68
Results ........................................................................................................................... 69
Discussion ..................................................................................................................... 71

Chapter Seven: Summary and Conclusions ...................................................................... 73
Objectives ..................................................................................................................... 73
Spectral Analysis .......................................................................................................... 74
Future Directions .......................................................................................................... 75

Appendix A ....................................................................................................................... 78
Definition of a soundscape ............................................................................................ 81
Classification of sounds in the soundscape ................................................................... 82
Quantifying acoustic samples from the environment ................................................... 84
Case Studies .................................................................................................................. 89
Conclusions ................................................................................................................. l 00

viii

Appendix B: List of species used in the Spectral Analysis ............................................ 102

References ....................................................................................................................... l 12

ix

List of Tables

Table 1. Formulae for the calculation of activity concentration'values for the three
primary regions of the soundscape. The column on the left depicts the formulae for
the calculation of ratio values in terms of the entire spectrum, while the column on
the right lists the formulae for the calculation of values in terms of percentage of the
entire spectrum (from Gage et.al. 2003). .................................................................. 21

Table 2. Numbers of files generated by a single automated and manual recording system
and the subsequent analysis. ..................................................................................... 25

Table 3. File Name template used by the analysis system for the different site types. 27

Table 4. The results of the spectral analysis of the acoustic signals of a) all 354 organisms
sampled; b) 264 species in the Class Aves; c) 49 species in the Class Insecta; and d)
41 species in the Class Amphibia. The value in the column “Band” is the high end of
that frequency band (i.e. Band 1 = 0 — 1 kHz), and SE refers to the standard error of
the means. ................................................................................................................. 47

Table 5. Results of a mixed ANOVA of the spectral properties of three Classes of
organisms (Insecta, Amphibia and Aves) across 1 1 frequency bands (0 — 11 kHz).
The effect labeled C1ass*Band indicates the analysis of the weighted means by Class
and frequency band. .................................................................................................. 50

Table 6. Comparisons of Least Squares Means by Class at the 11 frequency bands. p < a
indicates a significant difference (LSD, at = 0.05). ................................................... 51

Table 7. ANOVA results for the test of Location effects over 14 days at two locations
(a=0.05). The main effect Location is significant, while the interaction effect,
Location*Time, is not. .............................................................................................. 59

Table 8. ANOVA results for the two data sets from PP and FS (a = 0.05). ..................... 62

Table 9. Summary information on correlation analysis between anthrophony and
biophony values and population density. .................................................................. 69

Table 10. Formulae for the calculation of activity concentration values for the three
primary regions of the soundscape. The column on the lefi depicts the formulae for
the calculation of ratio values in terms of the entire spectrum, while the column on
the right lists the formulae for the calculation of values in terms of percentage of the
entire spectrum. ......................................................................................................... 88

Table 11. Summary statistics of the means and standard deviations of p, 0t, and [3 values
calculated for three consecutive days at four different locations. p represents the
ratio of biological to anthropogenic activity, while 0t and [3 represent the
anthropogenic and biological activity respectively ................................................... 90

Table 12. Pearson’s correlations of p, 0t, and B means over the three-day period. All sites
except Paris Park exhibit a significant correlation of p values. ................................ 91

Table 13. Summary statistics for the three sample classes (Lakes, Streams, and Wetlands)
gathered in the Muskegon River Watershed. The Streams class had the highest

calculated mean, the Lakes class had the lowest, and the Wetlands class had the
broadest range of values ............................................................................................ 95

Table 14. Mean value of activity in band 4 (3-4 kHz) at 22:00 on four different days with

four different temperatures. The lower temperatures appear to correlate with lower
activity means, and higher temperatures with higher means. ................................. 100

xi

List of Figures

Figure 1. Conceptual classification schematic of the soundscape and the three principal
components. Several hypothetical subclasses are also depicted below these
components. While the Anthrophony and Biophony tend to have discrete spectral
ranges, the Geophony tends to occur across all spectral bands, but is more diffuse. 15

Figure 2. A spectrogram with a maximum frequency of 1 1.025 kHz and a 90 dB black
and white palette. This spectrogram also illustrates the division of biological and
anthropogenic signals into discrete frequency bands ................................................ 18

Figure 4. The order and hierarchy of operational steps programmed into the EAS code
logic. Each analytical option is handled with an if. . .then procedure. ...................... 36

Figure 5. A hierarchical representation of the code logic within the histogram extraction
utility. Again, the various options are handled with if. . .then operators. .................. 37

Figure 6. An explanation of the file name metadata for each of the three site types ........ 38

Figure 7. Spectral bandwidth distribution of all three Classes (354 samples). Avg
Weighted MN represents the mean concentration of activity for each frequency
band, which is the respective 1 kHz spectral band. .................................................. 48

Figure 8. Spectral bandwidth distribution of acoustic samples from 264 avian species
(150 Passerines, 114 Non-passerines), with maximum activity from 2 - 5 kHz. ..... 48

Figure 9. Spectral bandwidth distribution of acoustic samples from 49 species of insects,
with two activity peaks. The highest activity concentration in a single band is at 6 —
7 kHz. ........................................................................................................................ 49

Figure 10. Spectral bandwidth distribution of acoustic samples from 41 species of
amphibians. The highest concentration of activity is at l - 2 kHz. .......................... 49

Figure 11. An overview of the dual role of sound in ecological research. The two graphs
on the bottom depict the total activity within two different locations over 24 hours.
................................................................................................................................... 55

Figure 12. p plotted against time for PP and MS 2001-06-11 through 2001-06-24.
Overall, MS has a higher value, indicating a stronger biophonic contribution. ....... 57

Figure 13. a plotted against time for PP and MS 2001-06-11 through 2001-06-24. With a
few exceptions, PP has a higher alpha value on each day. ....................................... 58

Figure 14. ,6 plotted against time for PP and MS 2001-06-11 through 2001-06-24. The
beta values are closer than the alpha. ........................................................................ 58

xii

Figure 15. p plotted over time for PP (Paris Park) and F S (Ferris State) from 2002-05-20
to 2002-06-20 (dates are in Julian format). Overall, PP has a relatively stronger
biophony component than F S. .................................................................................. 61

Figure 16. p plotted over time for PP (Paris Park) and FS (Ferris State) from 2002-06-25
to 2002-08-01. Again, PP has a relatively stronger biophony component than F S. . 62

Figure 17. Map of three different classes of sampling locations (Lakes, Streams, and
Wetlands) in the Muskegon River Watershed sampled in the summers of 2001 and
2002. 80-minute recordings were made at each sample site during the time of
sampling (also appears in Appendix A). ................................................................... 65

Figure 18. Graph of the linear relationship between the anthrophony (y-axis) and the
population density (x-axis) ........................................................................................ 70

Figure 19. Graph of the linear relationship between the biophony (y-axis) and the
population density (x-axis) ........................................................................................ 71

Figure 20. Conceptual classification schema of the soundscape and its three primary
components and their typical frequency ranges. Several hypothetical subclasses are
also depicted. While the Anthrophony and Biophony tend to have discrete ranges,
the Geophony tends to occur across the spectrum. ................................................... 83

Figure 21. The frequency windowing procedure. Each spectrogram is divided into 11
frequency bands and the mean amplitude is calculated for each band. This process
allows a relative comparison of the frequency bands with the highest concentration
of acoustic activity. ................................................................................................... 85

Figure 22. Frequency range comparison of multiple signals and signal types. A)
Brachypterous cricket (Bailey, 2001); B) Pycnonotus sinensis (Ping, 1996); C1)
Cyclochila australasiae, inward (Bennet-Clark, 1997); C2) Cyclochila australasiae,
outward (Bennet-Clark, 1997); D) Poecile atricapillus (Otter, 2002); E) Zenaida
macroura (Krause, 1987); F) Chaetura martinica (Krause, 1987); G) Terpsiphone
paradisi (Krause, 1987); H) Locomotive; I) Motor Boat; J) Air Conditioning. As the
distribution indicates, the anthropogenic signals generally tend to occur at lower
frequency ranges than the biological signals. ........................................................... 87

Figure 23. Biological and Anthropogenic activity curves over three 24-hour periods at
four sampling sites. A) Equestrian Center; B) Haymarsh Wetlands; C) Cooper
Ranch; D) Paris Park. The column of graphs on the left is the biophony curves, and
the column on the right is the anthrophony curves. Sites C and D appear to show the
least similarity between the temporal distribution of anthropogenic and biological
activity ....................................................................................................................... 92

Figure 24. Diurnal activity trends of the entire analyzed spectrum at a single site (Gage
Home) at four different times of year, representing four different seasons. While a
single day is a rather small representation of an entire season, the figures appear to
illustrate seasonal variations in activity trends. ........................................................ 93

xiii

Figure 25. Map of three different classes of sampling locations (Lakes, Streams, and
Wetlands) in the Muskegon River Watershed sampled in the summers of 2001 and
2002. 80-minute recordings were made at each sample site during the time of
sampling. ................................................................................................................... 94

Figure 26. 0t (anthrophony) region plotted against B (biophony) region for the three
classes of sampling sites (Lakes, Streams, and Wetlands) in the Muskegon River
Watershed. The maximum range of each class of samples is indicated in by the
boundary lines. The wetlands encompass the largest range of activity, while the
lakes encompass the smallest. ................................................................................... 96

Figure 27. Acoustic Intensity of band 4 (3 — 4 kHz, 3 highly biological band) plotted
against temperature throughout April 2002 at 22:00. These plots indicate that the
intensity of activity in band 4 increases with rising temperature .............................. 98

Figure 28. Acoustic Intensity of band 4 (3 — 4 kHz, a highly biological band) plotted
against temperature throughout September 2002 at 22:00. These plots indicate an
even sharper increase in activity once a temperature threshold of roughly 65° F is
surpassed. .................................................................................................................. 99

xiv

Introduction
The Need for a Dynamic Ecological Variable

General Perspective

Presently, the human population is expanding development throughout the globe. In
terms of both overall population and land use intensity, we are rapidly altering the
structure of Earth’s ecosystems. From changes in species composition to habitat structure
and type, we are altering the operation of ecosystems on multiple scales. Fortunately, we
are also working to understand our impacts on these ecosystems that we occupy. By
enhancing our understanding of the ways in which we affect ecological functions, we
may be able to mitigate or minimize the severity of the negative consequences of our
continued development. For instance, the National Academies' National Research
Council outlined eight critical research topics for the next generation of environmental
studies (National Research Council 2001):

o Biogeochemical Cycles

0 Biodiversity and Ecosystem Functioning

- Climate Variability

o Hydrologic Forecasting

0 Infectious Disease and the Environment

0 Institutions and Resource Use

0 Land-Use Dynamics

0 Reinventing the Use of Materials

Inherent in the study of Biodiversity and Ecosystem Functioning and their relationship to
land use and land cover patterns is the need for ecological variables that reflect the
dynamic nature of ecosystems. We are beginning to realize that ecosystems function in
dynamic equilibrium, with constant variations and changes, rather than as steady-state
systems (Odum 1963; Laszlo 1996). This understanding has led to the need for a
measurable variable that reflects these spatiotemporal dynamics of patterns in ecological

activity.

Sound as a dynamic variable

Sound, being one of the five basic senses, has historically been an overlooked variable in
ecological studies. In the past, this is likely due to the complexity of an acoustic signal
and the equipment necessary to capture and analyze the signal. However, the present
availability of relatively inexpensive computer technology has made the digitization and
quantification of acoustic signals more feasible. In terms of key criteria for ecological
indicators and biocomplexity assessment (Dale and Beyler 2001), sound is an optimal
variable as it meets the following requirements:
0 The analysis of sound simplifies the interpretation of complex biological
measurements by integrating several factors into a single variable.
0 Analysis of acoustic features integrates different biocomplexity measures because
the signals are tied to multiple other variables.
0 Continuous stationary acoustic monitoring reveals spatiotemporal patterns that

cannot be captured in single-point site-by-site observations.

0 Ecological acoustics can be measured automatically with minimal human
interference.
Therefore, I have decided to examine the role of sound as an indicator of ecological
quality and examine derived analytical relationships to other features using data from the

Muskegon River Watershed.

Muskegon River Watershed

The Muskegon River Watershed is the second largest watersheds in the state of
Michigan, draining a total surface area of 2,723 square miles. It was formerly part of the
vast timber operations in Michigan, and much of the forests present now are secondary or
early successional. Partially due to its size (219 miles in length), it is a diverse watershed
that encompasses many of Michigan’s different habitat types. This region has also been
the subject of several recent intensive ecological studies as part of a proposed model
watershed assessment for the state. The data for my thesis is derived from one such study
that is examining multiple variables in an attempt to assess the quality of the watershed
and outline necessary steps for restoration (Stevenson et a1. 2001). Given the topic of this
study, it has generated a significant volume of data on the Muskegon River Watershed,
including multiple acoustic samples throughout the watershed. These factors make the
Muskegon River Watershed an ideal test bed for the establishment and implementation of

a large-scale study of ecological acoustics.

Goals and Objectives

The overarching objective of this thesis is to develop and verify an analytical approach to
the quantification and interpretation of ecological acoustics. To accomplish this, I
focused on the following three objectives:

0 Design and implement an analytical framework to assess the ecological features
of acoustic signals and to handle large-scale acoustic data for comparison and
statistical analysis.

0 Verify the apparent spectral distribution of biological signals and enhance our
understanding of biophony.

0 Integrate the theoretical and analytical tools to a real-world scenario by relating

acoustic signals to other ecological variables in the Muskegon River Watershed.

Methods and Outline

To accomplish the above objectives, 1 have divided this work into five sections, which
comprise the five chapters of this thesis. They begin with the general principles and
framework for the interpretation of ecological acoustic signals, and develop into an
analysis of the acoustic properties and correlated variables in the Muskegon River
Watershed. The chapters are organized as outlined below:

Chapter One: Introduction to Ecological Acoustics

This chapter comprises a general overview of the current status of acoustics research, and
reviews the background information necessary for the development of an interpretive

framework.

Chapter Two: Development of an interpretive framework

This chapter reviews the fundamental principles of the standardized framework for
interpretation of ecological acoustics developed by Gage (Gage et a1. 2003). This
includes the justification of acoustic measurements and the process involved in
quantifying the signals

Chapter Three: Automation System to Facilitate Analysis of A caustic Signals at the
Landscape Level

This chapter describes the necessity of automation when working with large-scale
acoustic data and reviews the automation system I developed to handle large-scale
analysis.

Chapter Four: Spectral analysis of the acoustic signals of three groups of organisms in
the Northeastern United States

This is a spectral analysis of the vocalizations of three groups of organisms and their
spectral frequency distributions. This study is an attempt to enhance our understanding of
the biophony and its constituent members, as well as to verify the spectral frequency
distribution of acoustic signals in the environment.

Chapter Five: Acoustic Signatures of Different Locations

The temporal properties of different land cover types are compared to discern differences
in the acoustic signatures of land cover types.

Chapter Six: Establishment of baseline relationships between acoustic signals and

population density in the M uskegon River Watershed

After establishing an operational analytical and interpretive system, I applied the
principles to several sets of acoustic observations from the Muskegon River Watershed

and correlated the results to census block data from the 2000 census.

Chapter Seven: Summary and Conclusions
The research described in this thesis uncovers the fundamental principles of the analysis
of acoustic signals from an ecological perspective. Specifically, it describes the analytical

processes involved in quantifying large-scale samples from the soundscape.

Chapter One:
Introduction to Ecological Acoustics

Introduction

Acoustic information plays a significant role in the life history and behavior of many
animals, including humans. While we receive constant auditory input, most of this
information goes unnoticed unless it becomes a nuisance or ceases suddenly.
Unconsciously, however, sound has a dramatic impact on an individual’s emotional state
and decision-making. The increasing prevalence of personal music devices such as MP3
players and portable CD players reflect our desire to optimize our acoustic environment.
In addition to the psychological element in humans, sound functions in ecology in a dual
role as both an indicator of ecosystem activity, particularly when related to human
disturbance and biological diversity, and a stressor on ecosystem function and services.
As an indicator, the array of acoustic signals in a location represents features such as the
biological composition, the intensity of anthropogenic activity and the diversity of vocal
organisms. An examination of a location’s acoustics from the perspective of stressors
implies that certain acoustic signals may inhibit or degrade certain ecological services
deemed critical or relevant to human health. A thorough understanding of the properties
of ecological acoustics from both perspectives will allow for the quantification and
characterization of ecological features that may otherwise be too difficult or costly to
measure. To accomplish this, these properties must be studied within the context of a
comprehensive framework that accurately standardizes the interpretation, quantification

and synthesis of acoustic signals.

Physical Characters of Sound

In physical terms, a sound wave is a flow of energy in the form of lateral vibrations
through a medium capable of oscillation. A sound wave may be defined by these physical
components of oscillation, including the frequency (measured in Hertz), the amplitude,
which translates into the intensity (often in dB SPL, or Decibels Sound Pressure Level),
and the temporal attributes of signals length, periodicity and change. Sound waves also
exhibit more complicated physical properties, such as harmonics, which are essentially
integer multiples of a pure tone’s fundamental frequency (Hartmann 1998), and the
patterns of reflection and deterioration of signals in a three-dimensional setting. As a
sound wave is a pressure wave, many of these physical properties are largely contingent
upon the characteristics of the medium as well as the origin. Its status as a pressure wave
also implies that a sound signal will not propagate in a vacuum, as there is no vibrational
medium. When a medium is present, the energy of a sound wave decays at the same rate

as other pressure waves, one over the square of the spherical distance proportional to the
propagation properties of the medium (decay oc k (l/d 2 ) ). These physical characteristics

of sound waves have been an intensive field of study and research, including analyses of
the speed of sound waves in an array of media at various temperatures (Lide 2004). These
basic physical attributes of acoustic signals'should be key components of any interpretive

framework.

Human Perception of Sound

In addition to its physical elements, an entire field of physics, referred to as
Psychoacoustics, has arisen that examines the human perception of sound. This field of
study examines sound as it relates to human physiology and psychology. Psychoacoustics
does incorporate the physical components of sounds and tones, but focuses on different
features and utilizes a broader terminology than that applied to the basic physical
analysis. For instance, human perception of pitch is related to frequency, but not in a one-
to-one ratio. Therefore, an understanding of the ways in which humans perceive changes
in the pitch of a signal requires knowledge of both the signals actual change in frequency
and knowledge of the abilities of the human ear and brain to discriminate and interpret
such changes in signal frequency. The musical scale is structured around a combination

of both of these elements of signal structure and perception (Hartmann 1998).

While the ideal human range of frequency perception ranges from 20 Hz to 20,000 Hz,
the average human can typically discern frequencies ranging from 30 to 17,000 Hz.
Given the broad array of sounds in this range, the hearing of many vocal organisms
(particularly birds) also falls in this range, albeit with a significant degree of variation
(Heffner and Masterton 1980; Bailey 1991; Dooling et a1. 2002). Although the audible
spectrum extends to 20 kHz, information carried in human speech usually only requires
the lower portion up to 4 kHz (which is, incidentally, the maximum frequency detectable
by most telephone transducers). The dynamic range of the human voice, however,
extends across the audible spectrum. Music also fills the audible spectrum, but the
majority of information-carrying energy is still concentrated at the lower end. Human
perception of sound tends to be compressive, where slight changes in frequency are

significant at low pitches, and become increasingly difficult to perceive as the frequency

rises (Masterton et a1. 1969). Other vocal organisms tend to utilize different portions of
the spectrum, such as birds, which tend to concentrate their signals at a range between 3

and 7 kHz.

