living among giants: habitat modelling of harbour … · living among giants: habitat modelling of...

Living among giants:

Habitat modelling of Harbour porpoise (Phocoena phocoena)

in the Northern Gulf of St. Lawrence

By

Raquel Soley Calvet

Masters Research in Marine Mammal Science

Supervision by:

Dr. Sonja Heinrich

Dr. Debbie J. Russell

Sea Mammal Research Unit, August 2011

Table of contents

Abstract.................................................................................................................................i

1. Introduction

1.1 Cetacean habitat modelling....................................................................................1

1.2 Ocean Models........................................................................................................2

1.3 The Gulf of St. Lawrence......................................................................................3

1.4 Biology of Harbour Porpoises...............................................................................4

1.5 Aims.......................................................................................................................6

2. Materials and Methods

2.1 Study Area..............................................................................................................7

2.2 Data collection........................................................................................................8

2.3 Generating pseudo-absences.................................................................................10

2.4 Analysis.................................................................................................................11

2.4. a) Generalized linear models and Generalized estimating equations.......... 12

2.4. b) Habitat use models...................................................................................13

2.4. c) Group size models....................................................................................14

3. Results

3.1 Survey data............................................................................................................15

3.2 Habitat use models................................................................................................16

3.3 Group size models.................................................................................................21

4. Discussion....................................................................................................................22

Acknowledgements.......................................................................................................28

References..........................................................................................................................29

Appendix A – Methods...............................................................................................34

Appendix B – Results……………………………......................................................36

Abstract

Habitat preference studies increase the knowledge on the ecology and biology of cetacean

species by analysing environmental proxies to achieve an insight into the underlying reasons

for using a specific area. The presence of harbour porpoises in the eastern Canadian waters

has been studied extensively in the Bay of Fundy but little is known about the Gulf of St.

Lawrence stock. A summer population exists along the Quebec shore where the latest by-

catch estimations indicate that a large number of porpoises are killed by the fishing industry.

Non- systematic survey data collected in the summer seasons from 1997-2003 allowed the

study of harbour porpoise distribution in the western Jacques Cartier Strait. Generalized

Estimating Equations (GEEs) were used to analyse habitat use in relation to environmental

variables derived from an ocean model. Both spatial and temporal variations were found,

although porpoise’s occurrence was highly influenced by local upwelling currents. A

predictive model was also built using GEEs to investigate the presence of large aggregations.

The results suggest that the extent around the Mingan Archipelago in conjunction with

southern sea currents and low turbulent waters enhances the probability of occurrence of big

groups. However, the confidence in the model is dubious due to poor goodness of fit in the

graphical validation, which could not predict more than seven individuals per sighting while

groups made up of twenty individuals were observed. Spatial and temporal occurrence of

harbour porpoises in the study area raises the hypothesis that it may be a calving and nursing

ground. However, further studies with this dataset using different modelling frameworks

could obtain better results than those presented here. Future data collection should include

population-related data such as behaviour or presence of calves for confirmation of this

hypothesis.

Keywords: habitat modelling, Phocoena phocoena, harbour porpoise, generalized estimating

equations.

Mres Marine Mammal Science 2010/ 2011

1

Introduction

1.1 Cetacean habitat modelling

The marine environment is a complex, dynamic and three-dimensional (3D) system variable

at both spatial and temporal scales. Biotic and abiotic factors define the areas animals will

exploit to maximize their fitness (given their physical, behavioural and metabolic

adaptations). The habitats used by the species and individuals will change according to the

requirements needed such as feeding, mating or calving (Acevedo-Gutiérrez 2002). The study

of cetaceans’ distribution represents a huge challenge for the researchers due to its pelagic

nature and the hardly accessible and dynamic environment they live in. Hence, describing

cetacean habitats requires careful selection of the variables and the scale used to explain and

predict their distribution and abundance (Redfern et al. 2006; Redfern et al. 2008).

Several human activities overlap cetacean species distribution affecting negatively on their

habitats (Jefferson et al. 1993; Evans and Raga 2001; Kaschner et al. 2006). Fisheries

(Botsford et al. 1997), by-catch (Read 2006), pollution (Tanabe et al. 1983) or noise

(Richardson et al. 1998) are among the principal threats faced by this taxonomical order. In

order to minimize the potential disturbance and interaction with marine mammals a better

understanding of the habitats they use and select must be achieved.

Habitats in the terrestrial environment have been described and modelled extensively and

some of these analytical approaches used for the marine realm (Guisan and Zimmermann

2000). Identification of cetacean habitats have been made based on sighting locations from

surveys, defining areas of high encounter frequency and density (extended in (Hamazaki

2002) or by relating the observations to environmental variables (Gaskin 1968; Watts and

Gaskin 1985; Morris 1987). These variables can either be static or dynamic (Ross et al. 1987;

Selzer and Payne 1988; Forney 2000). Since cetaceans are highly mobile species and spend

most of their time underwater, descriptions of the habitats where they occur may not suffice

for conservation and management purposes. The development of technologies in the last

decades (satellites, dataloggers or global positioning systems, hereafter GPS), the

improvement of computational power and the implementation of new statistical techniques

from other scientific branches (Guisan and Zimmermann 2000; Guisan et al. 2002; Redfern et

al. 2006) have led to the exponential growth of cetacean habitat modelling studies (Cañadas


2

et al. 2005; Doniol-Valcroze et al. 2007; Panigada et al. 2008; Skov and Thomsen 2008;

Marubini et al. 2009; Žydelis et al. 2011). The design of statistical models predicting

potential habitats is a step forward in the support of management decisions.

The data available, the methodological constraints faced by the researchers or the ecology

and biology of the species should guide the choice of model. Once the description of the

habitat preferences and the potential habitat uses for the target animals to be studied is

explained, increasing knowledge of the ecology of the species can be achieved in order to

better define their ranges (Kaschner et al. 2006), create marine protected areas (Hooker et al.

1999; Cañadas et al. 2005; Hooker et al. 2011) or minimize fisheries interactions (Kaschner

2004).

1.2 Ocean models

The data used for cetacean habitat modelling may come from different survey techniques

(Forney 2002) or by tracking the animals (Baumgartner and Mate 2005; Johnston et al. 2005;

Redfern et al. 2006) and generally includes physical, biological or chemical information from

the environment. Commonly used variables can be distinguished as (1) static, which will not

change over time (water depths or slope) and (2) dynamic, where a spatial or temporal

variation may occur (temperature or salinity). These data can be collected in situ during the

surveys or extracted a posteriori from available datasets (bathymetric maps or remote

sensing).

Another approach obtains habitat data derived from oceanographic-process models, which

include spatial and temporal variability for dynamic features (tidal currents, temperature or

salinity) at different scales (Redfern et al. 2006). Ocean models are based on the hydrostatic

primitive equation introduced by Vilhelm Bjerknes (1910, Paul N. Edwards). Nowadays, two

and 3D ocean models can provide practical data for climate-warming studies (Kamenkovich

et al. 2002), ice-modelling behaviour (Meier 2002), current patterns (Saucier et al. 2003),

krill aggregations (Sourisseau et al. 2006) and model primary production (Skogen et al.

1995). They can also offer dynamic ocean data to marine vertebrate studies (Doniol-Valcroze

et al. 2007; Skov and Thomsen 2008).


3

Since marine mammals are not always available to be observed during surveys, satellite

tracking (using animal-borne tags) can indicate where the animals are, how much time they

spend on the selected areas and the what are the conditions they experience below the surface

(Johnston et al. 2005; Biuw et al. 2007; Sveegaard et al. 2011). However, these can come at

an important economic cost. The development of highly complex but realistic 3D ocean

models, can offer the possibility of investigating the vertical environment where the animals

were sighted, without the handicap of handling and tagging. Nevertheless, with this approach,

we are not specifying whether the animal was using the entire water column but a general

idea can be obtained on the motivation underlying the use of that particular area not only on

the surface, but at different depths.

1.3 The Gulf of St. Lawrence

The Gulf of St. Lawrence is located in the eastern shore of Canada on the North-West

Atlantic Ocean (Fig1). It is considered a semi-enclosed sea due to its geography, surrounded

by the Quebec shore on the north, Newfoundland on the East and Nova Scotia throughout the

West and South-West. It opens to the ocean by Cabot Strait (the biggest entrance to the Gulf)

and the Strait of Belle Isle and covers a total area of 236,000 km2 making it the world’s

largest estuary (Dickie and Trites 1983).

Fig. 1: Location of the Gulf of St Lawrence, North-West Atlantic (Canada). The blue rectangle indicates the study area which is described in further detail in section 2.1


4

The physical and oceanographic features of the gulf, makes it a particularly rich ecosystem

(FOC 2005; Morissette et al. 2006). It is ice-covered in winter, involving a horizontal two-

layer system (Saucier et al. 2003) where deep rich Labrador waters enter through the Strait of

Belle Isle (Han et al. 1999). In spring the rise in temperatures and the ice melting in both land

(freshwater run-offs all along the coast) and sea creates a new structure based on a three layer

system (superficial, intermediate and deep waters). At the same time, the major freshwater

supplier from the gulf (the St Lawrence River), also receives an abundant income from the

Great Lakes in the United States of America (USA). These non-advective processes

(diffusion and mixing) are responsible for the major circulation all over the gulf (FOC 2005;

Larouche et al. 2006). These processes of vertical mixing and the combination of deep and

shallow topography creates unique conditions for high primary production (phytoplankton

blooms, (Herman et al. 1991; FOC 2005) and consequently primary consumers (zooplankton)

such as for krill (Sourisseau et al. 2006).

