Content uploaded by Simon David Hoyle
Author content
All content in this area was uploaded by Simon David Hoyle on Oct 26, 2016
Content may be subject to copyright.
CoP17 Inf. 63 – p. 1
Original language: English CoP17 Inf. 63
(English only / Únicamente en inglés / Seulement en anglais)
CONVENTION ON INTERNATIONAL TRADE IN ENDANGERED SPECIES
OF WILD FAUNA AND FLORA
____________________
Seventeenth meeting of the Conference of the Parties
Johannesburg (South Africa), 24 September – 5 October 2016
PACIFIC-WIDE SUSTAINABILITY RISK ASSESSMENT
OF BIGEYE THRESHER SHARK (ALOPIAS SUPERCILIOSUS)
This document has been submitted by the Secretariat on behalf of the ABNJ Tuna Project (Common
Oceans)*, in relation to agenda items 88 on Proposals to amend Appendices I and II.
* The geographical designations employed in this document do not imply the expression of any opinion whatsoever on the part of the
CITES Secretariat (or the United Nations Environment Programme) concerning the legal status of any country, territory, or area, or
concerning the delimitation of its frontiers or boundaries. The responsibility for the contents of the document rests exclusively with its
author.
Pacific-wide sustainability risk assessment of bigeye thresher shark (Alopias superciliosus)
Pacific-wide sustainability risk
assessment of bigeye thresher
shark (Alopias superciliosus)
Final Report
Prepared for Western and Central Pacific Fisheries Commission
September 2016
Pacific-wide sustainability risk assessment of bigeye thresher shark (Alopias superciliosus)
Prepared by:
Dan Fu1, Marie-Julie Roux1, Shelley Clarke2, Malcolm Francis1, Alistair Dunn1 and Simon Hoyle3
1 National Institute of Water and Atmospheric Research, Wellington, New Zealand
2 ABNJ Tuna Project (Common Oceans), Western and Central Pacific Fisheries Commission, Pohnpei,
Federated States of Micronesia
3 National Institute of Water and Atmospheric Research, Nelson, New Zealand
For any information regarding this report please contact:
Rosemary Hurst
Chief Scientist
Fisheries
+64-4-386 0867
rosemary.hurst@niwa.co.nz
National Institute of Water & Atmospheric Research Ltd
Private Bag 14901
Kilbirnie
Wellington 6241
Phone +64 4 386 0300
NIWA CLIENT REPORT No: 2016089WN
Report date: September 2016
NIWA Project: WCP16301
Quality Assurance Statement
Reviewed by: Dr. Rosemary Hurst
Formatting checked by: Chloe Hauraki
Approved for release by: Dr. Rosemary Hurst
Pacific-wide sustainability risk assessment of bigeye thresher shark (Alopias superciliosus)
Contents
EXECUTIVE SUMMARY ......................................................................................................... 1
LIST OF ACRONYMS ............................................................................................................. 3
1 INTRODUCTION ......................................................................................................... 4
2 DATASETS .................................................................................................................. 8
2.1 CES longline logsheet (commercial effort) data ....................................................... 8
2.2 SPC observer data ..................................................................................................... 9
2.3 US observer data ..................................................................................................... 12
2.4 Japanese observer data .......................................................................................... 14
2.5 Composite dataset .................................................................................................. 15
3 APPROACH AND METHODS ...................................................................................... 18
3.1 Analytical approach ................................................................................................ 18
3.2 Spatial and temporal domains of the assessment .................................................. 19
3.3 Catch groups and fishery groups definition ............................................................ 20
3.4 Species distribution estimation .............................................................................. 23
3.5 Catchability estimation ........................................................................................... 26
3.6 Impact estimation (fishing mortality) ..................................................................... 31
3.7 Population productivity and MIST estimation ........................................................ 32
3.8 Sustainability risk calculations ................................................................................ 34
4 ASSESSMENT RESULTS ............................................................................................. 34
4.1 Species distribution................................................................................................. 34
4.2 Catchability ............................................................................................................. 38
4.3 Fishing impacts ....................................................................................................... 45
4.4 Sustainability risk .................................................................................................... 50
5 DISCUSSION AND RECOMMENDATIONS ................................................................... 56
5.1 Fishery groups ......................................................................................................... 57
5.2 Species distribution................................................................................................. 58
5.3 Catchability ............................................................................................................. 59
5.4 Post-capture survival .............................................................................................. 61
5.5 Maximum impact sustainable threshold (MIST) ..................................................... 62