Sound in Ecology

Traditionally, sound studies have held a somewhat limited application in ecological
analyses. A significant obstacle here has been the difficulty involved in gathering a
sufficient volume of usable acoustic information from a given environment. Prior to the
advent of digital technology, the only medium available for sound storage was a spool of
magnetic tape run along a steel head (Hopp et al. 1998). These cassettes generally have a
less than optimal signal to noise ratio and their fidelity deteriorates with each successive
playback. Moreover, a single cassette can typically hold no more than 80 minutes of
reasonable quality audio information. While these technological limitations have limited
the feasibility of large-scale ecological studies of acoustics, they have allowed
researchers to examine the acoustic behavior of vocal organisms in depth. Commonly,
sound is used to track populations of vocal organisms, or studied as a means of organism
communication (Heffner and Masterton 1980; Shackleton et al. 1991; Bukhvalova and
Zhantiyev 1994; Greenwood 1996; Kroodsma and Miller 1996; Buskirk 1997; Hopp et
a1. 1998; Bailey et al. 2001; Slabbekoom and Smith 2002). This has yielded a large
amount of information about the vocalizations of specific organisms and their '
physiology, but relatively little information about sound in the environment itself. One
significant exception to this trend is, of course, the Navies of various nations. Submarine

warfare in particular, which relies largely on the ability of the submarine to remain

10

concealed and passively collect information, has made significant progress in the analysis
and interpretation of acoustic signals. These techniques have been applied successfully to
the detection and identification of aquatic mammals such as whales and dolphins. While
these techniques are theoretically applicable to terrestrial acoustics, the transition is by no
means straightforward. To begin with, the propagational properties of water and the
atmosphere differ significantly, and water is a much more conducive medium (the speed
of sound in seawater at 25°C is 1,535 m/s as opposed to dry air at the same temperature,
346.3 m/s (Lide 2004)). This implies that acoustic signals must be sampled at a higher
spatial resolution in terrestrial studies to compensate for the faster signal decay and
obtain an equitable volume of information. Partially due to differences in the conductivity
of these media, as well as the intended application, the transducers utilized in marine and
terrestrial acoustic sampling differ. Finally, naval applications focus primarily on the
detection and identification of specific signals. While such a system is applicable to
automated species identification for species richness assays, it still lacks the broader
scope of an ecological observation system designed to measure acoustics at the landscape

scale.

In both marine and terrestrial acoustics, the advent of digital technology has resolved
several of the fidelity and storage limitations incurred by traditional analog recordings.
Digital devices such as MiniDisc and flash memory recorders allow researchers to deploy
large-scale observation systems that collect data at a sufficiently higher fidelity and larger
concentration to allow an accurate analysis and interpretation of the information. As

technology has improved the availability of acoustic information, ecologists have

11

developed new applications for acoustic studies. For instance, the larger volume of
available information allows researchers to examine assemblages of vocal organisms at
the community rather than individual level (Bailey et al. 2001; Nischk and Riede 2001;
Schwartz et al. 2001). Additionally, researchers have begun to examine novel
applications of sound studies beyond the observation and tracking of individuals or
species complexes. Several initial and proposed studies indicate that acoustic signals may
be used to derive species richness estimates for the extrapolation of biodiversity indices
(Home 2000). For instance, Reide ([Riede, 1993 #96]) used the frequency modulation
across species of Amazonian rainforest crickets to estimate richness of cricket species in
various portions of the forest. Of course, acoustic identification of species has
traditionally been used in bird surveys, particularly in forested habitats where visual
identification ranges from difficult to impossible (Gill 1995). Another potential
significant contribution of acoustic studies is in systematics. Several researchers have
suggested and begun to apply analyses of the variations of acoustic signals to the
identification and recognition of individual species (Sandborn and Phillips 2001; Helbig
et al. 2002; Freeberg et a1. 2003). Finally, another important aspect of ecological
acoustics is the relationship between the physical environment and the acoustic signals.
Features such as vegetation, topography and meteorology will all affect the propagation
and assemblage of acoustic signals in a landscape (Roffler and Butler 1967; Aylor 1971;

Morton 1975; Wollerman 1999; Benoit-Bird and Au 2001; Nischk and Riede 2001).

Another branch of acoustic studies, focused primarily on the examination of human

sound as a stressor, has begun to examine the increasingly significant anthropogenic

12

mechanical contributions to the assemblages of signals in a region. This increasing
presence of mechanical signals and its effect on human health and behavior is a large
focus of the field of acoustic ecology. The implications of mechanical noise extend, of
course, beyond the realm of human health and influence anthropogenic degradation of
ecological services and conditions (Schafer 1977, 1994; Krause 1999; Truax 1999;
Wollerman 1999; Krause 2001). The Federal Aviation Administration, for instance,
conducted a large-scale survey of the effects of sonic booms on humans and wildlife
(Federal Aviation Administration 1985). Taken together, the examination of acoustic
signals as indicators and stressors enables the development of an interpretive framework
that yields information about both elements of the assemblage of signals present within a

region.

While this array of research has provided a great deal of usefiil information about the
various ecological elements of acoustics, a large-scale interpretation requires a
standardized framework for a universal interpretation of ecological acoustics. This
research was the foundation upon which the interpretative framework described in the

next chapter was constructed.

l3

Chapter Two:
Development of an Interpretive Framework to Assess
Ecological Features of Acoustic Signals1

Introduction

A large-scale interpretation of a region’s acoustic features (or its soundscape) requires
standardized implementation of observation techniques and interpretative analyses. Given
the wide array of acoustic studies that have already been conducted, an ideal interpretive
approach would integrate these studies as well as future studies into its framework. The
framework described here attempts to do this by building upon some of the initial work in
acoustic studies. While the interpretation described in this framework does attempt to
unify standing concepts in acoustics, the quantification and statistical approaches
described in detail in the next chapter represent a novel application of existing

techniques.

Definition of the Soundscape

The term “soundscape” occurs frequently in the field of acoustic ecology. The working
definition of soundscape in this field is any collection of sounds specified as an area of
study (Schafer 1977, 1994; Truax 1999). This implies that the term soundscape may
apply to anything from a musical composition to the entire planet. In the study of the

ecological aspects of acoustics, I have defined the area of study as the set of sounds

 

' An article similar to this chapter was originally written with me as the second author. Because I was not
the first author, I have attempted to revise the original paper to reflect my own input. I did include segments
of the original paper where I was largely responsible for the text (primarily the segment on the analysis of
signals). For the sake of clarity and honesty, I have included the original paper as it was initially submitted
in Appendix 1.

l4

generated by the biophysical and social interactions and activities within a landscape,

where the landscape is a heterogeneous land area composed of a cluster of interacting

ecosystem patches (Turner et al. 2001). The working definition of the soundscape, then,

is the acoustic signals associated with the landscape and its constituent habitats.

Classification of Signals in the Soundscape

A useful characterization of an ecosystem’s soundscape involves a classification system

 

Soundscape
Acoustic Spectrum from 40 Hz to 11 050 kHz

Geophony

WNWM
Ann-mush 11.050“:

Anthrophony

40Hzto2kHz

Biophony

2.5kHZto 11.050 kHz

Intentional
Conmnication
Among

     
  
  

 

 
    
 
   

   
 

Oral Incidental
' Signals Caused

by Organisms

 
 

Mechanical
mm
um

i Stationary
‘ Consistent or Periodic

‘ Temporal
Non-oenodic or Briei

       

 

 

 

Figure 1. Conceptual classification schematic of the soundscape and the
three principal components. Several hypothetical subclasses are also
depicted below these components. While the Anthrophony and
Biophony tend to have discrete spectral ranges. the Geophony tends to
occur across all spectral bands, but is more diffuse.

that describes the signals
in terms of their
biophysical components.
Generally, various signals
in the acoustic spectrum
are thought of as
originating from either
natural processes or
human activity. This
analytical framework
distinguishes three main
categories of sounds that

occur in the soundscape:

biophony, anthrophony, and geophony. The term biophony describes the complex chorus

of ambient and prominent biological sounds encountered in a region. In the analytical

framework, this category encompasses only the natural sounds produced by organisms

other than humans, including birds, amphibians, insects and bats. This is the class of
sounds most extensively studied in ecological acoustics (Bosch et al. 2000; Bailey et al.
2001). While human oral signals would technically be considered a component of the
biophony, I have decided to classify them separately for two reasons. The first is that one
of the primary objectives of this research is to isolate and quantify the degree of
anthropogenic disturbance within an ecosystem. To do so, the assemblage of
anthropogenic signals, including oral, must be treated separately from other biological
organisms. Similarly, if my objective were to quantify the impact of ducks’ acoustic
activity in the soundscape, I would group anthropogenic activity with other organic
activity, and place duck signals in a separate category (perhaps Anserophony?) However,
as anthropogenic activity is my current focus, anthrophony refers to the collection of
anthropogenic signals in the soundscape. The simplest classification of the anthrophony
divides anthropogenic signals into either vocalizations or signals produced by mechanical
or technological means. Of these two subclasses, the mechanically induced signals
comprise a majority of anthropogenic signals in most samples, with negligible oral
components. This is the second reason I have opted to place anthropogenic vocalizations
in a separate category. While the theoretical basis may be debatable, the practical
implications are merely an increased conceptual convenience. Finally, the third category,
geophony, refers to the pattern of signals present within the soundscape generated by
physical (primarily geological) processes occurring in the region. Examples of these

classes of signals are those emanating from waterfalls, river flow, wind or rain.

16

This classification system represents a very simplistic and rudimentary approach to signal
categorization. As understanding of the characteristics of sounds in these categories
increases, this classification system may be refined. For instance, subdivisions in the
classification can be made based on the persistence of the signal (stationary versus
temporal), the function of the signal (intentional versus incidental) or the periodicity of
the signal (periodic versus random). However, for the purpose of this research, I focus
my analysis and discussion on the three major classes mentioned above, while still
outlining some of the more complicated subunits in Figure l for illustrative purposes.
Moreover, the simplicity of the approach has expedited the interpretation and enabled the
development of the quantitative framework built on a minimal number of assumptions
that may be corrected or modified without necessitating the redevelopment of the entire

framework.

Quantification of Acoustic Signals

The analytical system I have helped to develop is designed to analyze acoustic samples
with strictly standardized parameters. While a more flexible system would be preferable,
the volume of information and my extremely limited programming skills necessitated the
establishment of a relatively inflexible system. The analytical parameters can be adjusted,
but such an undertaking would require substantial modification of the software system.
Therefore, the initial analysis utilized slightly broader parameters that were narrowed as
the necessary minimal standards were determined. After examining the amount of
information distributed across the audible spectrum (20 — 20,000 Hz), it was determined

that the majority of information is concentrated below 1 1 kHz. This results in a sampling

l7

rate of 22.050 kHz, twice the maximum frequency to be analyzedz. To optimize signal

fidelity while minimizing file size, single-channel (monaural) 16-bit samples are used.

These sampling parameters are relatively high quality, although not quite on par with

standard CD quality (CD standards are 44.1 kHz 16-bit stereo samples).

Visual Representation of A caustic Samples

To maximize the amount of information available while simultaneously minimizing the

size of the data files, the acoustic samples were converted into visual spectrograms.

These spectrograms are essentially 3-dimensional representations of the original acoustic

signals. The frequency of the signal is plotted on the y-axis, time is plotted on the x-axis

 

11<r

O
.
v

Frequency (kHz)

N 0!
. a
u -

 

 

Tlmo (mace)

 

 

 

Figure 2. A spectrogram with a maximum frequency of I 1.025
kHz and a 90 dB black and white palette. This spectrogram also
illustrates the division of biological and anthropogenic signals
into discrete frequency bands.

and the z-axis represents the
amplitude (Figure 2). The
generation of a spectrogram
utilizes a series of Fast
Fourier Transforms (FFT),
which calculate the
amplitudes of the integrated
frequencies over time. The

spectrograms utilized in this

analysis were generated by a program called Spectrogram® (Home 2001) with a F FT

size of 4,096 and a frequency resolution of 5.4 Hz. The theoretical minimum frequency

was 0 kHz (although the minimum frequency of most microphones used was 20 Hz), and

 

2 This sampling rate is dictated by the Nyquist Sampling Theorem, which states that the minimum sampling

rate must be twice the maximum frequency.

the maximum was 1 1.025 kHz. The amplitude was scaled across 90 dB with a blue color
palette. Spectrogram® also generated all of the spectrograms with identical dimensions

(500 pixels high x 1,000 pixels wide).

Quantification of the Spectrogram

The functionality of IDRISI® (Clark Labs 2000) allowed me to analyze the spectrograms
in terms of topological and x, y spatial features within the signal, i.e. the spectrogram was
treated as a multiphase map of the acoustic signal. Sound intensity values were converted
to an 8-bit series with a range of 0-255 possible values related to the initial dB values of
the signal. The maximum dB value that the microphone could relay to the computer had a
corresponding analysis value of 255. On the other hand, an acoustically silent (i.e. no
energy above 10'12 watts/metre2 was captured by a microphone) recording would receive

a corollary value of zero.

The spectral ranges of acoustic signals in the environment tend to aggregate within two
primary regions consistent with the anthrophony and biophony. The first region occurs at
the lower frequencies of the sound spectrum (Schafer 1977, 1994). This band typically
extends from 0.4 to 2 kHz and consists primarily of mechanical signals (i.e. trains, cars,
air conditioners, etc.), and is therefore aptly referred to as the anthrophonic region. The
second band of concentration begins in the range of 3 kHz and is prevalent up to 8 kHz,
though it may extend to the top of the spectral range of the recorded signal (1 1 kHz). This
realm of acoustic activity consists primarily of signals generated by biological organisms,

and is therefore referred to as the biophonic region. This frequency band is classified as

19

the biological band based on observations of the data collected and the frequency ranges
referred to in the literature (Shackleton et a1. 1991; Naguib 1996; Ping et al. 1996;
Bennet-Clark 1997, 1998; Bennet-Clark 1999; Bosch et al. 2000; Bailey et al. 2001;
Rundus and Hart 2002; Freeberg et al. 2003). These two bands correspond to two of the
three taxonomic categories of the soundscape described above, but do not cover acoustics
emanating from the physical (i.e. wind, rain, etc.) or geophonic component. This is
because the geophony, when present, occurs as a signal that is diffuse throughout the
entire spectrum. Generally, when a geophonic signal is present, the frequency bands
above 8 kHz will exhibit greater signal intensity than in signals without geophony. When
geophony is present, it may be detected, therefore, by its tendency to generate stronger
signals at the higher frequencies above the predominant range of the biophonic spectrum
(that is, strong signals above 8 kHz). Using this partitioning of the acoustic spectrum, I
helped to develop a methodology to quantify the three primary acoustic elements, the
anthrophony (a), biophony (fl), and geophony (y), by calculating the mean value of
acoustic intensity in the spectral frequency range allocated to each of these regions. In
some cases, the mean value of these spectral ranges was divided by the mean activity of
the entire signal (a) in an attempt to “normalize” the intensity of the signals so that same
bands of signals with different overall intensities could be compared. We calculated the
mean values of the a, ,6, rand abands by assigning a numeric value to each pixel (0 —
255, mentioned above) in the specified range and calculating the mean value of the z-
axis. The a, ,6 and yactivity ratios were calculated using the equations in Table l. A
value > 1 for any of the indices indicated that the mean concentration of acoustic activity

in the analyzed region was greater than the average value for the entire signal. Therefore,

20

the region with the highest value was the predominant source of acoustic activity in the

signal. For example, if the ,6, had the highest value, then biological activity was

predominant, while a larger 0', value indicated dominant anthropogenic activity. As Table

1 indicates, the range of the geophony actually falls within the biophony range. This is

because the geophony is typically a diffuse signal without a particular spectral range.

When a significant geophonic component is present in the signal, however, there is a high

concentration of activity from 8 to l 1 kHz.

 

 

 

 

 

 

Index Ratio Percentage
a
r—[aj a”: M x100
21 lLeve/s
Anthrophony
a=Meanfrom0toZkHz . . .
a, : Ratio of anthropogenic at, = Percentage of actrvrty 1n the
activity mean to grand mean anthrophony band
p 23an
= _ : -——— X 100
3' (0 fl" leLevels
Biophony
:6 2 Mean from 2 to 11 kHz ,6” 2 Percentage of activity in the
,8, 2 Ratio of biological activity biophony band
mean to grand mean
7 L81 L11
7, =(—] ,z 21—3—— x100
0' ’ 21 1 Levels
Geophony '
7: Mean from 8 to l 1 kHz x, 2 Percentage of activity in the
)4 2 Ratio of geological activity geophony band
mean to grand mean
p = (g) Global Variables
, , a L = 1 kHz level
Actrvzty

 

p = Ratio of biological to
anthropogenic activity

 

0': Mean value of entire signal
(Grand Mean)

 

 

Table I. Formulae for the calculation of activity concentration values for the three primary regions
of the soundscape. The column on the left depicts the formulae for the calculation of ratio values in

21

terms of the entire spectrum, while the column on the right lists the formulae for the calculation of
values in terms of percentage of the entire spectrum (from Gage et.al. 2003).

In some cases, we divided the ,6 value by the avalue to calculate p, the ratio of biological
to anthropogenic activity, or the Index Value to emphasize the comparison of biological
and anthropogenic activity. In addition to computing the ratios of activity from our
classification system, we also determined the percentage of total activity a single band
contributes to the total signal (Left-hand column of Table l). A )f, value near 100%
coincident with a ,3}, value of approximately the same value indicated that the primary
signal source in the sound sample was geophony (geo-physical) activity. When the 0;,
value was greater than 50%, it indicated that the primary signal source was anthrophony
(anthropogenic) activity, whereas a value of ,4, greater than 50% indicated that biophony

(biological) activity was the dominant source.

Index Value Assumptions

The implicit assumption underlying the interpretation of the anthrophony and biophony
index values is that regions with higher biophony values and lower anthrophony values
represent systems with less anthropogenic stressors than systems exhibiting the opposite
features. This rests on the twofold assumption that regions with fewer disturbances will
retain a larger concentration of vocal organisms and will exhibit lower concentrations of
anthropogenic activity. These two assumptions generally hold true in environments that
are not in geophysical extremes (i.e. desert or arctic systems), and less disturbed systems

do generally exhibit more biological activity (Krause 1998, 2002). These derived index

22

values then represent an indirect method to quantify the degree of anthropogenic stress in
a system. Therefore, a large-scale assessment and analysis of ecological acoustic signals
should provide information about the condition of and stress on various habitats. While it
may require some refinement to handle regions outside the implicit assumptions, this
analysis presents a basic methodology to enable the analysis of acoustic signals in terms

of anthropogenic stressors and valued ecological attributes.

23

Chapter Three:
Application of the interpretive framework to landscape-
Ievel acoustic data

° Introduction

The interpretive framework described in chapter two establishes the fundamental
principles of the analysis of acoustic signals from the environment. While this framework
is an accurate method of analysis for individual samples, the labor and processing
required prevents the application of this analysis to landscape-level or large-scale
acoustic surveys without the implementation of a system capable of tracking and
organizing multiple samples. Unfortunately, the software to process multiple image files
through a GIS program, extract, and organize the statistical results was not readily
available. Therefore, I aided in the development of an interface layer that handles the file
processing and automation component. This automation system is referred to as the
Ecological Soundscape Analysis System (EAS), and represents the first step in the

automation of ecological soundscape assessment.

The Need for Automation

The analytical processes described in chapter two apply to a single file. However, an
accurate assessment of a region’s acoustic properties requires multiple samples over both
space and time. As Gage et al. (submitted, see Appendix A) demonstrate, the

infrastructure to acquire acoustic samples of a sufficient spatial-temporal scale is feasible,

24

but requires a significant data-management component. A series of sensors deployed in
the Muskegon River Watershed have been gathering high temporal resolution samples for
the past three years. The acoustic data analyzed from the Muskegon River Watershed (see
Chapters 5 and 6) was obtained from some of these sensors, as well as from several
manual and volunteer recording networks in the watershed. Each of the recording
instruments that were part of the cyber infrastructure recorded signals at half-hour
intervals throughout the day. This resulted in 48 files from each instrument every day, or
17,520 files per year. Samples from manual or volunteer recordings were generally from
80-minute Minidisk recordings. While the analysis of an 80-minute sample is feasible,
the time and processing required by a computer to perform such an analysis would
rapidly swamp the resources available to this project. Therefore, these samples were
broken into twelve 30-second subsamples. This process both alleviates the processing
requirements and adjusts the Minidisk recordings to the 30-second standard utilized by
the rest of the computational infrastructure. This also implies, however, that each
Minidisk sample translates into 12 actual samples to be processed and analyzed. In
addition to these initial files, several steps in the analytical process generate multiple new
files based on these originals. Table 2 below lists the various numbers of files generated
by the analysis described in the previous chapter.

Manual

Recording Instrument Stations

Month Year
# of Files 48 1 440 17
12 Bands 576 17280 21
Index Bands 144
Totals 720 21 ' 180

 

Table 2. Numbers of files generated by a single automated and manual recording system and the
subsequent analysis.

25

As the values in this table indicate, the analysis of acoustic signals at a landscape level
requires a significant volume of file processing, if each file is to be included in the

analysis. This need for an automation system led to the development of EAS.

