Fine scale distribution and habitat use of harbour ... - Clyde …€¦ · The Firth of Clyde is a...
Transcript of Fine scale distribution and habitat use of harbour ... - Clyde …€¦ · The Firth of Clyde is a...
1
Fine scale distribution and habitat use of harbour porpoises
(Phocoena phocoena) in the Firth of Clyde
William Brown
170023659
Supervised by Dr Jonathan Gordon and Dr Simon Northridge
A dissertation submitted in partial fulfilment of the requirements for the degree of Masters of Science at the University of St Andrews
Date of submission: 15th August 2018
2
I hereby certify that this dissertation, which is approximately 8600 words in length, has been
composed by me, that it is the record of work carried out by me and that it has not been
submitted in any previous application for a degree. This project was conducted by me at the
University of St Andrews from September 2017 to August 2018 towards fulfilment of the
requirements of the University of St Andrews for the degree of MSc Marine Mammal
Science under supervision of Dr Jonathan Gordon and Dr Simon Northridge.
15th August 2018
3
Contents
1. Abstract 4
2. Acknowledgments 5
3. Introduction 6
4. Materials and methods 13
4.1. Data collection 13
4.1.1. Acoustic analysis 13
4.1.2. Covariate data 15
4.2. Data analysis 17
4.2.1. Pre-statistical analysis 17
4.2.2. Habitat modelling 17
4.2.3. Hotspot analysis 18
4.2.4. Cluster analysis 19
5. Results 20
5.1. Habitat modelling 21
5.2. Hotspot analysis 23
5.3. Cluster analysis 24
6. Discussion 25
7. References 31
Appendix I 38
Data Use Agreement
Statement of Ethics
4
1. Abstract
High densities of harbour porpoises (Phocoena phocoena) have been repeatably reported
on the west coast of Scotland on coarse spatial scales (>10km). However, fine scale (<10km)
information is often lacking. Consequently, in this study, 25 passive acoustic monitoring
(PAM) surveys covering a small focal area in the north of the Firth of Clyde, undertaken as
part of larger survey network in the wider Clyde area, were analysed acoustically.
Generalised Additive Models (GAMs) fitted with Generalised Estimating Equations (GEEs)
were used to identify relationships between habitat variables and porpoise presence.
Further analysis was undertaken to identify the presence of hotspots (areas of temporally
persistent high use) within the focal area and what degree of clustering detections had in
these areas. Depth was the most important habitat covariate predicting porpoise
distribution with an increasing level of detection probability at depths between 40 and
150m. The final model also found that month, tidal range, slope, vessel speed and the
distance to buoys and sewage outfalls were important in predicting porpoise distribution.
Predictions made from the final habitat model successfully identified areas of high porpoise
use. These results were consistent with the results of hotspot analysis which was able to
identify fine scale (<2km) areas of higher use over a 250m grid. Analysis of the level of
clustering of porpoise detections also identified that clustering occurred at distances less
than 2km. The identification of hotspots on this scale, in combination with the low
parsimony of the final model, suggests that different covariates are likely important for
every hotspot. These results have important consequences for the design of future fine
scale surveys for harbour porpoises using static PAM devices. Furthermore, the identified
hotspots could aid in implicating conservation measures for harbour porpoises in the Firth
of Clyde, which is a highly anthropogenically influenced environment.
5
2. Acknowledgments
I would like to thank my lead supervisor Dr Jonathan Gordon for his expertise, guidance, and
impeccable patience with me throughout the completion of this project. His suggestions for
improvements in my introduction, methods and results substantially improved the
coherence and consistency of these sections. I would also like to thank my co-supervisor Dr
Simon Northridge for the expertise, guidance and time that he provided. Special thanks
must go to Mr David Nairn of the Clyde Porpoise c.i.c for providing the data and for whom
this research would not have been possible. His dedication and enthusiasm for the
conservation of harbour porpoises in the Clyde is second to none. I would also like to thank
Mr Gui Bortolotto, who went above and beyond to help provide expert statistical advice, Mr
Clint Blight, for advice on GIS analysis and provision of R code, Mr Jamie MacAulay, for
advice on acoustic analysis, and Mr Alex Coram for putting up with my constant questions.
Lastly, but by far not least, I would like to thank my girlfriend Sarah for providing more
support and encouragement than anyone could ever ask for throughout the completion of
this project.
6
3. Introduction
The harbour porpoise (Phocoena phocoena) is a small odontocete found in
temperate, coastal waters of the North Pacific, the North Atlantic and the Baltic Sea (Reid et
al. 2003, Bjørge, A and Tolley 2017). Three subspecies have currently been named, one in
each of these ocean basins (Rosel et al. 1995). Within northern European waters it is one of
the most widely distributed and abundant cetacean species present, with population
estimates in the hundreds of thousands (Hammond et al. 2013). However, this high
abundance is a false representation of the conservation importance of this species, whose
relatively short life span of 8-10 years (Bjørge and Tolley 2017), low annual population
growth rate of as little as 4% a year (Woodley and Read 1991), and low reproductive output
of as little as five calves in a lifetime (Lockyer 2003), makes them vulnerable to a variety of
anthropogenic threats (Reynolds III et al. 2009, Nabe-Nielson et al. 2014, Kesselring et al.
2017). Furthermore, like other cetaceans, as major consumers at several trophic levels,
harbour porpoise may play an important role in the health of the ecosystems they inhabit
(Bowen, 1997). Species do not have to be rare to be of high conservation priority, as is the
case for keystone, umbrella and indicator species (Mace et al. 2007, Gaston and Fuller
2008).
The widely held conservation concerns for harbour porpoises are reflected in their
legislation. In Europe, they are listed on Annex II and IV of the EU Habitats Directive (Council
Directive 92/43/EEC). This means that, like all other cetacean species, they are classed as a
European Protected Species meaning it is an offence to kill, injure or disturb them
throughout their whole range. It is also a requirement for member nations to create marine
protected areas for harbour porpoises called special areas of conservation (SAC’s). However,
creating successful SAC’s is difficult, especially for highly mobile species such as cetaceans
which can easily cross marine protected area boundaries (Perez-Jorge et al. 2015).
Nonetheless, in response to this legislation, in the United Kingdom (UK) there has been
substantial efforts to create SAC’s for harbour porpoise in recent years (IAMMWG 2016,
SNH 2016), and currently six potential sites are being analysed by the European Commission
Directorate-General Environment.
7
One of the proposed SAC’s, the Inner Hebrides and the Minches SAC, is on the west
coast of Scotland (SNH 2016). This is an area shown to have consistently high, year-round,
abundance of harbour porpoises (Reid et al. 2003, Hammond et al. 2013, Hammond et al.
2017). It is a highly variable and dynamic part of the Scottish coastline with many areas
having habitat features commonly associated with porpoise presence, such as low tidal
currents and steep slopes (Embling et al. 2010, Booth et al. 2013). However, this is likely not
the only area with these habitat features on the west coast of Scotland. It is possible that
other areas, outside the boundaries of the proposed SAC, that have received less or no
survey effort, may also contain high densities of porpoises. Also, as harbour porpoises are
highly mobile, having been found to travel up to 300 km in three weeks on the east coast of
the United States of America (Read and Westgate 1997), it is possible that the same animals
that frequent the proposed SAC may also spend considerable amounts of time elsewhere.
This becomes problematic when these other areas have high levels of anthropogenic
impacts (Hooker et al. 2011) or are important foraging or breeding sites (Ashe et al. 2010).
This is a commonly identified caveat of static marine protected areas (Perez-Jorge et al.
2015).
One area that falls outside the boundaries of the Inner Hebrides and the Minches
SAC that may play an important role in the conservation of harbour porpoises in Scotland, is
the Firth of Clyde (Figure.1). The Firth of Clyde is a large inlet with high freshwater input,
characterized by deep channels and unique sediment types. An ecosystem review of the
Clyde reports that 15 species of cetaceans have been recorded since 1980 (McIntyre et al.
2012). Few reliable abundance estimates for harbour porpoises are available, however, a
relative abundance estimate of 1645 individuals was provided for the month of July in 2003
(Goodwin and Speedie 2008). This relatively high number is supported by recent acoustic
surveys in the Clyde and the SCANS II survey results in this area (Hammond et al. 2013,
Lawrence et al. 2016). The most recent estimate of harbour porpoise abundance from the
SCANS III is currently not available for the south west coast of Scotland. However, it may be
assumed that the small changes seen in the rest of the west coast of Scotland are consistent
with the Clyde, considering it is likely the same population of animals (Hammond et al.
2017).