Pacific-wide sustainability risk assessment of bigeye thresher shark (Alopias superciliosus)
5.6 Sustainability risk .................................................................................................... 63
5.7 Recommendations for future developments and implementations ...................... 64
6 ACKNOWLEDGMENTS .............................................................................................. 65
7 REFERENCES ............................................................................................................ 66
Appendix A – CPUEΩ estimation and derivation ................................................................. 71
Appendix B – BDM description .......................................................................................... 72
Appendix C – q adjustment by fishery groups and spatial scaling ........................................ 74
Appendix D – Observer data characterisation ..................................................................... 75
Appendix E – Spatial standardisations ................................................................................ 77
Appendix F – q calibration Results and sensitivities ............................................................ 85
Appendix G – Catch history ................................................................................................ 90
Appendix H – Year effects standardisations ........................................................................ 91
Appendix I - Impact sensitivity ........................................................................................... 95
Appendix J - Supporting information .................................................................................. 98
Pacific-wide sustainability risk assessment of bigeye thresher shark 1
EXECUTIVE SUMMARY
The bigeye thresher shark, Alopias superciliosus, has been identified as one of the least productive
pelagic sharks and there is concern about its conservation status. Although it is one of three thresher
sharks designated by Western and Central Pacific Fisheries Commission as key shark species, no
Pacific Ocean stock assessment has been conducted. Information gaps and changes in reporting and
observer coverage over time and space, make traditional approaches to stock assessment
impractical. As an alternative and to gain new insights into the sustainability status of bigeye
thresher shark, this study applies a spatially explicit and quantitative sustainability risk assessment to
available data. The analytical framework evaluates sustainability risk as the ratio of current impacts
from fisheries (spatially-explicit and cumulative fishing mortality F) to a maximum impact sustainable
threshold (MIST) reference point based on population productivity. This approach differs from
traditional stock assessment because it evaluates F in terms of whether the population’s ability to
withstand fishing pressure is exceeded, rather than evaluating biomass (B) and whether the
population is overfished.
Key components (and analytical procedures) included: 1) estimation of the species distribution or
relative abundance in space; 2) calibration of population and fishery groups catchability; and 3)
estimation of the maximum intrinsic population growth rate r for the species, using available life
history data. The first two components were used in conjunction with commercial effort (logsheet)
data to quantify fishing impact. The third was used to define the MIST reference point. A scenario-
based approach to sustainability risk evaluation was implemented, with scenarios ranging from more
to less precautionary and representing different species distribution, initial population status,
maximum density and post-capture survival assumptions. This approach served to cope with
currently high levels of uncertainty in population status, movements and biology, and limited
information about some aspects of the available datasets.
Observer data from the Pacific Community (SPC), United States (US) and Japan were standardized
with two models, a zero-inflated negative binomial (ZINB) model and a geo-statistical delta-
generalised linear mixed (delta-GLMM) model, which permitted derivation of spatial indices of
relative abundance over different but overlapping areas. Population catchability (q) was statistically
calibrated using a Bayesian state-space biomass dynamics model (BDM) fitted to time series of
relative abundance and annual catch estimates obtained from a representative subset of the
observer data. This approach assumed that although the available data were insufficient to estimate
absolute catchability, they could be used to calibrate a relative catchability parameter for use in
spatially-explicit impact estimation. A range of plausible q values were estimated, with uncertainty,
and adjusted spatially by fishing season and catch group (i.e., ‘fishery groups’), as well as for the
occurrence of post-capture survival. Fishing mortality was calculated as the sum product of total
effort and fishery-group specific catchability in 5x5 degree cells, weighted by the relative density of
bigeye thresher shark in each cell, as obtained from the spatial standardization.
The distribution of the maximum population growth rate r had a median value of 0.03, which is
higher than previously reported for the species, and was used to define the MIST. Analyses
performed assuming 100% capture mortality produced median F values ranging from 0.02 to 0.04
among base case scenarios for the period 2000-2014. Sustainability risk, corresponding to the ratio
of total impact to the MIST, ranged from 0.6 to 1.2. The average probability that fishing impact
exceeded the MIST was 0.4 across years and scenarios. Analyses performed assuming a range of
post-capture survival rates produced median F values ranging from 0.01 to 0.03 and median
sustainability risk between 0.4 and 1.0, with an average probability of 0.20 of total fishing impact
exceeding the MIST.