The marine mammal biodiversity in the Gulf of St Lawrence is large. Thirteen different

species of cetaceans and four pinniped species inhabit or seasonally visit the gulf (Kingsley

and Reeves 1998; FOC 2005). In the early 1990’s, stocks depletion for some of the main

commercial species lead to the decline of the fishing industry (DFO 2001; DFO 2011). The

reduction in fish biomass raised an on-growing concern about competition between marine

mammals and fisheries, with the former accused for the non-recovery of the stocks.

Morissette (2006) revealed the important role of these animals (especially seals) for the

correct balance of the ecosystem. Analyzed in a food web context the effect of predators can

have beneficial effects.

1.4 Biology of harbour porpoises

In the Gulf of St Lawrence along with the biggest animal on the planet, coexist one of the

smallest species of cetacean, the harbour porpoise (Phocoena phocoena). It belongs to the

Phocoenidae family which contains seven different species (Folkens and Reeves 2002). The

harbour porpoise comprises four sub-species, all of them inhabiting cold temperate and sub-

arctic waters in the northern hemisphere (Jefferson et al. 1993) geographically isolated from

one another (Bjørge and Tolley 2002). For the Western North Atlantic, Wang (1996)

supported the early classification by Gaskin (1984) for P.p. phocoena into three management


5

units: (1) Gulf of Maine-Bay of Fundy, (2) Gulf of St Lawrence (with an estimated northern

population of 21,000 individuals, Kingsley et al. 1998) and (3) Newfoundland.

Reaching between 160-170cm in length and an average of 70kg harbour porpoises are small

stocky animals without a pronounced beak (Read 1999). It is characterized by a dark-grey

dorsal body and diminishing light pattern towards white to the ventral side. They have a

small and falcate dorsal fin which is the most common feature of a sighting (Fig 2). Their life

span is over 20 years, reaching sexual maturity between 3-5 years old (Lockyer 1995).

Mating occurs shortly after giving birth (between May and June) and pregnancy is around 11

months. Lactation follows for 8-12 months post-partum, meaning that females are capable of

being pregnant while still feeding their calves (Gaskin 1984; Lockyer 1995; Read 1999).

Small groups of three individuals might be formed by a female with the older and a new calf

(Read 1999). Group sizes are normally small (1-8) but larger aggregations have been

recorded occasionally probably for migrating and feeding (Hoek 1992; Jefferson et al. 1993;

Van Waerebeek et al. 2002).

The diet of harbour porpoises varies on a regional and seasonal scale (Jefferson et al. 1993)

and is dominated by gadoids, clupeids and squid in the Western North Atlantic (Smith and

Gaskin 1974; Recchia and Read 1989). In the Gulf of St Lawrence important species such as

cod (Gadus morhua) or hake (Merlucidae) represents a minor percentage of harbour porpoise

intakes. Instead, small schooling fish such as capelin (Mallotus villosus) and herring (Clupea

harengus) are the most important prey items making up to 80% of their diet (Fontaine et al.

1994a). Harbour porpoises require food equivalent to approximately 3.5% of their body

weight to be consumed daily and this percentage can increase in lactating females (Yasui and

Gaskin 1986). These fish species satisfy their daily caloric requirements as described for

similar nearby areas such as the Bay of Fundy (Fontaine et al. 1994a). Therefore, harbour

porpoise occurrence is thought to be linked to their prey distribution (Santos and Pierce

Fig. 2: Draw of the external morphology of harbour porpoise. © Maurizio Würtz (http://www.artescienza.org/)


6

2003). Tidal currents, sea bottom depths, temperature and topography are some of the

physical features described as distribution proxies which favours prey concentration,

enhances primary production and thus determine predator distribution (Watts and Gaskin

1985; Lockyer 1995; Weir and O’Brian 2000; Johnston et al. 2005; Johnston and Read 2007;

Skov and Thomsen 2008; Sveegaard 2011).

Due to their energy requirements, harbour porpoise distribution overlap with fisheries. One of

the major threats affecting this species is by-catch and for the late 1980’s Fontaine (1994b)

estimated mortalities close to 2000 porpoises per year in some portions of Gulf of St.

Lawrence. Most of the incidental entanglements of porpoises seemed to occur in groundfish

nets set for Atlantic cod while foraging on capelin and herring (Fontaine et al. 1994a). A

reduction in incidental catches was suspected in the 1990’s after fisheries moratoria due to

fish stock depletion and was confirmed in the Bay of Fundy (Trippel and Shepherd 2004).

The latest estimations for the Gulf of St Lawrence (based on questionnaires to the fisherman

and at-sea observers) indicate a decrease in harbour porpoise by-catch although the authors

recognize imprecise results (Lesage et al. 2004).

1.5 Aims

Harbour porpoise distribution has been related to many different physical features according

to their prey. While in some areas tidal currents seem to be the principal characteristic, in

others underwater topography, temperature or depths are the patterns that better describe their

occurrence. Since diet is slightly different across regions, distribution could be reflected in

the application of different feeding behaviour techniques in order to maximize the energy

income. Harbour porpoises are not socially interacting animals but large aggregations have

been described most likely for migration and feeding purposes. Therefore, the aims of this

study will be:

• Analyse the habitat use of harbour porpoise from the northern Gulf of St.

Lawrence,

• Investigate the relationship between environmental conditions and harbour

porpoise group sizes in the study area.


7

Fig. 3: Study area with the main base at Mingan (Canada) and the Minimum Convex Polygon (MCP, grey line) which contains 95% of the total effort for the study period (1997-2003). Projection: Mercator (centered at -63°longitude). Datum: World Geodetic 1984 (WGS84). Bathymetry: GEBCO_08 (www.gebco.net)

Heights/

Depths (m)

Methods

2.1 Study area

The study area was located on the northern section of the Gulf of St Lawrence, between the

Mingan Archipelago and Anticosti Island along the Jacques Cartier Passage (Fig.3). The

geographic and oceanographic characteristics of this area make it particularly important for

marine mammals (Kingsley and Reeves 1998; Naud et al. 2003).

Around the Mingan Archipelago several physical features such as canyons, dunes systems or

rock shelves and deltaic and prodeltaic regions define its topography (Naud et al. 2003). This

strait presents a channel-like bathymetry, reaching maximum depths of 200m with cold and

icy surface waters in winter (Han et al. 1999; Saucier et al. 2003). The presence of Anticosti

Island opposite to the archipelago and the currents from the North-western gulf creates a gyre

between the head of the Jacques Cartier Passage and the lower St. Lawrence Estuary where

plankton aggregations build in winter months (Saucier et al. 2003; Sourisseau et al. 2006).

The Jacques Cartier Strait shows a persistent cold band in summer due to wind driven

upwellings (Koutitonsky and Budgen 1991; Bourque and Kelley 1995), as reported for other

regions in the gulf which has an important impact on fisheries (Rose and Legget 1988). The

cold intermediate layer (CIL) below the warmer surface waters is rich in nutrients and


8

contributes to the shrimp industry as well as the phytoplankton and nekton communities

(Bourque and Kelley 1995; Naud et al. 2003; Sourisseau et al. 2006). Krill aggregations

remain in this deep layer closely related to bathymetry where they accumulate on most slopes

at the margin of the basin. These aggregations are produced on the eastern side of the strait

where dense Labrador waters from the Strait of Belle Isle pushes organisms along the north

shore of Anticosti Island (Sourisseau et al. 2006). Given the winds and currents occurring in

summer (Han et al. 1999; Saucier et al. 2003), the rich plankton community may attract fish

and thus, the bigger predators.

2.2 Data collection

Data collection was carried out by Mingan Island Cetacean Study (MICS, Longue-Pointe-de-

Mingan, Québec, Canada) as part of their long-term research on baleen whales1. A 7m long

rigid-hulled inflatable boat with outboard engines was employed during the study period

between June and November from 1997 to 2003. However, survey efforts varied across

seasons and are summarized in Table1.

Survey speed was around 15-20 knots and every marine mammal sighting (regardless

taxonomic order) was recorded with a unique ID code, along with position and biological

information such as species identification, presence of calves and number of animals. 1 http://www.rorqual.com/englisch/index.html

Table 1. Total survey effort in days for the study period (1997-2003)

Year

Month

June July August September October November Total

1997 11 21 20 16 6 - 74

1998 9 19 22 11 7 - 68

1999 8 16 16 12 1 - 53

2000 4 18 23 13 2 - 60

2001 10 17 21 13 5 1 67

2002 13 15 19 5 2 - 54

2003 10 5 - - - - 15

Total 65 112 121 70 23 1 391


9

In the early years of this dataset, GPS technology was not available and some sightings were

defined using physical features and subjective measures of distance from shore. Approximate

sighting coordinates were then derived from nautical charts. Areas with numerous small

groups of harbour porpoises were summed over several minutes, resulting in large to very

large group sizes (up to 500 individuals). Initial and ending time and their corresponding

positions were recorded for every track to account for survey effort.

