Dynamic habitat suitability modelling reveals rapid poleward distribution shift in a mobile apex ...

37 Pages·2015·2.56 MB·English

ResearchOnline@JCU
This is the Accepted Version of a paper to be published in
the journal Global Change Biology:
Hill, Nicholas J., Tobin, Andrew J., Reside, April E.,
Pepperell, Julian G., and Bridge, Tom C.L. Dynamic
habitat suitability modelling reveals rapid poleward
distribution shift in a mobile apex predator. Global
Change Biology, 22 (3). pp. 1086-1096
http://dx.doi.org/10.1111/gcb.13129
Article type: Primary Research Articles
e
Title: Dynamic habitat suitability modelling reveals rapid poleward distribution shift in
l
a mobile apex predator
c
Running head: Rapid distribution shift in pelagic predator
i
t Nicholas J. Hill1*, Andrew J. Tobin1, April E. Reside2, Julian G. Pepperell3, Tom C.L.
r Bridge4,5
A 1 Centre for Sustainable Tropical Fisheries and Aquaculture, College of Marine and
Environmental Sciences, James Cook University, Townsville, Queensland 4811, Australia
2 Centre for Tropical Environmental & Sustainability Sciences, College of Marine and
d Environmental Sciences, James Cook University, Townsville, Queensland 4811, Australia
3 Pepperell Research and Consulting Pty Ltd., P.O. Box 1475, Noosaville DC, Queensland
e
4566, Australia
t
4ARC Centre of Excellence for Coral Reef Studies, James Cook University, Townsville,
p
Queensland 4811, Australia
5Australian Institute of Marine Science, PMB #3, Townsville MC, Queensland 4810, Australia
e
*Corresponding author email: [email protected]
c
Phone: [phone]
c
A
This article has been accepted for publication and undergone full peer review but has not been
through the copyediting, typesetting, pagination and proofreading process, which may lead to
differences between this version and the Version of Record. Please cite this article as doi:
10.1111/gcb.13129
This article is protected by copyright. All rights reserved.
Keywords: distribution shift, climate change, apex predator, MaxEnt, habitat suitability,
e black marlin (Istiompax indica), tunas & billfishes, species distribution modelling, boundary
current
l
c
Primary research article
i
Abstract
t
r Many taxa are undergoing distribution shifts in response to anthropogenic climate change.
However, detecting a climate signal in mobile species is difficult due to their wide-ranging,
A
patchy distributions, often driven by natural climate variability. For example, difficulties
associated with assessing pelagic fish distributions has rendered fisheries management ill-
equipped to adapt to the challenges posed by climate change, leaving pelagic species and
d ecosystems vulnerable. Here we demonstrate the value of citizen science data for modelling
the dynamic habitat suitability of a mobile pelagic predator (black marlin, Istiopmax indica)
e within the south-west Pacific Ocean. The extensive spatial and temporal coverage of our
occurrence data set (n=18717), collected at high resolution (~1.85km2), enabled identification
t
of suitable habitat at monthly time-steps over a 16-year period (1998-2013). We identified
p considerable monthly, seasonal and inter-annual variability in the extent and distribution of
suitable habitat, predominately driven by chlorophyll-a and sea surface height. Inter-annual
e variability correlated with El Nino Southern Oscillation (ENSO) events, with suitable habitat
extending up to ~300 km further south during La Nina events. Despite the strong influence of
c
ENSO, our model revealed a rapid poleward shift in the geometric mean of black marlin
habitat, occurring at 88.2 km decade-1. By incorporating multiple environmental factors at
c
monthly time-steps, we were able to demonstrate a rapid distribution shift in a mobile pelagic
A
species. Our findings suggest that the rapid velocity of climate change in the south-west
This article is protected by copyright. All rights reserved.
Pacific Ocean is likely affecting mobile pelagic species, indicating that they may be more
e vulnerable to climate change than previously thought.
l
c
Introduction
i
t Rapid and extensive distribution shifts in response to global climate change have been
detected in numerous marine and terrestrial taxa (Parmesan & Yohe, 2003; Chen et al., 2011;
r
Pinsky et al., 2013). These shifts are predicted to continue in coming decades, resulting in a
A
reassembly of current ecological communities (Loarie et al., 2009; Hazen et al., 2012;
Burrows et al., 2014). Although numerous shifts have been predicted, empirical evidence via
retrospective detection has proven elusive for the majority of species due to inadequate spatial
and/or temporal coverage of baseline occurence data (Booth et al., 2011; Hobday & Evans,
d
2013). This has inhibited the development and implememntation of appropriate management
e strategies to assess species vulnerability to climate change (Conroy et al., 2011).
Where species occurrence data is patchy, modelling changes in habitat suitability has proven
t
useful for identifying distribution shifts (Elith et al., 2006, 2010). However, habitat suitability
p
metrics are often focused on temperature (Loarie et al., 2009; Burrows et al., 2014) which can
result in underestimation of the rate and magnitude of distribution shifts expected to occur
e
(VanDerWal et al., 2012). Species distributions are driven by multiple, interacting factors and
therefore, their response to climate change is likely to be more complex and region specific
c
than simple models based solely on temperature suggest (Ackerly et al., 2010; Bell et al.,
c 2013a). Data paucity is particularly prevalent in pelagic ecosystems, which are difficult to
survey at fine spatial and temporal scales (Hobday & Evans, 2013). Increasing atmospheric
A
CO concentrations have caused fundamental changes to the world’s oceans, including
2
increased sea surface temperatures and shifts in oceanic circulation and primary productivity
This article is protected by copyright. All rights reserved.
patterns (Boyce et al., 2010; IPCC, 2014). However, our understanding of the impacts these
e physical changes will have on pelagic species, and the resulting consequences, is poorly
understood.
l
c
Apex predators exert important top-down control of food webs, and are critical for
maintaining ecosystem function (Myers et al., 2007; Estes et al., 2011). Many large-bodied
i
pelagic predators are highly-mobile, and regularly undertake migrations of hundreds to
t
thousands of kilometres at annual and inter-annual timescales (Block et al., 2011). Despite
r
their widespread distributions, mobile pelagic predators have declined in abundance due to
A
overfishing (Baum & Worm, 2009), causing changes to open ocean food webs (Myers et al.,
2007; Worm & Tittensor, 2011). Climate-induced distribution shifts are likely to alter the
functioning of pelagic ecosystems already under pressure from anthropogenic stressors