8
The Firth of Clyde is a highly anthropogenically influenced environment. It has faced
extensive exploitation and overfishing over the last two centuries and has been labelled as
being in “Ecological Meltdown” due to numerous fishery collapses (Thurstan and Roberts
2010). This has resulted in a shift in the size composition of fish within the Clyde, where
small fish species, such as whiting (Merlangius merlangus), are dominant and large ones are
rare (Heath and Spiers 2011). This is a common response to overfishing (Pauly et al. 1998).
Considering this collapse, it may be expected that the ecosystem within the Firth of Clyde
would be suffering. However, a subsequent ecosystem review of the Firth of Clyde disagrees
with this assumption, and instead of being in “Meltdown” the ecosystem has simply
Figure 1. Vessel tracks and harbour porpoise detections from 25 passive
acoustic monitoring surveys in the Firth of Clyde.
9
changed, with the biomass of fish species within the Clyde being the same as, if not greater
than, it was prior to the fishery collapses (McIntyre et al. 2012). The impact this has had on
harbour porpoises is unclear. However, dietary studies have shown that they have a higher
preference for small fish species in general, and in Scottish waters whiting and sand eels
(Ammodytidae) are the most important prey types (Santos et al. 2004). Consequently, the
shift in size composition of fish species within the Firth of Clyde may in fact be beneficial for
harbour porpoises with an abundance of available prey at their disposal.
The fishery collapses within the Firth of Clyde may have other benefits for harbour
porpoises, such as reducing fisheries interactions. Bycatch has been widely reported as one
of, if not the most, significant anthropogenic threat to harbour porpoise populations (Read
et al. 2006). Therefore, any reduction in bycatch would have a positive impact on their
population. However, it is unknown what bycatch levels harbour porpoise faced prior to
over-exploitation in the Firth of Clyde. Furthermore, bottom set gillnets are the major cause
of bycatch of harbour porpoise globally (Jefferson and Curry 1994, Tregenza et al. 1997),
however, most fishing undertaken in the Firth of Clyde in the past was using bottom
trawlers (Thurstan and Roberts 2010). As a result, irrespective of whether past over-
exploitation is having a positive or negative effect on harbour porpoises in the Firth of Clyde,
both scenarios could be used to argue that they are of higher conservation concern than
they are currently and need protecting.
As a highly industrialised area there are several sources of pollution in the Clyde.
These include discharge from sewage outfalls, run-off from urban areas, oil spills, high levels
of shipping traffic, marine litter, extensive fishing activity and aquaculture (McIntyre et al.
2012). Harbour porpoise are most likely to be affected by increased levels of anthropogenic
noise and contaminant chemicals, such as persistent organic pollutants (POP’s).
Anthropogenic noise is widely regarded as a major potential threat to cetaceans. It can mask
communication, result in avoidance of critical habitat and cause temporary, or even
permanent hearing damage (Weilgart 2007). Harbour porpoises, like other echolocating
odontocetes, are particularly sensitive to high frequency sounds within their hearing range
((10 - 120 kHz (<55 dB re 1 μPa); Kastelein et al. 2010, Dyndo et al. 2015). Consequently,
anthropogenic activities that may produce sounds within this range could disturb harbour
porpoises and has been shown to be the case for both acoustic deterrent devices (ADD’s),
10
used on salmon farms to deter seals (Götz and Janik 2013), and vessel noise (Dyndo et al.
2015). It is likely that disturbance from both ADD’s and vessel noise is a problem in the Firth
of Clyde. Unlike anthropogenic noise, the effects of contaminants on harbour porpoises are
indirect and are difficult to prove. However, samples from stranded animals show that high
levels of POP’s may accumulate in the blubber of porpoises resulting in negative effects such
as reproductive failure or immunosuppression (Murphy et al. 2015). This is relevant as high
levels of POP’s have been recorded within the Firth of Clyde (Kelly 1995, Hussy et al. 2012).
Consequently, harbour porpoises may be at greater threat from the negative effects of
pollution in the Firth of Clyde than in adjacent areas.
The Firth of Clyde was not included in the Inner Hebrides and Minches proposed SAC
as it was not found to contain the top densities of harbour porpoises in at least two spatial
models (SNH 2016). The first of these models (the West Coast Shelf Model) did identify an
area within the Firth of Clyde containing the top 10% of harbour porpoise density (Heinänen
and Skov, 2015). However, this was at a coarse scale (>10km) meaning that no differences in
harbour porpoise density within the Firth of Clyde can be identified. The second of these
models referenced (the West Coast Model), does not contain any survey effort within the
Firth of Clyde at all (Booth et al. 2013, SNH 2016), and is the likely reason for it not being
included in the SAC. Consequently, the lack of survey effort, especially on a fine scale
(<10km) within the Firth of Clyde is apparent.
It has been widely acknowledged that research on the distribution and habitat use of
animals should be undertaken on many spatial scales (Wiens 1989). This is to allow different
trends and patterns to be identified. Research on cetaceans at coarse scales usually involves
large scale surveys along dedicated transect lines to identify areas where animals are
present and absent. When undertaken over large temporal scales, this can then be powerful
in identifying distributional changes in species presence. For example, the SCANS surveys
identified a distinct southerly shift in harbour porpoise distribution in the North Sea over 11
years (Hammond et al. 2013). Furthermore, statistical methods, such as Generalised
Additive Models (GAM’s), can be used on this data to identify habitat features animals are
associated with (Bailey and Thompson 2009, Pirotta et al. 2011). For example, extensive
coarse scale surveys for northern bottlenose whales in eastern Canada identified a high
preference for depth and an association with a submarine canyon (Hooker et al. 1999).
11
However, on coarse scales, such relationships are regularly limited to being with static or
easily predictable, habitat variables, such as depth, slope and sea surface temperature. This
is problematic in dynamic environments, especially when areas are only visited once, as only
a ‘snapshot view’ is provided. As a result, fine scale studies, where areas can be repeatably
surveyed, may be preferable when answering certain ecological questions (Guisan and
Thuiller 2005).
Research on harbour porpoise distribution and habitat use has been undertaken on a
variety of spatial scales. On coarse scales line transect surveys are widely undertaken to
detect porpoise presence. These usually involve following pre-determined transect lines
throughout a sampling area and counting any animals that are detected, either visually or
acoustically using passive acoustic monitoring (PAM) techniques. Using standardised
distance sampling techniques detection functions can then be calculated and the number of
animals that were missed by the observers can be accounted for to get an accurate
abundance estimate (Buckland et al. 2012). Spatial models analysing data collected in this
way have found that important habitat features predicting harbour porpoise presence may
include depth, slope, tidal currents, distance to land and sea surface temperature (MacLeod
et al. 2007, Marubini et al. 2009, Embling et al. 2010, Booth et al. 2013). On fine scales a
higher variety of survey techniques have been used from shore-based visual surveys
(Waggitt et al. 2018), satellite telemetry (van Beest et al. 2018) and static PAM (Nuuttila et
al. 2017). One shore-based study benefiting from the high resolution provided by fine scales
studies, found that porpoise presence increased by three times when the tide started to ebb
over steep slopes (Isojunno et al. 2012). An interaction of this kind between covariates
would be difficult to identify on a coarser scale.
Both visual and acoustic methodologies have been have used for studies on cetacean
distribution and habitat use. Both have their advantages and disadvantages, however, with
continued technological developments PAM is becoming an effective survey method for
detecting harbour porpoises. Harbour porpoises produce characteristic high frequency,
narrow band (most energy between 125 and 140kHz) echolocation clicks (Au et al. 1999),
with high source levels of 178 to 205 dB peak to peak re:1µPa @1m (Villadsgaard et al.
2007). Furthermore, they echolocate frequently (Linnenschmidt et al. 2013), meaning there
is a low probability that a porpoise will not be detected if it passes the hydrophone. Most
12
fine scale studies using acoustic techniques use static PAM devices (Kyhn et al. 2012). One
of the major advantages of static devices is their ability to monitor continuously throughout
the day and for several weeks, if not months at a time. Such temporal coverage is difficult to
achieve with coarse scale transect surveys, unless there is a high repeatability of tracks,
which requires extensive, costly, survey effort (Booth et al. 2013). However, static PAM
devices do not provide the spatial coverage that transect surveys can. Also, it can be difficult
to determine where to position static PAM devices to reliably detect harbour porpoises.
Consequently, compromises are usually made in terms of spatial or temporal resolution.
One solution to this problem may be to use an acoustic array towed behind a vessel, to
cover a fine scale area repeatably. There have been several successful projects using this
methodology, although at coarse scales (Gillespie et al. 2005, Embling et al. 2010, Booth et
al. 2013, Cucknell et al. 2017). They are also preferable to visual survey techniques on this
scale, with porpoises able to be detected in poor sea states, at fast vessel speeds (needed to
prevent counting the same animals twice) and with a small research team (Teilmann 2003,
Embling et al. 2010). Here, for these reasons a towed array is used.