On transect, animal counts were quite precise and recounts of the same animal or group of

animals were considered unlikely. Given the size and behaviour of harbour porpoises, most

of the sightings were within 50m of the boat. In some cases, if animals were spotted further

away, a compass was used to obtain a bearing and an estimated distance was given. Based on

this information a coordinate position was calculated. Overall, it was assumed that harbour

porpoise sightings were contained within 500m from the boat, as their detection decreases

with distance (Palka 1995b).

In the event of a baleen whale sighting, the boat followed the individual to obtain images for

photo-identification purposes. Biopsy samples were also collected or implantable tags

attached for telemetry. However, during these procedures, other marine mammal sightings

were still recorded, but not with the same precision due to the focus on the main activities

mentioned above. This could represent a bias in the data as a result of imprecise counts and

possible re-sampling of harbour porpoise groups while following the baleen whales.

The survey effort was highly dependent on weather conditions, which influenced the area

covered daily and the time spent at sea. In order to avoid potential biases from distant

locations visited only once, a Minimum Convex Polygon (MCP) was generated. The MCP

contained 95% of the total area visited during the entire study period (Appendix A).

Locations falling outside the MCP were not considered for analysis in this study.

Following Doniol-Valcroze et al. (2007; 2008) the environmental data for every sighting was

extracted from the oceanographic model developed by Saucier et al. (2003). This 3D coupled

ice-ocean model is gridded with a horizontal resolution of 5x5km, vertical resolution of 5m

and 5min time-step, offering realistic data for tides, atmospheric, hydrologic and ocean forcing.

These physical variables were derived from pre-defined locations in the study area (model nodes,

App. A). Observed measurements are taken as initial conditions at different depths (to develop


10

the seasonal model) at specifically designated stations across the Gulf. These stations are located

on the open boundaries, on the Strait of Belle Isle, St. Lawrence Estuary and Cabot Strait (Fig. 4).

Harbour porpoise occurrences were plotted using ArcView 3.1 software and environmental

data was associated by linking the sighting to the nearest 3D node. Temperature (T°), salinity

(S‰), U (East-West), V (North-South) and W (Up/Downwelling) components of water

current (m.s-1) and turbulent kinetic energy (m2s-2) were then calculated from the ocean model

for the specific date-time of the sighting, covering each 5m layer from surface to the

maximum depth of the specific ocean node.

Approximately, 62% of the observation points were lost in the environmental extraction

process. Most of the missing sightings vanished from the dataset for unknown reasons (but

most likely due to the large amount of data) and others were removed because their position

fell outside the grid of the ocean model.

2.3 Generating pseudo-absences

The data obtained during the survey period was solely represented by presences. Different

analysis methods (such as Ecological Niche Factor Analysis, ENFA, or Principal Component

Fig. 4: Open boundaries where oceanographic stations (red circles) retrieve data for the 3D ocean model. From Saucier et al. 2003


11

Analysis, PCA) rely on this type of data, but in order to improve the habitat use analysis

(Brotons et al. 2004; MacLeod et al. 2008), pseudo-absences are created. These false

absences represent locations in the effort area where the target species could have been seen

(a potential suitable area) but assumed they were not at that particular instant in space and

time. Statistically there is no evidence to attribute an absence following this procedure

because of its inherent uncertainty, but it has been proved to be a valuable resource (Doniol-

Valcroze et al. 2007; Žydelis et al. 2011).

For every unique sighting with its correspondent environmental data, a random point was

generated for the same day and time to create an absence location. Pseudo-absences were

distributed randomly within the 95% MCP. Predictive models using presence only data tend

to enlarge the suitable areas for the studied species. Applying this method, explanatory

variables can be better represented in the model and the predictive power of the model highly

improved (Chefaoui and Lobo 2008).

The full dataset containing presences/absences of harbour porpoises with their associated

biological and environmental data produced and personally provided by Thomas Doniol-

Valcroze.

2.4 Analysis

The statistical analysis on this study was conducted with R software (R Core Development

Team 2011) using geepack (Yan and Fine 2004) and car (Fox and Weisberg 2011)

libraries depending on the models and tests to be performed. Maps used to plot results were

generated using Manifold system 8.0 (© CDA International Ltd., 2008). For the present

study, observations with positions not given by a GPS and group sizes over 20 individuals

were dismissed in order to avoid possible biases from the data collection. Given the nature of

the data set an initial standard exploratory analysis was carried out to check for linearity or

patterns in the covariates for both habitat use and group size models.


12

2.4. a) Generalized Linear Models and Generalized Estimating Equations

Generalized linear models (GLMs) are extensions of linear models where the response

variable follows a non-normal distribution (it follows any of the exponential family

distributions instead) and where the mean of the response is a function of the explanatory

variable via a link function (Zuur et al. 2009).

To build a model three steps must be detailed:

(1) The distribution of the response. Let Yi be the response variable, the notation is:

Y i ~ Binomial (ni,p)

(2) The logistic regression model:

logit (ηi)= α+ β1*variable1+…+ βn*variablei

And (3) the link function (logit, ensures model predictions are between 0-1):

�� = log��

1 − ��

Since GLMs allow the response variable to follow a non-normal distribution and non-

constant variance, these models are more flexible and better fitted for ecological data than

standard linear models (Guisan et al. 2002).

However, biological studies frequently result in spatial or temporal correlated data.

Generalized estimating equations (GEEs) are used for nested data where the independence

assumption is violated (Liang and Zeger 1986). As with GLMs, this statistical approach can

be used to model any of the exponential family distributions (Zuur et al. 2009). The steps to

specify a GEE are the same as in a GLM with the addition of a correlation structure.

(1) The systematic part (where i= nested data and s= measure ):

logit (ηis)= α+ β1*variableis+…+ βn*variableis

(2) The random part: Measurements on different groups (or panels) are assumed to be

independent but correlated within the same group. The variance is then related to the


13

mean by a variance function which allows for over-dispersion by estimating a

dispersion parameter:

Var(Yis) = ϕV(pis)

(3) The correlation structure: Needed for the repeated measures and is given in the shape

of a correlation matrix (R (α)). The correlation matrix will describe as much as

possible the correlation between groups. Four different types can be fitted: AR(1),

Unstructured, Independent or Exchangeable (Hardin and Hilbe 2003; Zuur et al.

2009) depending on the data.

2.4. b) Habitat use models

The data supplied by MICS contained sightings of harbour porpoises with associated physical

data and their corresponding pseudo-absence points, with equivalent information. Depending

on the depth of the presence/absence locations, fewer or more environmental data were

available from the different water layers given by the 3D ocean model. A large number of

sites were found to be in shallow areas and thus fewer layers available for analysis. In order

to evaluate and compare the maximum number of observations, only the surface layer was

entered in the computation of the following models.

For the presence/absence distribution a GLM within the binomial family and logit link

function was performed. A model with all the explanatory variables was reduced to the best

fitting model running hypothesis tests performed by Analysis of Deviance (Anova type II

test, using the “car” library) to obtain p-values from a Chi-Square distribution. The covariates

were dropped from the fully saturated model using backwards selection. Once all the

covariates were significant (at a level of significance, α, of 0.05) the explained Deviance (D)

for binary outcomes was calculated to identify the proportion of deviance explained by the

model.

In order to account for possible non-independence in the data, a Binomial-GEE was built as a

fully saturated model and reduced to the best fitting model in a backwards selection process.

The GEE was specified by day as panel with an independent correlation structure. The

underlying reason for the selection of this panel was driven by the physical conditions of the


14

area, influence of winds and currents (Bourque and Kelley 1995; Han et al. 1999), being

more likely the presence of harbour porpoises to be correlated on a daily basis, but

independent across days. The significance of the covariates was determined by Wald chi-

square hypothesis testing set at a 95% confidence level (Rotnitzky and Jewell 1990). This test

performs its analysis adding terms sequentially. Each covariate occupied the last position in

the model to test for its significance and once all the explanatory variables were tested the

one with the highest p-value was dropped from the model. This procedure ran for several

rounds, in which one covariate was removed once the significance of all the variables was

tested. If no existing correlation arises, the model should be the same as the one performed by

the GLM. On the other hand, if a difference between models was detected, the GEE model

was given priority as it would account for the correlation, and thus the biological meaning of

the model would be more accurate.

2.4. c) Group size models

Two Poisson-GLMs with a log link function were performed to investigate the relationship

between the response variables (group size) and the environmental variables. On the first

model all the observed group sizes entered in the analysis whereas in the second model only

the group sizes above the observed mean group size were considered. Due to the highly right-

skewed distribution on the counts, the former was intended to relate group sizes to

environmental conditions while the latest was projected to explore for the likelihood of large

group sizes (up to 20 individuals). To account for possible over-dispersion in the data a full

model with a quasi-poisson link was refitted before the covariates selection. The dispersion

parameter (ϕ) appeared to be higher than one (1.63) and so the models with extra variance

were selected. The fully saturated models were reduced to the best fitting models following

the same method used for the habitat use analysis mentioned above. The D for the Poisson-

GLMs was calculated to quantify the amount of deviance explained by the models in each

case.