2 Pacific-wide sustainability risk assessment of bigeye thresher shark
Earlier studies indicated that the species is vulnerable to exploitation owing to limited productivity,
even at relatively low levels of fishing mortality. Sustainability risk results presented here, which
incorporate considerable uncertainty both within and among scenarios, are not inconsistent with
this view. They suggest that total impacts from pelagic longline fisheries in the Pacific since 2000 are
generally low (<5%), but have exceeded the maximum impact sustainable threshold for bigeye
thresher in some years.
Risk outcomes were sensitive to q calibration assumptions used in the Biomass Dynamic Model
(BDM), namely values of the prior bounds for the unfished biomass at equilibrium (K), initial stock
status (biomass in the first year of the model relative to K), and process error inclusion. The
implications of such assumptions and sensitivities are discussed in the report, along with potential
means of refining impact estimation in future work. Better information on initial stock status,
biomass at unfished equilibrium and post-capture survival assumptions, would serve to weight
alternative scenarios and improve the accuracy of sustainability risk estimation.
The strengths and value of a spatially-explicit, sustainability risk assessment framework reside in
data integration from multiple sources and the ability to map relative fishing impact and
sustainability risk spatially and among fishery sectors, with uncertainty.
Pacific-wide sustainability risk assessment of bigeye thresher shark 3
LIST OF ACRONYMS
ABNJ Areas Beyond National Jurisdiction (or Common Oceans)
AFFRC Agriculture, Forestry and Fisheries Research Council, Japan
ALB Albacore Tuna (Thunnus alalunga)
BET Bigeye Tuna (Thunnus obesus)
BTH Bigeye Thresher Shark (Alopias superciliosus)
CES Tuna Fishery Catch and Effort Query System
HBF number of hooks between floats
IATTC Inter-American Tropical Tuna Commission
ICCAT International Convention for the Conservation of Atlantic Tunas
IOTC Indian Ocean Tuna Commission
JP Japan
MIST Maximum impact sustainable threshold
MLS Striped marlin (Kajikia audax)
NOAA National Oceanographic and Atmospheric Administration
ROP Regional Observer Program
SPC The Pacific Community
SST Sea surface temperature
SWO Broadbill Swordfish (Xiphias gladius)
TCSB Tuna Project Technical Coordinator Sharks and Bycatch
TUBS Tuna Fisheries Observer System
US United States
YFT Yellowfin Tuna (Thunnus albacares)
WCPFC Western and Central Pacific Fisheries Commission
WCPO Western and Central Pacific Ocean
4 Pacific-wide sustainability risk assessment of bigeye thresher shark
1 INTRODUCTION
The Western and Central Pacific Fisheries Commission (WCPFC) is one of five tuna Regional Fisheries
Management Organizations (t-RFMOs) responsible for the sustainable use, conservation and
management of highly migratory species taken by tuna fisheries. Unlike some of the other t-RFMOs, the
WCPFC has explicit responsibility for assessing and managing not only tuna species, but also dependent
and associated species under Articles 5(d) and 10.1(c) of its Convention. Recognition by the WCPFC of
sharks as dependent and associated species in need of conservation and management has resulted in a
list of thirteen shark species found in the Western and Central Pacific Ocean (WCPO) for which both data
provision and assessment are required (WCPFC 2012). The three thresher shark species of the family
Alopiidae (Alopias superciliosus, bigeye thresher; A. pelagicus, pelagic thresher; and A. vulpinus,
common thresher) have been included in this list since its original formulation in 2008. Thus far, the
WCPFC has conducted stock assessments for three of the shark species on the key shark list: oceanic
whitetip shark (Carcharhinus longimanus), silky shark (Carcharhinus falciformis) and North Pacific blue
shark (Prionace glauca) (Rice & Harley 2012, 2013; Rice et al. 2014). A stock assessment for South Pacific
blue shark is currently underway.
Indicator analyses for the thresher sharks were conducted by the WCPFC’s Scientific Services Provider,
the Pacific Community (SPC), in 2011 and 2015 (Clarke et al. 2011, Rice et al. 2015). In both cases, most
of the analyses were performed at the family level due to presence of a substantial number of non-
species specific observer records. The most recent of these analyses hinted at a declining index of
abundance for the thresher group as a whole based on decreased catch rates in 2012-2014 and an
overall decline since 2003 (Rice et al. 2015). On this basis, the WCPFC Scientific Committee in August
2015 recognized assessment of thresher sharks as a priority.