This study aims to use spatial modelling to predict the fine scale (<10km) distribution
and habitat use of harbour porpoises within the northern section of the Firth of Clyde
(Figure.1). This area has received extensive survey effort making analysis of porpoise
distribution and habitat possible at this scale. Considering the scale at which this study is
undertaken potential unique predicator variables are included that are often not possible to
include at coarser scales. These include sources of pollution, in the form of sewage outfall
locations, and potential fishery aggregating devices (FADs), such as buoys and ship wrecks.
There is reliable anecdotal evidence that within the focal area there are predictable
‘hotpots’ (areas of temporally persistent high use) of harbour porpoise presence. Two
methods of analysis will be undertaken to identify whether these hotspots exist and to what
extent porpoises are clustered within them. These hotspots will be compared to the results
of habitat modelling to see if favoured features exist. The results of this analysis will then be
discussed in relation to what this means for the future conservation of harbour porpoises in
the Firth of Clyde and the potential applications of the methodological approaches used for
future research.
13
4. Materials and methods
4.1. Data collection
PAM surveys were carried out by the Clyde Porpoise c.i.c from the 40’ yacht ‘Sarosa’.
During 2016 and 2017 most of the Firth of Clyde received fairly even coverage. In addition,
areas closer to the vessels home base, Largs, received additional coverage resulting in a high
number of repeat tracks. In this study 25 survey tracks completed between the 4th October
2016 and the 9th November 2017, covering this localised ‘focal area’ in the North of the Firth
of Clyde have been analysed (Figure.1).
PAM surveys were undertaken using a 100m towed hydrophone array consisting of
two high frequency hydrophones, and a depth sensor mounted in an oil-filled polyurethane
tube towed on a strengthened cable. The hydrophones and corresponding preamplifiers
have a flat sensitivity between 2kHz and 150 kHz. Continuous recordings were made to a
computer running PAMGuard which also collected GPS data and ran a real time click
detector (Gillespie et al. 2009). The computer system was customised to operate on a small
research vessel running of a 12V DC power supply.
4.1.1. Acoustic analysis
To achieve a consistently analysed dataset recordings (.wav files) from each trip
were all analysed by the author using PAMguard Beta (2.00.12c) running the porpoise click
detector to produce a new set of binary files (the click detection and classification settings
were consistent with those used by researchers on the west coast of Scotland (Embling et al.
2010, Booth et al. 2013)). Binary files were further analysed in PAMGuard viewer mode. The
author scrolled through each section of data and ‘marked-up’ porpoise clicks and labelled
14
these as “Events”. Standard event parameters, which are based on those used in the SCANS
II acoustic surveys (Swift et al. 2008), and widely used elsewhere (Booth et al. 2013, Cucknell
et al. 2017) were assigned. These were porpoise certain click trains (PCTR; at least 7
porpoise clicks in a train with clear directionality (Figure.2)), porpoise likely click trains (PlTr;
a less clear click train that may either have less than 7 clicks or less clear directionality),
porpoise certain events (PcEv; a group of clicks that are clearly porpoise but have no
directionality) and porpoise likely events (PlEv; a group of at least three porpoise clicks that
may be poorer in quality). Unlike the earlier studies mentioned single porpoise clicks where
not included in this analysis. This is because they are more difficult to differentiate from
false positive clicks. To reduce the number of false positive clicks in each event, each click
was assessed to see if it had the characteristic frequency range (as described), waveform
(Villadsgaard et al. 2007) and Wigner plot (Papandreou-Suppappola and Antonelli, 2001) of
a harbour porpoise (Figure.2). Clicks which did not have these features were removed from
events. An amplitude selector was used to aid in this process, as low amplitude clicks were
often echoes or false detections. Sections of track containing identifiable vessel noise, echo-
sounders or sonar where deleted from the final effort data. ‘Marking up’ clicks in this way is
subject to a certain level of subjectivity and as a result a single acoustic analyst (the author)
analysed all 25 days of data.
Figure 2. PAMGuard viewer display showing a characteristic harbour porpoise click train (red triangles
are porpoise clicks identified by the detector). The characteristic waveform, click spectrum and Wigner
plot of a porpoise are shown in the bottom panels.
15
All porpoise certain click trains identified during acoustic analysis were localised
using the PAMGuard Target Motion algorithm to provide a perpendicular distance to the
animal on the assumption that it was stationary as the vessel passed. This algorithm
measures the bearing of every porpoise click within an event using the difference in time of
arrival between the two hydrophone elements. The location of the porpoise is where the
calculated bearings cross. Target motion analysis performed in this way resulted in a left -
right ambiguity for many of the porpoise locations because the bearings calculated are the
same on both sides of the boat. The PAMGuard algorithm solves this ambiguity by
identifying how well the bearing from different clicks converge on each side of the vessel.
This is most successful if the vessel track varies over the course of an event. Here, it is
assumed that PAMGuard solves this correctly. The result of acoustic analysis was a database
including all the ‘marked-up’ events, GPS and acoustic acquisition information. Purpose
written Microsoft Access queries were used to assign each event to the corresponding GPS
locations.
4.1.2. Covariate data
Data was acquired on 12 covariates to predict the distribution and habitat use of
harbour porpoises (Table.1). Tidal data (time and height of low and high tide) was acquired
for the town of Millport, Great Cumbrae, located in the centre of the study area. Purpose
written Microsoft Access queries were used to calculate the position in the tidal cycle and
tidal range for every GPS point. Position in the tidal cycle was subsequently converted into
five factors. This was due to the added complication of using cyclical data in habitat models.
High resolution depth data was available for most of the study area in the form of point
depth samples. This data was interpolated and had a resolution from 5-50m. Where point
depth data was not available, Edina depth data was used. This had a 30m resolution. Ideally,
Edina data would have been used where it has a higher resolution than the point depth
data. However, it was not possible to identify the areas where the point depth data had a
lower resolution. Slope was calculated from both the point depth data and Edina depth data
in QGIS (3.2.0; QGIS Development Team (2018)) using the inbuilt slope calculation tool.
Eleven habitat types were located within the focal survey area according to the EUNIS
habitat classification (Table.1; (Davies et al. 2004)). Four of the habitat types contained few
GPS points and were consequently combined into an ‘other’ habitat category. Depth, slope
16
and habitat type were acquired for every GPS location using the Point Sampling Tool plugin
in QGIS.
Locations of ferry routes and buoys were digitized in QGIS from marine charts. Ferry
routes were traced and then converted into a line of points (10m apart). All marker and
cardinal buoys present on the charts were digitized. It is assumed all buoys were permanent
and were not moved. Coordinates for ship wrecks and sewage outfalls were acquired from
Marine Scotland NMPI. Wreck locations originated from Historic Environment Scotland.
Sewage outfall locations originated from the Scottish Environment Protection Agency. The
shortest distance from each GPS point to each of these covariates was calculated with the
distance to nearest hub tool in QGIS. Vessel speed was not recorded during surveys. Instead,
it was calculated manually for each segment (unit for habitat modelling; see section 2.2.1)
by dividing segment length (m) by segment duration (s).
Table 1. Covariate data used in habitat model selection. Data source, resolution, unit and range is provided where available. When mentioned, segments are the units for habitat modelling. UKHO point bathymetry data was
collected by the Royal Navy, as commissioned by the Maritime and Coastguard agency. Edina data acquired from: http://digimap.edina.ac.uk
Covariate Source Resolution Unit Range
Year NA NA NA 2016-2017
Month NA NA NA January, March-November
Vessel speed Average over segment Every 250m segment Metres per second 1.08 – 4.30
Position in tidal cycle POLTIPS Every GPS location NA High2, Dropping, Low, Rising, High2
Tidal range POLTIPS Every GPS location Metres 1.89 – 3.33
Depth UKHO & Edina 5-50m 5.61 – 174.80
Slope Calculated from Depth Every GPS location Degrees 0.00 – 30.16
Habitat EMODnet 250m EUNIS classification 7 types*
Distance from detection to:
Buoys Edina chart (1:20 000 to 1:75 000 scale)
NA Metres 59.70 - 12544.80
Ferry routes Edina chart (1:20 000 to 1:75 000 scale)
Point taken every 10m Metres 6.30- 15990.30
Ship wrecks Marine Scotland NMPI
NA Metres 53.52 - 8639.84
Sewage outfalls Marine Scotland NMPI
NA Metres 74.84 - 12911.19
*Habitat types: Circalittoral mud, Circalittoral rock and biogenic reef, Circalittoral sand, Offshore circalittoral mud, Offshore circalittoral rock and biogenic reef, Offshore circalittoral sand, Other
17
4.2. Data analysis
4.2.1. Pre-statistical analysis
A detection function was created using only the localised events in R (version 1.1.383; R
Core Team 2017) using the ‘Distance’ package (Miller 2017; Appendix I). It was assumed that
each localised event was unique. Localised events with perpendicular distances over 1000m
were excluded from analysis, as they are assumed to be errors (Cucknell et al. 2017). The
largest 5% of distances were truncated (Buckland et al. 2012).