GEEs can be considered as an extension of quasi–likelihood models for independent

measurements (Ballinger 2004). Both count GLMs had their own coupled Poisson-GEE

models to be compared so that non-independence in the data could be detected while already

accounting for over-dispersion. The panels and the correlation structures were set to day and


15

independent respectively for both models. Several rounds using Wald chi-squares p-values

(Rotnitzky and Jewell 1990) determined the variables that would be included in the model at

a 0.05 level of significance.

Results

3.1 Survey data

Total distances covered from 1997 to 2003 are summarized in App. B, and on average

8622km (SE ±7522km) were surveyed each year. The number of days spent at sea and the

total number of harbour porpoise sightings recorded is given in Table 1 and App. B

respectively. The dataset offered by MICS included 3751 harbour porpoise sightings and

their linked absences containing seven different environmental variables (Table 2) for the

surface layer given by the 3D ocean model from Saucier et al. (2003).

Table 2. Physical and oceanographic explanatory variables used in the models. The variable Depth refers to the closest ocean model node to the sightings for which the environmental data was derived. The first term in U, V and W component of the water current are described as the positive values. Mean (95% confidence

intervals) and standard deviations (SD) from the sightings, derived from the ocean model nodes are shown.

Explanatory variable Type Units Mean (95% CI) ±SD

Depth Static m 98.05

(96.53-99.58) 43.46

Temperature Dynamic T° 9.12

(9.028-9.215) 2.65

Salinity Dynamic S‰ 28.95

(28.90-29.01) 1.51

U component of water current (East-West)

Dynamic m.s-1 0.04

(0.034- 0.06) 0.36

V component of water current (North-South)

Dynamic m.s-1 -0.06

(-0.07/ -0.05) 0.21

W component of water current (Up-Downwelling)

Dynamic m.s-1 1.98 e-05

(1.628 e-05-2.326 e-05) 9.93 e-05

TKE turbulent kinetic energy Dynamic m2s-2 0.00022

(0.00013-0.00030) 0.0024


16

Along with this information, the GPS position of the sighting, date, time, month, year and

group size were also specified for every observation. However, because of the

aforementioned potential biases in the data collection (non GPS positions and group sizes

over 20 individuals), 637 records were removed leaving a total of 3114 presences (plus their

linked absences), making up a total of 6228 points entered for the analysis.

3.2 Habitat use models

The GLM-GEE based models differed slightly in some covariates. The meaning of these

differences reflects the temporal autocorrelation in the data. The panel used in the GEE

model was day and so model error correlation was permitted on sightings from the same

days. As a result, the GEE model was the one used for predictions for more feasible

biological meaning. The Wald-statistic test retained the factor Year (χ2 =30.717; p-

value<0.001) although two of these levels (2002, 2003) were not significant. The rest of the

coefficients were all significant at the 0.05 level. The estimated correlation parameter (α) was 0

since the correlation structure was chosen as independent.

The goodness-of-fit of the model was assessed by the Deviance (D) for binary data which

offers an equivalent measure of an R2. The D explained by the GLM model was 0.3 (30%).

Since D is based on a likelihood measure, it cannot be applied to the GEE model. As an

alternative, graphical model validation was performed which returned patterns in the errors

Caution is needed when data are treated with a GEE. Depending on the number of panels

used (usually small), estimates may vary using different correlation structures as well as the

significance of the variables. It is normally recommended to have more than 20 clusters

(Prentice 1988; Ballinger 2004). In this analysis, the number of panels was 282 with a

maximum cluster size of 170. The parameter coefficients for the model are summarized in Table 3.

Based on these estimates, there is compelling evidence that presence of harbour porpoise

groups highly increased with upwelling currents, increased with Longitude and Latitude and

slightly increased with depths. In contrast, lower probabilities of porpoise presence are

associated to warmed water masses or northward currents.


17

Although the predicted probabilities of encounter rose in deeper waters, this estimate was

Month dependent (Fig. 5). June appears to be significantly different from the rest of months

in each year. In June, porpoises are more likely to be seen in shallow waters than in deeper

waters, while the opposite is true for the other months.

As temperature and salinity gradients are often linked and determine water masses, a T/S

diagram was created to predict occurrence subject to these two variables (Fig. 6). The model

suggests that the likelihood of harbour porpoise encounters increase in cold and slightly

saline (estuarine) waters. No differences are appreciated between months, indicating that

physical water preferences are not modified along the season and porpoise’s distribution is

related to areas with a considerable input of freshwater. In this model, harbour porpoises

Table 3. Parameter coefficients for the covariates retained in the binomial GEE model for the habitat use analysis. Year and Month estimates differed across the survey period and the minimum and maximum values of the estimates are given. P-values for these covariates (*) are extracted from the hypothesis testing used with the anova function in the geepack library.

Parametric coefficients Estimate Std. error p-value (Wald-statistic)

Intercept

-199.1000

38.3900

<0.001

Longitude 0.6066 0.1520 <0.001

Latitude 4.7560 0.7995 <0.001

Year ((-0.44)-0.6)* - <0.001*

Month ((-1.34)-(-1.84))* - 0.34*

Temperature -0.1409 0.0500 0.005

Salinity 0.0764 0.0361 0.034

V current -0.7228 0.3368 0.032

W current 2598.0000 675.3000 <0.001

Depth 0.1436 0.0333 <0.001

Depth:Month (0.017-0.021)* - <0.001*

Depth:T5 -0.0024 0.0006 <0.001

Depth:S5 -0.0045 0.0010 <0.001


18

Fig. 5: Predicted occurrences by depths across the season for the years entered in the model. Note than June appears to be the month where sighting probability is higher at shallow waters. Annual differences are suspected but were not tested.

appear to be highly related to upwelling currents and no differences are observed across

seasons or years.

Fig. 6: Predicted hydrographic conditions. The model suggests that the sighting probability increases with lower water temperatures and increasing concentrations in the salinity. Maximum probabilities in the flat top area of this graph, with predicted values of 0.99 in extremely low salinity conditions, should not be considered and will be covered in the next section (Discussion)


19

No differences are appreciated taking into account only for the coordinate position (Fig 7).

This graph suggests that the probabilities of sighting are higher in the north-east region of the

study area (close to the Mingan archipelago). The computational power to develop this graph

did not permit the interaction of more variables and this fact may have caused these similar

results.

The only variable in the model which reflects any difference in the predicted occurrences

across the season is depth. Prediction maps for two months (June and August) were created to

check for spatial distribution. In June, the GEE model predicted the highest probabilities of

porpoise occurrence in shallow waters, near the coast and over the underwater seamounts

(Fig 8). Predicted sightings in August are localized in the underwater valley north of

Anticosti Island (close to the tail of the Anticosti gyre) and the northern region of the MCP.

The outcome given by the model did not reflect the observed locations for that month,

focusing most of the sightings in the inner, but deeper area of the Jacques Cartier Strait.

Fig. 7: Predicted occurrence based on the geographical position for two months across the season, with mean values for the rest of the variables in the mode. North-east region corresponds to the location of the Mingan archipelago.


20

Fig

. 8

: a

) S

igh

ting

s co

llect

ed

in J

un

e a

cro

ss t

he

su

rve

y p

eri

od

(19

97-

20

03

); b

) P

red

icte

d s

igh

ting

s g

ive

n b

y th

e b

ino

mia

l G

EE

mo

de

l in

Ju

ne

fo

r th

e M

CP

are

a;

c)

Sig

htin

gs

colle

cte

d in

Au

gu

st a

cro

ss t

he

su

rve

y p

er

iod

(1

99

7-2

00

3);

d)

Pre

dic

ted

sig

htin

gs

giv

en

by

the

bin

om

ial G

EE

mo

de

l in

Aug

ust

fo

r th

e M

CP

are

a

a b

c d


21

3.3 Group sizes model

Mean group sizes were similar for the entire study period and were found to be around 2-3

individuals per sighting. The group sizes are highly right skewed being groups bigger than 5

individuals rare. Given the previous analysis of occurrence a coupled GLM-GEE model of

density with Poisson distribution from these observations was modelled. Before the variable

selection process began a quasi-poisson model was fitted to check for over-dispersion in the

data, which resulted in a ϕ value of 1.64. The GLM-GEE pair returned the same variables for

the model, indicating no significant temporal autocorrelation and GLM-Poisson model

preferred. The covariates retained in the model are Latitude and the interaction of year and

depths. The D returned by the model could only explain 2.5% of the variation in the data,

demonstrating a poor fit of the model. Due to this deficient proportion, no more than 3

individuals per group could be predicted while groups of 20 were observed. This

underestimation of the data could be most likely to the scarce large groups observed (groups

5-20 individuals only represented 16.4% of the sightings).

For the last model, only sightings with group sizes over the mean average (more than three

individuals) were modelled. The GLM and the GEE Poisson models differed in the selection

of the most significant variables, representing significant temporal autocorrelation. Under this

circumstance, TKE appeared to be significant in the number of porpoises seen in a group for

the first time. Any of the previous models showed significance for this variable. The GEE

selection process included TKE, V and the interactions formed by depth with Year and

Longitude. Graphical validation revealed weak predictive power with no more than groups of

seven porpoises expected by this model.