The WCPFC, along with the four t-RFMOs, is a partner in the Areas Beyond National Jurisdiction (ABNJ) –
also referred to as Common Oceans – Tuna Project (www.commonoceans.org). The objective of the
ABNJ Tuna Project is to achieve efficient and sustainable management of fisheries resources and
biodiversity conservation in marine areas that do not fall under the responsibility of any one country.
One set of activities of the GEF-funded ABNJ Tuna Project aims at reducing the impact of tuna fisheries
on biodiversity by improving data and assessment methods for sharks and thereby promoting their
sustainable management. Within this set of activities WCPFC has committed to leading four new stock
status assessment studies for Pacific-wide shark stocks. The bigeye thresher shark was identified as the
thresher species with the widest distribution and the greatest number of catch records from the WCPO
(Matsunaga and Yokawa 2013, Rice et al. 2015), and it is likely to be the most vulnerable of the three
threshers to longline fishing (WCPFC 2006, IOTC 2012, ICCAT 2015), so it was chosen as the best
candidate for assessment. A bigeye thresher shark stock status assessment meets the criteria for ABNJ
funding as this species has a Pacific-wide distribution, was identified as a priority assessment by at least
one of the t-RFMOs, and provides an opportunity to further develop methods for data-poor species.
Biology and distribution
In the Pacific, the bigeye thresher shark primarily occurs in tropical waters, however its habitat ranges as
far north as central Japan and Baja California and as far south as the North Island of New Zealand and
the southern coast of Peru (Matsunaga & Yokawa 2013). This species is found near the surface at night
and makes deep dives to experience temperatures of 6-11oC (up to 500 m depth) during the day,
Pacific-wide sustainability risk assessment of bigeye thresher shark 5
perhaps aided by its rete mirabile, a structure within the orbital sinus believed to help stabilize brain and
eye temperatures (Nakano et al. 2003, Weng & Block 2004). Studies from the Atlantic suggest that
juveniles concentrate primarily in the tropical North Atlantic, and pregnant females are found at higher
latitudes off West Africa and Brazil (Fernandez-Carvalho et al. 2015). Findings from the Pacific suggest a
slightly different pattern: neonates and juveniles are clustered near 10oN and S latitude, with pregnant
females either also at 10oN or at higher latitudes (20-30oN) to the northeast. Few pregnant females have
been found south of the equator in the Pacific (Matsunaga & Yokawa 2013).
There is limited information from which to draw any conclusions regarding stock structure for any of the
thresher shark species. One unpublished study indicated no population structure in bigeye threshers
across what it considered to be the Indo-Pacific (samples from California, Gulf of California, Ecuador,
Hawaii, Taiwan and South Africa). However, the sample size was small (n=64) and it used only one type
of DNA (mitochondrial control region) (Trejo 2005). Tagging studies of bigeye thresher sharks off Hawaii
have reported movements in both northwesterly and easterly directions with a maximum linear
displacement of nearly 3,500 km over 240 days (Weng & Block 2004, Musyl et al. 2011).
The bigeye thresher shark is characterized by high juvenile survival and year-round reproduction (i.e.
there is no fixed mating or birthing season), but its low fecundity causes it to have low productivity
compared to other pelagic sharks and to be highly vulnerable to fisheries which that catch juveniles of
this species. In the Pacific, age at maturity was estimated at 12.3-13.4 years for females and 9-10 years
for males. The litter size is 2 pups per cycle with a 1:1 sex ratio and the reproductive cycle duration is
unknown (Clarke et al. 2015). In a recent ecological risk assessment conducted for pelagic sharks caught
by Atlantic longline tuna fisheries, the bigeye thresher was found to have the lowest intrinsic rate of
increase (0.009, confidence interval 0.001-0.018), in other words to be the least productive, of the 16
species considered (ICCAT 2012).
Review of population trends
As introduced above, standardized catch rate indicators for Alopias spp. have been produced from SPC
data holdings twice under the WCPFC’s Shark Research Plan (Clarke et al. 2011a, Rice et al. 2015).