Covariate data for each GPS point were joined together in R. Purpose written code was
then adapted to split the GPS tracks into 250m segments (Appendix I). This is the coarsest
resolution of the available covariates (Table.1). The mean value of each continuous
covariate was calculated for every segment in Microsoft Access. For factor variables the
value of the first GPS point in each segment was used. Segments missing covariate data
were deleted. A binomial detection column was added to the segment data marking the
presence or absence of an acoustic event in that segment.
4.2.2. Habitat modelling
Prior to model selection collinearity between covariates was tested in R using the
generalized variance inflation factor (GVIF) using the VIF function in the ‘car’ package (Fox
and Weisberg 2011). This was necessary as collinearity can cause inflated or underestimated
standard errors and p-values (Booth et al. 2013). Large GVIF scores indicate collinearity,
however, there is no standardised cut-off value. Here, a cut-off value of 5 was used where
larger values were dropped from model selection. A binary Generalised Additive Model
(GAM) (with a logit link function) was used to predict porpoise presence using the ‘mgcv’
package (Wood 2011). The level of autocorrelation was tested from visualising the
autocorrelation function plot (ACF ) (Appendix I). Subsequent models used a Generalised
Estimating Equation (GEE) framework to model autocorrelation (Liang and Zeger 1986). This
approach has been successful in accounting for autocorrelation in similar studies (Pirotta et
al. 2011, Booth et al. 2013, Pirotta et al. 2014). Here, each individual survey was used as the
blocking variable. An independence model was chosen to model the autocorrelation as the
underlying structure of autocorrelation was unknown. The modelling approach used
followed that outlined by Pirotta et al. 2011, using the ‘yags’ and ‘splines’ packages. The
18
‘yags’ package allows models to be compared using the Quasi-likelihood Information
Criterion (QICu) goodness of fit statistic. QICu is a model selection criteria designed especially
for GEEs as Akaike Information Criterion (AIC) cannot be used on non-likelihood based
methods (Pan 2001). Reduction in QICu was used to identify whether variables should be in
linear or smoothed form (B-splines with a knot at the mean value) within the final model
selection process.
Backwards stepwise model selection was used to identify the final model. This involved
dropping one variable at a time from a full model to see which resulted in the largest drop in
QICu. This was repeated for the next full model, which did not include this dropped
covariate. The final model was identified when no further decrease in QICu is seen when
further variables are dropped. The full model was fitted with the ‘geeglm’ function in the
‘geepack’ package in R (Halekoh et al. 2006). This allows the significance of each covariate to
be found. Model performance was assessed using a confusion matrix which summaries the
goodness of fit of the model by comparing the predictions of the model to the observed
detections. A Receiver Operating Characteristic (ROC) curve, was used to create the cut-off
probability for the confusion matrix. The area under the ROC curve (AUC) was used to test
model performance (values closer to 1 indicate a better model (Boyce et al. 2002)). The final
model was used to predict the probability of detecting porpoises in the original segment
data. For dynamic covariates (tidal range and vessel speed) the mean value from all
segments was used for prediction. October, the month containing the highest level of
survey effort, was used for all segments when predicting. Once predicted, the segments
were then joined to a 250m grid in QGIS. It was not possible to predict the model directly
over the grid. The model was predicted only over grid cells that intersected with two or
more vessel tracks. This is to remain consistent with the Hotspot Analysis (4.2.3).
4.2.3. Hotspot analysis
Hotspot analysis was undertaken in QGIS using the ‘HotSpotAnalysis’ plugin (Oxoli et
al. 2017) to predict porpoise distribution over different size grids. This uses Queen’s Case
Contiguity Matrix to associate z-scores and p-values with porpoise presence in each grid cell.
By accounting for these scores in neighbouring cells hotspots can be identified. Porpoise
presence was predicted over a 250, 500 and 1000m grid. Prior to analysis all of the porpoise
events were uploaded into QGIS. The latitude and longitude provided through localisation in
19
PAMGuard was used as the position for all localised events. For all other events porpoises
were assumed to be on the track line and GPS midpoints were used. The number of unique
(occurring on different surveys) porpoise detections in each grid cell was calculated.
Detections were then effort corrected by dividing each grid cell by the number of unique
tracks which intersected them. For the 1000m grid each cell had to have at least 19 GPS
points falling within it to say that it had been surveyed. For 500m, this limit was 4 points. For
the 250m grid there was no limit. Grid cells containing only one survey track or less were
deleted from the final output as we were only interested in areas that had repeated effort.
4.2.4. Cluster analysis
To identify whether harbour porpoises are clustered throughout the Firth of Clyde an
original analysis technique was applied to see if there was a difference in the distance
between porpoise detections observed to what would be expected by chance. The segment
data was inputted into QGIS. The midpoint of each segment was used as its location.
Distances were calculated from each segment containing a detection to all other segments
containing a detection (DD), in every other survey track, in QGIS using the Distance Matrix
tool. Distances between detections in the same track were not calculated. To provide a
comparison set of distances, which would be expected by chance, distances were calculated
between segments containing detections and segments not containing detections (DN) in
every other survey track. A 5km buffer was set over these distances and they were then split
into 10 bins of equal size in R. A higher number of distances were calculated between the
DN segments than the DD segments. Hence, the proportion of DN segments in each bin was
calculated. This was then multiplied by the sum of all detections in the DD segments in each
bin so the two groups of distances could be compared. A Chi-squared test was used to
compare these observed (DD) and expected (DN) values (For R code, see Appendix I).
20
5. Results
In all 25 surveys a total of 1089 events were “marked up”. Six hundred and twenty-one of
these events were identified as being either a PcTr, PlTr, PcEv or PlEv (Figure.1, Table.2).
Over half of events were localised (PcTr). One perpendicular distance was estimated as
being greater than 1000m, so this event was remarked as a PlTr. The detection function was
truncated at 450m (excluding 5 % of the largest localised distances (Figure.3)). The effective
half strip-width was found to be 231m. In total the GPS tracks were split into 5326
segments, all 250 ± 1.2m long. Out of these segments, 5126 contained data for every
covariate. Segments ranged in duration from 58 seconds to 231 seconds.
Table 2. Summary statistics of harbour porpoise events identified in both survey years (as identified in
PAMGuard). PcTr = porpoise certain click train. PcEv = porpoise certain event PlTr=Porpoise likely click train.
PlEv = porpoise likely event.
Year Number
of tracks
PcTr (localised)
PcEv PlTr PlEv Total
2016 4 90 14 48 14 166
2017 21 229 36 145 45 455
Total 25 319 50 193 59 621
Figure 3. Detection function fitted over a histogram of 319 localised harbour porpoise click trains
(as identified in PAMGuard). The detection function is truncated at 450m.
21
5.1. Habitat modelling
Initial GVIF tests showed that month was correlated to at least one of the other variables in
the model (GVIF=6.39). All other variables were not correlated (GVIF<5). Running GVIF tests
without month showed that it was most likely correlated with year (GVIF scores dropping
from 2.59 to 1.26). Removal of year resulted in the GVIF score for month dropping below
the threshold (GVIF scores dropping from 6.39 to 3.10). It was decided that month should be
included in model selection rather than year. This is because of the lack of uniformity in
survey effort between the two years represented in this data (Table.2), which would make
biological interpretations difficult.
From comparison of QICu scores, preliminary model selection showed that slope, vessel
speed, distance to ship wrecks and distance to sewage outfalls all performed better as linear
terms than as B-splines. These were then included in this form in final model selection. After
selection the final model retained seven covariates (Table.3).
Table 3. Final model results from model selection with GAM’s built within a GEE framework. Covariates
ordered by their importance in the model. χ2 = Wald Statistic. Arrows indicate direction of relationship (even if
small).
The final stage of model selection recommended that vessel speed should have been
dropped from the model. However, it was decided that it should remain nonetheless. This is
because of the high probability that it would have been impacting the detection of harbour
porpoises. Also, removing vessel speed did not have any effect on the relationship between
detection probability and any other covariate. Depth was the most important covariate for
identifying harbour porpoise detections with a higher detection probability at depths
between 40 and 150m. Relationships with other covariates were less clear due to the
realistic confidence intervals produced by the GEEs (Figure.4). One downfall of using the
Covariate df χ2 p
Depth 4 25 <0.001 Month 8 38190 <0.001 Range 4 33 <0.001 Slope 1 16 <0.001 Buoy distance 4 36 <0.001 Outfall distance
4 18 <0.01
Vessel speed 1 1 0.45
22
modelling approach here is that it is very difficult to plot the relationship between linear or
factor variables with the probability of detection (which was not achieved here).