Large groups were predicted to occur in southward currents and low turbulent waters. When

plotting this model for different depths, large interannual variability was discovered. Some

years had bigger groups concentrated in shallow waters while in others, porpoises

congregated in deeper regions. The same phenomenon occurred in a spatial scale (Fig. 9).

Smaller groups were predicted in 1997 while the biggest group size estimation was in 2001

reaching up to 7 individuals. The rest of years considered in the analysis had similar

predictions, but the overall trend was bigger groups occurring in the eastern zone of the study

area, between the Mingan archipelago and the north shore of Anticosti Island (Jacques Cartier

Strait).


22

Discussion

This study has investigated the habitat use of harbour porpoises in the northern section of the

Gulf of St. Lawrence and the relationship between number of individuals in a group and

environmental variables. The presence of this species in the study area is related to local

upwelling currents along the northern shore that varying seasonally. The number of porpoises

observed in a group was not related to environmental variables and differed through space

and time. However, when modelling groups larger than the mean size, this species was

distributed in relation to low turbulent waters and local currents in the eastern section of the

study area. This is the first habitat analysis of harbour porpoises in the northern Gulf of St.

Lawrence and offers a general insight into the distribution of this population, which suffers

from large by-catch rates in bottom fishing gillnets (Lesage et al. 2004).

Fig. 9: Predicted harbor porpoise group size occurring in a temporal and spatial scale. Note 1997 and 2002 where the group sizes were the lowest and the highest predicted. General trend is derived from this graph in which eastern areas will congregate big groups corresponding geographically to the Jacques Cartier strait, between the Mingan archipelago and the north shore of Anticosti Island.


23

The results of the habitat use model suggest a temporal correlation within days motivated by

local currents occurring in the north-eastern sector of the study area. The particular

oceanographic processes in the area, such as the wind-driven upwellings (Bourque and

Kelley 1995), may vary in intensity on a daily basis and thus affect the amount of

zooplankton at the surface. These microorganisms might attract the harbour porpoises’ main

prey to the upper layers, facilitating feeding activity and thus reducing physical effort. Under

favourable upwelling conditions, more sightings will be recorded on the same day. The

higher probabilities observed in the north-east region of the study area can identify two

effects. On the one hand, it offers sheltered areas in the complex topography of the Mingan

Archipelago (Naud et al. 2003) while on the other hand supplies large quantities of prey. The

breeding season of harbour porpoises is short and seasonal (Read 1990), with most of the

births occurring in a nearby area, the Bay of Fundy, in late May-early June (Smith and

Gaskin 1983). The higher predictions of sightings in June would support the hypothesis that

pregnant females search for protected areas to give birth in shallower waters. No information

on the breeding season has been documented for the Gulf of St. Lawrence population but may

become available in future studies. The demanding biological schedule followed by female

porpoises (simultaneous lactation, pregnancy and own survival) would require areas with a

high quantity of available resources to compensate this energy loss in milk production and

gestation. Fortunately, the low energetic requirements of the foetus, allow the mother to

perform both tasks (Read 1999). Within the same context, the porpoise’s main prey in the

study area (capelin and herring) have their spawning events in coastal waters coincident with

the arrival of the porpoises breeding season. Capelin has one large coastal spawn, but

sporadic hatching has been described due to interannual variations in temperature in the west

coast of Newfoundland (DFO 2001). Herring has two spawning stocks in the northern region

of the gulf, in spring and autumn. Tight schools of this pelagic fish start a shoreward

migration in April to spawn along beaches, but can also occur in deeper waters (125m, DFO

2011). This late spring productive peak, could supply the daily energy requirements of

harbour porpoises at a delicate time of the year (Yasui and Gaskin 1986). Following the

completion of capelin and herring first migration, the surviving individuals will undertake the

return journey to deep waters searching for zooplankton. This relocation of prey can be

related to the higher predicted occurrences of harbour porpoises in deeper waters in the

following months. According to previous studies, a large variability exists in an

oceanographic spatial and temporal scale in the Jacques Cartier Strait influencing the annual

primary production and the seasonal distribution of zooplankton (Saucier et al. 2003;


24

Sourisseau et al. 2006). This variability would explain the significance of the variables year

and month added to the model, causing a non-constant monthly and yearly distribution. The

predicted water properties preferred by porpoises did not differ across the season (Fig. 6).

However, it is worth mentioning that the graphical results obtained for this prediction do not

entirely describe the expected trend, and it is most likely an artefact of the predictions code

used for these joined variables. The upper and flat area of the graph retrieves maximum

probabilities in water properties inappropriate from a marine or estuarine environment

(<15‰). This error could not be fixed due to time constraints. Observed temperatures and

salinities are shown for proof of comparison (Fig. 10).

Skov et al. (2008) identified two different upwelling cells used at different temporal scales

(tidal dependent) for harbour porpoises in Eastern North Sea. The results presented here agree

with their findings in the use of upwelling areas, but in the current study, these were caused

by wind forces and topographic features.

The results derived from the group size models differed when treating the response variable,

number of individuals per group, in different frameworks. The number of individuals forming

a group was not determined by environmental factors. In fact, the GLM-GEE coupled

analysis retained the same variables in the model which were latitude, depth and year. Hence,

no significant differences were found for temporal autocorrelation within the panels (days)

Fig. 10: Histogram of temperature (a) and salinity (b) from the harbor porpoise dataset. The salinities observed and the mean value confirmed with 95% CI, does not suggest any potential preference in salinities under 20‰, while the maximum sighting probabilities were given for salinities between 0-20‰ as mentioned for Fig. 6.


25

but varied across years and location. However, the confidence in this model with a 2.5% of

explained deviance is unconvincing and should not be considered due to its poor quality.

The relationship between large group sizes and environmental variables significantly differed

from the results obtained in the previous Poisson-GLM model. The model for big groups

considered these aggregations as the sets formed by more than three individuals per sighting.

The model presented significant temporal autocorrelation within days and thus the GEE was

used. Bigger groups were predicted to occur in the eastern sector of the study area in southern

currents conditions but low turbulent waters. Hoek (1992) recorded a single event of 800

harbour porpoises in a site close to this study area. The presence of such large aggregations is

not rare (Gaskin 1992; Kingsley and Reeves 1998) in this region and has also been described

in the Bay of Fundy (Read 1990). Yet an improvement was obtained with this model, no

more than seven porpoises per group were predicted, while sightings of up to 20 individuals

were introduced in the analysis. Water depth preferences differed in an interannual mode and

with the exception of 1997 and 1998, the highest predicted group sizes occurred in shallow

waters. These conditions of shallow areas and low turbulence in the eastern sector would

reinforce the theory that there is a calving and nursing area around the Mingan Archipelago

and agrees with the higher observed abundance of harbour porpoise in the Jacques Cartier

Strait (Gaskin 1992; Kingsley and Reeves 1998). Time constraints did not allow investigating

this model in depth, or use the few and hard-computational statistical procedures to validate

the model or explain the importance of the variables could not be inferred.

The analysis carried out in this project employed non-systematic survey methods recording

occurrences of harbour porpoises comprising a wide range of habitat variability (Doniol-

Valcroze 2007). A large number of sightings (7722) were recorded during the study period,

but not the entire dataset could be included in the analysis. The study only focuses on the

harbour porpoise distribution within the MCP, and so the number of sightings missing due to

distant locations did not represent an important loss, but the same is not applied for the extra

lost values. Since the main focus by MICS was not on harbour porpoises, a bias on the

sightability might have been unintentionally introduced. If the emphasis is set to find baleen

whales, the localization of this smaller animal could have involuntary been missed by

inexperienced observers as their sightability decreases with distance (Palka 1995b). In active

search of baleen whales, time spent in some areas may differ from one another and changes

in position due to external factors, such as the weather, could be missed. If no observations


26

were recorded every time the boat changed direction, a bias on the total effort would be

introduced in its computation. Therefore, another approach was considered.

Generating pseudo-absences has proved to be biologically valuable in both land and sea

studies for presence-only data, by allowing some kind of measure of the available space to be

used but assumed it is not at that particular instant in time (Brotons et al. 2004; Doniol-

Valcroze et al. 2007; MacLeod et al. 2008). Creating random points over an area where

porpoises could be seen adds an inherent uncertainty in the data. The probability of an

occurrence at the exact time of a sighting but in a random location over a vast area was not

calculated but can be anticipated to be sufficiently small that would not affect the analysis.

Based on this hypothesis, the data presents locations that would be considered preferred by

the porpoises with other areas of no particular interest.

While focusing in other activities, misdetection and imprecise counts will enter non-realistic

data in the survey. Given the behaviour of this species, incorrect counts in this dataset could

be suspected for large groups due to the lack of an appropriate methodology. A major

concern in counting arises based on how to define a group for these frequently solitary

animals. Several small groups in a big area could be annotated as single events correlated in

time or as a unique observation of numerous individuals in a specific site. Due to equipment

constraints, MICS personnel opted for recording big aggregations as the sum of abundant

smaller groups over a period of time. If this occurrence is very closely related in space and

time, the sum of these smaller groups should be considered as a single event.