Japanese longline logbook and research and training vessel data catch rate series for threshers as a
group were also produced in the earlier round of analysis (Clarke et al. 2011b)1. In the 2011 analyses, no
strong trends in standardized catch rates were found for thresher sharks analysed as a group, although
the Japanese research and training vessel data indicated a slight increase in catch rates in the central
Pacific from the early 2000s through 2008 (the last available data point; Clarke et al. 2011a,b). The Rice
et al. (2015) update study, analyzing data through 2014 but excluding data from the US observer
programmes, noted that most catches were observed in the longline fishery in an area from 10oS to
20oN and east of 170oE, and the majority of observed individuals were immature. Catch rates rose from
1995-2001 but decreased slightly from 2003-2011 before falling more sharply in 2012-2014. That study
thus concluded that the thresher shark complex appeared to be declining though it was noted that the
last data point was based on relatively few data and may have exaggerated the trend in the final year
(Rice et al. 2015).
1 Note that while the Japanese research and training vessel data recorded the three thresher species separately, the
Japanese logbook data do not, and so for the sake of comparison between the two Japanese datasets, as well between the
Japanese datasets and the SPC datasets, threshers were analysed as a group.
6 Pacific-wide sustainability risk assessment of bigeye thresher shark
All three studies also examined trends in median size as a potential measure of fishing pressure. The first
SPC analysis considered threshers as a group and found statistically significant decreasing median sizes
in the central Pacific (Clarke et al. 2011a). The analysis of Japanese research and training vessel data
found declines in median size only for pelagic threshers and no trend for bigeye threshers (Clarke et al.
2011b) which suggests that the trends identified by Clarke et al. (2011a) may have been driven by
pelagic thresher shark. The Rice et al. (2015) update study noted that thresher sharks as a group showed
relatively stable size trends based on a sample of mostly immature females and immature and mature
males in the central Pacific (Rice et al. 2015).
The only consistent catch rate time series specific to bigeye thresher shark prior to the current study
was an analysis by the United States National Oceanic and Atmospheric Administration (NOAA) in
support of a decision regarding whether to list bigeye thresher sharks on the United States Endangered
Species Act. The analysis standardized catch rates based on the extensive Hawaii-based longline
observer data for 1995-2014. The catch rate in the final year of the series (2014) was nearly double that
of the previous year and was the highest on record. As a result, NOAA conducted a sensitivity test by
excluding the 2014 data point but concluded that the influence of the 2014 data point was negligible
and that abundance was relatively stable (Young et al. 2016).
At present there are no known stock status assessments for the bigeye thresher shark in any ocean, but
two studies of pelagic thresher in Taiwanese waters concluded that the stock was slightly over-exploited
(Liu et al. 2006, Tsai et al. 2010). NOAA also recently completed a stock assessment for the common
thresher shark (Alopias vulpinus) based primarily on data from California and Mexico. That assessment
found that fishing mortality for this primarily coastal stock was relatively low (0.08), well below the
overfishing threshold, and the stock was at 94% of its unexploited level and so substantially larger than
the minimum stock size threshold. Therefore, the assessment concluded that the common thresher
shark was unlikely to be in an overfished condition nor to be experiencing overfishing (Teo et al. 2016).
Finally, there have been a number of studies of thresher sharks in the Atlantic Ocean in recent years, but
most analyses have been conducted for Alopias species, i.e. at the family level. In this region, the most
consistent, comprehensive data sources are logbook and observer records from the United States’
longline fishery in the northwest Atlantic. Selecting the observer data as the more reliable dataset,
Young et al. (2016) re-analysed the time series from 1992-2013 for bigeye thresher shark per se. They
found no obvious change in the population trend over time and thus concluded that the northwest
Atlantic population had stabilized. One older analysis from the southwest Atlantic, quoted in Amorim et
al. (2009), indicated increasing catch rates from 1971-1989 and a gradual decrease from 1990-2001.
However, the authors noted that during this period a change in the depth of fishing operations also
occurred and this may have affected the time series (Amorim et al. 2009). There are no known available
catch rate time series for bigeye thresher sharks from the Indian Ocean.
Current conservation and management designations and measures
The IUCN Red List classifies all three thresher species as “Vulnerable” (IUCN 2015). The Red List
assessment for the bigeye thresher shark dates from 2007 and is supplemented by regional assessments
of “Vulnerable” in the eastern central Pacific, “Endangered” in the northwest and western central
Atlantic, “Near Threatened” in the southwest Atlantic, “Data Deficient” in the Mediterranean Sea; and
“Vulnerable” in the Indo-West Pacific (Amorim et al. 2009).