EAS as an interface layer

EAS is essentially an interface layer between the acoustic spectrograms and the programs
required to analyze and quantify their values. It incorporates both a file management and
a file processing component. The file management aspect tracks the file names and
locations and ensures they are associated with the proper metadata. The file processing
component reads the numerical values output by the statistical analyses and organizes
them with the metadata to allow for further interpretation and analysis of the quantitative
information from the acoustic signals. As the analytical and processing demands became
more complicated, these two aspects of EAS, initially two separate programs, were

integrated into the single program that became EAS.

To understand how EAS operates as this interface layer, however, a bit more explanation
of the two separate processes it links is needed. To generate a quantitatively accurate
representation of the sound file to be analyzed, a program called Spectrogram® reads the
wave file graphs the intensity of the signal against its frequency and temporal span. This
process is described in detail in Chapter 1. In the end, Spectrogram produces a
spectrograph for each sound file. The spectrographs used for this research were all

standardized with the following parameters:

26

0 Time Span: 30,000 milliseconds, corresponding to a thirty second sample

0 The input files, as mentioned in Chapter 1, were 16-bit monaural signals.

0 Decibel Scale: 90 dB with a bluescale color palette

0 Time Scale: 30 milliseconds (each pixel represented a 30-millisecond sample)

0 A linear frequency scale was used (as opposed to logarithmic, which expands the

lower frequencies and compresses the higher).

0 Fast Fourier Transform Size: 512 points

0 Frequency Resolution: 43.1 Hz

0 Low Band Limit: 0 Hz (i.e. lowest possible)

0 High Band Limit: 11,025 Hz (Highest possible while adhering to the Nyquist

Sampling Theorem)

The sizes of the spectrographs were standardized at 500 pixels high by 1,000 pixels wide.

Finally, the sonogram images needed to be resampled to an 8-bit scale in order to be read

by IDRISI®. To keep track of the large numbers of files this research generates,

important metadata was coded in the image file names. Each file name had a specific

template, based on the type of file. Table 3 below describes the different file name

templates and the code EAS assigns to them.

 

 

 

 

 

 

 

 

:12; Permanent Site Sampling Site Manual Recording
Desc Site with recording Part of the large-scale Recording made with an MD
' instrumentation sampling of the MRW or other recorder
Template AAYYYYMMDD_hhmmss CCCCCMMDD_hhmmFF AAYYYYMMDD_hhmmssFF
Code S A M

 

Table 3. File Name template used by the analysis system for the different site types.

27

 

IDRISI is the second half of EAS’ interface level. While it is a powerful spatial analysis
program, IDRISI was not designed primarily to perform repeated analyses on large
numbers of images.

To interface between the spectrograms and IDRISI, the GIS program, EAS utilizes both
Microsoft Windows and IDRISI API protocols. Based on the variables selected by a user,
EAS reads the list of input files and generates the proper macro command lines for
IDRISI to run to perform the spatial analysis. EAS then calls IDRISI and instructs it to
run the macro file at the command line level, and then waits for IDRISI to complete the
analysis. After IDRISI has finished, EAS reads each histogram generated by IDRISI and
copies the pertinent values into a single comma-delimited text file. This text file may then
be imported into a database or statistics program for further visualization and

interpretation of the data.

Stepwise Operation of EAS

To generate the statistical information from the spectrograms, EAS uses stepwise logic
that divides into nine sequential operations. This stepwise orientation of the program
arises both for the sake of programming simplicity, and because each step has been
integrated into the program as demand has necessitated, often resulting in a new version
of the original program. The section below describes the purpose and function of each

processing step in turn.

Steps I and 2: Input and Output Paths

28

In step 1, the user selects the directory path to the spectrograms from the directory list in
frame one. EAS then creates an internal file list of all the files in that directory. In step
two, the user tells EAS where to place the output from the analysis, and the desired name
of the macro file. Separating the input and output specifications allows greater
organization of the statistics, and is useful when one wishes to compare the statistics of
multiple sites. The output includes both the raster files and the histograms, so EAS
creates two subdirectories in the output directory. The ‘rasters’ folder contains all the
raster and histogram files, and the ‘results’ folder contains the final text file and any

orthographs created.

Step 3: Specification of Data Source

The samples used in these studies of acoustic characteristics are gathered in four
predominant manners: Scalable Modular Instruments (SMIs), Manual Recordings,
Volunteer Recordings, and samples gathered in conjunction with aquatic and other
measurements (Aquatic Samples). SMI sites are permanent, and record thirty seconds of
sound at every half-hour. Manual, Volunteer, and sampling recordings all use the Sony
Minidisk recorders to obtain a single 80-minute sample, which is then subdivided into 12
thirty-second subsets at five-minute intervals. The filename format is different for each of
these data types, so EAS must know which data type it is dealing with. With the
exception of the aquatic samples, all the recordings have a standard two-letter
abbreviation assigned to them. The aquatic samples use an ll-character site code, and so

the user inputs this code in step three if the data set is from an aquatic sample.

29

Step 4: Specification of Metadata

Files transferred from a minidisk or DAT (Digital Audio Tape) medium do not generally
transfer with a standard file name format. EAS reads the file metadata from the file name
string, however, so a correction is needed for these non-standard file names (See the
section regarding file names below). To correct this, a series of text boxes in step four
allow the user to enter the date and start time of these recordings, which will then replace
the metadata EAS finds in the file name. If the user checks the option box to use the
metadata, EAS will store the variables in the text boxes as strings, and insert them into
the file name parameters of the macro file at the appropriate position, so that IDRISI®
converts the file names for its rasters and histograms into the proper format. This entire
procedure does not apply to files from the SMIs, however, for two reasons. Foremost, the
SMIs do name the sound files with the standard format specified below, so EAS will not
need to change the metadata values. Second, files captured via SMIs are temporal data
sets, meaning they have multiple start times. EAS is capable of specifying only one start
time, so the SMI metadata would be distorted if EAS were to attempt to specify the same

time to all the data files.

Step 5: Two-Letter Site Abbreviation

Initially, EAS maintained an internal record of the entire list of site abbreviations that it

displayed in a drop down menu. However, the volume of sites established soon began to
increase faster then EAS could be updated. Therefore, the user now may either select an

abbreviation from the list or specify a new one in the text box below the list. In either

30

case, the program stores the two characters as a text string and enters them into the macro

file.

Step 6: Windowing Options

The ability to “window” the spectrographs into different frequency bands was the feature
of IDRISI that enabled bandwidth analyses of the acoustic signals. IDRISI will window a
spectrogram into rectangles based on the positions of the pixels in the top left and bottom
right comers of the rectangle. To work properly, however, all the input spectrograms
must have the same dimensions. After discussing this with Richard Home, the creator of
Spectrogram, he agreed to build a batch processor into Spectrogram that outputs
spectrographs of a predetermined size (1,000 pixels wide by 500 pixels high). Because
the frequency scale used in these spectrographs is linear, each pixel represents the same
range in the y-axis. Therefore, the dimensions of a given window need only be entered
once into EAS, and it can replicate this window across multiple spectrographs. To
maximize both legibility and computing resources, EAS instructs IDRISI to divide the
spectrograms into 11 bands with bandwidths of approximately 1 kHz each. IDRISI
requires that the x, y coordinates of its windows be in terms of pixels, so the number of
pixels in any given band may be calculated by dividing the number of pixels in the y-axis
(500 pixels) by the signal bandwidth (1 1 kHz). This yields a value of approximately 45
pixels for each 1 kHz band. EAS then instructs IDRISI to generate a histogram and
calculate the mean amplitude value for each 45-pixel band, which enables the comparison

of the distribution of acoustic activity across frequency bands.

31

A second windowing option, Temporal Windowing, offers the user the opportunity to
window the spectrogram horizontally, across the time domain. This option allows the
user to examine changes in the acoustic activity at a greater temporal resolution (300
msec per division). While this option is not as informative as the bandwidth analysis, it

does provide a basic indication of the variation in activity over time within the sample.

Step 7: Landscape and Other Analyses
The macro parameters in IDRISI to initiate its landscape analysis tools operate in a
manner similar to the basic IDRISI commands, so EAS may instruct IDRISI to apply
various landscape analysis algorithms to the spectrograms. The landscape analyses
offered in EAS are:

0 Diversity

0 Dominance

o Fragmentation

0 Relative Richness
For a description of these analyses and the mathematics behind them, consult Turner et
al.’s L_andscape Ecolggy (2001). If the user selects any of these analyses, EAS simply
writes the proper parameters into the macro file for IDRISI to perform upon execution of
the macro commands. While the interface to these analyses is relatively straightforward,
their translation to the different spatial parameters of the spectrogram is not, and these
analyses are not implemented frequently. Further examination of the results of various
landscape analysis algorithms and correlation between multiple variables may help to

elucidate the meaning of these analyses as they apply to the acoustic spectrum.

32

An analysis that does apply readily to the spectrogram is the Biophony Indexing. These
indices are essentially an extension of the approach used to perform the bandwidth
analysis, with different window sizes specified to generate windows for the anthrophony,
biophony, and geophony. EAS uses the same windowing macro and instructs IDRISI to
divide the fiill spectrum into the three windows based on the pixel coordinates at their

respective frequency positions.

Macro Generation and Implementation

Because EAS generates a list of macro commands in a separate file, the implementation
phase divides into two steps. In the first step, EAS writes the series of command lines for
the various processing and analyses that the user selected into the .iml (IDRISI Macro
Language) file. These command lines are based on internal processes built into the
IDRISI system, which IDRISI calls and performs based on the specified parameters.
However, IDRISI does not access these commands until the second step, when EAS calls
IDRISI and instructs it to run the macro file it created. This is done using an API
command called Run_Macro, which opens IDRISI and instructs it to run the macro file

designated in the command line parameters.

Step 8: Histogram Extraction Tool
Step 8 contains the entirety of EAS’ second function, the reading and extraction of the
pertinent statistical values from the multiple text histograms generated by the GIS

processing in IDRISI. This step utilizes Microsoft’s Scripting Runtime Library to open,

33

read and copy information from a series of text files sequentially. When EAS generates
the macro commands for IDRISI, it also creates a list of the histogram text files it expects
IDRISI to create when it runs the macro file. When the extraction tool begins operation, it
opens the first file in the list, scans it for the pertinent lines of text, copies these lines into
a text file and then closes the file. It then repeats this process for every file in the list. The
file name string in the macro file uses a series of alphanumeric codes to organize the files
based on the various analyses and processes performed on the bitmapped spectrograms.
The extraction tool also reads these strings and copies the important portions into the
same comma-delimited text file that it copies the values in the histograms into to ensure

the final data is organized and readable when entered into a database.

Step 9: Generate Orthographs

The Orthographic function is a recent add-on to the EAS system. The purpose of this

 

utility is to create three-dimensional
orthographic representations of the
bitmap spectrograms. These three-
dimensional images are particularly
useful for visualization of the acoustic

activity, as well as for creating slick

 

graphics for presentations and theses.

Figure 3. An example of a slick graphic
generated from the orthographs.

 

 

The orthograph utility generates and

 

runs a second macro file, similar to the first macro file that lists IDRISI’s commands for

the actual analysis. It uses the file names and paths listed in the same internal file list used

34

for the first macro file, but it generates a much simpler series of commands that instruct

IDRISI to generate the orthographs and export them as bitmap images.

Code Structure and Logic

The program code for EAS was written entirely in Microsoft Visual Basic 6.0® because
it is one of the easiest programming interfaces to work with and because IDRISl’s
external command library is designed for Visual Basic. The user interface modules are all
built into Microsoft Visual Studio®, and have been incorporated into the program code.
Partially because of its iterative nature, the code in EAS is largely modularized into
distinct components. The majority of operation takes place in the Macro Generation
procedure, which is where the program incorporates all the user-selected options into the
macro file. Prior to pressing the “Generate Macro” button, the user must press the “Lock
Parameters” button, which locks all the options selected into the program. Then, when the
user presses the “Generate Macro” button, EAS reads all the selected options and
translates them into the appropriate IDRISI® commands. The decision tree in Figure 3
attempts to describe the order in which EAS processes the options entered by the user

before generating the histograms.

35

 

Data Input Path

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

----l

 

i
Results Output Path
G
Data Source
, ....... i. ........
: Specify Metadata 5
........ I--------
i 7
Select Abb from
1 List or Specify New

 

 

 

 

 

, Frequency Windowing E— - - - — — — — — — — 9

 

This is the minimal analytical procedure. It yields
the mean total activity for each sample.

This analysis will yield the mean for each full
sample and each frequency window thereof.

This analysis will yield the mean for each firll
sample and each index window. It may be
performed in conjunction with the windowing.

One, all, or any combination of these analyses
may be performed on the full samples, each
fi'equency window, each index window, or any
combination thereof.

Figure 4. The order and hierarchy of operational steps programmed into the EAS code logic. Each
analytical option is handled with an if...then procedure.

The output processing is slightly less complicated. The numeric codes in the file names

attached to the histograms describe the frequency bands, analyses, and metadata attached

to each file, so that the histogram extraction process only needs to read these codes and

place the values in the appropriate columns. The decision tree in figure 4 below describes

the output procedure.

36

 

Output Histograms

l

’ After the Data Source has been specified, EAS can
Data Source ------

 

 

 

 

 

 

read the histogram values into its final text file.

 

',---------------.' ,---------------..: These twooptions allow the final text file to use
: : Standard Dates fi-fl Julian Dates : .h Standard date (MM/DD/YYYY), Julian Dates. or
:_‘_______,___,__,____________ _________' both.

  

.......... . . . ‘ . .
' Leap Year '5 This rs always good to know when working wrth
I ---------- . Julian dates.

Figure 5. A hierarchical representation of the code logic within the histogram extraction utility.
Again, the various options are handled with if...then operators.

EAS reads the histogram files iteratively into a single comma-delimited text file that
stores both the calculated values and the associated metadata. To track the metadata
through the several analytical steps, EAS uses certain characters at certain positions in the
file names. This implies, of course, that each file name string is identical in length, and
has the proper information at the correct positions in the string. As was mentioned in the
description of Step 4 above, EAS can generate these standard file name strings if they are
not already present. In either case, the histogram extraction component uses these file
name strings to assign the proper calculated values to their respective sources. Figure 5

below is an illustration of how these file name strings are structured.

37

 

 

 

Site Code 3-4 »———a Time- 14-19 nalysis Type- 2223
P cm Characters from Characters from left \ Characters from left
“mm left
. YYYYMMD
Site SLIAAI DIthmssIIMNN
/
Recording Type- 1 Date- 542 Frequency Band~ 20-21
Character from left Characters from left Characters from left

Site Code- 3-4 Time- 14-19 Analysis Type- 24-25

Characters from \ Characters from left Characters from left

'9" YYYYMMD
Sampling ALIAAI DLIhhrmnssIFFILLINN
Site Recording Type 1 Date- 512 FHI— 20-21 Frequency Band 22—
Character from left Characters from Characters from 23 Characters from
left left left

Site Code- 3—4 Time- 14-19 Analysis Type- 24—25

ual Characters from left Characters from left Characters from left

. YYYYMMD
Rclc'omrdm'g ALIAAI DI_Ih1unmss|FF|LL|NN

Recording Type- 1 Date- 5-12 Fllll- 20-21 Frequency Band- 22-
Character from left Characters from left Characters from left 23 Characters from left

 

Figure 6. An explanation of the file name metadata for each of the three site types.

Conclusion

This program arose from the need for a programming layer to interface with and allow
IDRISI® to handle multiple data files and automate the repetitive processing steps that
would otherwise require intensive manual operation. This automation inevitably reduces
the flexibility of the analysis by limiting the types of valid input data and the number of
analytical options available. I have attempted to circumvent this limitation by using
modular code segments that may be updated and modified with a minimal impact on the
overall program operation. Moreover, the analysis of time-series or landscape-scale
acoustic data sets would not be feasible without the automation that EAS introduces to
the process. EAS enabled the batch processing of both the bandwidth analysis presented
in the following chapter and the acoustic samples from the Muskegon River Watershed

presented thereafter. The simple program that began as a batch-processing extension to

38

IDRISI® became a critical tool in the analysis of acoustic data on both temporal and

landscape scales via the methods described in the previous chapters.

39

Chapter Four:
Spectral Analysis of the Acoustic Signals of Three
Classes of Organisms in the Northeastern United States

Introduction

Of the three spectral regions of the soundscape, the biophony typically carries the most
diverse array of ecological information. This is because it is the region utilized by the
majority of vocal organisms in the biosphere to relay and broadcast acoustic signals.
While the anthrophony may occasionally carry intentional signals (i.e. a train whistle that
signals a locomotive is indeed approaching), the majority of persistent signals in the
anthrophony are incidental (i.e. fans, automobile motors and jet engines). The geophony,
consisting exclusively of acoustic signals initiated by geophysical processes, may affect
the composition of signals within the biophony, but does not carry a great deal of
biological information itself. Therefore, given the concentration of signals in the
biophony, the question arises of how, if at all, the signals are organized in a manner that

allows each vocal member to communicate effectively.

Spectral bandwidth is an inherently limited resource, particularly in the biophony, where
communication is the primary objective and signal reception and accurate interpretation
is critical. Faced with a similar task of signal organization, the United States Federal
Communication Commission regulates and assigns high-frequency spectral bandwidth to
television, radio, and other operators to prevent the chaos that would ensue if signals

were allowed to overlap at random. Similarly, Dr. Bernard Krause has proposed that

40

evolution and natural selection regulates the spectral frequencies and temporal properties
of vocal organisms in a manner that allows each organism to communicate with minimal
signal interference. This signal regulation operates as a form of niche partitioning
wherein each species adjusts the frequency and temporal structure of its signal to an
unused portion of the acoustic spectrum (Krause 1987). As with most evolutionary
theories regarding character displacement at the community level, the cause of this niche
partitioning remains difficult to ascertain due to the existence of a variety of possible
alternatives to competition (i.e. Is signal structure more dependent on physiological
attributes, evolutionary history, or a combination of variables?) While an extensive study
of the phylogenetic history and ecosystem structure of vocal organisms may elucidate a
single explanation, this question is second to that of whether organisms actually do
partition their signals into different spectral locations. This, too, is a challenging research
question that has been approached in several studies of the acoustic behavior and spectral
response of organisms to inter and intraspecific interference (e. g. (Bukhvalova and
Zhantiyev 1994; Bosch et al. 2000; Schwartz et al. 2001)). I have attempted to approach
this question from a larger-scale perspective by analyzing and comparing the spectral
structure of samples from three taxonomic classes of organisms (Aves, Insecta and
Amphibia). Krause proposes two primary partitioning mechanisms (frequency and time)
in his acoustic niche hypothesis. While an analysis of the timing of signals remains
impractical (i.e. the only way to satisfactorily measure the timing would be to examine
the timing of organisms vocalizations in their habitats), an analysis of the frequency

partitioning simply requires a sizable proportion of different organisms’ vocalizations.

41

Therefore, this study in part examines the degree to which different groups of organisms

utilize different spectral frequency ranges.

Birds, Bugs and Bullfrogs

I selected samples from the Classes Aves, Insecta and Amphibia for analysis because
these three classes of organisms are some of the most prevalent members of the
biophony. Additionally, each class represents diverse members and unique features of the
biosphere. Birds comprise one of the most diverse classes of terrestrial vertebrates, with a
current species estimate totaling almost 10,000 (Gill 1995). Additionally, birds exhibit a
great diversity of speciation and niche partitioning amongst the vertebrates. Insects
comprise one of the most diverse classes of vocal invertebrates, with a current species
estimate ranging anywhere from 5- to 80-million (Romoser and John G. Stoffoloano
1998). Finally, amphibians, while perhaps not exhibiting the same degree of diversity of
niches or speciation as their vocal counterparts (estimate 5,500 species), occupy an
important position as an indicator of ecological change (Duellman and Trueb 1986).
Their unique life history makes them particularly vulnerable to detrimental impacts of
human activity and disturbance. Taken together, these three classes of organisms

comprise an informative and diverse component of the biophony.