Although based on a model and its uncertainty, the environmental data from this ocean model

should be considered adequate (80% goodness-of-fit, Saucier et al. 2003) as positive

economical and ethical aspects are resolved. To study the physical surroundings faced by an

underwater mammal, high technology tags are the only method to directly examine their

habitat. The costs of the tags are avoided here, as well as the capture and handling of the

individual. Nevertheless, an essential limitation is experienced since there is no technique to

recognize whether the individual is using the entire column of water, half of it or just the

surface.

To avoid problems with the non-systematic observations, the GEE model offered the

opportunity to correlate the sightings within days while considering independence across the

survey. This analytical method seemed satisfactory for this data given the diel environmental

variations that occur in the study area created by winds and semi-diurnal tides (Bourque and


27

Kelley 1995; Han et al. 1999). Since the number of strata for each location is determined by

the depth of the sea bottom, a comprehensive analysis of all the layers in the water column

will result in uneven sample sizes. An improved statistical approach could resolve this

problem but time constraints restricted the analysis to the surface layer. In addition,

scheduling did not allow for the performance of different correlation structures and panels for

both the habitat use and group size models. Validation of the model was only given by

graphical summaries where dubious patterns were observed. The non-systematic calculation

of the Quasilikelihood Information Criterion (QIC, Pan 2001) prevented the comparison

between models to account for the better variables to be introduced in the models. This

limitation was undertaken by repeated rounds of Anova, which represented a large amount of

time consumed reducing the full saturated models to the best fitting models. Furthermore,

since the GEEs are based on a quasilikelihood distribution the D explained by the model

could not be obtained and thus the proportion of variation in the data was not identified.

Zheng (2000) developed a formula for marginal models, for instance GEE’s, that he called

Marginal R2 which could be used but time constraints did not allow the computation of this

value. The utilization of this method has increased over the last years (Ballinger 2004;

Panigada et al. 2008; Fieberg et al. 2009) and better models have arisen.

Future work should be focused on the better collection of the biological variables during the

surveys and the systematic tracking of the areas visited per day. The data provided did not

allow for an inference of the effort and this conditioned the generation of the pseudo-

absences. Estimations from better quality data and more biologically descriptive information

(such as behaviour or presence of calves) would improve the analysis. On the early years,

GPS was not available but the development of this technology in the last decade and the

consequent reduction in prices now allows the tracking of all the movements made by the

research vessel. In a statistical context, the large amount of data provided by the ocean model

should be considered in upcoming analysis. The comparison between the significance of

presences in the surface and the adjacent water masses (CIL and deep waters) is encouraged

to be investigated as the zooplankton and prey items are localized at different depths

(Sourisseau et al. 2006). Better techniques to consider for this dataset should include GLMM

and GAMM given the gradual trend in environmental variables (Guisan et al. 2006). Since a

temporal variation was observed over a large scale, individual year models can be attempted

as in Panigada et al. (2008). The methodology and definition used for large groups should be

considered in future seasons in order to better define the causes of this events. The


28

distribution of harbour porpoises in the north shore of Anticosti Island should be considered

in both feeding and social perspectives, since the behaviour and social organisation of

harbour porpoise is poorly understood.

The forthcoming goal to be achieved is to estimate the actual population of harbour porpoises

in this sector of the gulf confirming the calving and nursing area hypothesis in such a rich

feeding ground for marine mammals. The significance of these results can raise the

willingness of a new dedicated survey of abundance (last described in Kingsley and Reeves

1998) in the Northern Gulf of St. Lawrence. The collection of biological parameters (such as

presence of calves) is recommended as the creation of improved habitat models.

Acknowledgements

The development of this project has followed the same life history as a harbour porpoise:

short and intense. Thanks to Richard Sears to kindly reply an e-mail before embarking on the

field. To Christian Ramp to convince me to undertake this long journey, before starting the

new season, Sonja Heinrich to accept supervising it and Clint Blight for the useful maps. To

Thomas Doniol-Valcroze for very valuable comments and kindly helping when everything

meant no-sense and for providing the dataset. To the Canadian Hydrographic Service, and

François Saucier as well as Denis Lefaivre (DFO Canada) for providing the environmental

data, and MICS for collecting data all this years. To Ainhoa Martin Balugo and La Caixa

(Post-graduate scholarships in the UK 2010/2011), for offering me the unique and

unbelievable chance to be part of this. For all the Mres students that shared this experience in

the good and the stressful moments, especially Elisa for making me work harder, Jo for all

the English teaching and Kathleen for all the laughs. For the ones home, supporting me in

every decision I make, and the one that left us too early. Last but not the least, deepest thanks

to Debbie Russell for all the teaching, the guidance, the support and the good vibrations in

every step of this project. You are just awesome!


29

References

Acevedo-Gutiérrez, A. (2002). Habitat use. Encyclopedia of marine mammals B. W. W. F. Perrin, and J. G. M. Thewissen (eds.), Academic Press: 524-528.

Ballinger, G. A. (2004). "Using Generalized Estimating Equations for Longitudinal Data Analysis." Organizational Research Methods 7(2): 127-150.

Baumgartner, M. F. and B. R. Mate (2005). "Summer and fall habitat of North Atlantic right whales (Eubalaena glacialis) inferred from satellite telemetry." Canadian Journal of Fisheries and Aquatic Sciences 62(3): 527-543.

Biuw, M., L. Boehme, C. Guinet, M. Hindell, D. Costa, J.-B. Charrassin, F. Roquet, F. Bailleul, M. Meredith, S. Thorpe, Y. Tremblay, B. McDonald, Y.-H. Park, S. R. Rintoul, N. Bindoff, M. Goebel, D. Crocker, P. Lovell, J. Nicholson, F. Monks and M. A. Fedak (2007). "Variations in behavior and condition of a Southern Ocean top predator in relation to in situ oceanographic conditions." Proceedings of the National Academy of Sciences of the United States of America 104(34): 13705–13710.

Bjerknes, V., T. Hesselberg and O. Devik (1910). Dynamic meteorology and hydrography, Carnegie Institution of Washington.

Bjørge, A. and K. A. Tolley (2002). The harbor porpoise. Encyclopedia of Marine Mammals. B. W. W. F. Perrin, and J. G. M. Thewissen (eds.), Academic Press.

Botsford, L. W., J. C. Castilla and C. H. Peterson (1997). "The Management of Fisheries and Marine Ecosystems." Science 277(5325): 509-515.

Bourque, M. C. and D. E. Kelley (1995). "Evidence of wind‐driyen upwelling in jacques‐cartier strait." Atmosphere-Ocean 33(4): 621-637.

Brotons, L., W. Thuiller, M. B. Araújo and A. H. Hirzel (2004). "Presence-absence versus presence-only modelling methods for predicting bird habitat suitability." Ecography 27(4): 437-448.

Cañadas, A., R. Sagarminaga, R. De Stephanis, E. Urquiola and P. S. Hammond (2005). "Habitat preference modelling as a conservation tool: proposals for marine protected areas for cetaceans in southern Spanish waters." Aquatic Conservation: Marine and Freshwater Ecosystems 15(5): 495-521.

Chefaoui, R. M. and J. M. Lobo (2008). "Assessing the effects of pseudo-absences on predictive distribution model performance." Ecological Modelling 210(4): 478-486.

DFO (2001). Capelin of the Estuary and Gulf of St. Lawrence. DFO Science. Stock Status Report B4-03.

DFO (2011). Assessment of the Quebec North Shore (Division 4S) herring stocks in 2010. Can. Sci. Advis. Sec., Sci. Advis. Rep. 2011/007.

Dickie, L. M. and R. W. Trites (1983). "The Gulf of St. Lawrence." Ecosyst. World 26: 403-425. Doniol-Valcroze, T. (2008). Habitat selection and niche characteristics of rorqual whales in the

Northern Gulf of St. Lawrence (Canada). Department of Natural Resource Sciences. Montreal, McGill University. PhD Thesis: 170.

Doniol-Valcroze, T., D. Berteaux, P. Larouche and R. Sears (2007). "Influence of thermal fronts on habitat selection by four rorqual whale species in the Gulf of St. Lawrence." Marine Ecology Progress Series 335: 207-216.

Evans, P. G. H. and J. A. Raga (2001). Marine mammals: biology and conservation, Kluwer Academic/Plenum Publishers.

Fieberg, J., R. H. Rieger, M. C. Zicus and J. S. Schildcrout (2009). "Regression modelling of correlated data in ecology: subject-specific and population averaged response patterns." Journal of Applied Ecology 46(5): 1018-1025.

FOC (2005). The Gulf of St Lawrence. A unique ecosystem. O. S. Branch, Fisheries and Oceans Canada. Cat. No. FS 104-2/2005.

Folkens, P. A. and R. R. Reeves (2002). Guide to marine mammals of the world, A. A. Knopf, National Audubon Society


30

Fontaine, P.-M., M. O. Hammill, C. Barrette and M. C. Kingsley (1994a). "Summer Diet of the Harbour Porpoise (Phocoena phocoena) in the Estuary and the Northern Gulf of St. Lawrence." Canadian Journal of Fisheries and Aquatic Sciences 51(1): 172-178.