Pacific-wide sustainability risk assessment of bigeye thresher shark 7
Two of the five t-RFMOs have adopted conservation and management measures which pertain to bigeye
thresher sharks. In 2009, ICCAT adopted a measure requiring all members to prohibit retention of bigeye
thresher sharks with the exception of Mexican small-scale coastal fisheries with catches of less than 110
fish (ICCAT Resolution 09-07). IOTC’s measure requires all members to prohibit retention of all species of
thresher shark (IOTC Resolution 13/06). In addition to these species-specific measures, starting with
ICCAT in 2004 (Recommendation 04-10), and followed by IATTC (Resolution C-05-03) and IOTC
(Resolution 05/05) in 2005, WCPFC in 2006 (CMM 2006-05) and CCSBT in 2008, all of the t-RFMOs have
adopted a 5% fins-to-carcass ratio as a means of controlling shark finning for all species including
thresher sharks (Clarke et al. 2014a).
All three species of thresher sharks were listed on Appendix II of the Convention on the Conservation of
Migratory Species of Wild Animals (CMS) in November 2014. CMS Appendix II listing encourages
international cooperation towards conservation of shared species. Subsequently, the three thresher
species were added to the Convention on Migratory Species (CMS) Memorandum of Understanding
(MOU) for Sharks in February 2016. The function of the MOU is to develop a Conservation Plan to guide
cooperation between the signatories to CMS Convention as well as other interested stakeholders.
A proposal to list the bigeye thresher shark, along with the pelagic and common threshers as look-alike
species, on Appendix II of the Convention on International Trade in Endangered Species of Wild Flora
and Fauna (CITES) was first posted on 2 May 2016 and revised on 1 June 2016. The proponents for the
proposal include Sri Lanka, the Bahamas, Bangladesh, Benin, Brazil, Burkina Faso, the Comoros, the
Dominican Republic, Egypt, the European Union, Fiji, Gabon, Ghana, Guinea, Guinea-Bissau, Kenya, the
Maldives, Mauritania, Palau, Panama, Samoa, Senegal, Seychelles and Ukraine. The proposal will be
considered at the 17th Conference of the Parties (COP) in Johannesburg, South Africa from 24
September-05 October 2016. If listed, all exports of thresher sharks, including landings in non-flag State
ports will require permits to be issued by the flag State CITES Management Authority. Export permits are
contingent upon legal acquisition and non-detriment findings (NDFs), the latter of which represents a
certification by an authorized CITES Scientific Authority that the proposed export is not detrimental to
the survival of the species (Clarke et al. 2014b).
Sustainability status evaluation
This report presents the preliminary results of a Pacific-wide, spatially-explicit sustainability risk
assessment of bigeye thresher shark. Risk assessment tools have been developed in response to data
limitation problems in the evaluation of fishing effects on non-target species, including sharks and other
elasmobranch species (Stobutski et al. 2002, Griffiths et al. 2006, Braccini et al. 2006, Zhou & Griffiths
2008, Cortés et al. 2010, Gallagher et al. 2012). Recent applications have used semi-quantitative
approaches (namely productivity-susceptibility analysis) and demographic methods to estimate
population productivity, without quantifying total impacts from fisheries or fishing-induced mortality.
Such risk assessments applied to pelagic sharks caught in Atlantic pelagic longline fisheries identified
bigeye thresher as one of the most vulnerable species to exploitation (Cortés et al. 2008, 2010, 2012).
Herein, we develop and apply a quantitative framework for estimating spatially-explicit fishing mortality
and derive a sustainability status for the species as the ratio of total impact to a maximum impact
sustainable threshold (MIST) reference point. Rather than following a traditional stock assessment
approach, which relies heavily on population processes that for sharks are often poorly understood, this
spatially-explicit approach is based on species productivity, inferred distribution and data on the
8 Pacific-wide sustainability risk assessment of bigeye thresher shark
occurrence, characteristics and intensity of fishing. The quantitative framework allows uncertainty to be
quantified and propagated throughout the assessment process. An important outcome is that impact,
sustainability risk and uncertainty can be partitioned spatially and among fishery sectors, allowing more
focused management.