The Use of Acoustic Signals
Most vocal organisms communicate similar forms of information through their acoustic
signals. The most common information communicated includes mating status, territory

claims, alarms and warnings, and coordination. Commonly, organisms use a variety of

42

 

signals contingent upon the type of information they are attempting to communicate. For
instance, many songbirds’ alarm calls have relatively higher frequencies than their mating
or territorial signals because higher frequency signals are more difficult to localize.
Organisms also adapt their signals to habitat features that influence the relative strength
and propagation of the call. For instance, birds that live in the equatorial rainforests tend
to have shorter and simpler vocalizations than those that live in open areas, as the
rainforest vegetation tends to distort and absorb vocalizations that are more complex (Gill
1995). Among insects, some species of crickets will dig burrows with dimensions that
resonate at the frequencies of their chirps to enhance the volume and range of their

signals (Bailey et al. 2001).

Objectives

The primary objective of the study described in this chapter is to examine and begin to
understand the complex structure of acoustic signals within the biophony. The theory of
evolution through natural selection leads to the intuitive conclusion that organisms will
attempt to organize their signals within the acoustic spectrum in a manner that minimizes
signal interference, thereby minimizing the amount of energy expended on acoustic
communication. Additionally, previous research, based on observations of a series of
acoustic samples (Gage et al. submitted), has indicated that the majority of organisms’
vocalizations lie in a frequency range of 2 to 11 kilohertz. This range was estimated by
examining a series of acoustic observations and attempting to delineate where the

majority of vocalizations occurred. I attempted to verify this conclusion by examining the

43

spectral frequency range of a collection of “pure” signals (i.e. signals from commercial

recordings with background noise filtered out).

Methods

1 used a series of commercially available compilations of organism vocalizations to
obtain an adequate data set for analysis. These compilations consisted of vocalizations of
265 species of birds (Bird Songs Eastern/Central 2002), 42 species of amphibians (Elliott
2004), and 52 species of insects (Rannels et al. 1998). Appendix B contains a list of the
Scientific and Common Names of the organisms sampled for analysis, as well as their
taxonomic classification. Each species’ sample consisted of a relatively “pure” (little or
no background noise in the sample) recording of the organism’s primary mating and
contact vocalizations. The statistical research hypothesis was that the primary frequency

bandwidths of the three Classes of organisms differed significantly from one another.

Sound Signal Analysis

Each signal was downsampled from a CD-quality track to 22.050 kHz and converted into
a single-channel monaural wave file prior to analysis. A spectrogram was then generated
from each wave file with a 90 dB scale and a 30-mi11isecond pixel resolution. The
frequency analysis parameters were on a linear scale with a Fast Fourier Transform size
of 4,096 points and a frequency resolution of 5.4 Hz. The theoretical low band limit was

0 Hz, and the high band was 11.025 kHz3. These signal parameters were the standard

 

3 This high band limit is constrained by the N yquist Sampling Theorem, which states that the maximum
signal frequency must be no greater than V2 the sampling rate, or Sampling Rate = 2 X f (max)

44

 

parameters outlined in Chapter 1. Each spectrogram was then read by a raster-based GIS
program that divided the image into 11 frequency bands of 1 kHz each, and calculated the

mean amplitude value in each band based on an 8-bit scale.

Statistical Analysis

Prior to statistical analysis, the mean of each frequency band of each sample was divided
by the mean of the entire signal to yield a proportioned mean for the frequency bands that
took into account the relative intensity of the signals, thereby allowing valid comparison
across signals. These weighted means were then averaged by Class and used to generate
bandwidth histograms that depict the average distribution of activity across the l 1
frequency bands. The three taxonomic classes of signals were then compared in SAS
based on the square roots4 of the weighted means of their 11 frequency bands in an
ANOVA using a standard linear model designed to handle fixed effects with or = 0.05. To
account for the lack of independence between the 1 1 frequency bands in each sample, the
ANOVA was run with a repeated heterogeneous first-order autoregressive structure. To
determine where the significant differences occurred, the differences of the least squares
means were compared. Using the square root of the mean for each band fit the data to the
AOV conditions (normal distribution, homogeneity of variance and using the repeated

statement compensated for the lack of independence between bands).

 

4 The square root was used to fit the data to the normal distribution of residuals assumption of the ANOVA
process.

45

Results

Spectral Bandwidth Histograms

The spectral bandwidth histogram for the entire series of organisms indicated the
strongest concentration of acoustic activity was in the 3 — 4 kHz band, with an average
proportioned mean of 1.79. The 2 — 3 and 4 — 5 kHz bands were also comparatively high,
with means of 1.60 and 1.61 respectively (Table 4a). In the Class Aves, the strongest
band was 3 — 4 kHz with a mean of 1.90, while the 4 — 5 kHz band also had a mean of
1.81 (Table 4b). In the Class Insecta, the band with the highest concentration was 6 — 7
kHz, with a mean of 1.41 (Table 4c). The activity in the Class Amphibia was at relatively

lower frequencies, with a maximum mean of 2.65 at l — 2 kHz (Table 4d).

46

 

 

 

 

 

 

 

 

 

a) All Samples b) Aves

Ave ' ' Ava

blgnd Weighted_M£~l , __SE Band Weighted_MN SE
1 1.05 _ 0.09 1 1.06 0.11
2 1.15 0.07 2 1.06 0.08
3 1.60 0.07 3 1.61 0.08
4 1.80 0.07 4 1.90 0.08
5 1.61 0.07 5 1.81 0.08
6 1.18 0.05 6 1.27 0.06
7 0.95 0.05 7 0.94 0.05
8 0.73 0.05 8 0.70 0.05
9 0.49 0.04 9 0.43 0.04
10 0.31 0.03 10 0.21 0.02
11 0.22 0.02 11 0.12 0.01

c) Insecta d) Amphibia

L ' Avg TE , Avg /

m d ng' hted_NlN . 7 .SE Band WeightedJAN SE
1 0.28 0.05 1 1.99 0.22
2 0.40 0.10 2 2.65 0.24
3 0.84 0.18 3 2.39 0.20
4 1.31 0.26 4 1.68 0.17
5 1.15 0.15 5 0.90 0.10
6 1.22 0.13 6 0.51 0.07
7 1.41 0.18 7 0.45 0.13
8 1.39 0.15 8 0.20 0.05
9 1.14 0.11 9 0.12 0.03
10 0.99 0.11 10 0.12 0.03
11 0.87 0.12 11 9425-02 2.51 E-02

 

Table 4. The results of the spectral analysis of the acoustic signals of a) all 354 organisms sampled; b)
264 species in the Class Aves; c) 49 species in the Class Insecta; and d) 41 species in the Class
Amphibia. The value in the column “Band” is the high end of that frequency band (i.e. Band I = 0 — 1
kHz), and SE refers to the standard error of the means.

Based solely on the histograms, the distribution of activity across the spectrum already
appears to differ between the three Classes of organisms. The spectral plots depicted
below provide a visual representation of the distribution of activity across the spectral

band.

47

 

All Samples Bandwidth Distribution

 

18* +5

Avg Weighted MN

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

1 2 3 4 5 6 7 8

Frequency Band

 

 

 

Figure 7. Spectral bandwidth distribution of all three Classes (354 samples). Avg Weighted MN

represents the mean concentration of activity for each frequency band, which is the respective 1 kHz
spectral band.

 

 

Aves Bandwidth Distribution i
2.5 l
1 5 '5 * '

Avg Weighted MN

 

 

 

 

 

 

 

 

 

 

 

   

1 2 3 4 5

Frequency Band

 

 

 

Figure 8. Spectral bandwidth distribution of acoustic samples from 264 avian species (150 Passerines,
114 Non-passerines), with maximum activity from 2 — 5 kHz.

48

‘ Insecta Bandwidth Distribution

 

 

 

 

4 5 6 7 8 9 10 11
L Frequency Band

' l

0.4 l
0.2 , E]

o . . . . . . . . . . . l

1 2 3 .

I

Figure 9. Spectral bandwidth distribution of acoustic samples from 49 species of insects, with two
activity peaks. The highest activity concentration in a single band Is at 6— 7 kHz.

Amphibia Bandwidth Distribution

l' ,7 __7 7 7,7 7, 77,, 77,7

1

3.5 , 7 , l
3 I l

2.5 ‘ ‘ i
. l
1 1 l
l

l

71

Avg Weighted MN

0:: . D, [Z] E71713Lfi 21:.

1 2 3 4 10 11

Frequency Band

 

 

Figure 10. Spectral bandwidth distribution of acoustic samples from 41 species of amphibians. The
highest concentration of activity is at I —- 2 kHz.

49

Analysis of Variance

The ANOVA results of the Class-level analysis indicated a statistically significant
difference between the spectral bandwidth distributions of the samples based on their
taxonomic Classes with a 0.05 probability of a Type I error (p-value < 0.0001, (1 = 0.05;
Table 5). This indicated that the spectral structure of three primary members of the
biophony (Avians, Insects, and Amphibians) differed significantly within the general
spectral range of the biophony (2 — 11 kHz, Figure 6). Based on the bandwidth
histograms, amphibians tended to utilize the lower segment of the spectrum (I — 3 kHz,
Figure 7), birds tended to utilize slightly higher frequencies (3 — 5 kHz, Figure 8) and
insects tended to utilize even higher frequencies (6 — 8 kHz, Figure 9). Insects also
appeared to have the most varied signals in terms of spectral structure (relatively high

concentrations of activity from 3 to 9 kHz).

 

Type 3 Tests of Fixed Effects
Num Den

 

Effect df df F Value Pr > F
Class 2 352 0.23 0.7945
Band 10 3520 30.05 < 0.0001

Class'Band 20 3520 12.58 < 0.0001

Table 5. Results of a mixed ANOVA of the spectral properties of three Classes of organisms (Insecta,
Amphibia and Aves) across 11 frequency bands (0 - II kHz). The effect labeled CIass*Band indicates
the analysis of the weighted means by Class and frequency band.

The ANOVA indicated that the Class*Band effect was significant (which, in this case,
was the desired outcome, as the interaction effect was the variable of interest), so 1
compared the mean differences of the three Classes by band using the Least Significant
Differences (a = 0.05). All but four of the bands were significantly different. The Class

Amphibia was not significantly different from Insecta at band 5, nor did it differ from

50

Aves at bands 4 and l l. The class Aves did not differ significantly from Insecta at band

6. All other bands were significantly different (Table 6).

 

 

 

Significant
Class 1 Class 2 Band Estimate SE t Value Pr > N Difference?
Amphibia Insecta 1 0.90 0.90 6.48 <.0001 Yes
Amphibia Insecta 2 1.12 0.15 7.59 <.0001 Yes
Amphibia Insecta 3 0.81 0.15 5.58 <.0001 Yes
Amphibia Insecta 4 0.33 0.14 2.38 0.0172 Yes
Amphibia Insecta 5 -0.07 0.12 -0.62 0.5384 No
Amphibia Insecta 6 -0.39 0.10 -4.00 <.0001 Yes
Amphibia Insecta 7 -0.56 0.09 -6.24 <.0001 Yes
Amphibia Insecta 8 -0.72 0.08 -8.99 <.0001 Yes
Amphibia Insecta 9 -0.70 0.07 -10.25 <.0001 Yes
Amphibia Insecta 10 -0.64 0.05 -1 1.81 <.0001 Yes
Amphibia Insecta 11 -0.58 0.05 -11.51 <.0001 Yes
Aves Insecta 1 0.38 0.10 3.70 0.0002 Yes
Aves Insecta 2 0.38 0.1 1 3.44 0.0006 Yes
Aves Insecta 3 0.46 0.1 1 4.24 <.0001 Yes
Aves Insecta 4 0.38 0.10 3.65 0.0003 Yes
Aves Insecta 5 0.28 0.09 3.23 0.0012 Yes
Aves Insecta 6 0.00 0.07 0.06 0.9503 No
Aves Insecta 7 -0.23 0.07 -3.38 0.0007 Yes
Aves Insecta 8 -0.36 0.06 -6.10 <.0001 Yes
Aves Insecta 9 -0.44 0.05 -8.66 <.0001 Yes
Aves Insecta 10 -0.53 0.04 -13.13 <.0001 Yes
Aves Insecta 1 1 -0.53 0.04 -14.44 <.0001 Yes
Amphibia Aves 1 0.52 0.1 1 4.74 <.0001 Yes
Amphibia Aves 2 0.75 0.75 6.39 <.0001 Yes
Amphibia Aves 3 0.36 0.1 1 3.09 0.002 Yes
Amphibia Aves 4 -0.04 0.1 1 -0.40 0.6896 No
Amphibia Aves 5 -0.37 0.09 -3.81 0.0001 Yes
Amphibia Aves 6 -0.39 0.08 -5.12 <.0001 Yes
Amphibia Aves 7 -0.34 0.07 -4.73 <.0001 Yes
Amphibia Aves 8 -0.36 0.06 -5.66 <.0001 Yes
Amphibia Aves 9 -0.26 0.05 -4.87 <.0001 Yes
Amphibia Aves 10 -0.1 1 0.04 —2.65 0.008 Yes
Amphibia Aves 1 1 -0.04 0.04 -1.05 0.2948 No

Table 6. Comparisons of Least Squares Means by Class at the II frequency bands. p < a indicates a
significant difference (LSD, a = 0.05).

51

Discussion

The ANOVA results and the means comparisons indicated a significant differentiation of
frequency structures of the samples. The overlap between Aves and Amphibia at band 1 l
is probably due to the low concentration of activity in that band for both orders. The other
three overlapping bands (4, 5 and 6) are likely regions where signals may coincide
spectrally. An analysis of the temporal features of these three bands fi'om field recordings
where the three Classes of organisms occur would enable researchers to determine
whether temporal modulation occurs in regions of spectral competition. While these
results do not conclusively prove Krause’s Niche Hypothesis (Krause 1987), they do
indicate a significant degree of signal variation between groups of vocal organisms. The
bandwidth distribution of the samples also indicates that the majority of organic
vocalizations, at least in the Northeastern United States, are concentrated in a frequency
range between 2 and 5 kHz. This, of course, is a generalization, and several organisms
utilize spectral bands outside this approximate region of concentration. Of particular
significance is the relatively low frequency class of signals generated by amphibians. The
spectral bandwidth utilized by these organisms appears to overlap with the variety of
mechanical “noises” generated by human activity. This may place amphibians at a
relatively higher risk of acoustic interference from human activity than other organisms.
Human mechanical signals tend to be stronger and more continuous than organic
vocalizations, thereby “masking” organic signals when they overlap at the critical band
(Hopp et al. 1998). This region of overlap merits further investigation, as it may help, in

part, to explain recent observations of declines in amphibian populations (Alford and

52

Richards 1999). Further investigation of the manner in which organic vocalizations are
partitioned in the biophony would also reinforce the concept of frequency modulation
within the biophony. Specifically, analyses of the frequency structure of vocalizations
within the same habitats would indicate whether the signals were modulated to avoid
acoustic interference, or whether the modulation was more a byproduct of physiological
constraints. Additionally, an analysis of the temporal features of acoustic signals would
indicate whether temporal modulation occurs in conjunction with frequency modulation
in organisms with limited capacities to modulate the critical frequency bands of their
signals. The results of this investigation do indicate at the least a rough degree of signal

modulation within the biophony based on the taxonomic Classes of vocal organisms.

53

Chapter Five:
Acoustic Signatures of Different Locations

Introduction

An interesting question that arises in the examination of acoustic signals is whether
different locations and consequent land use/ land cover types exhibit unique acoustic
“signatures,” or defining characteristics. Intuitively, the differences in biophysical
elements between a wetland and a city, for instance, could lead one to conclude that the
acoustic signatures would also differ significantly. In addition to this spatial element of
variation, the acoustic activity within a region should exhibit temporal variation on both a
seasonal and diurnal scale. For instance, the Dawn Chorus is a documented event in bird
communities, and several species engage in song in the evening (Gill 1995). This vibrant
chorus of song in the morning will significantly influence the overall diurnal pattern of
acoustic activity at a location. Moreover, in the mid to late summer of most temperate
regions, the daily biophony transitions from birds throughout the day to amphibians and
crickets throughout the night. Similarly, the intensity and composition of biophonic
vocalizations varies across seasons. In the temperate regions, the winter generally
exhibits a significantly lesser degree of organic activity than does the spring, summer or
fall. The second major component of the soundscape, the anthrophony, also exhibits
some predictable spatiotemporal variations. Sounds of traffic and other human activity
are generally less intense at night, particularly as one travels farther from major urban
centers. The typical times of the highest concentration of traffic noise correlate to the

traditional morning ‘Rush Hour’, the afiemoon ‘Lunch Break’ and the evening ‘Rush

54

Hour’. While the qualitative features of a landscape’s soundscape may differ

dramatically, a quantitative measure is needed to determine the degree to which

anthropogenic activity is the causal agent for the observed differences between two given

land cover types.

The derived ratio (p) between biophonic and anthrophonic activity described in chapters

two and three should be a usefiIl quantification that retains the expected spatiotemporal

The Role of Sound

As an Ecological lndr'cafor— As a Stressor—

The integrity and dynamics of Organisms require

an ecosystem may be communication for their survwal.

correlated to the complexity of Organism population may be

that ecosystems soundscape. inversely proportional to the
degree of acoustic disruption.

AgriculturalLULC 7 , ,, Urban LUiLCW 7

Figure II. An overview of the dual role of sound in ecological research. The two
graphs on the bottom depict a summary of the total activity within two different
locations over 24 hours.

 

patterns.
Additionally, a
single-region
analysis of both
the anthrophony
and the
biophony may
elucidate

spatiotemporal

patterns in these

two features. A

visual interpretation of the temporal patterns of the ratio value may help to demonstrate

any potential exchange between anthropogenic and biophonic activity within or between

various land cover types.

55

Objectives

This chapter summarizes two separate analyses conducted at different times over the past
three years in attempts to determine whether a, ,8 and the derived p value could be used to
detect spatiotemporal differences in the data. General observations indicate that the
biophysical and human constituents within a given location tend to affect the temporal
patterns of the soundscape on a diurnal scale (Sound as an Ecological Indicator and
Stressor Slide, Figure 10). However, these initial observations are both rudimentary and
based on overall acoustic activity, rather than focused on the differential contributions by
biological and anthropogenic sources. Therefore, the two separate analyses described
here represent somewhat more comprehensive examinations of this phenomenon. By
using p in addition to a and ,6, the analysis focuses on the differential contribution of one

of the two signal classes, in addition to the overall activity.

Study One: 14 Days, 2 Locations

This study utilized data gathered at half-hour intervals over 14 days (2001-06-1 l to 2001-
06-24) at two different locations with permanent recording stations in the summer of
2001. One location, Meadows (MS), was a small private ranch located a few miles
outside of Big Rapids, MI. The second location, Paris Park (PP), was a county park
located in Paris, MI. Meadows was situated in the midst of a low-intensity agricultural
complex, while Paris Park was a fish hatchery, public campground, and the headquarters
for the Mecosta County Parks Commission. In the 14-day analysis, the signals from
07:30 and 19:30 were analyzed using the standard spectrograph analysis with the

automation system described in Chapter 3, and then the ratio of biophony to anthrophony

56

was plotted against time for both locations (Figure l I), followed by the average values

for anthrophony (Figure 12) and biophony (Figure 13).

 

p against time for PP and MS 2001-06-11 through 2001-06-24

 

1.0 ‘

‘6-
o.9 1
. \fl
0.3 4
0.7 1 ‘

0.6 \/

 

 

 

 

 

 

 

0.5 ‘
~~0—— MS.rho
+ PPJ‘hO
0.4 F r r ' T ' r 1 * I * 1
160.5 163.0 165.5 168.0 170.5 173.0 175.5

Modifieddulian
X.0 indicates 07:30, X.5 indicates 19:30

 

 

 

Figure 12. p plotted against time for PP and MS 2001-06-11 through 2001-06—24. Overall, MS has a
higher value, indicating a stronger biophonic contribution.

57

 

 

 

 

 

 

 

 

 

Alpha (anthrophony) against time for PP and MS 2001-06-11 through 2001-06-24
100 . ‘
n '
90 1
70
60 1
——0~- MS alpha

5° «4— PP alpha

160.5 163.0 165.5 168.0 170.5 173.0 175.5

M_Jrlian

 

 

 

Figure 13. a plotted against time for PP and MS 2001-06-11 through 200l-06-24. With a few
exceptions, PP has a higher alpha value on each day.

 

Beta (bi0phony) against time for PP and MS 2001-06-11 through 2001-06-24

 

90“

80‘

70‘
60*
50* v

 

 

 

 

 

 

 

40 i
+ MS bee
--‘— PP beta
30 T 7 r ‘ r ' I ‘ r ' r ' I
1605 163.0 165.5 168.0 170.5 173.0 175.5

M_Juian

 

 

 

Figure 14. fl plotted against time for PP and MS 2001-064] through 200I-06-24. The beta values are
closer than the alpha.