Fontaine, P. M., C. Barrette, M. O. Hammill and M. C. S. Kingsley (1994b). Incidental catches of harbour porpoises (Phocoena phocoena) in the Gulf of St. Lawrence and the St. Lawrence River Estuary, Québec, Canada. Rep. Int. Whaling Comm. Spec. Issue No. 15: 159–163.

Forney, K. (2002). Surveys. Encyclopedia of marine mammals B. W. W. F. Perrin, and J. G. M. Thewissen (eds.), Academic Press: 1203-1205.

Forney, K. A. (2000). "Environmental Models of Cetacean Abundance: Reducing Uncertainty in Population Trends." Conservation Biology 14(5): 1271-1286.

Fox, J. and S. Weisberg (2011). An {R} Companion to Applied Regression, Second Edition. Thousand Oaks CA: Sage. URL: http://socserv.socsci.mcmaster.ca/jfox/Books/Companion.

Gaskin, D. E. (1968). "Distribution of Delphinidae (Cetacea) in relation to sea surface temperatures off Eastern and Southern New Zealand." New Zealand Journal of Marine and Freshwater Research 2(3): 527-534.

Gaskin, D. E. (1984). The harbour porpoise Phocoena phocoena (L.): regional populations, status, and information on direct and indirect catches. Rep Int Whal Commn 34: 569-584.

Gaskin, D. E. (1992). "Status of the harbour porpoise, Phocoena phocoena, in Canada." Can. Field-Nat 106: 36-54.

Guisan, A., T. C. Edwards and T. Hastie (2002). "Generalized linear and generalized additive models in studies of species distributions: setting the scene." Ecological Modelling 157(2-3): 89-100.

Guisan, A., A. Lehmann, S. Ferrier, M. Austin, J. M. C. Overton, R. Aspinall and T. Hastie (2006). "Making better biogeographical predictions of species’ distributions." Journal of Applied Ecology 43(3): 386-392.

Guisan, A. and N. E. Zimmermann (2000). "Predictive habitat distribution models in ecology." Ecological Modelling 135(2-3): 147-186.

Hamazaki, T. (2002). "Spatio-temporal prediction models of cetacean habitats in the mid-western North Atlantic Ocean (from Cape Hatteras, North Carolina, USA to Nova Scotia, Canada)." Marine Mammal Science 18(4): 920-939.

Han, G., J. W. Loder and P. C. Smith (1999). "Seasonal-Mean Hydrography and Circulation in the Gulf of St. Lawrence and on the Eastern Scotian and Southern Newfoundland Shelves." Journal of Physical Oceanography 29(6): 1279-1301.

Hardin, J. W. and J. Hilbe (2003). Generalized estimating equations, Chapman & Hall/CRC. Herman, A. W., D. D. Sameoto, C. Shunnian, M. R. Mitchell, B. Petrie and N. Cochrane (1991).

"Sources of zooplankton on the Nova Scotia shelf and their aggregations within deep-shelf basins." Continental Shelf Research 11(3): 211-238.

Hoek, W. (1992). "AN UNUSUAL AGGREGATION OF HARBOR PORPOISES (PHOCOENA PHOCOENA)." Marine Mammal Science 8(2): 152-155.

Hooker, S. K., A. Cañadas, K. D. Hyrenbach, C. Corrigan, J. J. Polovina and R. R. Reeves (2011). "Making protected area networks effective for marine top predators." Endangered Species Research 13(3): 203-218.

Hooker, S. K., H. Whitehead and S. Gowans (1999). "Marine Protected Area Design and the Spatial and Temporal Distribution of Cetaceans in a Submarine Canyon." Conservation Biology 13(3): 592-602.

Jefferson, T. A., S. Leatherwood and M. A. Webber (1993). FAO Species Identification Guide. Marine mammals of the world. Rome.

Johnston, D. W. and A. J. Read (2007). "Flow-field observations of a tidally driven island wake used by marine mammals in the Bay of Fundy, Canada." Fisheries Oceanography 16(5): 422-435.

Johnston, D. W., A. J. Westgate and A. J. Read (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: 279-293.

Kamenkovich, Sokolov and Stone (2002). "An efficient climate model with a 3D ocean and statistical-dynamical atmosphere*." Climate Dynamics 19(7): 585-598.

Kaschner, K. (2004). Modelling and mapping of resource overlap between marine mammals and fisheries on a global scale. Vancouver, University of British Columbia. PhD Thesis: 240.


31

Kaschner, K., R. Watson, A. W. Trites and D. Pauly (2006). "Mapping world-wide distributions of marine mammal species using a relative environmental suitability (RES) model." Marine Ecology Progress series 316: 285:310.

Kingsley, M. and R. R. Reeves (1998). "Aerial surveys of cetaceans in the Gulf of St. Lawrence in 1995 and 1996." Canadian Journal of Zoology 76(8): 1529-1550.

Koutitonsky, V. and G. Budgen (1991). The physical oceanography of the Gulf of St. Lawrence; A review with emphasis on the synoptic variablitiy of the motion. The Gulf of St. Lawrence: Small ocean or big estuary? J. C. Therriault, Can. Spec. Pucl. Fish. Aquat. Sci. 113: 57-90.

Larouche, P., J. M. Boucher and É.Thomassin (2006). Development of a surface feature motion estimation system for the Gulf of St. Lawrence using SST images. Canadian Technical Report of Hydrography and Ocean Sciences 250, Ocean Sciences Branch (Fisheries and Oceans Canada): 27.

Lesage, V., J. Keays, S. Turgeon and S. Hurtubise (2004). Incidental catches of harbour porpoises (Phocoena phocoena) in the gillnet fishery of the Estuary and Gulf of St. Lawrence in 2000−2002. Can. Tech. Rep. Fish. Aquat. Sci.2552. 37 p.

Liang, K.-Y. and S. L. Zeger (1986). "Longitudinal data analysis using generalized linear models." Biometrika 73(1): 13-22.

Lockyer, C. (1995). Aspects of the biology of the harbour porpoise, Phocoena phocoena, from British waters. Developments in Marine Biology. L. W. Arnoldus Schytte Blix and U. Øyvind, Elsevier Science. Volume 4: 443-457.

MacLeod, C., L. Mandleberg, C. Schweder, S. Bannon and G. Pierce (2008). "A comparison of approaches for modelling the occurrence of marine animals." Hydrobiologia 612(1): 21-32.

Marubini, F., A. Gimona, P. Evans, P. Wright and G. Pierce (2009). "Habitat preferences and interannual variability in occurrence of the harbour porpoise Phocoena phocoena off northwest Scotland." Marine Ecology Progress Series 381: 297-310.

Meier (2002). "Regional ocean climate simulations with a 3D ice-ocean model for the Baltic Sea. Part 1: model experiments and results for temperature and salinity." Climate Dynamics 19(3): 237-253.

Morissette, L., M. O. Hammill and C. Savenkoff (2006). "THE TROPHIC ROLE OF MARINE MAMMALS IN THE NORTHERN GULF OF ST. LAWRENCE." Marine Mammal Science 22(1): 74-103.

Morris, D. W. (1987). "Ecological Scale and Habitat Use." Ecology 68(2): 362-369. Naud, M.-J., eacute, B. Long, Br, ecirc, J.-C. thes and R. Sears (2003). "Influences of underwater

bottom topography and geomorphology on minke whale (Balaenoptera acutorostrata) distribution in the Mingan Islands (Canada)." Journal of the Marine Biological Association of the United Kingdom 83(04): 889-896.

Palka, D. (1995b). Abundance estimates of the Gulf of Maine harbour porpoise. Reports of the International Whaling Commission, Special Issue 16, 27-50.

Pan, W. (2001). "Akaike's Information Criterion in Generalized Estimating Equations." Biometrics 57(1): 120-125.

Panigada, S., M. Zanardelli, M. MacKenzie, C. Donovan, F. Mélin and P. S. Hammond (2008). "Modelling habitat preferences for fin whales and striped dolphins in the Pelagos Sanctuary (Western Mediterranean Sea) with physiographic and remote sensing variables." Remote Sensing of Environment 112(8): 3400-3412.

Paul N. Edwards "Before 1955: Numerical models and the prehistory of AGCMs". Atmospheric general circulation modeling: A participatory history. The American Institute of Physics. http://www.aip.org/history/sloan/gcm/prehistory.html.".

Prentice, R. L. (1988). "Correlated Binary Regression with Covariates Specific to Each Binary Observation." Biometrics 44(4): 1033-1048.

R Core Development Team (2011). R: A language and environment for statistical computing. Vienna, Austria, R Foundation for Statistical Computing ISBN 3-900051-07-0, URL http://www.R-project.org.

Read, A. J. (1990). "Reproductive seasonality in harbour porpoises, Phocoena phocoena, from the Bay of Fundy." Canadian Journal of Zoology 68(2): 284-288.

Read, A. J. (1999). Porpoises, Voyageur Press.


32

Recchia, C. A. and A. J. Read (1989). "Stomach contents of harbour porpoises, Phocoena phocoena (L.), from the Bay of Fundy." Canadian Journal of Zoology 67(9): 2140-2146.