(Beaugrand et al., 2008; Hazen et al., 2012; Robinson et al., 2014). Furthermore, mobile
d
pelagic predators represent 20% of total economic value in global marine capture fisheries
(FAO, 2012). Distribution shifts in commercially important species will have serious
e
implications for food security and human welfare globally (Brander, 2010; Cheung et al.,
t
2010; Madin et al., 2012). Early detection and characterisation of species responses to climate
p change is therefore vital for bolstering the resilience and adaptive capacity of fisheries,
allowing appropriate management contingencies to be implemented (Hobday & Evans, 2013;
e Holbrook & Johnson, 2014; Maxwell et al., 2015).
Detecting distribution shifts in mobile marine species has proven difficult due to naturally low
c
densities, high mobility and the remoteness of preferred habitats. Acoustic and archival
c
tagging technologies have improved understanding of species movements across space and
time (Schaefer et al., 2014; Maxwell et al., 2015), but are costly and can be logistically
A
difficult to undertake. In the absence of sophisticated tagging data, species distribution
This article is protected by copyright. All rights reserved.
modelling has proven useful for illustrating the dynamic habitat suitability of mobile species,
e and presents itself as a potential tool for dynamic ocean management (Elith et al., 2006;
Lehodey et al., 2006; Maxwell et al., 2015). Habitat suitability of mobile species is
l
determined by a range of environmental (eg: temperature; Boyce et al., 2008) and ecological
c
parameters (eg: prey abundance; Griffiths et al., 2010) which vary across both space (metres –
i
1000’s km) and time (minutes – multi-decadal). Therefore, modelling distribution shifts using
t
environmental data averaged over broad climatic scales (eg: Perry et al., 2005; Montero-Serra
r
et al., 2015), may inaccurately represent habitat suitability for mobile species (Reside et al.,
A 2010). Low spatial resolution data and use of inappropriate modelling techniques has
hindered the detection of underlying long-term shifts in some mobile species (Hobday &
Evans, 2013).