58

In addition to the temporal plot, an Analysis of Variance was performed to determine
whether the two sites differed significantly. Because replication was not feasible, the
separate sampling days served as pseudoreplicates. The lack of a clear linear progression
of p over time at both locations allowed this approximation to fulfill the random sample
requirement, and the ANOVA was performed with a first-order autoregressive matrix to
account for the lack of independence between samples. The ANOVA is still not entirely
statistically valid, but it does provide a rough indication of whether the two sites differ

significantly.

 

 

Effect ”3;" 05," F- Value Pr > F
Location 1 26 68.54 <.0001
Time 1 26 0.27 0.6068
Location‘Time 1 26 0.46 0.5046

 

Table 7. ANOVA results for the test of Location effects over I4 days at two locations (a=0.05). The
main effect Location is significant, while the interaction effect, Location*Time, is not.

As the ANOVA results indicate, neither the Time nor the Location * Time Interaction is
significant. Therefore, the data appears to indicate that the Location main effect yields a

significant difference between the two data sets (Table 7).

Study Two: 3 Months, 2 Locations

This study represents a more comprehensive and long-term analysis. In this case, the full
set of acoustic data was analyzed over three months at two locations. Paris Park was at
the same location described above. Ferris State was a recording site situated on the Ferris

State University Campus in downtown Big Rapids, MI. The administration at Ferris State

59

University had agreed to host the site and provide housing for me while I conducted
research in the Big Rapids area. Unfortunately, logistical difficulties with the campus
water supply shut down the delivery system in the summer housing, and we relocated
midway through the summer. As the second location consisted of a slightly different
acoustic structure, I have analyzed the data from the two Ferris State locations separately.
Set one extends from (2002-05-20) May 20 to (2002-06-20) June 20, and set two runs
from June 25 (2002-06-25) to July 17 (2002-07-17). All 48 samples from each day were
analyzed and incorporated into the time series for both locations. Again, the p value was
calculated as the ratio of biophony to anthrophony, and was plotted over time (Figure 13).
The Friedman Supersmooth (a smoothing function that uses an iterative process to
determine basic trends) was plotted to highlight any clear temporal variations in signal

strength.

60

05-20-2002 through 06-24-2002 Freidman Supersmooth

 

 

 

   

 

 

—— FS.rho
—— PP.rho
1.0 7 a
5 3 2 I
_ .. V. a
0.8 < . . I; :5 fl 8. C
r i: 7‘ ,‘ ' e V "
C n '3 3 375 g s Q 2' g ,
2 ‘5: 2 - -r ’ 8 - D .9, 5
06 - 7 an I O“ 0 (. .1 9 $ ,
.» is +.—°ui°‘°-.
- a] 2 J: . -' ‘ iii :2 2 . ”“3
0‘4 7" .51!§;_',.__. . 553's,
gtijg’ifiigfig
' - I- ~ I. g
, a» » , g. [r . A
0.2 - i ‘ f g: 7i ’i
_ $840 I ”
0.0 i
I I H I I I
245242000000 245244000000 245246000000
245243000000 245245000000 245247000000
CorrectedJulian

Figure 15. p plotted over time for PP (Paris Park) and FS (Ferris State) from 2002-05-20 to 2002-06-
20 (dates are in Julian format). Overall, PP has a relatively stronger biophony component than FS.

For both temporal sets, Paris Park exhibits a higher degree of biophony than Ferris State.
The first set shows a greater degree of variation and oscillation over time in Paris Park
than Ferris State while both sites oscillate more frequently in the second set.

Interestingly, both sites follow similar patterns of oscillation in the second set.

61

06-25-2002 through 08-01-2002 Freidman Supersmooth

j —«— Fsmo
1.0 ‘ PPII‘IO

 

 

 

 

 

0.0 T r I I
245246000000 245247000000 245248001XJOO 245249000000
CorrectedJulian

Figure 16. p plotted over time for PP (Paris Park) and FS (Ferris State) from 2002-06—25 to 2002-08~
0t. Again, PP has a relatively stronger biophony component than FS.

An ANOVA was performed using the days as pseudoreplicates, and the results indicated
a significant interaction effect (Table 3). However, the temporal plots indicate a

difference between the biophonic intensity of the two locations.

 

 

 

 

Set One
Effect Ngfm 0;" F- Value Pr > F
Location 1 62.1 318.26 <.0001
Time 47 2771 5.33 <.0001
Location*Time 47 2771 2.28 <.0001
Set Two
Effect Ngf’" 0:," F- Value Pr > F
Location 1 44 212.4 <.0001
Time 47 1947 9.47 <.0001
Location*Time 47 1947 4.3_2 <.0001

 

Table 8. ANOVA results for the two data sets from PP and FS (a = 0.05).

62

Discussion

The two studies described above represent similar approaches to the determination of a
location’s acoustic signature. The underlying assumption that the biophony and
anthrophony values represent biological and anthropogenic activity remains untested,
however. Overall, Meadow’s low-intensity agricultural cover type could conceivably
represent a lower degree of anthropogenic activity than the Paris Park outdoor
recreational facility. This conceptual difference does correlate to the acoustic signatures
insofar as the biophony is higher in the Meadow site than Paris Park. However, there are
no grounds to attribute the land cover type as the causal mechanism behind the observed
differences. Alternate explanations, such as relative distances of the microphones from
signal sources or the positions of the microphones, cannot be eliminated based on the
information available. Similarly, Paris Parkappears to represent a significantly lower
concentration of anthropogenic acoustic activity than Ferris State, an urban cover type.
While the number of humans in the landscape appears to be the causal factor for the

degree of anthrophony in the analysis, I did not include this in the study.

63

Chapter Six: A
Correlation Analysis of Acoustic Signals and Ecological
Features in the Muskegon River Watershed

Introduction

As our capacity to measure and interpret ecological acoustics increases, so too does our
desire to understand how these signals influence and represent dynamic and continuous
elements of the Earth’s biosphere. In order to be a useful measure of ecological activity,
the relationships between acoustic analyses and other ecological indicators and
measurements must be examined. To this end, I have begun to research correlations
between the derived acoustic indices (anthrophony and biophony) and population density

in the Muskegon River Watershed.

The Muskegon River Watershed has been a region of intensive ecological study within
the past five years. Various research projects and assessments, including a large-scale
collaborative assessment (Stevenson et a1. 2001), have been conducted in an attempt to
ascertain and preserve the ecological integrity of the watershed. This watershed
encompasses a significant portion of Michigan’s natural resources, including freshwater
supplies, agricultural production and fisheries (Stevenson et al. 2001). These research
projects have yielded a large array of data on multiple ecological features of the
Muskegon River Watershed. The datasets include a significant archive of acoustic
samples gathered from a variety of habitats (Gage 2004). Several of these acoustic

samples were gathered in conjunction with other variables including water chemistry and

64

land cover information. As the map below illustrates, these samples were gathered

throughout the watershed.

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 17. Map of three different classes of sampling locations (Lakes, Streams, and Wetlands) in the
Muskegon River Watershed sampled in the summers of 2001 and 2002. 80-minute recordings were
made at each sample site during the time of sampling (also appears in Appendix A).

I used the analytical framework described in Chapter 1 in conjunction with the

automation system described in Chapter 3 to quantify the acoustic samples in terms of the

65

biological and human mechanical activity, and compared these values to population
density information derived from the 2000 national census (US. Census Bureau). These
acoustic index values were based on the spectral distribution of anthropogenic and

biological activity described in the first two chapters and investigated in Chapter 3.

Objectives

The objective of the analysis described in this chapter was to examine whether
relationships exists between acoustic samples, as quantified by the index values, and
corresponding ecological variables. This assessment arose from the need to understand
and quantify a baseline relationship among variables to guide the use of acoustic
sampling in ecological assessment. Previous research had indicated that physical
constituents of the ecosystem will affect acoustic properties, but a framework to integrate
these variables into the acoustic features in a manner that also yields information about
anthropogenic impacts on ecological processes was not available until our proposed
investigation. While many studies have reported on how physical properties such as
vegetation and land cover would affect the transmission and propagation of acoustic
signals (e.g. Aylor 1971; Rundus and Hart 2002), the links in the causal chain that
connected the resultant acoustic features to anthropogenic activity did not appear to be
thoroughly documented. This thesis, as a component of the study proposed by Stevenson
et al. (2001), attempted to fill some of these gaps in our knowledge of acoustics by
examining the relationship between the composition of the acoustic spectrum as
represented by the sampling regime and the degree of human activity within the system,

as represented by the population density.

66

Methods

The three steps in this study encompassed the data gathering techniques, the analysis of
the signals, and the correlation techniques. These are described as Data Gathering, index

Derivation and Regression Analysis.

Data Gathering

As part of the collaborative assessment of the MRW, teams from Stevenson’s laboratory
sampled a series of lakes, streams and wetlands in the watershed. In addition to gathering
water quality samples, each team was equipped with a Sony MZ-R700 Recording
Minidisk Walkman and several 80-minute minidisks. At each site, the researchers used
the MD Recorder with a Sony ECM-MS907 Stereo Microphone set to a 120° pickup
range, and recorded for 80 minutes. In addition to these recordings, they recorded water
temperature, pH, DO, and other variables. The acoustic samples were stored on the
minidisks in a Sony proprietary format. The MD recorders stored 44.100 kHz l6-bit
Stereo samples. The minidisks were then recorded to 80-minute audio CDs in the same
format as Compact Disk Audio (CDA) files via a Philips CDR 775/ l 7 Compact Disk
Recorder. These CDs were then ‘ripped’ into a computer via Roxio’s Soundstream
program. Soundstream converted the CDA files to Windows Waveform (wav) files (the
standard file format used in the analytical framework), but retained the 44.100 kHz l6-bit
Stereo format. Therefore, I split the files into 12 30-second sub-samples, and output them
sequentially as separate files. The program then resampled each file at 22.050 kHz, and

spliced the two channels into a monauralsample. This process generated the 12 30-

67

second 22.050 kHz 16-bit wav files, which were the standard parameters for spectral
analysis. Prior to statistical analysis, recordings in which the sampling crews were

audible were excluded.

Index Derivation

The wav files from the data set were then used to generate standard series’ of
spectrograms with F FT (Fast Fourier Transform) sizes of 4,096 points, a frequency
resolution of 5.4 Hz and a linear frequency scale. A GIS program was used to examine
the three-dimensional spectrograms and divided each one into its anthropogenic (a) and
biological (,8) region, based on the spectral delineation of these two regions as described
in Chapters 1 and 2. The mean amplitude of these two regions was then calculated and
used as a measure of anthropogenic acoustic activity (anthrophony) and biological

acoustic activity (biophony).

Regression Analysis

The data was analyzed in SAS® to test the correlation between the derived anthrophony
and biophony values and the population density (based on census blocks) of the sampling
regions with predictive linear regression models. These models followed the basic
formula y = mx + b, where y = the mean of the variable of interest, x = the derived index
value, m = the slope of the best-fit line, and b = the y-intercept. The value for
anthrophony or biophony was used as the dependent variable, as the hypothetical
relationship proposes that population density affects the intensity of anthropogenic or

biological activity within a system. The natural log (In) of the anthrophony, biophony and

68

density values was used to minimize error in the distribution of points and bring the

values closer to a Gaussian distribution.

Results

In terms of both the strength and direction of correlation, neither the anthrophony value
nor the biophony value appeared to exhibit a significantly strong correlation to population
density (Table 9). The anthrophony had an r2 value of 0.02 and a slope of 0.39. The
biophony had a slightly higher r2 value of 0.04, and a slope of 0.24. Overall, the graphs

appeared to have a large degree of scatter among the data points (Figures 18 and 19).

2

 

Variable y-lntercept Slope r
Anthrophony 4.79 0.39 0.02
Biophony 0.02 0.24 0.04

Table 9. Summary information on correlation analysis between anthrophony and biophony values
and population density.

69

looalpha - 4.7879 00.389? loodannlty

 

 

 

 

l2
+ +
101 +
+
+
+
+
+
0* t """
+ + ,,,,,,,,,,,
__ ,, .,—-- +
+ + ..‘——"""
o + _..-"”
.s: + -_-_,’I" +
o. + “ ,.~
— -4 + fl'---P'
o 6 + ,-- "’
c» ,,,,,,, t’"
o ____
_ ,,,,,,,,,, + + +
+
4.
+ + +
4‘ + +
+
+
+ +
+
z-i
+
0 l r I r l i
l 2 3 4 S 6 7 8
loodanalty

Figure [8. Graph of the linear relationship between the anthrophony (y-axis) and the population

density (x-axis).

70

ladle

ta - 0.0223 00.2395 loodorulty

 

41

31

logbela

-2"

-3‘

-
-—
--
--
--
P”-a
r
’0'
_"

0'-
,’a”'

 

1"
.”
-"
-—-'
,a
,a
’—
,a

a
,v
—-
_--
-'--
,,,,,
,
a’
,’

-—
-
-_--‘-
.a'
—
’I
,a

 

 

loodensity

err
355%

Figure 19. Graph of the linear relationship between the biophony (y-axis) and the population density
(x-axis).

Discussion

Several possible explanations exist for the relatively low correlations between the

anthrophony and biophony and the population density. Perhaps the most pertinent factor

is that acoustic signals are much more dynamic than other variables insofar as they

change with a greater frequency and reflect the immediate activity in the ecosystem. The

samples analyzed represented single-point samples taken once at each site visit.

Therefore, external conditions such as the time of day and season influence the acoustic

signals in a manner that would distort the correlation. Of particular relevance is the fact

71

that none of these samples incorporates the Dawn Chorus (Gill 1995). If the time of
sampling is a significant factor, it would be a rationale for long-term continuous
monitoring of acoustic signals. This long temporal-scale sampling requirement limits the
effectiveness of acoustics as ecological indicators, however, as it implies that using them
as such would require a rather extensive observation network. Examining the biophony as
a valued ecological attribute may prove more fruitful, as it may reflect the concentration
of vocal organisms, which may be considered a favorable ecological condition.
Moreover, the anthrophony should be examined as an environmental stressor, considering
the degree to which human activity is modifying the biosphere. The relationship between
these acoustic samples and the other ecological attributes, however, may not be as
straightforward as a simple linear correlation would elucidate. A distributed sensing
network that simultaneously measures acoustic and other features over extended
spatiotemporal scales may provide a richer dataset that would elucidate stronger
correlations between biophysical ecological variables and their acoustic features.
Additionally, the species recognition system being pursued by Gage et al. (2004) may
help to develop and index of biodiversity that would prove to be an invaluable ecological
assessment tool that utilizes acoustic analysis. The results of this research provide some
insight into the possible ways acoustic samples represent (or fail to represent) ecological
processes occurring within the system. The dynamic nature of these signals implies that
they will continue to yield a significant quantity of information on the dynamic features

of the Earth’s biosphere, provided they are analyzed systematically and thoroughly.

Chapter Seven:
Summary and Conclusions

Objectives

Sound is a multidimensional and complex variable that encompasses a wide array of
information about the structure and composition of the habitat from which it emanates.
The analytical approach applied in this research focused on the examination of the
spectral features of acoustic signals and their relationship to the environment. To
summarize, the objectives stated in the introduction are restated with summaries of the
research findings.
Objective One: Implement an analytical framework that reflects the ecological features of
acoustic signals and is capable of handling large-scale acoustic data for comparisons and
statistical analysis.
The analytical framework used in this study focused on the spectral properties of
acoustic signals. The natural tendency of anthropogenic and biological signals to
occur at two different spectral bands enabled the development of an analytical
system that quantifies the biological and anthropogenic activity in the system.
This analysis simplified the interpretation of the acoustic signals and, with further
refinement, may yield a meaningful interpretation of the degree of anthropogenic
disturbance within an ecosystem. The automation system designed to handle
acoustic spectrograms enabled the analysis of acoustic data from larger

spatiotemporal scales.

73

Objective Two: Verify the apparent spectral distribution of biological signals and

enhance our understanding of the biophony.
A spectral analysis of a representative sample of birds, insects and amphibians
indicated a statistically significant degree of signal partitioning within the
biophony. At the taxonomic level of Class, amphibians generally vocalize at the
lower frequencies (1 — 3 kHz), birds (particularly passerines) generally vocalize at
slightly higher frequencies (3 — 5 kHz) and insects tend to utilize the higher
frequencies (6 — 8 kHz). Overall, the majority of organic signals were
concentrated between 2 and 5 kHz, and almost all the signals were within the
delineated range of 2 to 11 kHz.

Objective Three: Tie the theoretical and analytical tools to a real-world scenario by

relating acoustic signals to other ecological variables in the Muskegon River Watershed.
Comparisons amongst a variety of sites and variables indicated that further
research and refinement is required of the analytical framework in terms of both
data collection techniques and statistical analysis. The absence of a strong
correlation of both the biophony and the anthrophony to population density may
be a result of the sampling approach, the correlation technique or a combination

of factors.

Spectral Analysis

The acoustic analysis framework developed in this research was unique in its focus on the
spectral properties of the acoustic signals. While traditional terrestrial acoustic studies

(such as noise assessment) focus primarily on overall signal intensity, the focus on the

74

spectral features yielded more information about the source and nature of the signals that
were analyzed. The drawback to this analysis was the initial complexity of the analysis,
and the computational resources required. However, the automation program described in
Chapter 3 made the analytical process possible by improving the computational
efficiency to process significant volumes of acoustic data relatively quickly. This spectral
band analysis led to the development of the anthrophony and biophony measurements,
which enable straightforward assessments of the proportion of biological activity and
anthropogenic activity in a signal. The challenge is to relate these measures of

anthropogenic and biological activity in a sound signal to the corollary system.

Future Directions

This research has initiated several interesting enquiries into the ecological aspects of
acoustic signals. Future undertakings should enhance our understanding of the
information about ecological processes and dynamics that acoustic signals reveal. First is
an approach that is already beginning to be examined by Gage et al. (2004) that utilizes
pattern recognition systems to identify species specific vocalizations in acoustic samples.
This system may be used to perform species richness assessments of various habitats.
Such a system could be used to fiirther our understanding of the patterns of and
anthropogenic impacts on the Earth’s biodiversity. Potential research projects include an
examination of the correlation between land use and disturbance intensity and species
richness, and the ways these correlations vary with latitude, elevation and climatic

conditions.

75

Second, the bandwidth analysis of amphibian vocalizations described in Chapter 4
indicate that amphibians vocalize at a lower frequencies that often overlap the
frequencies of mechanical anthropogenic signals. The interference incurred by this signal
overlap may be partially responsible for observed declines of amphibian populations,
particularly where development has fragmented wetland habitats. Specifically, the
interference introduced by mechanical signals may impair the mating success of
amphibians, thereby reducing the overall population. Different intensities of critical band
interference would be introduced to mating groups of amphibians, with a control group
that is has no introduced interference. Resultant numbers of successful copulations would
then be observed and compared across treatments to determine whether signal

interference detrimentally affects reproductive success.

Third, it is necessary to pursue the Acoustic Niche Hypothesis further by sampling
biological vocalizations across gradients of both disturbance and land cover (i.e. forest,
grassland, wetland, built) and examining the spectral bandwidth compositions to
determine whether the same species of organisms will use different bandwidths in
different habitats. The replication of observations over several instances of the same
disturbance intensity and land cover type would begin to reveal whether any frequency
modulation observed is a result of competition for spectral space or adaptation of signal
structure to differential propagation in different habitat types. If the former were the
stronger pressure, then frequency modulation would have a stronger correlation to
bandwidth availability, and lower disturbance intensities would likely yield a greater

proportion of frequency modulation. If the latter were a stronger pressure, then

76

modulation would vary more with land cover than with population sizes, and may not be

as strongly correlated to disturbance intensity.

The last proposed project is to increase the spatiotemporal resolution of the sampling
network to gather a larger representative sample of acoustic signals from different
habitats. Simultaneous samples of acoustics with established variables that are considered
ecological indicators (Chapter 6) would begin to reveal how the biophysical conditions of
an area are reflected in its acoustic signature. The variables that demonstrate the highest
and most consistent correlations would be targeted for initial analysis. Additional
variables could also be sampled, such as energy availability (i.e. net primary production
or potential evapotranspiration) to test the hypothesis that the intensity of biological
acoustic activity (as well as species richness) is influenced by energy availability and

habitat disturbance.