Redfern, J. V., J. Barlow, L. T. Ballance, T. Gerrodette and E. A. Becker (2008). "Absence of scale dependence in dolphin–habitat models for the eastern tropical Pacific Ocean." Marine Ecology Progress Series 363: 1-14.

Redfern, J. V., M. C. Ferguson, E. A. Becker, K. D. Hyrenbach, C. Good, J. Barlow, K. Kaschner, M. F. Baumgartner, K. A. Forney, L. T. Ballance, P. Fauchald, P. Halpin, T. Hamazaki, A. J. Pershing, S. S. Qian, A. Read, S. B. Reilly, L. Torres and F. Werner (2006). "Techniques for cetacean-habitat modeling." Marine Ecology Progress Series 310: 271-295.

Richardson, W. J., Greene, C.R., C. I. Malme and D. H. Thomson (1998). Marine Mammals and Noise, Elsevier Science.

Rose, G. and W. Legget (1988). "Atmosphere-ocean coupling in the northern Gulf Of St. Lawrence: frequency-dependent wind-forced variatinos in near shore temperatures and currents." Canadian Journal of Fisheries and Aquatic Sciences 45: 1222-1233.

Ross, G. J. B., V. G. Cockcroft and D. S. Butterworth (1987). "Offshore distribution of bottlenosed dolphins in Natal coastal waters and Algoa Bay, Eastern Cape. ." South African Journal of Zoology 22: 50-56.

Rotnitzky, A. and N. P. Jewell (1990). "Hypothesis testing of regression parameters in semiparametric generalized linear models for cluster correlated data." Biometrika 77(3): 485-497.

Santos, M. B. and G. J. Pierce (2003). The diet of harbour porpoise (Phocoena phocoena) in the northeast Atlantic. Oceanogr. Mar. Biol. R. N. G. a. R. J. A. Atkinson. Annu Rev 41: 355-390.

Saucier, F. J., F. Roy, D. Gilbert, P. Pellerin and H. Ritchie (2003). "Modeling the formation and circulation processes of water masses and sea ice in the Gulf of St. Lawrence, Canada." J. Geophys. Res. 108(C8): 3269.

Selzer, L. A. and P. M. Payne (1988). "The distribution of white-sided (lagenorhynchus acutus) and common dolphins (delphinus delphis) vs. environmental features of the continental shelf of the northeastern united states." Marine Mammal Science 4(2): 141-153.

Skogen, M. D., E. Svendsen, J. Berntsen, D. Aksnes and K. B. Ulvestad (1995). "Modelling the primary production in the North Sea using a coupled three-dimensional physical-chemical-biological ocean model." Estuarine, Coastal and Shelf Science 41(5): 545-565.

Skov, H. and F. Thomsen (2008). "Resolving fine-scale spatio-temporal dynamics in the harbour porpoise Phocoena phocoena." Marine Ecology Progress Series 373: 173-186.

Smith, G. J. D. and D. E. Gaskin (1974). "The diet of harbor porpoises (Phocoena phocoena (L.)) in coastal waters of Eastern Canada, with special reference to the Bay of Fundy." Canadian Journal of Zoology 52(6): 777-782.

Smith, G. J. D. and D. E. Gaskin (1983). "An environmental index for habitat utilization by female harbour porpoises with calves near Deer Island, Bay of Fundy. ." Ophelia 22(1): 1-13.

Sourisseau, M., Y. Simard and F. J. Saucier (2006). "Krill aggregation in the St. Lawrence system, and supply of krill to the whale feeding grounds in the estuary from the gulf." Marine Ecology Progress Series 314: 257-270.

Sveegaard, S. (2011). Spatial and temporal distribution of harbour porpoises in relation to their prey. Dep. of Arctic Environment, NERI. National Environmental Research Institute. Denmark, Aarhus University: 128 pp.

Sveegaard, S., J. Teilmann, J. Tougaard, R. Dietz, K. N. Mouritsen, G. Desportes and U. Siebert (2011). "High-density areas for harbor porpoises (Phocoena phocoena) identified by satellite tracking." Marine Mammal Science 27(1): 230-246.

Tanabe, S., T. Mori, R. Tatsukawa and N. Miyazaki (1983). "Global pollution of marine mammals by PCBs, DDTs and HCHs (BHCs)." Chemosphere 12(9-10): 1269-1275.

Trippel, E. A. and T. D. Shepherd (2004). By-catch of harbour porpoise (Phocoena phocoena) in the lower Bay of Fundy gillnet fishery 1998–2001. Can. Tech. Rep. Fish. Aquat. Sci 2521: 33 p.

Van Waerebeek, K., L. Santillán and J. C. Reyes (2002). "An unusually large aggregation of Burmeister's porpoises Phocoena spinipinnis off Peru, with a review of sightings from the eastern South Pacific." Noticiario Mensual 350: 12–17.


33

Wang, J. Y., D. E. Gaskin and B. N. White (1996). "Mitochondrial DNA analysis of the harbour porpoise, (Phocoena phocoena), subpopulations in North American waters." Canadian Journal of Fisheries and Aquatic Sciences 53(7): 1632-1645.

Watts, P. and D. E. Gaskin (1985). "Habitat Index Analysis of the Harbor Porpoise (Phocoena phocoena) in the Southern Coastal Bay of Fundy, Canada." Journal of Mammalogy 66(4): 733-744.

Weir, C. and S. O’Brian (2000). Association of the Harbour Porpoise (Phocoena phocoena) with the Western Irish Sea front. Proceedings of the Fourteenth Annual Conference of the European Cetacean Society, Cork, 2-5 April Ireland. pp:61-65

Yan, J. and J. Fine (2004). "Estimating equations for association structures." Statistics in Medicine 23(6): 859-874.

Yasui, W. Y. and D. E. Gaskin (1986). "Energy budget of a small cetacean: The harbour porpoise Phocoena phocoena." L. Ophelia 25: 183–198.

Zheng, B. (2000). "Summarizing the goodness of fit of generalized linear models for longitudinal data." Statistics in Medicine 19(10): 1265-1275.

Zuur, A. F., E. N. Ieno, N. J. Walker, A. A. Saveliev and G. M. Smith (2009). Mixed effects models and extensions in ecology with R, Springer.

Žydelis, R., R. L. Lewison, S. A. Shaffer, J. E. Moore, A. M. Boustany, J. J. Roberts, M. Sims, D. C. Dunn, B. D. Best, Y. Tremblay, M. A. Kappes, P. N. Halpin, D. P. Costa and L. B. Crowder (2011). "Dynamic habitat models: using telemetry data to project fisheries bycatch." Proceedings of the Royal Society B: Biological Sciences.


34

APPENDIX A - METHODS

Fig. A-1: Reconstructed survey effort tracks across the study period from the entire sightings dataset provided by MICS in order to create the MCP containing the 95 % survey effort .From Doniol-Valcroze (2007, 2008)

Fig. A-2: 3D ocean model nodes in the study area where environmental data for sightings and pseudo-absences were retrieved. From Doniol-Valcroze (2008) and Saucier et al (2003)

EFFORT AND OCEAN MODEL NODES


35

Fig. A-3: Harbour porpoise sightings across the study period (1997-2003 provided by MICS

Fig. A-4: Random points (pseudo-absences) created from harbor porpoise sightings within the MCP. Provided by Doniol-Valcroze.

DATA/ RANDOM POINTS


36

APPENDIX B – RESULTS

Table B-1: Total distances from the reconstructed effort tracks in kilometres surveyed per month and year across the study period (1997-2003)

Year

Month


1997 729.14 1889.51 3386.93 2286.13 655.33 - 8947.04

1998 922.43 1776.66 2530.23 1055.15 699.93 - 6984.4

1999 607.19 1328.81 2923.45 1648.32 148.83 - 6656.6

2000 447.61 3262.54 4532.83 1604.57 176.8 - 10024.35

2001 911.61 2020.81 2915 1761.66 282.26 158.52 8049.86

2002 1316.14 3022.38 3768.36 843.85 197.26 - 9147.99

2003 1009.9 911.2 - - - - 1921.1

Total 5944.02 14211.91 20056.8 9199.68 2160.41 158.52 51731.34

Table B-2. Total number of harbour porpoise sightings per month and year across the survey period (1997-2003)

Year

Month


1997 13 52 415 208 146 - 834

1998 4 135 309 39 29 - 516

1999 22 266 470 203 5 - 966

2000 19 846 1073 339 31 - 2308

2001 47 377 819 504 40 3 1790

2002 46 564 436 170 19 - 1235

2003 45 28 - - - - 73

Total 196 2268 3522 1463 270 3 7722


37

Fig. B-3: Observed annual mean group sizes obtained from the dataset provided by MICS used in this analysis. Note: large aggregations which could infer potential biases (<20 individuals per sighting) were removed for the analysis).

Fig. B-3: Observed monthly mean group sizes obtained from the dataset provided by MICS used in this analysis. Note: large aggregations which could infer potential biases (<20 individuals per sighting) were removed for the analysis).

living among giants: habitat modelling of harbour … · living among giants: habitat modelling of...

Documents