d Utilizing citizen science to assist with the spatial and temporal coverage of data sampling can
help overcome problems associated with data paucity. For example, citizen science has
e proven critical in establishing the projected impacts of climate change on the distribution and
volume of remaining viable habitat for mobile bird species (VanDerWal et al., 2012;
t
Abolafya et al., 2013). In Australia, a large-scale tagging program utilising recreational
p fishermen to collect data on targeted pelagic species has been operating since 1974 in
collaboration with fisheries management (NSW DPI, 2014). Participants apply a conventional
e streamer tag to captured fish before release and record the capture date, release location and
approximate length and weight of each fish. Under the program, recreational anglers have
c
tagged over 419 000 individual fish across 25 different species, providing a spatially and
temporally extensive dataset of the distribution of pelagic fishes (NSW DPI, 2014).
c
Black marlin (Istiopmax indica) are a common target of recreational anglers in many
A
locations throughout the south-west Pacific Ocean, and are regularly recorded in the NSW
This article is protected by copyright. All rights reserved.
DPI Gamefish Tagging Program (NSW DPI, 2014). The distribution of black marlin extends
e throughout tropical and sub-tropical regions of the Pacific and Indian Oceans predominately
found in epipelagic waters (Pepperell, 1990; Domeier & Speare, 2012). Although
l
geographically widespread and capable of crossing ocean basins (Pepperell, 1990), black
c
marlin show a seasonal affinity for continental margins and sea-mounts (Campbell et al.,
i
2003; Gunn et al., 2003), increasing their accessibility to recreational anglers. Recently
t
Williams et al. (2015) identified three distinct genetic populations in the south-west Pacific,
r
eastern Indian Ocean and the south China Sea. In the south-west Pacific population examined
A in this study, spawning occurs from September-November adjacent to the continental shelf of
north-east Australia (Leis et al., 1987; Domeier & Speare, 2012). From September-April,
juveniles 1–4 years old undertake a southerly migration along Australia’s eastern continental

margin (17-34oS) (Pepperell, 1990). Although the south-west Pacific population is targeted by
d
a substantial recreational fishery, little is known about population status (Collette et al., 2011)
or where else the species might occur.
e
Here, we demonstrate the value of citizen science collected data for investigating distribution
t
shifts in the south-west Pacific population of black marlin. We model the dynamic habitat
p suitability of black marlin using high resolution spatial and temporal data to investigate 1)
environmental factors that characterise suitable habitat of black marlin; 2) the variation in
e location of suitable habitat across seasonal and inter-annual timescales in relation to natural
climate oscillations; and 3) whether a long-term distribution shift has occurred and is
c
consistent with the effects of climate change observed in the south-west Pacific Ocean.
c

A

This article is protected by copyright. All rights reserved.
Materials and methods
e
Study region
l
The study was conducted in the south-west Pacific Ocean (3-39oS/142-180oE; Fig. 1a). The
c
study area encompasses the ‘core’ range of the south-west Pacific black marlin population
i
identified using genetic analysis, archival tagging and historical commercial catch data
t
(Williams et al., 1994; Domeier & Speare, 2012; Williams et al., 2015). The dominant
r
oceanographic feature of this region is the East Australian Current (EAC), a poleward-
A flowing western boundary current that transports warm, oligotrophic waters along Australia’s
east coast (Ridgway, 2007). The EAC originates from the westward-flowing South Equatorial
Current (SEC), which bifurcates at the Australian continental margin at 17-19oS (Brinkman et