77

Appendix A

Gage, Stuart H., B.M. Napoletano, M. Colunga-Garcia, J. Qi. (2003) An Analytical
Framework to Interpret Acoustic Observations in Heterogeneous Landscapes. Ecological
Applications. In Review

78

Introduction

4.5 Million years of evolution has refined humans’ auditory senses to the point where we
can localize, interpret and react to acoustic signals virtually instantaneously (Masterton et
al. 1969). However, as is often the case, the sense of familiarity imbued by these abilities
has allowed us to overlook the volume and diversity of information we extract from the
acoustic world. Sounds produced by the environment can be valuable but are an
untapped resource to assess the health of the Earth’s ecosystems. The arrays of sounds in
a place depend on a complex array of circumstances including the type of habitat (e. g.
wetland, desert, forest, grassland), the time of day and the season of the year. The
diversity of sounds at a place depends on the heterogeneity of the landscape and on the
status of its ecosystems. Many groups of animals, including birds, mammals, amphibians
and insects, produce sound and use it to communicate (i.e. Schwartz et al. 2001, Hopp et
al. 1998).

With the growth and expansion of human populations has come a greater need to
understand the dynamics of ecosystems and their complex interactions (Michener et al
2001). As humans increasingly expand the urban infrastructure, we tend to disrupt the
very ecosystem services that are critical to us (Daily 1997). In addition, the fragmentation
of natural ecosystems reduces habitat available for wildlife (Krause 1998). At the onset
of our quest to understand environmental acoustics, we postulated that the consideration
of sound as a variable in ecosystem studies would help to increase our understanding of
ecosystem change due to human disturbance as well as to indicate biological dynamics
over time.

Patterns of acoustic signals reflect the dynamics of physical, biological, and social

components of ecosystems (see Aylor 1971). Changes in the spatial and temporal

79

distribution of acoustic signal patterns reflect changes in those dynamics. The exact
meaning of these signals, in terms of the processes and interactions they represent
between social and bio-physical systems is a challenging area of study. This “heartbeat of
the ecosystem” as represented by its acoustics holds information about the dynamic
processes and the changing states of the ecosystem. We have found that sound is a
relatively easy variable to sample, and our increasingly extensive study of the temporal
and spatial distribution of acoustic signals has produced a rich volume of information
about ecosystems. Repeated discrete sampling of the sounds in a given area, however,
rapidly generates a volume of data that is difficult to manage. Moreover, the
multidimensional aspect of acoustic signals necessitates a complex analysis that examines
multiple variables simultaneously.

We have found that the proper examination of an acoustic signal in an ecological context
requires, at a minimum, information about its frequency, intensity, and temporal
properties. These issues makes sound a variable complex to manage, analyze and
interpret. This study began with the objective of developing quantitative approaches to
assess ecological processes within a watershed through the application of environmental
acoustics. During this process, we developed a framework for the study of patch-level
acoustic signals in an ecosystem. This paper describes this framework, including: a) the
definition of a soundscape, b) the classification of signals in the soundscape based on
their physical, biological, and social origins, and c) an analytical approach to quantify the
components of acoustic samples taken from the environment. The analysis of acoustical

signatures within the context of this framework will enable the use of sound as a means

80

of comparing different types of places or different times within the same place under

varying environmental conditions.

Definition of a soundscape
Sound is, in physical terms, the transmission of vibrations of a certain frequency range

through a medium (Hartmann 1998). Both air and water provide the media for most
sound transmissions, although some sounds are also transmitted through the ground. The
soundscape is a common term in the field of study known as acoustic ecology. The
working definition of soundscape in this field is any group of sounds specified as an area
of study (Schafer 1977, 1994, Truax 1999). This implies that the term soundscape may
apply to anything from a musical composition to the entire planet. For the purposes of
our study of the ecological aspects of acoustics, we define the area of study as the set of
sounds generated by the biophysical and social interactions within a landscape, where the
landscape is a heterogeneous land area composed of a cluster of interacting ecosystem
patches (Forman and Godron 1986).

Our defined soundscape, then, is the acoustic signals associated with the landscape.
Research to measure and understand soundscapes builds on three areas of ecological
acoustics: a) communication and acoustic behavior of organisms, b) effects of “noise
pollution” on humans and other organisms, and c) acoustic design in large-scale human
systems and structural acoustics of architecture. Studies of communication of organisms
in the environment focus on the behavior, social structures and evolutionary trends of
organisms (Kroodsma and Miller 1996, Bailey et al. 2001, Schwartz et al. 2001, Fischer

et al. 2002). These studies entail general population surveys and provide warning signals

81

of major problems in the system (Carson 1962). Noise pollution studies (e.g. sonic
booms or aircraft disturbances) have investigated how human development affects natural
systems and social behavior (Federal Aviation Administration 1985, Komanoff 2000).
Studies in “Acoustic Design” examine the effects of different sound types on stress in
humans and other organisms and provide valuable insight into the relationships between
sound and behavior (Schafer 1977, 1994). Our research on environmental acoustics has
produced methods to characterize acoustics in human dominated ecosystems. Our
findings are in three areas: soundscape classification; measurement of diurnal patterns of

acoustics, and the development of indices relating human and biophysical acoustics.

Classification of sounds in the soundscape
Generally, researchers consider the various signals in the acoustic spectrum as

originating from either natural processes or human activity. For the purpose of our
framework we have distinguished three main categories of sounds that occur in the
soundscape: biophony, anthrophony, and geophony. Krause (1987), in his studies of
natural soundscapes, devised the term biophony for the complex chorus of ambient
natural sounds. In our framework, this category encompasses only the natural sounds
produced by organisms, including birds, amphibians, insects and bats. This is the class of
sounds most extensively studied in ecological acoustics (Bosch et al. 2000, Bailey et al.
2001). The term anthrophony refers to the human-induced sounds within an ambient
soundscape. Human induced sounds are primarily oral (i.e. speaking, singing, whistling
or shouting) or mechanical (use of technology). Because humans are organisms, their oral

signaling would technically fall into the realm of the biophony. However, we need to

82

specify a separate class for human derived signals to measure and quantify the impact of
human activity on the soundscape. Moreover, mechanical sounds are the most dominant
anthrophonic sounds in the soundscape and oral activity comprises a negligible
proportion of anthropogenic signals in our investigations. Finally, the third category,
geophony, refers to the pattern of signals present within the soundscape as a result of

physical processes occurring in the region. Examples of these classes of signals are those

 

emanating from
Sou ndscape .
Acousbc Spectrum from 40 Hz to 11 050 kHz waterfalls, r1ver flow,
Geophony - ‘
N wmd or rain.
Annamaria mason-u
As our understanding

   
 

Anthrophony Biophony
4O

HztoZkHz 2.5kHzto 11.050kHz
Mechanical
mm

Intentional
Corrmunication
Among
Stationary
Consistent or Periodic
we may firrther refine
Tem poral
Non-oenodic or Brief

our classification. For

of the characteristics of

  
           

Oral Incidental _
Sim“ Caused sounds in these

by Organisms

     

categories increases,

 

 

 

Figure 20. Conceptual classification schema of the soundscape and its
three primary components and their typical frequency ranges. Several instance subdivisions in
hypothetical subclasses are also depicted. While the Anthrophony and ’

Biophony tend to have discrete ranges, the Geophony tends to occur _ .
across the spectrum, the class1fication can be

made based on the persistence of the signal (stationary versus temporal), the function of
the signal (intentional versus incidental) or the periodicity of the signal (periodic versus
random). However, for the purpose of this paper, we focus our analysis and discussion on
the three major classes mentioned above, while still outlining some of the more

complicated subunits in Figure 20 for illustrative purposes.

83

Quantifying acoustic samples from the environment
An acoustic signal is characterized by multiple physical attributes, among which are

timing, frequency, and intensity (Hartmann 1998). To process and synthesize acoustic
observations gathered under field conditions we have developed the following
methodology. First, we convert the time, sound frequency and sound intensity into
spectrograms. Spectrograms are three-dimensional grids where the x-axis denotes time,
the y-axis represents the frequency of the signal (Hz), and the z-axis represents the
intensity (dB SPL as received by the microphone). We standardized the dimensions of
our spectrograms in pixels as 500 x 1000 in the y and x axes respectively. In the y-axis,
46 pixels represent approximately 1,000 Hz, which we refer to in this paper as a single
Frequency Band. In the x-axis one pixel represents 30 milliseconds of a Fast Fourier
Transformation of the original signal. We used the functionality of IDRISI® (Clark Labs
2000) to analyze the spectrograms in terms of topological and x, y spatial features within
the signal, i.e. the spectrogram was treated as a multiphase map of the acoustic signal.
Sound intensity values were converted to 8-bit series of values with a range of 0-255
possible values. For example, if a spectrogram represented a signal that entirely filled the
spectrum with the maximum sound intensity, then the average value would be 256. On
the other hand, if the signal was acoustically silent (i.e. no energy above 10"2

watts/metre2 was detected by a microphone), then its average value would be 0.

84

 

    
  
  
  
   
 

-—--—---------------------—--

IDRISl dlvldes the
spectrogram Into

 

I----------------------------
r.-

  
 
 

p---------- ---------------

 

 

This enables
comparison
of actMty
across

Frequency Band different
frequencies

 

 

 

 

 

Figure 21. The frequency windowing procedure. Each spectrogram is divided into 11 frequency
bands and the mean amplitude is calculated for each band. This process allows a relative
comparison of the frequency bands with the highest concentration of acoustic activity.

To determine the distribution of activity over the spectral frequency range of the
spectrogram, we windowed the image into multiple spectral bands based on the 1,000 Hz
Frequency Bands. We averaged the intensity values in each Frequency Band to obtain
what we call the Relative Sonic Amplitude (RSA). This allows us to quantify the amount
of acoustic activity in the entire sample, as well as within each frequency level. The
spectral properties of acoustic signals in the environment tend to aggregate within two
primary regions. The first region occurs at the lower frequencies of the sound spectrum
(Schafer 1977, 1994). This band typically extends from 0.2 to 1.5 kHz and consists
primarily of mechanical signals (i.e. trains, cars, air conditioners, etc.), and is therefore

aptly referred to as the anthrophonic region. The second band of concentration begins in

85

the range of 3 kHz and is prevalent up to 8 kHz, though it may on occasion reach the top
of the spectral range of the recorded signal. This realm of acoustic activity consists
primarily of signals generated by biological organisms, and is therefore referred to as the
biophonic region. We have delineated this frequency band as the biological band based
on our observations and the frequency ranges referred to in the literature (Shackleton et
al. 1991, Naguib I996, Ping et al. 1996, Bennet-Clark 1997, 1998, Bennet-Clark 1999,
Bosch et al. 2000, Bailey et al. 2001, Rundus and Hart 2002, F reeberg et al. 2003). These
two bands correspond to two of the three taxonomic categories of the soundscape
described above, but do not cover acoustics emanating from the physical (i.e. wind, rain,
etc.) or geophonic component. This is because the geophony, when present, occurs as a
signal that is diffuse throughout the entire spectrum. The geophony is a diffuse signal
that is strongest at the lowest frequencies but that continues with a relatively high
intensity into the higher frequencies. Generally, when a geophonic signal is present, the
frequency bands above 8 kHz will exhibit greater signal intensity than in signals without
geophony. (When geophony is present, it may be detected, therefore, by its tendency to
generate stronger signals at the higher frequencies above the predominant range of the

biophonic spectrum (that is, strong signals above 8 kHz).)

86

 

I:

Q
‘—
&
v

0 (D

@Q@ q; ©

0 was

'1.

Frequency (kHz)

 

 

 

llll

Figure 22. Frequency range comparison of multiple signals
and signal types. A) Brachypterous cricket (Bailey, 2001); B)
Pycnonotus sinensis (Ping, 1996); C l) Cyclochila
australasiae, inward (Bennet-Clark, 1997); C2) Cyclochila
australasiae, outward (Bonnet-Clark, 1997); D) Poecile
atricapillus (Otter, 2002); E) Zenaida macroura (Krause,
1987); F) Chaetura martinica (Krause, 1987): G)
Terpsiphone paradisi (Krause, 1987); H) Locomotive; l)
Motor Boat; J) Air Conditioning. As the distribution
indicates, the anthropogenic signals generally tend to occur
at lower frequency ranges than the biological signals.

 

 

Using this partitioning of the
acoustic spectrum, we
developed a methodology to
quantify the three primary
acoustic elements, the
anthrophony (0t), biophony
(l3). and gCOPhony (Y), by
calculating the mean value of
acoustic intensity in the
spectral frequency range we
allocated to each of these

regions, and comparing their

values to the mean activity of the entire signal (0'). We calculated the mean values of the

on, B, y, and 0 bands by assigning a numeric value to each pixel in the specified range and

calculating the mean value of the z-axis. The or, B, and 7 activity ratios were then

calculated using the equations in Table 10. A value > 1 for any of the indices indicated

that the concentration of acoustic activity in the analyzed region was greater than the

value for the entire signal. Therefore, the region with the highest value was the

predominant source of acoustic activity in the signal. For example, if the [3, had the

highest value, then biological activity was predominant, while a larger car value indicated

dominant anthropogenic activity. To emphasize the comparison of biological and

87

anthropogenic activity, we divided the B value by the Qt value to calculate p, the ratio of

biological to anthropogenic activity.

 

 

 

 

 

 

Index Ratio Percentage
a
r—(a) a”: M x100
Zr lLevels
A nthrophony
0L=Meanfrom0t02kHz .. .
Ctr 2 Ratio of anthropogenic or? = Percentage of actrvrty 1n the
activity mean to grand mean anthrophony band
3] 2L3“) L11
= — = —— x 100
'8’ (0' fl” 2] lLevels
Biophony
B 2 Mean from 2 to 11 kHz Bp = Percentage of activity in the
Br = Ratio of biological activity biophony band
mean to grand mean
7 L8t L1 1
a " Zr lLevels
Geophony
Y : Mean from 8 to 11 kHz VP 2 Percentage of activity in the
y, = Ratio of geological activity geophony band
mean to grand mean
,0 = (g) Global Variables
Activi a L = 1 kHz level
ty 0' I Mean value of entire signal
p 2 Ratio of biological to (Grand Mean)
anthropogenic activity

 

 

 

 

Table 10. Formulae for the calculation of activity concentration values for the three primary regions
of the soundscape. The column on the left depicts the formulae for the calculation of ratio values in

terms of the entire spectrum, while the column on the right lists the formulae for the calculation of
values in terms of percentage of the entire spectrum.

In addition to computing the ratios of activity from our classification system, we also
determined the percentage of total activity a single band contributes to the total signal
(Left-hand column of Table l). A yp value near 100% coincident with a Bp value of

approximately the same value indicated that the primary signal source in the sound

88

sample was Biophony (gee-physical) activity. When the (1,, value was greater than 50%,
it indicated that the primary signal source was Anthrophony (anthropogenic) activity,
whereas a value of Bp greater than 50% indicated that Biophony (biological) activity was

the dominant source.

Case Studies
We present three case studies from our research that represent characteristics of acoustic

signals and the applications to which we have applied our environmental acoustic
framework. In the first case we compared the diurnal soundscape patterns at four
different locations. In the second case study we determined the acoustic quality of a
location and compared the acoustic quality at different locations. The third case
exemplifies the combination of acoustic measurements with other environmental data
(e. g. temperature) to examine the complexity of environmental acoustics, particularly

when measuring acoustic signals resulting from communication by organisms.

Case study I. Diurnal Soundscape Patterns

We compared the diurnal acoustical signals from four different locations using an
automated system to sample and capture acoustic signals from the same location at
frequent intervals. This procedure allowed us to sample acoustics at regular times of the
day and night, thus enabling the characterization of diurnal patterns of sound. To obtain
high temporal resolution acoustic data, we deployed an automatic digital recording
system using a Bird Bug® 100M parabolic microphone that sampled acoustic signals at a

22.050 kHz 16 bit sampling rate for thirty seconds every half-hour. We sampled the

89

acoustic signals for three contiguous days from four locations in the Muskegon River
Watershed located in mid-west Michigan. These locations included an equestrian center,

a private ranch, a wetland, and a public park and campground. Tables 1 l and 12 list the

summary statistics and Pearson’s Correlations respectively.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Abbreviation EC

Name West Michigan Equestrian Center

Land Use Agriculture/ Outdoor Recreation

Date 2002-08-23 2002-08-24 2002-08-25
9 1.52 1 0.79 1.62 1 0.71 1.48 1 0.51

Mean 0 26.26 1 8.49 25.67 1 8.13 22.84 1 6.61
B 34.61 1 7.51 36.85 1 6.59 31.34 1 6.31

Abbreviation GC

Name Haymarsh Wetlands

Land Use Wetland/ Residential

Date 2002-08—23 2002-08-24 2002-08-25
p 0.66 1 0.22 0.73 1 0.21 0.65 1 0.18

Mean (1 15.16 1 4.02 17.32 1 5.11 14.35 11.96
B 10.13 1 4.31 12.21 1 3.78 9.18 1 2.31

Abbreviation PP

Name Paris County Park

Land Use Outdoor Recreation

Date 2002-08-23 2002-08-24 2002-08-25
p 0.78 1 0.16 0.76 1 0.14 0.71 1 0.25

Mean (1 45.23 1 9.06 42.97 1 6.21 44.56 1 6.96
B 34.13 1 5.59 32.41 1 7.20 31.10 1 10.40

Abbreviation MS

Name COOper Residence and Ranch

Land Use Agriculture/ Pasture and Livestock

Date 2002-08-23 2002-08-24 2002-08-25
p 1.72 1 1.62 2.01 1 2.28 1.61 1 1.33

Mean 0 0.59 1 0.22 0.60 1 0.44 0.84 1 0.71
B 0.97 1 0.89 1.02 1 0.94 0.92 1 0.55

 

 

 

 

 

Table II. Summary statistics of the means and standard deviations of p, a, and B values calculated
for three consecutive days at four different locations. p represents the ratio of biological to
anthropogenic activity, while a and B represent the anthropogenic and biological activity
respectively.

90

 

 

 

 

 

 

 

Equestrian Center Haymarsh Cooper Ranch Paris Park
23- 24, 23- 24- 23- 24- 23- 24-
Aug M AuL Aug Ag M AuL AuL
24-
Aug 0.7617 0.7703 0.8159 0.1673
p-value 0.0000 0.0000 0.0000 0.2558
Afr: 0.5824 0.5771 0.4833 0.5242 0.7464 0.7291 0.1187 0.5027
p-value 0.0000 0.0000 0.0005 0.0001 0.0000 0.0000 0.4217 0.0003
24- -
Aug 0.5204 0.1122 0.0408 0.4620
p-value 0.0001 0.4476 0.7832 0.0009
25- -
Aug 0.0761 0.1746 0.137., 0.3606 0.1950 0.0313 0.3278 0.2076
p-value 0.6072 0.2353 0.3505 0.0118 0.1840 0.8329 0.0230 0.1568
24-
Aug 0.4185 0.2712 0.8374 0.6518
p-value 0.0031 0.0623 0.0000 0.0000
A1: 0.5479 0.4571 0.4740 0.4524 0.7219 0.7736 0.5919 0.7624
p-value 0.0001 0.0011 0.0007 0.0012 0.0000 0.0000 0.0000 0.0000

 

 

 

 

 

 

Table 12. Pearson’s correlations of p, a, and B means over the three-day period. All sites except
Paris Park exhibit a significant correlation of p values.

The private ranch (Figure 23, C) exhibited diurnal patterns of acoustic activity that
remained relatively low at night and during mid-day. Acoustic activity peaked around
0600, coinciding with the well-known ‘Dawn Chorus’. The anthropogenic activity was
minimal during the 144 '/2 hour samples from this location. The other three locations
however, exhibited different patterns and none of the acoustics peaked at dawn (Figure
23). The equestrian center (Fig 23A) and the public park (Fig 23D) were outdoor
recreational facilities, which explained why the peak in activity observed during the
midday. The wetland site (Fig 23 B) contained relatively large populations of crickets,
cicadas, and other acoustically active insects, which are active in mid to late summer.
Insects were responsible for the higher concentration of diurnal activity at the wetland

site, as well as for the higher concentrations of nocturnal activity at the public park, the

91

equestrian center, and the wetland site. Insect vocalization is more regular and is

sustained for longer time spans than are vocalizations from birds or amphibians.