2 DATASETS
Review of the potential sources of catch, effort and size data for bigeye thresher in the Pacific identified
the following as key data sets:
Non-public domain longline catch and effort data for the entire Pacific maintained in the SPC CES
database and accessible to the ABNJ TCSB via the WCPFC Secretariat (“CES longline logsheet data”);
Non-public domain longline observer data maintained by SPC as part of the ROP and on behalf of
Australia, the Cook Islands, the Federated States of Micronesia, Fiji, French Polynesia, the Republic
of the Marshall Islands, New Caledonia, New Zealand, Samoa, Solomon Islands, Tonga and Vanuatu
and accessible to the ABNJ TCSB through data confidentiality agreements with each country for use
in the ABNJ Tuna Project (“SPC observer data”);
Non-public domain United States longline observer data provided directly to the ABNJ TCSB for use
in the ABNJ Tuna Project under a data confidentiality agreement (“US observer data”);
Non-public domain Japan longline observer data provided to the ABNJ TCSB and to NIWA under a
data confidentiality agreement specific to this BTH assessment (“Japan observer data”).
Each of these datasets is described separately below. Data confidentiality agreements necessary to
obtain access to the data required for this study have precluded the provision of the majority of datasets
described in this report to NIWA. As a result, the ABNJ (Common Oceans) Tuna Project Technical
Coordinator-Sharks and Bycatch (ABNJ TCSB) has taken on the role of data manager and has served as
an intermediary between NIWA and the raw datasets.
2.1 CES longline logsheet (commercial effort) data
The data were downloaded by the ABNJ TCSB from CES on 11 March 2016 and again on 14 April 2016 as
there was an update to the data by SPC. The downloaded data consisted of 269,702 records aggregated
by year (1950-2014), month (1-12), flag2, and 5 degree latitude by 5 degree longitude (5x5) cell (ranges:
-82.5 to 62.5 latitude; 7.5 to 362.5 longitude). The coordinates for each grid represent the southwest
corner of each 5x5 cell. Catch data were provided for albacore (ALB), bigeye (BET), Pacific bluefin,
skipjack (SKJ), southern bluefin, and yellowfin tunas (YFT); black, blue and striped marlin; Indo-Pacific
sailfish; shortbilled spearfish; broadbill swordfish (SWO); blue, “mako”, silky, oceanic whitetip,
“thresher” and “other” sharks; and “other”.
Annual effort totalled 1.3-1.4 billion hooks in 2011-2013, with lower effort recorded for 2014 likely as a
result of incomplete reporting at the time of writing (Figure 1). Overall trends in effort and target
species catch in the WCPO longline fishery through 2014 were reviewed by Williams & Terawasi (2015).
2 Flags (countries and fishing entities) include AU, BZ, CK, CN, ES, FJ, FM, GU, ID, JP, KI, KR, MH, NC, NU, NZ, PF, PG, PH, PT, PW,
SB, SN, TO, TV, TW, US, VN, VU and WS (see http://www.nationsonline.org/oneworld/country_code_list.htm for code and
country name matching)
Pacific-wide sustainability risk assessment of bigeye thresher shark 9
Catch was downloaded in number of sharks as that is the unit used in the observer datasets and is likely
to be more accurate than weight-based measures. The total number of “thresher” sharks in the dataset
was 129,933 with an annual high of 28,991 in 2014 (data for 2015 were likely incomplete at the time of
writing). The first “thresher” shark to be recorded on a logsheet was by Papua New Guinea in 1997;
other flags’ first reporting was in 1998 (Samoa), 2000 (US), 2002 (Fiji), 2006 (Spain), 2007 (Australia and
New Zealand), 2008 (Japan and Taiwan), 2010 (Korea and New Caledonia), 2011 (Cook Islands), 2013
(FSM and Vanuatu), 2014 (Kiribati) and 2015 (China). These dates probably reflect the year in which the
logsheets first provided a space for recording thresher sharks rather than the actual first encounter of a
thresher shark by each flag’s fishing vessels.
The CES longline logsheet data were aggregated by year, month, 5x5 cell and flag to obtain the total
effort in hooks fished per strata.
Figure 1: Total longline effort for the Pacific Ocean, 1995-2014 as downloaded from the SPC Catch Effort Query
System (CES as of April 2016).