Nonetheless, a small preference for steeper slopes can be identified, whilst tidal range,
distance to buoys, distance to ferry routes and vessel speed all have lower detection
probabilities as they increase in value. There were also highest detection rates during the
month of October.
Despite the large number of covariates retained, and the large confidence intervals, the
final model performed reasonably well with the confusion matrix correctly predicting 76.6%
of presences and 48.4% of absences. Ideally both predictions would be equally high. Here
there is a high false positive rate meaning many absences are being falsely predicted as
presences. The area under the ROC curve suggests the model performed well with an AUC
score of 0.69. When predicted over a 250m grid the final model was able to locate areas
with a higher probability of detecting porpoises (Figure.5). Four areas of markedly higher
probability include those to the east and west of Little Cumbrae (the smallest island in the
centre of the map), an extensive area north of great Cumbrae (the island north of little
Cumbrae) and a large area to the north of the survey area.
Figure 4. Harbour porpoise presence as a function of the smoothed terms depth (a), tidal range (b), distance to buoy (c) and
outfall distance (d). The final model also contained month as a factor and slope and vessel speed as linear terms. Shaded
areas represent the GEE based 95% confidence intervals. A rug plot at the base of the plot shows the actual data values.
23
5.2. Hotspot analysis
Harbour porpoise hotspots were identified on all grid sizes tested. However, over
both the 500m and 1000m grid these were less defined and a high number of grid cells were
wrongly identified as being significant, due to their closer proximity to other cells containing
porpoises. The prediction over the 250m grid performed well (Figure.6). It predicted a group
of small hotspots either side of the Cumbraes (little and great), a large hotspot north of
Great Cumbrae, a large group of small hotspots to the north of the survey area and a few
small hotspots around the island of Bute (the large island to the left of the Cumbraes).
Figure 5. Predicted distribution of harbour porpoises in the Firth of Clyde from the final habit model
(Detection ~ bs(Depth) + Month + bs(Tidal Range) + Slope + bs(Buoys) + bs(Sewage Outfall) + Vessel Speed).
The mean tidal range and vessel speed from all segments was used for prediction. October was used for
month. Grid cells containing only one track (NA) were excluded from predictions.
24
5.3. Cluster analysis
Over 100,000 distances were calculated between DD segments whilst over 2,600,000
distances were calculated between DN segments. Application of the 5km buffer reduced
these values to 19,198 and 347,237 respectively. Putting these distances into bins enabled
comparison. DD segments had a marginally higher proportion of distances less than 2,000m
whilst DN had a marginally higher proportion above this value (Figure.7). A Chi-squared test
performed over all bins suggested that there was no significant difference in the
distributions of each group of distances (χ2= 37.107, df=9, p<0.001).
Figure 6. Predicted distribution of harbour porpoise hotspots in the Firth of Clyde from hot
spot analysis. Grid cells containing only one track (NA) were excluded from predictions.
25
6. Discussion
The results of this study show that alternative analysis techniques can be
incorporated to identify predictable harbour porpoise hotspots within the Firth of Clyde on
a fine scale (<2km). Both the results of the habitat models and hotspot analysis identified
similar areas of higher porpoise presence. Higher resolution results are provided by the
hotspot analysis, whilst the final habitat model gives an indication as to why harbour
porpoises are using these areas. Furthermore, porpoise detections were discovered to be
more highly clustered, at distances less than 2km. This result corroborates with the hotspot
analysis which identified several smaller hotspots at this scale.
Depth was the most important covariate in the final habitat model. As expected,
detection rates increased as depth increased up to 40m. Depth has consistently been found
to be an important predictor of harbour porpoise presence (MacLeod et al. 2007, Goodwin
and Speedie 2008, Isojunno et al. 2012). However, unlike previous studies no peak depth
value was discovered and the detection rates stayed nearly constant for values over 40m.
One potential explanation for this is that the topography of the Firth of Clyde is highly
variable, with deep trenches (<100m) in close proximity to the shoreline, meaning that any
potential preference for certain depths by harbour porpoises is not detected. Alternatively,
Figure 7. The frequency of distances from each segment containing a detection to all other
segments containing a detection in every other track line (Observed), and from each segment
containing a detection to each segment not containing a detection in every other track line
(Expected) in 10 bins of equal size.
26
the relationship seen here is due to the depth preferences of their prey within the Clyde. For
example, whiting have been shown to prefer depths between 40 and 200m (Persohn 2009),
which is exactly the range of depths the final model predicts. Whiting is one of the dominant
species in the Firth of Clyde, that has thrived since larger competitors were overfished
(Heath and Spiers 2011). As a result, it may be uniformly abundant throughout this whole
depth range within the Firth of Clyde, so any particular depth preferences by harbour
porpoises are not detected.
Other important covariates in the final model included month, tidal range, slope,
vessel speed and distance to buoys and sewage outfalls. However, the confidence in these
relationships is low due to the use of GEEs to realistically model autocorrelation.
Nonetheless, the model performance tests undertaken indicated that this model performed
adequately and as a result the relationships seen between each covariate stand if
interpreted with caution. The conclusion that the model performed well was based on the
combined results of both model performance tests. The likely high level of false-positives
predicted, as indicated by a low percentage of absences predicted by the confusion matrix,
suggests that the model did not perform as well as it could have done. However, the high
percentage of positives predicted by the confusion matrix contradicts this conclusion. The
high AUC score given swayed opinion towards accepting the results of this model.
The importance of month in the final model suggested that there may be a seasonal
difference in harbour porpoise presence within the Firth of Clyde as shown elsewhere (Gilles
et al. 2016). However, a trend in the data collection was identified where longer surveys,
covering larger parts of the survey area, were undertaken in later and earlier months of the
year. In summer months, shorter surveys were more common. As a result, the relationship
seen may be not accurately represent the true seasonal changes in porpoise densities in the
Firth of Clyde. Still, the identification of a seasonal trend is likely correct. Tidal range and
slope were expected to be important for the detection of porpoises as has been previously
shown on coarser scales (Marubini et al. 2009, Isojunno et al. 2012). Here, both of these
variables were highly significant in the final model (Table.3) with porpoise detections found
to occur more frequently over steeper slopes and low tidal ranges. It is thought that
porpoises do not prefer steep slopes themselves, but instead the dynamic variables
associated with steeply sloped areas such as increased current flow (Skov and Thomsen
27
2008). Areas of increased current flow against steep slopes cause upwellings and as a result
may have higher levels of primary productivity (Arístegui et al. 1997). In coastal areas it is
possible that tidal range may play a role in the formation of these upwellings by effecting
the amount of water that is moved through them. However, usually, tidal range itself is not
important in the habitat preferences of harbour porpoises. Instead, the position in the tidal
cycle is more important with higher densities previously discovered during flood tides in
other areas (Johnston et al. 2005). Interestingly, in this study, the position in the tidal cycle
was not included in the final model output. One reason for this may be due to the effects of
grouping the tide times into a factor variable. As a continuous variable, these preferences
may have been detected.
One of the original aspects of the habitat modelling presented here is the inclusion
of distances to features that are not usually included in such models. As a result, it is most
interesting that the distance to buoys and the distance to sewage outfalls came out as being
important in the final habitat model. It was hypothesised that buoys may act like fishery
aggregating devices (FADs) and here it was found that porpoise detection increased closer
to buoys. FADs are a common fishing method for several different commercial fish species
and work by using man-made structures or objects to create artificial reefs to attract fish
(Baine 2001). Therefore, in the same way, but on a much smaller scale, permanent buoys
may attract fish as a place to shelter within the Firth of Clyde. Harbour porpoises may then
be attracted to buoys as a result. There was a small increase in the probability of detecting
porpoises at distances less than a 1000m from sewage outfalls. This suggests that they may
be attracted towards them at this scale. It is possible that they are attracted to these
locations due the presence of prey (potentially aggregating around outfalls due to higher
levels of nutrients or warmer temperatures). This finding has important conservation
implications in regard to managing the discharge from such outfalls.
Distance to ship wrecks and ferry routes were not included in the final model. It was
hypothesised, that like buoys, ship wrecks may act as FADs, attracting high abundances of
fish seeking shelter. The influence of ship wrecks may have played a greater role in the
distribution of porpoises if they had a high preference for benthic fish within the Firth of
Clyde. However, although they do form part of their diet (Santos et al. 2004), in the Firth of
Clyde whiting are likely to be more important due to their high abundance. It was also
28
expected that harbour porpoises may have shown a preference for areas further away from
ferry routes due to avoidance of the noise ferries produce. However, this may not have
been detected for either one of three reasons; the ferries using these routes are not regular
enough to result in porpoises permanently avoiding these areas, the harbour porpoises have
become habituated to the vessel noise, or the effect of avoidance was lost by removal of
segments containing vessel noise (including that caused by ferries) from the acoustic data
(due to the inability of detecting porpoise clicks in these periods).