Rdflw Sonic Ampfitudo

Biophony

 

10'

““1744th

A

 

 

 

O
the

10

 

 

 

 

 

23

 

 

 

 

 

«J
«i
so?
20-
u.

 

 

 

 

 

 

 

 

 

Anthrophony
u
so
u
so 11.
as '
r

" A
0 . f fl . v

o s u a u as
as
so
as
:0
1s
10
3 B
o . - . - .

o s a a u u

 

 

 

 

 

 

 

 

I
I
I
I

Aug 230— Aug 24+ Aug 25+

Figure 23. Biological and Anthropogenic activity curves over three 24-hour periods at four sampling
sites. A) Equestrian Center; B) Haymarsh Wetlands; C) Cooper Ranch; D) Paris Park. The column
of graphs on the left is the biophony curves, and the column on the right is the anthrophony curves.

Sites C and D appear to show the least similarity between the temporal distribution of anthropogenic
and biological activity.

92

After examining the differences in diurnal acoustic trends at the four locations, we
examined seasonal variations in these trends by selecting a day to represent each of mid-
Michigan’s four seasons, and plotted the acoustic activity (mean RSA) at ‘/2 hour
intervals for each day (Figure 24). The results exhibited different acoustic activity trends
in each season. The seasonal variation that occurs in Michigan has a significant impact on
biological activity and species composition of the landscape, which may affect the trends
of diurnal activity.

Case study 2. Activity Concentration Analysis of Multiple Locations

 

 

10 Feb 2002 10 May 2002 This case illustrates our
'° 4 . . ability to detect

differences in the acoustic

 

 

 

 

composition of multiple

 

 

 

I
‘r r f - f—v ' v 1- 4

Time Time
10 July 2002 GageHome 12 Oct 2002 locations. This allows us

 

 

.0;

to establish different

reference levels of

 

disturbance based on the

 

 

 

   

 

 

Figure 24. Diurnal activity trends of the entire analyzed spectrum at acoustic properties ofa
a single site (Gage Home) at four different times of year, representing

four different seasons. While a single day is a rather small wide gradient of habitats.
representation of an entire season, the figures appear to illustrate
seasonal variations in activity trends. To elucidate the different

acoustic character of multiple locations, we computed pZB/a from recordings made in a
series of streams, lakes and wetlands in the Muskegon River Watershed over a three-year
period as part of an ecological assessment and restoration initiative (Figure 6). Acoustic

recordings from 41 different sites throughout the watershed were collected using a Sony

93

M inidiskR‘ recorder and a Sony EC M-MS907'TR; Stereo Microphone. All streams were
sampled in late August, while all lakes were sampled in the late spring through early
summer, a time of peak activity for many organisms. Wetlands were sampled throughout

the mid-summer, from early June to late July.

 

 

 

I Lake

at.

“it ........
if "r

Stream

 

 

a Wetland

 

 

 

 

 

Figure 25. Map of three different classes of sampling locations (Lakes, Streams, and Wetlands) in the
Muskegon River Watershed sampled in the summers of 2001 and 2002. 80-minute recordings were
made at each sample site during the time of sampling.

94

We digitized the recordings from these samples into 12 30-second sub-samples and

computed p=B/a for each location to calculate the ratio of biological to anthropogenic

acoustic activity as described in the section on Quantifying Acoustic Signals described

above. We then compared the ranges of values after sorting them into groups of streams,

lakes, or wetlands. This allowed us to not only compare multiple locations, but also to

compare the properties of the different location types (i.e. did lakes exhibit more

anthropogenic activity than did wetlands?) In this analysis streams exhibited the highest

 

Streams Lakes Wetlands
Mean 0.1210 0.0386 0.0824
n 5 1 1 25
SD 0.0802 0.0402 0.0613
Max 0.2486 0.1311 0.2739
Min 0.0505 0.0001 0.0000
Range 0.1980 0.1310 0.2739

Table 13. Summary statistics for the three

 

sample classes (Lakes, Streams, and

Wetlands) gathered in the Muskegon
Riv" water-“‘9": The Streams “‘35 m" Table 4 lists the statistics of the ratio values for
the highest calculated mean, the Lakes
class had the lowest, and the Wetlands
class had the broadest range of values.

ratio of biophony and lakes exhibited the highest
ratio of anthrophony. The maximum and
minimum p—values were from wetland sites

(0.274 and 0 respectively, where 0 is the smallest

possible value, indicating no detectable Biphony).

the three classes of locations.

Lakes were all sampled during a period of higher biological activity. The low ratios

encountered in the lake samples indicate that lakes were the most heavily impacted

landscapes in terms of anthropogenic disturbance. Additionally, all lake samples were

taken at boat launches for logistical reasons. The streams yielded the highest mean p-

value, indicating that these sites generally demonstrated the least concentration of

anthropogenic activity. Finally, the wetlands exhibited the largest range of ratio values,

with one location registering no biological activity (Figure 26).

95

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

0-—Laite ‘ T
b-"Stream
20‘ \
15‘”""‘"TT ””" ’
m104P—‘— ibué- ~ “-1, fire-— . /
: ’ '
1 1
5 ' T
1 ',I'.fi”‘ I
O 1.... -11. m ”11.11
10 30 50 70 90 110
alpha

Figure 26. a (anthrophony) region plotted against B (biophony) region for the three classes of
sampling sites (Lakes, Streams, and Wetlands) in the Muskegon River Watershed. The maximum
range of each class of samples is indicated in by the boundary lines. The wetlands encompass the
largest range of activity, while the lakes encompass the smallest.

Case study 3. Temperature — Acoustics Relationships

This study exemplifies the combination of acoustic measurements with other
environmental data to examine the complexity of environmental acoustics, particularly
when measuring acoustic signals that emanate from organisms that respond to physical
factors such as temperature. The occurrence and intensity of sounds made by some
organisms (ex. amphibians, insects) are ofien influenced by temperature (Bailey 1991).
We investigated the interrelationship between acoustic signals and temperature,
illustrating our ability to extract specific components of the acoustic signal (Biophony)
and relate that component to temperature readings recorded at the same time period. The
example we selected illustrates the relationship between acoustic signals and temperature

recorded at '/2 hourly intervals during April and September, 2002. Recordings of acoustic

96

signals were made using the Bird-Bug microphone connected to a digitizing sound card
(Creative Labs) in a microcomputer. Recordings of 30-second duration were written to
disk using Total Recorder. Air temperature measurements were transmitted to the
microcomputer using a RainWise wireless weather station and recorded at the same time
as recordings were captured. These observations were made in Meridian Township,
Michigan (N42° 43.46, W840 22.57) in a rural-suburban habitat. April and September
were selected, as they are seasonal transitions when temperature and acoustics signals
vary. The time of 22:00 hrs was selected as amphibians and crickets are often signaling at
this time in this landscape. A spectrogram was produced for each acoustic signal sampled
at 22:00 hrs each day of each month. The mean acoustic intensity was calculated for each
of 11 frequency bands at l KHz intervals. Band 4 (3-4 KHz) was selected to represent
biological events in the signal. The acoustic intensity in band 4 was compared to
temperature at the time of recording for each day in April and September, 2002. There
was a significant correlation between temperature and the acoustic amplitude in Band 4
in both April and in September at 22:00 hrs. The relationship was stronger in September
than in April (Figures 27 and 28). The acoustic signals in April were generated by Spring
Peepers and the September signals were generated by tree crickets. Table 14 shows
numerical values of acoustic intensity and recorded temperature at 22:00 hours for two

consecutive days in April and September.

97

 

45‘

Acoustic Intensity
3 9 3’ F l? S
0 o
8
o
O

i5

00 00

g
o 0(1) 0m 000

O ,

O

 

 

 

N 2' a '35 40' 45 6'0 “5'5 0 65 7° 7;
' TEMPERATURE
‘ April 2002 22200 Band4'

Figure 27. Acoustic Intensity of band 4 (3 - 4 kHz, a highly biological band) plotted against
temperature throughout April 2002 at 22:00. These plots indicate that the Intensity of activity in
band 4 increases‘with rising temperature. '

98

 

Acoustic intensity
3 r» .3

A

00‘ 0 0.

 

 

 

m 0.9 0 o 00°C 0 O

a r v r fl
I

40 ‘45 50. 55 so P 65 f as . 75
' TEMPERATURE ‘
September 2002 22:00 Band 4

Figure 28. Acoustic Intensity of band 4 (3 — 4 kHz, a highly biological band) plotted against
temperature throughout September 2002 at 22:00. These plots indicate an even sharper increase in -
activity once a temperature threshold of roughly 65° F is surpassed.

99

 

 

 

 

 

% (tango in % Change in
Date 84 (3 - 4 kHz) Temperature
Temp Activity

April 7 2002 00 91 43

25.58 2931407
April 8. 21.102 2? 01 54
Sept 14. 21:02 40 o? 71

-21.13 -99.66
Sept 15.2002 00 lo 50

 

 

 

 

 

 

 

Table 14. Mean value of activity in band 4 (3-4 kHz) at 22:00 on four different days with four
different temperatures. The lower temperatures appear to correlate with lower activity means, and
higher temperatures with higher means.

Conclusions

Our investigation of ecological acoustics revealed that a framework for the collection,
analysis and interpretation of environmental acoustics on a landscape scale had not been
developed. This paper provides a framework to facilitate the scientific community’s
ability to communicate and quantitatively interpret environmental acoustics. We consider
the study of acoustic signals to be a valuable resource that will allow us to interpret
ecological change. We developed a simple classification system of acoustic signals that
was based on the frequency domains of different components of these signals to enable a
quantitative interpretation of the soundscape in terms of its human and biophysical
components. Earlier work by Krause (Krause 1998) provided an excellent start by
characterizing the biophony, but in order to quantify the ecological impacts of our
increasingly mechanized society, we found it necessary to separate mechanized
anthropogenic activity into the anthrophony. We built our interpretation on the

foundation of Schaffer’s initial soundscape concept (Schafer 1977, 1994), enhancing it by

100

attempting to quantify the activity by means of the spectrogram and statistical analyses.
Many studies have been conducted on acoustic signals of specific organisms and how
organisms use acoustics to communicate (sea mammals, insects, birds, amphibians), as
well as acoustic surveys to assess the occurrence of these organisms. We did not find,
however, any examples of methods to obtain data at regular intervals over long time
periods that would enable the examination of temporal patterns of soundscapes. Thus we
developed new analytical methods to examine soundscapes’ various temporal features.
The analyses provided in the case studies represent the diversity of information our
approach to ecological acoustics may yield. These cases were our first steps to
demonstrate that it is possible to extract vital ecological information by examining the
temporal and frequency aspects of ecosystems and examining relationships between
acoustic signals and other ecological variables. We have developed a sizeable digital
library of acoustic signals at regular intervals from several locations in Michigan and at a
few other places in the United States (California, New Mexico, Colorado). We are
developing a cyber-infrastructure to enable automated processing and analysis of
incoming signals from different landscapes in the United States for near real-time display
on our “Clickable Ecosystem” web site (http://envirosonic.cevl.msu.edu/). We will use

this site to communicate the information we gather and analyze from acoustic signals.

101

 

Appendix B

List of species used in the Spectral Analysis

 

 

 

 

 

-102-

 

mu [Subfamlly 103m god-s nailsh Name
ANURA
BUFONIDAE
Bufo amertcanus American Toad
Bufo cognatus Great Plains Toad
Bufo fowlerl Fowler's Toad
Bufo quercicus Oak Toad
Bufo terrestrls Southern Toad
Bufo valllceps Gulf Coast Toad
HYLIDAE
Hyllnae
Acrls crepttans Northern Cricket Frog
Acrls gryllus Southern Cricket Frog
Hyla andersonll Pine Barrens Treefrog
Hyla arborea Green Treefrog
Hyla avivoca Bird-voiced Treefrog
Hyla chrysoscells Cope's Gray Treefrog
Hyla femoralls Pinewoods Treefrog
Hyla gratlosa Barking Treefrog
Hyla squirella Squirrel Treefrog
Hyla verslcolor Gray Treefrog
Pseudacrts brachyphona Mountain Chorus Frog
Pseudacrls brlmleyl Brimley's Chorus Frog
Pseudacn's cruclfer Spring Peeper
Pseudacris nigrita Southern Chorus Frog
Pseudacrls ocularis Little Grass Frog
3 Pseudacris ornate Ornate Chorus Frog
E Pseudacrls streckeri Streckers Chorus Frog
3 Pseudacris trlserlata Western Chorus Frog
MlCROl-IYLIDAE
Mlcrohylinae
Gasa'ophryne carollnensis Eastern Narrowmouth Toad
Gastrophryne ollvacea Great Plains Narrowmouth Toad
MYOBATRACl-IIDAE
Limnodynastinae
lenodynastes dorsalls Bullfrog
PELOBATIDAE
Pelobates syrlacus Eastern Spadefoot
Spea bombifrons Plains Spadefoot
RANIDAE
Rana areolata Gopher Frog
Rana areolata Crawfish Frog
Rana blalri Plains Leopard Frog
Rana clamitans Green Frog
Rana daemeli Wood Frog
Rana gryllo Pig Frog
Rana heckscherl River Frog
Rana okaloosae Florida Bog Frog
Rana palustrls Pickerel Frog

 

 

 

 

 

 

 

 

Caprlmulglnae
Caprlmulgus caroltnensls
Caprlmulgus vociferus

Chordeillnae
Chordelles gundlachll
Chordelles minor

CHARADRllFORMES
CHARADRIIDAE

Charadrilnae
Charadrlus alexandrlnus
Charadrlus melodus
Charadrius semlpalmatus
Charadrlus vociferus
Charadrius wilsonia
Pluvlalis dominica
Pluvlalts squatarola

LARIDAE

Larinae
Lams argentatus
Larus atricilla

-103-

 

F Y [Summlly [Genus species ngllsh Nams
Rana pipiens Northern leopard Frog
Rana septentrionalls Mink Frog
Rana sphenocephala Southern Leopard Frog
i Rana vlgatlpes Carpenter Frog
3} ANSERIFORMES
i ANATlDAE
Anatlnae
Alx sponsa Wood Duck
Anas acuta Northern Pintail
Anas amerlcana American Wigeon
Anas clypeata Northern Shoveler
Anas crecca Green-winged Teal
Anas platyrhynchos Mallard
Anas strepera Gadwall
Clangula hyemalls Long-tailed Duck
Oxyura jamalcensls Ruddy Duck
Somaterla molllsslma Common Eider
Anserinae
Branta bemlcla Brant
Branta canadensls Canada Goose
Chen caerulescens Snow Goose
Cygnus columblanus Tundra Swan
APODIFORMES
APODIDAE
Chaeturlnae
Chaetura pelaglca Chimney Swift
TROCHILIDAE
Trochilinae
Archllochus colubrls Ruby-throated Hummingbird
CAPRIMULGIFORMES
CAPRIMULGIDAE

Chuck-wills-widow
Whip-poor-will

Antillean Nighthawk
Common Nighthawk

Snowy Plover

Piping Plover
Semipalmated Plover
Killdeer

Wilsons Plover
American Golden-Plover
Black-bellied Plover

Herring Gull
Laughing Gull

 

 

 

 

 

 

F Y Subfamlly [Germ species Name
Sternlnae
Chlldonlas nlger Black Tern
Stema antlllamm Least Tem
Stema caspla Caspian Tern
Stema dougallll Roseate Tern
Stema forsteri Forster’s Tern
Stema hirundo Common Tern
Stema maxtma Royal Tern
Stema nilotlca Gull-billed Tern
Stema paradlsaea Arctic Tern
RECURVIROSTRIDAE
Himantopus mexlcanus Black-necked Stilt
SCOLOPACIDAE
Phalaropodlnae

Phalaropus tricolor
Scolopaclnae
Actltls macularla
Arenarla Interpres
Bartramla Ionglcauda
Calldrls alba
Calidris alpine
Calidrls maurl
Calldris minutllla
Calldris pusllla
Catoptmphorus semlpalmatus
Galllnago galllnago
lenodromus grlseus
lenodromus scolopaceus
Numenius phaeopus
Scolopax minor
Tringa flavipes
Trlnga melanoleuca
Trlnga solitaria
ClCONlIFORMES
ARDEIDAE
Ardea herodlas
Botaurus Ientlglnosus
Butarldes virescens
lxobrychus exllls
Nyctanassa vlolacea
Nycticorax nyctlcorax
COLUMBIFORMES
COLUMBIDAE
Columbine passerlna
Zenaida macroura
CORACIIFORMES
ALCEDINIDAE
Cerylinae
Ceryle alcyon
CUCULIFORMES
CUCULIDAE

-104-

Wilsons Phalarope

Spotted Sandpiper
Ruddy Tumstone
Upland Sandpiper
Sanderiing

Dunlin

Western Sandpiper
Least Sandpiper
Semipalmated Sandpiper
Willet

Common Snipe
Short-billed Dowitcher
Long-billed Dowitcher
Whimbrel

American Woodcock
Lesser Yellowiegs
Greater Yellowiegs
Solitary Sandpiper

Great Blue Heron

American Bittem

Green Heron

Least Bittem
Yellow-crowned Night-Heron
Black-crowned Night-Heron

Common Ground-Dove
Mourning Dove

Belted Kingfisher

 

 

 

F Y3ubfamlly Genusspeclea

ngllshName

 

 

 

Coccyzlnae
Coccyzus americanus
Coccyzus erythropthalmus
Coccyzus minor
FALCONIFORMES
ACClPlTRlDAE
Acclpltrinae
Acclplter cooperfi
Acclplter gentllls
Buteo jamalcensis
Buteo Iineatus
Buteo platypterus
Hallaeetus Ieucocephalus
FALCONIDAE
Falconlnae
Falco peregrinus
Falco sparverlus
GALLIFORM ES
ODONTOPHORIDAE
Collnus vlrglnlanus
PHASlANlDAE
Meleagrldlnae
Meleagrls gallopavo
Phaslanlnae
Phaslanus colchlcus
Tetraoninae
Bonasa umbellus
Falclpennls canadensls
GAVIIFORMES
GAVIIDAE
Gavla Immer
GRUlFORM ES
ARAMIDAE
Aramus guarauna
GRUIDAE
Gruinae
Grus canadensls
RALLIDAE
Coturnlcops noveboracensis
Fullca amerlcana
Galllnula chloropus
Laterallus jamalcensls
Porzana carollna
Rallus elegans
Rallus Iimlcola
Rallus Ionglrostrls
PASSERIFORMES
ALAUDIDAE
Enemaphlla alpestris
BOMBYCILLIDAE
PASSERIFORMI Bombycllla cedrorum

-105-

 

Yellow-billed Cuckoo
Black-billed Cuckoo
Mangrove Cuckoo

Coopers Hawk
Northern Goshawk
Red-tailed Hawk
Red-shouldered Hawk
Broad-winged Hawk
Bald Eagle

Peregrine Falcon
American Kestrel

Northem Bobwhite

Wild Turkey
Ring-necked Pheasant

Ruffed Grouse
Spruce Grouse

Common Loon

Limpkin

Sandhill Crane

Yellow Rail
American Coot
Common Moorhen
Black Rail

Sora

King Rail

Virginia Rail
Clapper Rail

Horned Lark

Cedar Waxwing

 

 

 

 

 

 

Pinlcola enucleator

-106-

 