al., 2001). A seasonal strengthening in the EAC occurs from September-April (Luick et al.,
d
2007). The EAC departs from the east coast at 32-34oS, flowing east towards New Caledonia
and New Zealand, forming the Tasman Front (Baird et al., 2008; Suthers et al., 2011). The
e
Tasman Front is a transition zone, representing the collision of cold (Tasman Sea) and warm
water (Coral Sea) bodies, often exhibiting strong thermal gradients of >2oC (Baird et al.,
t
2008). Over winter (April-August), subantarctic cold waters push north, forcing the EAC and
p
Tasman Front to retreat towards the equator.
e These major oceanographic features vary across numerous temporal scales, attributed to the
influence of climate oscillations. El Nino Southern Oscillation is the dominant force, which
c
drives strong inter-annual variability in the oceanography of the south-west Pacific Ocean
(Holbrook et al., 2009). El Nino events are characterised by anomalously high sea surface
c
temperatures adjacent to Australia’s south-east coast and anomalously cool temperatures in
A
the north as the West Pacific Warm Pool disperses east. In contrast, during La Nina events
trade winds strengthen, forcing West Pacific Warm Pool water to remain in the south-west
This article is protected by copyright. All rights reserved.
Pacific (Holbrook & Bindoff, 1997; Holbrook et al., 2009). Decadal variability is also
e present, characterised by extended warm, El Nino like conditions or conversely cold, La Nina
like conditions. Combination of these various climatic influences create highly variable
l
oceanographic conditions within the south-west Pacific Ocean. Globally, western boundary
c
currents such as the EAC are warming 2-3 times faster than the global mean (Wu et al., 2012;
i
Hobday & Pecl, 2013; Hu et al., 2015). Despite this substantial natural variability in
t
oceanographic features, a long term incessant poleward shift in these features has been
r
recorded (Cai et al., 2006; Ridgway, 2007) and is predicted to continue (Cai et al., 2005;
A Ridgway & Hill, 2012), with subsequent shifts in numerous marine taxa documented
(Frusher et al., 2014; Verges et al., 2014).
Occurrence data
d
Occurrence records were obtained from black marlin tagged by recreational anglers in the
New South Wales Department of Primary Industries Tagging Program (NSWDPI, 2014). The
e
database covers the period from 1974-present. However, in this study we only used a subset
from the period 1998-2013. This period was chosen due to the availability of high-resolution
t
environmental data, and also to account for the increased fishing effort present in the database
p
in the 1970s and 1980s. Occurrence data was binned into monthly time-steps, leaving a total
of 24344 occurrence records available at a spatial resolution of one minute of
e
latitude/longitude (~1.85km2) (Fig. 1b). Although the distribution of tag records is
predominately restricted to within close proximity of coastlines, the use of this data enabled
c
analysis at a much finer spatial scale than would be possible using commercial catch per-unit
c effort data, which is often collected at a resolution of 5x5o (~556.63km2) (eg: Su et al., 2011).
A

This article is protected by copyright. All rights reserved.
Environmental data
e
Environmental factors (Table S1) were chosen as potential explanatory variables based on
l
their availability and demonstrated influence on pelagic species distributions (Su et al., 2011).
c
Spatial layers for environmental factors were acquired using the Marine Geospatial Ecology
Tool (MGET) (Roberts et al., 2010) in ArcGIS. Daily measurements for each factor were
i
t averaged at monthly time-steps (n=192) from 1998-2013 to reduce the influence local
r minima/maxima and no-data cells (e.g. due to cloud cover). No-data cells were interpolated
using the del2a method within MGET which uses laplacian interpolation (D’errico, 2005). All
A
environmental layers were resampled to a common spatial resolution (4 km2) to satisfy
requirements of MaxEnt. Mixed layer depth and bathymetry were also investigated, but were
not included in the final model as they led to over-fitting of the data.
d
Species distribution model
e We used the species distribution modelling algorithm MaxEnt, which estimates the
probability distribution of a species occurrence based on constraints from biologically
t
relevant environmental factors (Phillips et al., 2006). MaxEnt is a robust technique that
p performs well against similar methods when modelling non-systematically collected
occurrence data even if sample sizes are small (Elith et al., 2006). This technique accurately
e models species distributions, despite data limitations and biases (Elith et al., 2006; Pearson et
al., 2007). MaxEnt generates a continuous layer of habitat suitability (ranging from 0-1)
c
across a specified domain by distinguishing the distribution of a species occurrence from the
available surrounding environment (Phillips et al., 2006; Elith et al., 2011).
c
We generated a background data set to representatively sample the environment surrounding
A
each occurrence record. MaxEnt assumes the occurrence data was randomly sampled, with all
This article is protected by copyright. All rights reserved.