The final covariate included in the final model was vessel speed, although it was
found to not have a significant effect (Table.3), and removal of it may have improved the
final model’s performance. The reason for keeping it in the model was that it was expected
that the speed of the vessel would have an important effect on the detection of harbour
porpoises. As vessel speed was directly correlated with the duration of each segment (being
calculated from it), higher vessel speeds are found in segments surveyed for a shorter
period of time (due to each segment being the same size). Segment duration is likely to have
an important impact on the detection of harbour porpoises in each segment simply because
longer segments have a higher probability of detecting a porpoise. If this was the only
problem with vessel speed then it may have been possible to include it as an offset in the
final model rather than a covariate. However, it is also thought that increased vessel speeds
may result in the increased avoidance of harbour porpoises from the vessel, due to the
increased levels of noise produced. Therefore, porpoises swimming away from the vessel
are unlikely to be detected. This is due to the echolocation beam of harbour porpoises being
highly directional (Au et al. 1999). The avoidance of porpoises from the vessel is seen from
the lower number of detections localised within 25m of the vessel as shown by the
detection function (Figure.3). This pattern has been regularly reported in other PAM
transect surveys and makes calculating abundances from this data difficult using standard
distance sampling protocols (as this assumes the probability of detection at 0m is 1)
(Cucknell et al. 2017).
Despite the relationships discussed, and the acknowledgment that caution should be
used during interpretation, the low confidence and parsimony in the final model suggests
that it does not completely explain the distribution of harbour porpoises in the Firth of
Clyde. This suggests that there may be other important covariates, that were not included in
29
this model, that are also important. One likely important covariate, as discussed in relation
to depth, is the distribution of prey throughout the survey area. However, accurately
predicting prey distribution and how it changes over time is difficult to achieve. A recent
study, that was also undertaken in the Firth of Clyde, did attempt to measure fish
abundance using an echosounder whilst undertaking PAM surveys for porpoises (Lawrence
et al. 2016). No preferences for higher prey numbers were discovered on a fine scale but
porpoises did avoid large areas (>5km) with low fish biomass. However, these results were
only from a single transect survey. There is potential that a study with the repeatability of
tracks, such as the one presented here, might be able to identify preferences on a finer
scale, assuming that porpoises do not show avoidance of the echosounder. Other important
covariates that were not included in modelling that may have been important are latitude
and longitude. These are often included on coarser scale studies and account for general
large-scale spatial differences in distribution (Pirrota et al. 2011). However, in highly
topographically variable areas, like the Firth of Clyde, incorporating them properly is
difficult. This is important as areas that have the same latitude or longitude may be
separated by an island or other geographic barrier which breaks the assumption that
animals can move freely throughout their environment. Recent statistical developments
however, have started to come up with ways to deal with this problem, such as using the
CReSS package in R. Future studies on this scale should try to incorporate this approach. An
alternative explanation for the low parsimony in the final model could be because different
porpoise hotspots are in fact associated with different habitat covariates.
The results of both the model prediction and the hotspot analysis show that porpoise
hotspots do exist within the Firth of Clyde. Furthermore, both methods predict that the
location of these hotspots are in similar places (Figure.5, Figure.6). This shows that habitat
modelling can successfully predict the distribution of animals on fine scales. Furthermore, it
provides more evidence that the final model performs well and that the habitat variables it
includes are important. The hotspot analysis plugin uses spatial clustering statistics to
identify where areas of higher abundance are whilst accounting for neighbouring values.
Similar statistics have had success in identifying distribution of animals (Baruch‐Mordo et al.
2008). However, using it to identify areas of higher abundance of harbour porpoises is a
novel approach. Here, it was able to identify hotspots on a very fine scale (250m grid). The
30
hotspots identified here likely represent the areas within the Firth of Clyde with the highest
productivity. Furthermore, two of the areas identified (the areas around the Cumbraes and
the area in the north of the survey area) corroborate with biodiversity hotspots previously
identified in the Firth of Clyde (Langmead et al. 2008).
The final statistical approach undertaken also showed that porpoises show a level of
clustering at fine scales with detections closer together at distances less than 2000m than
what would be expected by chance. However, the chi-squared test did suggest that there
was no overall difference in the distribution of distances. Nonetheless, this, along with the
identification of hotspots, has important consequences for future research on harbour
porpoises as it shows that they are not uniformly distributed throughout their environment.
In particular, this has important consequences for the design of future static PAM surveys.
This is because they assume that the probability of detecting porpoises is the same in
different areas in which they are placed. However, if devices are placed in either hotspots,
or coldspots, then they may be providing inaccurate information on porpoise abundance.
This is especially true if devices are widely spaced, as is commonly the case. Consequently,
to solve this issue surveys are recommended to be undertaken, similar to the one presented
here, before the deployment of static devices to locate where these hotspots are.
Alternatively, many more devices would need to be deployed to adequately cover an area,
as has been undertaken in many large scale projects (Taylor et al. 2017).
In this study three unique, but comparable, analysis techniques have been used to
predict the fine scale distribution and habitat use of harbour porpoises in the Firth of Clyde
between the end of 2016 and 2017. Through the identification of 621 distinct click events,
over 25 days of survey effort, the high numbers of porpoises that must inhabit the Firth the
Clyde is clearly demonstrated. This has important consequences for the future conservation
of harbour porpoises in the Firth of Clyde, who have yet to receive the attention they
require. It is therefore recommended that abundance estimates for this harbour porpoise
population are provided as a matter of urgency on a fine scale. This would be to assess the
importance of the Firth of Clyde for harbour porpoises on the west coast of Scotland and to
identify the areas where harbour porpoises are coming into contact with anthropogenic
threats.
31
6. References
Arístegui, J., Tett, P., Hernández-Guerra, A., Basterretxea, G., Montero, M.F., Wild, K., Sangrá, P.,
Hernández-León, S., Cantón, M., García-Braun, J.A. and Pacheco, M. 1997. The influence of
island-generated eddies on chlorophyll distribution: a study of mesoscale variation around
Gran Canaria. Deep Sea Research Part I: Oceanographic Research Papers, 44(1), pp.71-96.
Ashe, E., Noren, D.P. and Williams, R. 2010. Animal behaviour and marine protected areas:
incorporating behavioural data into the selection of marine protected areas for an
endangered killer whale population. Animal Conservation, 13(2), pp.196-203.
Au, W.W., Kastelein, R.A., Rippe, T. and Schooneman, N.M. 1999. Transmission beam pattern and
echolocation signals of a harbor porpoise (Phocoena phocoena). The Journal of the
Acoustical Society of America, 106(6), pp.3699-3705.
Bailey, H. and Thompson, P.M. 2009. Using marine mammal habitat modelling to identify priority
conservation zones within a marine protected area. Marine Ecology Progress Series, 378,
pp.279-287.
Baine, M. 2001. Artificial reefs: a review of their design, application, management and
performance. Ocean & Coastal Management, 44(3-4), pp.241-259.
Baruch‐Mordo, S., Breck, S.W., Wilson, K.R. and Theobald, D.M. 2008. Spatiotemporal distribution
of black bear‐human conflicts in Colorado, USA. The Journal of Wildlife Management, 72(8),
pp.1853-1862.
Boyce, M.S., Vernier, P.R., Nielsen, S.E. and Schmiegelow, F.K. 2002. Evaluating resource selection
functions. Ecological modelling, 157(2-3), pp.281-300.
Bjørge, A and Tolley, K. A. 2017. Harbour Porpoise. In Encyclopedia of marine mammals (Wursig,
B, Thewissen, J.G.M. and Kovacs, K.M., eds), pp.448-451, Elsevier Academic Press.
Booth, C.G., Embling, C., Gordon, J., Calderan, S.V. and Hammond, P.S. 2013. Habitat preferences
and distribution of the harbour porpoise Phocoena phocoena west of Scotland. Marine
Ecology Progress Series, 478, pp.273-285.
Bowen, W.D. 1997. Role of marine mammals in aquatic ecosystems. Marine Ecology Progress
Series, 158, pp.267-274.
Buckland, S.T., Anderson, D.R., Burnham, K.P. and Laake, J.L. 2012. Distance sampling: estimating
abundance of biological populations. Springer Science & Business Media.