MY [Subtemlly [Genus species nglish Name
CARDINALIDAE
Cardlnalls cardinalls Northern Cardinal
Passerina caerulea Blue Grosbeak
Passerina clrls Painted Bunting
Passerina cyanea Indigo Bunting
Pheuctlcus Iudoviclanus Rose-breasted Grosbeak
Splza amerlcana Dickcissel
CERTHIIDAE
Certhiinae
Certhla amerlcana Brown Creeper
CORVIDAE
Corvus brachyrhynchos American Crow
Corvus corax Common Raven
Corvus osslfragus Fish Crow
Cyanocitta cristata Blue Jay
Perlsoreus canadensls Gray Jay
EMBERIZIDAE
Atmophlla aestivalls Bachmans Spanow
Ammodramus caudacutus Saltmarsh Sharp-tailed Sparrow
Ammodramus henslowll Henslows Sparrow
Ammodramus Ieconteli Le Contes Spanow
Ammodramus marltlmus Seaside Sparrow
Calcartus lapponlcus Lapland Longspur
Chondestes grammacus Lark Spanow
Junco hyemalts Dark-eyed Junco
Melosplza georglana Swamp Sparrow
Melospiza IIncoInll Lincolns Sparrow
fl,’ Melosplza melodic Song Sparrow
> Passerculus sandwlchensls Savannah Sparrow
<
Passerella lllaca Fox Sparrow
Plpllo erythrophthalmus Eastern Towhee
Plectrophenax nlvalls Snow Bunting
Pooecetes gramlneus Vesper Sparrow
Splzella arborea American Tree Sparrow
Splzella palllda Clay-colored Sparrow
Splzella passerina Chipping Sparrow
Splzella pusllla Field Sparrow
Zonotrlchla albicollis White-'throated Sparrow
Zonotrichla Ieucophrys White-crowned Spanow
FRINGILLIDAE
Carduellnae
Carduells flammea Common Redpoll
Carduells plnus Pine Siskin
Carduelis trlsfls American Goldfinch
Carpodacus mexlcanus House Finch
Carpodacus purpureus Purple Finch
Coccothraustes vespertlnus Evening Grosbeak
Loxla curvlrostra Red Crossbill
Loxla leucoptera White-winged Crossbill

Pine Grosbeak

 

 

 

 

 

F Y [Genus epochs nailsh Name
HlRUNDlNlDAE
l-llrundlnlnae
leundo rusflca Barn Swallow
Petrochelldon pyn'honota Cliff Swallow
Progne subls Purple Martin
Riparla riparla Bank Swallow
Stelgidopteryx serripennls Northern Rough-winged Swallow
Tachyclneta blcolor Tree Swallow
ICTERIDAE
Agelalus phoenlceus Red-winged Blackbird
Dollchonyx oryzlvorus Bobolink
Euphagus carolinus Rusty Blackbird
Euphagus cyanocephalus Brewers Blackbird
Icterus bullockli Bullock's Oriole
Icterus galbula Baltimore Oriole
Icterus spurlus Orchard Oriole
Molothrus ater Brown-headed Cowbird
Qulscalus major Boat-tailed Grackle
Qulscalus mexlcanus Great-tailed Grackle
Qulscalus quiscula Common Grackle
Stumella magna Eastern Meadowlark
Stumella neglecta Western Meadowlark
Xanthocephalus xanthocephalus Yellow-headed Blackbird
LANllDAE
Lanlus ludovlclanus Loggerhead Shrike
MlMlDAE
Dumetella carollnensls Gray Catbird
Mlmus polyglottos Northern Mockingbird
Toxostoma rufum Brown Thrasher
MOTACILLIDAE
Anthus rubescens American Pipit
PARIDAE
Baeolophus blcolor Tufted Trtmouse
Poecile atricapillus Black-capped Chickadee
Poecile carollnensls Carolina Chickadee
Poecile hudsonlca Boreal Chickadee
PARULIDAE

Dendrolca caerulescens
Dendroica castanea
Dendrolca cerulea
Dendroica coronata
Dendrolca discolor
Dendroica dominica
Dendrolca fusca
Dendroica magnolia
Dendrolca palmarum
Dendroica pensylvanlca
Dendroica petechia
Dendrolca pinus
Dendrolca striata

-107-

 

Black-throated Blue Warbler
Bay-breasted Warbler
Cerulean Warbler
Yellow-rumped Warbler
Prairie Warbler
Yellow-throated Warbler
Blackbumian Warbler
Magnolia Warbler

Palm Warbler
Chestnut-sided Warbler
Yellow Warbler

Pine Warbler

Blackpoll Warbler

 

 

 

 

 

 

FAMILY Subfamlly Gurus specles ndlsh Name
Dendrolca tlgrlna Cape May Warbler
Dendrolca vlrens Black-throated Green Warbler
Geothlypls trlchas Common Yellowthroat

REGULIDAE

SIT'l'lDAE
Sittlnae

SYLVIIDAE

Helmltheros vermlvorus
lcteda vlrens

Mnlotllta varia
Oporomls agllls
Oporomls fonnosus
Oporomls phlladelphla
Parula amerlcana
Protonotarla cltnea
Selurus aurocapllla
Selurus motacllla
Selurus noveboracensls
Setophaga mtlcllla
Vermlvora chrysoptera
Vermlvora peregrine
Vermlvora plnus
Vermlvora ruflcapllla
Wllsonla canadensls
Wllsonla cltrlna

Regulus calendula
Regulus satrapa

Sltta canadensls
Sltta carollnensls
Slita pusllla

Polioptillnae

THRAUPIDAE

TROGLODYTIDAE

TURDIDAE

Polloptlla caerulea

Plranga ollvacea
Plranga rubra

Clstothorus palustrls
Clstothorus platensls
Thryomanes bewlckll
Thryothorus ludovlclanus
Troglodytes aedon

Troglodytes troglodytes

Catharus fuscescens
Cadrerus guttatus
Catharus mlnlmus
Catharus ustulatus
Hyloclchla mustellna
Slalla slalls

Turdus mlgratorlus

-108-

 

Wonn-eating Warbler
Yellow-breasted Chat
BIack-and-white Warbler
Connecticut Warbler
Kentucky Warbler
Mouming Warbler
Northern Parula
Prothonotary Warbler
Ovenbird

Louisiana Waterthrush
Northern Waterthrush
American Redstart
Golden-winged Warbler
Tennessee Warbler
Blue-winged Warbler
Nashville Warbler
Canada Warbler
Hooded Warbler

Ruby-crowned Kinglet
Golden-crowned Kinglet

Red-breasted Nuthatch
White-breasted Nuthatch
Brown-headed Nuthatch

Blue-gray Gnatcatcher

Scariet Tanager
Summer Tanager

Marsh Wren
Sedge Wren
Bewicks Wren
Carolina Wren
House Wren
Winter Wren

Veery

Hermit Thrush
Gray-cheeked Thrush
Swainsons Thrush
Wood Thrush
Eastern Bluebird
American Robin

 

 

r visitor-nay [imam

nglish Name

 

 

 

TYRANNIDAE

Fluvlcolinae

Contopus cooperl
Contopus vlrens
Empldonax alnorum
Empidonax fla vlventrls
Empldonax mlnlmus
Empldonax tralllll
Empldonax vlrescens
Sayornls phoebe

Tyrannlnae

VIREONIDAE

PlClFORMES
PICIDAE
Plclnae

PODICIPEDIFORMES
PODlClPEDlDAE

STRIGIFORMES
STRIGIDAE

Mylarchus crlnltus
Tyrannus domlnlcensls
Tyrannus tyrannus
Tyrannus verticalls

Vireo altlloquus
Vireo atricapllla
Vireo bellll

Vireo flavlfrons
Vireo gllvus

Vireo grlseus

Vireo ollvaceus
Vlreo philadelphlcus
Vireo solitarlus

Colaptes auratus
Dryocopus plleatus
Melanerpes carollnus
Melanerpes erythrocephalus
Plcoldes arctlcus

Plcoldes borealls

Plcoldes dorsalls
Plcoldes pubescens
Picoldes villosus
Sphyraplcus varlus

Podllymbus podlceps

Aegollus acadlcus
Aegollus funereus
Aslo flammeus
Aslo otus

Adrene cunicularla
Bubo vlrginlanus

Megascops aslo

-109-

 

Olive-sided Flycatcher
Eastern Wood-Pewee
Alder Flycatcher
Yellow-bellied Flycatcher
Least Flycatcher

Willow Flycatcher
Acadian Flycatcher
Eastern Phoebe

Great Crested Flycatcher
Gray Kingbird

Eastern Kingbird
Western Kingbird

BIack-whiskered Vireo
Black-capped Vireo
Bells Vireo
Yellow-throated Vireo
Warbling Vireo
White-eyed Wreo
Red-eyed Vireo
Philadelphia Vireo
Blue-headed Vireo

Northern Flicker

Pileated Woodpecker
Red-bellied Woodpecker
Red-headed Woodpecker
Black-backed Woodpecker
Red-cockaded Woodpecker
American Three-toed
Woodpecker

Downy Woodpecker

Hairy Woodpecker
Yellow-bellied Sapsucker

Pied-billed Grebe

Northern Saw-whet Owl
Boreal Owl

Short-eared Owl
Long-eared Owl
Burrowing Owl

Great Horned Owl
Eastern Screech-Owl

 

 

 

 

 

 

Conocephallnae
Conocephaius fasciatus
Conocephaius spartinae
Orcheiimion nigripes
Orcheiimion vulgare
Orcheiimum nemoralis
Orcheiimum robustus

Copiphorinae

Neoconcephaius ensiger

-110-

 

FAMILY [Germs species nélishName
Str'ix nebuiosa Great Gray Owl
Str'ix varia Barred Owl
3 TYTONIDAE
: Tyto aiba Barn Owl
HOMOPTERA
CICADIDAE
Diceroprocta vitripennis Diceroprocta
Magicicada septendecim Magicicada
Okanagana canadensis Okanagana 1
Okanagana rimosa Okanagana 2
Tibicen auietes Tibicen 1
Tibicen canicuiaris Tibicen 2
Tibicen chloromera Tibicen 3
Tibicen linnei Tibicen 4
Tibicen Iyricen Tibicen 5
Tibicen pruinosa Tibicen 6
ORTHOPTERA '
GRYLLIDAE
Eneopterinae
Anaxipha exigua Say's Bush Cricket
Orocharis saltator Jumping Bush Cricket
Phyiiopaipus puicheiius Red-headed Bush Cricket
Gryllinae
Gryiius veietis Northern Spring Field Cricket
Nemobilnae
Aiionembius tinnulus Tinkling Ground Cricket
Alionemobius aliardi Allard's Ground Cricket
Aiionemobius fasciatus Striped Ground Cricket
Eunemobius carolinus Carolina Ground Cricket
Oecanthlnae
Neoxabea bipunctata Two-spotted Tree Cricket
Oecanthus ceierinictus Fast-calling Tree Cricket
Oecanthus exclamationls Davis Tree Cricket
Oecanthus fuitoni Snowy Tree Cricket
Oecanthus Iatipennis Broad-winged Tree Cricket
( ‘ Oecanthus nigricomis Black-named Tree Cricket
5 Oecanthus niveus Narrow-winged Tree Cricket
3‘, Oecanthus pini Pine Tree Cricket
5 GRYLLOTALPIDAE
Neocurtiiia hexadectyia Northern Mole Cricket
TETTIGONIIDAE

Slender Meadow Katydid
Saltmarsh Meadow Katydid
Black-legged Meadow Katydid
Common Meadow Katydid
Lesser Pine Katydid
Woodland Meadow Katydid

Sword-bearer Conehead Katydid

 

 

ORDER FAMILY Sum! [600088906108

ngllshName

 

 

 

Neoconcephalus exiiiscanoms
Neoconcephalus nebrascensis

Neoconcephaius retusus
Neoconcephalus robustus

Phaneropterlnae

Ambiycorypha obiongifoiia
Ambiycorypha rotundifolia
Microcentrum retinerve
Microcentrum rhombifolium
Scudderia curvicauda
Scudderia furcata
Scudderie septentrionalis
Scudderia texensis

Pseudophyllinae

Pieryphylia cameiiifoiia

Tettlgonllnae

Atlanticus testaceus
Metrioptera roeselii

 

Long-beaked Conehead Katydid
Nebraska Conehead Katydid

Round-tipped Conehead Katydid
Robust Conehead Katydid

Oblong-winged Katydid
Round-winged Katydid
Lesser Angle-winged Katydid
Greater Angle-winged Katydid
Curve-tailed Bush Katydid
Fork-tailed Bush Katydid
Northem Bush Katydid

Texas Bush Katydid

Northern True Katydid

Short-legged Shield-bearer
Roesel's Decticid

 

-111-

 

References

Alford, R. A. and S. J. Richards (1999). Global amphibian declines: A problem in
applied ecology. Annual Review of Ecology and Systematics 30: 133-165.

Aylor, D. (1971). Noise reduction by vegetation and ground. Journal of the Acoustical
Society of America 51: 197-205.

Bailey, W. J. (1991). Acoustic behgviour of in_sects. New York, Chapman and Hall.

Bailey, W. J ., H. C. Bennet-Clark and N. H. Fletcher (2001). Acoustics of a small
australian burrowing cricket: The control of low-frequency pure-tone songs. IL;
Journi of Experimental Biology 204: 2827-2841.

Bennet-Clark, H. (1999). Resonators in insect sound production: How insects produce
loud pure-tone songs. Journal of Experimental Biology 202(23): 3347-3357.

Bennet-Clark, H. C. (1997). Tymbal mechanics and the control of song frequency in the
cicada cyclochila australasiae. Joumgl of Experimental Biology 200(1 1): 168 l -
1694.

Bennet-Clark, H. C. (1998). Size and scale effects as constraints in insect sound
communication. Philosophical Transaction_s of the Royal Society of London B
Biological Sciences. 353(1367): 407-419.

 

Benoit-Bird, K. J,. and W. W. L. Au (2001). Target strength measurements of hawaiian

mesopelagic boundary community animals. J ourngl of the Acoustical SocietLof
America 110(2): 812-819.

Bird Songs Eastern/Central (2002). Peterson Field Guide Audio Series. R. T. Peterson.
Cornell Laboratory of Ornithology, Houghton Mifflin.

Bosch, J ., A. S. Rand and M. J. Ryan (2000). Acoustic competition in physalaemus

pustulosus, a differential respOnse to calls of relative frequency. Ethology 106:
865-871. '

Bukhvalova, M. A. and R. D. Zhantiyev (1994). Acoustic signals in grasshopper

communities (orthoptera, acrididae, gomphocerinae). Entomological Review
73(2): 121-136.

Buskirk, J. v. (1997). Independent evolution of song structure and note structure in
american wood warblers. Proc. R. Soc. Lond. 264: 755-761.

_ Dale, V. H. and S. C. Beyler (2001). Challenges in the development and use of ecological
indicators. Ecological Indicators 1: 3-10.

Dooling, R. J ., M. R. Leek, O. Gleich and M. L. Dent (2002). Auditory temporal
resolution in birds: Discrimination of harmonic complexes. J oumIal of the
Acoustical Society of America 112(2): 748-759.

Duellman, W. E. and L. Trueb (1986). Biology of Amphibians. Baltimore, The Johns
Hopkins University Press.

112

Elliott, L. (2004). The Calls of Frogs and Toads, Stackpole Books.

Federal Aviation Administration (1985). Aviation noise effects. Washington, DC, US.
Department of Commerce.

Freeberg, T. M., J. R. Lucas and B. Clucas (2003). Variation in chick-a-dee calls of a
carolina Chickadee population, poecile carolinensis: Identity and redundancy
within note types. Journal of the Acoustical Society of America 113(4): 2127-
2136.

Gage, S. H., B. M. Napoletano, M. Colunga-Garcia and J. Qi (2003). An analytical
framework to interpret acoustic observations in heterogeneous landscapes.
Ecological Applications.

Gill, F. B. (1995). Ornithology. New York, W.H. Freeman and Company.

Greenwood, D. D. (1996). Comparing octaves, frequency ranges, and cochlear-map
curvature across species. Hearing Research 94: 157-162.

Hartmann, W. M. (1998). Sigaals, sound, and sensation. New York, Springer-Verlag.

Heffner, H. and B. Masterton (1980). Hearing in glires: Domestic rabbit, cotton rat, feral

house mouse, and kangaroo rat. Journal of the Acoustical Society of America
68(6): 1584-1599.

Helbig, A. J ., A. G. Knox, D. T. Parkin, G. Sangster and M. Collinson (2002). Guidelines
for assigning species rank. QLS 144(3): 518-525.

Hopp, S. L., M. J. Owren and C. S. Evans (1998). Animal acoustic communication:
Sound analysis and research methods. Germany, Springer-Verlag Berlin
Heidelberg.

Home, J. K. (2000). Acoustic approaches to remote species identification:A review.
Fisheries Oceanography 9(4): 356-371.

Home, R. S. (2001) Spectrogram 6.5c. Visualization Software LLC,

Krause, B. (1987). The niche hypothesis: How animals taught us to dance and sing.
Wholegth Review 57(Winter).

Krause, B. (1999). How loss of natural sound causes stress in humans and other
creatures. Acoustical Society of America: [-3.

Krause, B. (2001). Loss of natural soundscape: Global implications of its effect on
humans and other creatures. San Francisco World Affairs Council. San Francisco.

Kroodsma, D. E. and E. H. Miller, Eds. (1996). Ecolggy and evolution of acoustic
communication in birds, Comstock Publishing Associates.

Laszlo, E. (1996). The systems view of the world: A holistic vision for our time.
Cresskill, NJ, Hampton Press, Inc.

Lide, D. R., Ed. (2004). Cifllandbook of chemistry and physics. New York, CRC Press.

Masterton, B., H. Heffiier and R. Ravizza (1969). The evolution of human hearing.
Journal of the Acoustical Society of America 45(4): 966-985.

113

Morton, E. S. (1975). Ecological sources of selection on avian sounds. The American
Naturalist 109(965): 17-34.

Naguib, M. (1996). Ranging by song in carolina wrens thryothorus ludovicianus: Effects
of environmental acoustics and strength of song degradation. Behaviour 133:
541-559.

National Research Council (2001). Grand challenges in environmental sciences.
Washington, DC, Committee on Grand Challenges in Environmental Sciences

Oversight Commission for the Committee on Grand Challenges in Environmental
Sciences.

Nischk, F. and K. Riede (2001). Bioacoustic of two cloud forest ecosystems in ecuador

comparedto a lowland rainforest with special emphasis on singing cricket species:
217-242.

Odum, E. P. ( 1963). Ecology. New York, Holt, Rinehart and Winston.

Ping, J. S. R. D., Z. Yang and W. Y. Chun (1996). Characteristics of songs of the chinese

bulbul (pycnonotus sinensis) in the breeding season. Acta Zoologica Sinica 42(3):
253-259.

Rannels, S., W. Hershberger and J. Dillon (1998). Songs of Crickets and Katydids of the
Mid-Atlantic States, Wil Hershberger and Steve Rannels.

Riede, K. ([993). Monitoring biodiversity: Analysis of amazonian rainforest sounds.
Ambio 22(8): 546-549. '

Roffler, S. K. and R. A. Butler (1967). Factors that influence the localization of sound in
the vertical plane. The Journal of Acoustical Society of America 43(6): 1255-
1259.

Romoser, W. S. and J. John G. Stoffoloano (1998). The Science of Entomology. Boston, ~
Massachusets, McGraw-Hill.

Rundus, A. S. and L. A. Hart (2002). Overview: Animal acoustic communication and the

role of the physical environment. Jouraal of Comparative Psychology 116(2):
120-122.

Sandbom, A. F. and P. K. Phillips (2001). Re-evaluation of the diceroprocta delicata

(homoptera: C icadidae) species complex. Annals of the Entomological Society of
America 94(2): 159-165.

Schafer, R. M. (1977, 1994). The soundscape: Our sonic environment and the tuning of
the world. Rochester, Vermont, Destiny Books.

Schwartz, J. J ., B. W. Buchana and H. C. Gerhardt (2001). Female mate choice in the
gray treefrog (hyla versicolor) in three experimental environments. Behavioral
Ecologyand Sociobiology 49: 443-455.

Shackleton, S. A., L. Ratcliffe, A. G. Horn and C. T. Naugler (1991). Song repertoires of
harris' sparrows zonotrichia querula. Canadian Journal of Zoology 69(7): 1867-
1874.

114

Slabbekoom, H. and T. B. Smith (2002). Bird song, ecologv and speciation. The Royal
Society 357: 493-503.

Stevenson, R. J., S. H. Gage, T. Hough, D. T. Long, B. Pijanowski, J. Qi, M. Wiley, P.
Bonnell, R. Bowman and D. Denison (2001). An ecological assessment of the
muskegon river watershed to solve and prevent environmental problems. East
Lansing, Great Lakes Fisheries Trust.

Truax, B. (1999). Handbook for acoustic ecology, Cambridge Street Publishers.

Turner, M. G., R. H. Gardner and R. V. O'Neill (2001). Landscape ecologLin theoryzm
pLactice: Pattern and process. New York, New York, Springer-Verlag New York,
Inc.

Wollerman, L. (I999). Acoustic interference limits call detection in a neotropical frog
hyla ebraccata. Animal Behaviour 57: 529-536.

ll5

 

   

llllilllllllll