Cucknell, A.C., Boisseau, O., Leaper, R., Mclanaghan, R. and Moscrop, A. 2017. Harbour porpoise
(Phocoena phocoena) presence, abundance and distribution over the Dogger Bank, North
Sea, in winter. Journal of the Marine Biological Association of the United Kingdom, 97(7),
1455-1465.
Davies, C.E., Moss, D. and Hill, M.O. 2004. EUNIS habitat classification revised 2004. Report to:
European Environment Agency-European Topic Centre on Nature Protection and
Biodiversity, pp.127-143.
32
Dyndo, M., Wiśniewska, D.M., Rojano-Doñate, L. and Madsen, P.T. 2015. Harbour porpoises react
to low levels of high frequency vessel noise. Scientific reports, 5, p.11083.
Embling, C.B., Gillibrand, P.A., Gordon, J., Shrimpton, J., Stevick, P.T. and Hammond, P.S. 2010.
Using habitat models to identify suitable sites for marine protected areas for harbour
porpoises (Phocoena phocoena). Biological Conservation, 143(2), pp.267-279.
Fox, J. and Weisberg, S., 2011. An {R} Companion to Applied Regression, Second Edition. Thousand
Oaks CA: Sage. URL: http://socserv.socsci.mcmaster.ca/jfox/Books/Companion.
Gaston, K.J. and Fuller, R.A. 2008. Commonness, population depletion and conservation biology.
Trends in Ecology & Evolution, 23(1), pp.14-19.
Gilles, A., Viquerat, S., Becker, E.A., Forney, K.A., Geelhoed, S.C.V., Haelters, J., Nabe‐Nielsen, J.,
Scheidat, M., Siebert, U., Sveegaard, S. and Beest, F.M. 2016. Seasonal habitat‐based density
models for a marine top predator, the harbor porpoise, in a dynamic environment.
Ecosphere, 7(6). pp.1-22.
Gillespie, D., Berggren, P., Brown, S., Kuklik, I., Lacey, C., Lewis, T., Matthews, J., McLanaghan, R.,
Moscrop, A. and Tregenza, N. 2005. Relative abundance of harbour porpoises (Phocoena
phocoena) from acoustic and visual surveys of the Baltic Sea and adjacent waters during
2001 and 2002. Journal of Cetacean Research and Management, 7(1), p.51-57.
Gillespie, D., Mellinger, D.K., Gordon, J., McLaren, D., Redmond, P., McHugh, R., Trinder, P., Deng,
X.Y. and Thode, A. 2009. PAMGUARD: Semiautomated, open source software for real‐time
acoustic detection and localization of cetaceans. The Journal of the Acoustical Society of
America, 125(4), pp.2547-2547.
Goodwin, L. and Speedie, C. 2008. Relative abundance, density and distribution of the harbour
porpoise (Phocoena phocoena) along the west coast of the UK. Journal of the Marine
Biological Association of the United Kingdom, 88(6), pp.1221-1228.
Götz, T. and Janik, V.M. 2013. Acoustic deterrent devices to prevent pinniped depredation:
efficiency, conservation concerns and possible solutions. Marine Ecology Progress Series,
492, pp.285-302.
Guisan, A. and Thuiller, W. 2005. Predicting species distribution: offering more than simple habitat
models. Ecology letters, 8(9), pp.993-1009.
Halekoh, U., Højsgaard, S. and Yan, J. 2006. The R package geepack for generalized estimating
equations. Journal of Statistical Software, 15(2), pp.1-11.
Hammond, P.S. et al. 2013. Cetacean abundance and distribution in European Atlantic shelf
waters to inform conservation and management. Biological Conservation, 164, pp.107-122.
Hammond, P. S. et al. 2017. Estimates of cetacean abundance in European Atlantic waters in
summer 2016 from the SCANS-III aerial and shipboard surveys. Available at:
https://synergy.st-andrews.ac.uk/scans3/files/2017/04/SCANS-IIIdesign-based-estimates-
2017-04-28-final.pdf.
33
Heath, M.R. and Speirs, D.C. 2011. Changes in species diversity and size composition in the Firth of
Clyde demersal fish community (1927–2009). Proceedings of the Royal Society of London B:
Biological Sciences.
Heinänen, S. and Skov, H. 2015. The identification of discrete and persistent areas of relatively
high harbour porpoise density in the wider UK marine area, JNCC Report No.544 JNCC,
Peterborough.
Hooker, S.K., Whitehead, H. and Gowans, S. 1999. Marine protected area design and the spatial
and temporal distribution of cetaceans in a submarine canyon. Conservation Biology, 13(3),
pp.592-602.
Hooker, S.K., Cañadas, A., Hyrenbach, K.D., Corrigan, C., Polovina, J.J. and Reeves, R.R. 2011.
Making protected area networks effective for marine top predators. Endangered Species
Research, 13(3), pp.203-218.
Hussy, I., Webster, L., Russell, M. and Moffat, C. 2012. Determination of chlorinated paraffins in
sediments from the Firth of Clyde by gas chromatography with electron capture negative
ionisation mass spectrometry and carbon skeleton analysis by gas chromatography with
flame ionisation detection. Chemosphere, 88(3), pp.292-299.
IAMMWG. 2016. Consultation on possible Special Areas of Conservation for Harbour Porpoise:
Post-Consultation Report. JNCC Report number 597. JNCC, Peterborough.
Isojunno, S., Matthiopoulos, J. and Evans, P.G. 2012. Harbour porpoise habitat preferences: robust
spatio-temporal inferences from opportunistic data. Marine Ecology Progress Series, 448,
pp.155-170.
Johnston, D.W., Westgate, A.J. and Read, A.J. 2005. Effects of fine-scale oceanographic features
on the distribution and movements of harbour porpoises Phocoena phocoena in the Bay of
Fundy. Marine Ecology Progress Series, 295, pp.279-293.
Jefferson, T.A. and Curry, B.E. 1994. A global review of porpoise (Cetacea: Phocoenidae) mortality
in gillnets. Biological Conservation, 67(2), pp.167-183.
Kastelein, R.A., Hoek, L., de Jong, C.A. and Wensveen, P.J. 2010. The effect of signal duration on
the underwater detection thresholds of a harbor porpoise (Phocoena phocoena) for single
frequency-modulated tonal signals between 0.25 and 160 kHz. The Journal of the Acoustical
Society of America, 128(5), pp.3211-3222.
Kyhn, L.A., Tougaard, J., Thomas, L., Duve, L.R., Stenback, J., Amundin, M., Desportes, G. and
Teilmann, J. 2012. From echolocation clicks to animal density—Acoustic sampling of harbor
porpoises with static dataloggers. The Journal of the Acoustical Society of America, 131(1),
pp.550-560.
Kelly, A.G., 1995. Accumulation and persistence of chlorobiphenyls, organochlorine pesticides and
faecal sterols at the Garroch Head sewage sludge disposal site, Firth of Clyde. Environmental
Pollution, 88(2), pp.207-217.
34
Kesselring, T., Viquerat, S., Brehm, R. and Siebert, U. 2017. Coming of age:-Do female harbour
porpoises (Phocoena phocoena) from the North Sea and Baltic Sea have sufficient time to
reproduce in a human influenced environment?. PloS one, 12(10): p.e0186951.
Langmead, O., Jackson, E., Lear, D., Evans, J., Seeley, B. Ellis, R., Mieszkowska, N. and Tyler-
Walters, H. 2008. The Review of Biodiversity for Marine Spatial Planning within the Firth of
Clyde. Report to the SSMEI Clyde Pilot from the Marine Life Information Network (MarLIN).
Plymouth, Marine Biological Association of the United Kingdom.
Lawrence, J.M., Armstrong, E., Gordon, J., Lusseau, S.M. and Fernandes, P.G. 2016. Passive and
active, predator and prey: using acoustics to study interactions between cetaceans and
forage fish. ICES Journal of Marine Science, 73(8), pp.2075-2084.
Liang, K.Y. and Zeger, S.L. 1986. Longitudinal data analysis using generalized linear models.
Biometrika, 73(1), pp.13-22.
Linnenschmidt, M., Teilmann, J., Akamatsu, T., Dietz, R. and Miller, L.A. 2013. Biosonar, dive, and
foraging activity of satellite tracked harbor porpoises (Phocoena phocoena). Marine
Mammal Science, 29(2), pp.77-97.
Lockyer, C. 2003. Harbour porpoises (Phocoena phocoena) in the North Atlantic: Biological
parameters. NAMMCO Sci Pub, 5, pp.71-89
Mace, G.M. et al. 2007 Prioritizing choices in conservation. In Key Topics in Conservation Biology
(Macdonald, D. and Service, K., eds), pp. 17–34, Blackwell Publishing
MacLeod, C.D., Weir, C.R., Pierpoint, C. and Harland, E.J. 2007. The habitat preferences of marine
mammals west of Scotland (UK). Journal of the Marine Biological Association of the United
Kingdom, 87(1), pp.157-164.
Marubini, F., Gimona, A., Evans, P.G., Wright, P.J. and Pierce, G.J. 2009. Habitat preferences and
interannual variability in occurrence of the harbour porpoise Phocoena phocoena off
northwest Scotland. Marine Ecology Progress Series, 381, pp.297-310.
McIntyre, F., Fernandes, P.G. and Turrell, W.R. 2012. Scottish Marine and Freshwater Science
Report Volume 3 Number 3: Clyde Ecosystem Review, Scottish Marine and Freshwater
Science Report. Marine Scotland Science, 1–123.
Miller, D.L. 2017. Distance: Distance Sampling Detection Function and Abundance Estimation. R
package version 0.9.7. https://CRAN.R-project.org/package=Distance
Murphy, S., Barber, J.L., Learmonth, J.A., Read, F.L., Deaville, R., Perkins, M.W., Brownlow, A.,
Davison, N., Penrose, R., Pierce, G.J. and Law, R.J. 2015. Reproductive failure in UK harbour
porpoises Phocoena phocoena: legacy of pollutant exposure?. PLoS One, 10(7), p.e0131085.
Nabe-Nielsen, J., Sibly, R.M., Tougaard, J., Teilmann, J. and Sveegaard, S. 2014. Effects of noise
and by-catch on a Danish harbour porpoise population. Ecological Modelling, 272, pp.242-
251.
35
Nuuttila, H.K., Courtene-Jones, W., Baulch, S., Simon, M. and Evans, P.G. 2017. Don’t forget the
porpoise: acoustic monitoring reveals fine scale temporal variation between bottlenose
dolphin and harbour porpoise in Cardigan Bay SAC. Marine biology, 164(3), pp.1-16.
Oxoli, D., Prestifilippo, G., Bertocchi, D., Zurbaràn, M. 2017. Enabling spatial autocorrelation
mapping in QGIS: The Hotspot Analysis Plugin.
Pan, W., 2001. Akaike's information criterion in generalized estimating equations. Biometrics,
57(1), pp.120-125.
Papandreou-Suppappola A. and Antonelli L. 2001. Use of quadratic time-frequency
representations to analyse cetacean mammal sounds. Newport, Rhode Island: NUWC-NPT
Technical Report 11, 284. pp. 1-129.
Pauly, D., Christensen, V., Dalsgaard, J., Froese, R. and Torres, F. 1998. Fishing down marine food
webs. Science, 279(5352), pp.860-863.
Pérez-Jorge, S., Pereira, T., Corne, C., Wijtten, Z., Omar, M., Katello, J., Kinyua, M., Oro, D. and
Louzao, M. 2015. Can static habitat protection encompass critical areas for highly mobile
marine top predators? Insights from coastal East Africa. PloS one, 10(7), p.e0133265.
Persohn, C., Lorance, P. and Trenkel, V.M. 2009. Habitat preferences of selected demersal fish
species in the Bay of Biscay and Celtic Sea, North‐East Atlantic. Fisheries Oceanography,
18(4). pp.268-285.
Pirotta, E., Matthiopoulos, J., MacKenzie, M., Scott-Hayward, L. and Rendell, L. 2011. Modelling
sperm whale habitat preference: a novel approach combining transect and follow data.
Marine Ecology Progress Series, 436, pp.257-272.
Pirotta, E., Thompson, P.M., Miller, P.I., Brookes, K.L., Cheney, B., Barton, T.R., Graham, I.M. and
Lusseau, D. 2014. Scale‐dependent foraging ecology of a marine top predator modelled
using passive acoustic data. Functional Ecology, 28(1), pp.206-217.
QGIS Development Team. 2018. QGIS Geographic Information System. Open Source Geospatial
Foundation Project. http://qgis.osgeo.org”.
R Core Team. 2017. R: A language and environment for statistical computing. R Foundation for
Statistical Computing, Vienna, Austria. URL https://www.R-project.org/.
Read, A.J. and Westgate, A.J. 1997. Monitoring the movements of harbour porpoises (Phocoena
phocoena) with satellite telemetry. Marine Biology, 130(2), pp.315-322.
Read, A.J., Drinker, P. and Northridge, S., 2006. Bycatch of marine mammals in US and global
fisheries. Conservation Biology, 20(1), pp.163-169.
Reid, J.B., Evans, P.G.H., Northridge, S.P. 2003. Atlas of Cetacean Distribution in North-West
European Waters. Joint Nature Conservation Committee, Peterborough. pp. 1-82
Reynolds III, J.E., Marsh, H. and Ragen, T.J., 2009. Marine mammal conservation. Endangered
Species Research, 7(1), pp.23-28.
36
Rosel, P.E., Haygood, M.G. and Perrin, W.F. 1995. Phylogenetic relationships among the true
porpoises (Cetacea: Phocoenidae). Molecular Phylogenetics and Evolution, 4(4), pp.463-474.
Santos, M.B., Pierce, G.J., Learmonth, J.A., Reid, R.J., Ross, H.M., Patterson, I.A.P., Reid, D.G. and
Beare, D. 2004. Variability in the diet of harbor porpoises (Phocoena phocoena) in Scottish
waters 1992–2003. Marine Mammal Science, 20(1), pp.1-27.
Skov, H. and Thomsen, F. 2008. Resolving fine-scale spatio-temporal dynamics in the harbour
porpoise Phocoena phocoena. Marine Ecology Progress Series, 373, pp.173-186.
SNH. 2016. Inner Hebrides and the Minches proposed Special Area of Conservation. Consultation
Report. SNH.
Swift RJ, Caillat M, Gillespie D. 2008. Appendix D2.3: analysis of acoustic data from SCANS-II
survey. In:SCANS-II. Small cetaceans in the European Atlantic and North Sea. Final Report to
the European Commission under project LIFE04NAT/GB/000245. Sea Mammal Research
Unit, Gatty Marine Laboratory, University of St Andrews, St Andrews, pp. 55.
Taylor, B.L., Rojas‐Bracho, L., Moore, J., Jaramillo‐Legorreta, A., Ver Hoef, J.M., Cardenas‐Hinojosa,
G., Nieto‐Garcia, E., Barlow, J., Gerrodette, T., Tregenza, N. and Thomas, L. 2017. Extinction
is imminent for Mexico's endemic porpoise unless fishery bycatch is eliminated.
Conservation Letters, 10(5), pp.588-595.
Teilmann, J. 2003. Influence of sea state on density estimates of harbour porpoises (Phocoena
phocoena). Journal of Cetacean Research and Management, 5(1), pp.85-92.
Thurstan, R.H. and Roberts, C.M. 2010. Ecological meltdown in the Firth of Clyde, Scotland: two
centuries of change in a coastal marine ecosystem. PloS one, 5(7), p. 11767.
Tregenza, N.J.C., Berrow, S.D., Hammond, P.S. and Leaper, R. 1997. Harbour porpoise (Phocoena
phocoena L.) by-catch in set gillnets in the Celtic Sea. ICES Journal of Marine Science, 54(5),
pp.896-904.
van Beest, F.M., Teilmann, J., Dietz, R., Galatius, A., Mikkelsen, L., Stalder, D., Sveegaard, S. and
Nabe-Nielsen, J. 2018. Environmental drivers of harbour porpoise fine-scale movements.
Marine biology, 165(5), p.95.
Villadsgaard, A., Wahlberg, M. and Tougaard, J. 2007. Echolocation signals of wild harbour
porpoises, Phocoena phocoena. Journal of Experimental Biology, 210(1), pp.56-64.
Waggitt, J.J., Dunn, H.K., Evans, P.G., Hiddink, J.G., Holmes, L.J., Keen, E., Murcott, B.D., Piano, M.,
Robins, P.E., Scott, B.E. and Whitmore, J. 2017. Regional-scale patterns in harbour porpoise
occupancy of tidal stream environments. ICES Journal of Marine Science, 75(2), pp.701-710.
Weilgart, L.S. 2007. The impacts of anthropogenic ocean noise on cetaceans and implications for
management. Canadian Journal of Zoology, 85(11), pp.1091-1116.
Wiens, J.A., 1989. Spatial scaling in ecology. Functional Ecology, 3(4), pp.385-397.
37
Wood, S.N. 2011. Fast stable restricted maximum likelihood and marginal likelihood estimation of
semiparametric generalized linear models. Journal of the Royal Statistical Society (B).
73(1):3-36
Woodley, T.H. and Read, A.J. 1991. Potential rates of increase of a harbour porpoise (Phocoena
phocoena) population subjected to incidental mortality in commercial fisheries. Canadian
Journal of Fisheries and Aquatic Sciences, 48(12), pp.2429-2435.