2.2 SPC observer data
These data were downloaded by the ABNJ TCSB on 3 March 2016 through a special TUBS interface for
SPC and WCPFC Secretariat staff. Some issues with large files sizes were encountered which prevented
remote downloading of all necessary files at that time; the remaining large data files were received on 8
March 2016. Downloaded data consisted of two files for each fleet and year: one file that contained set-
level information with one row per set and one file that contained catch records for individual sharks
with one row per shark or ray caught.
Length data were provided in some datasets (i.e. SPC and Japan data), but were not formatted for use3.
Length data can be used to distinguish life stages of the species, potentially allowing for fishing impacts
to be evaluated for different life stage groups, but this requires further development of the
3 Length data presumably exist in the US observer programme data but were not included in the extract provided by the US for this study.
10 Pacific-wide sustainability risk assessment of bigeye thresher shark
methodology in this assessment which has not been undertaken. Fate and condition data were provided
and used to distinguish between BTH which were and were not alive upon release. This was
accomplished by first removing all BTH which were recorded as unknown either at landing or upon
release. Then those with fate codes beginning with R (retained) or DFR (discarded, fins retained), or
condition codes A3 (alive but dying) or D (dead) were considered dead and all others were considered to
be alive at release. These data could be used to examine the trend in the post-capture survival. Data on
BTH sex exist in the SPC observer dataset (Clarke et al. 2011, Rice et al. 2015) but were not included in
the subset of data downloadable through the TUBS interface.
To link each catch record to its set characteristics, a unique identifier was created by combining set
identifiers and trip identifiers in the set database. At this step, 522 set records shared identifiers with
another set. As it was impossible to know which, if any, of these set records were correct, all 522 were
removed. From the remaining number of sets (n=41,048), containing 3,388 BTH, the following number
of sets (and BTH records) were removed sequentially:
Removed due to missing lat/long information (1,947 sets and 180 BTH);
Removed due to not being within the year range 1995-2014 (4,791 sets and 51 BTH);
Removed due to missing hooks fished values (715 sets and no BTH);
Removed due to missing hooks between floats (68 sets and no BTH);
Removed due to too many or too few hooks (965 sets and 34 BTH);
Removed due to too many or too few hooks between baskets (220 sets and 7 BTH); and
Removed due to being outside the spatial boundaries of the assessment (4,226 sets and 4 BTH) (see
Section 3.1 for the spatial range criteria applied).
Removals related to missing values (hooks between floats, latitude, longitude and number of hooks
fished) were necessary because these values are likely to be very important in the standardizations and
missing values may interfere with coefficient estimation. Extreme values of hooks fished (i.e. <500 or
>4000) were considered to represent abnormal fishing operations and were also thus removed.
Similarly, sets recording fewer than four, or more than 45 hooks between baskets were considered
dubious and were removed. Finally, sets before 1995 (the year when the SPC regional observer program
began in earnest) were removed due to expected poorer data quality in the initial years, and sets after
2014 were removed to avoid biases associated with incomplete reporting.
Pacific-wide sustainability risk assessment of bigeye thresher shark 11
A number of other filters applied or discussed in Rice et al. (2015) were considered but not applied as
follows:
sets from fisheries known to be targeting sharks (e.g. Papua New Guinea) and those sets for which
the set header field target_shk_yn=yes (Table 3), were not removed a priori as it was considered
that any shark targeting effect could be addressed through the catch rate standardization;
removing sets from small national observer programs with < 100 sets each was not considered
necessary as this analysis will not be using the observer program identifier in lieu of actual (lat/long)
location;
removing records considered to be outside the sea surface temperature (SST) range of species was
not done due to doubts about the certainty of bigeye thresher species’ SST range and a preference
to address habitat issues through a lat/long exclusion criterion; and
removing records where the catch rate of BTH was greater than the 97.5th percentile of nominal
mean CPUE for the dataset as a whole was not done because BTH may exhibit schooling behaviour
and thus we might expect to see rare large catches.
In total 12,932 sets were removed from the analysis, containing 276 BTH, leaving 28,116 sets and 3,112
BTH4. The annual number of sets observed and number of BTH caught per year in the SPC observer
dataset are shown in Table 1.
4 There were n=2,001 sets with 183 BTH that had date or time errors (missing values, or Haul Start before Set Start) but these
were retained pending a decision about whether time of day, soak time, hours of set during night, or other time-related
variables would be used in the standardization model.
12 Pacific-wide sustainability risk assessment of bigeye thresher shark
Table 1: Summary of BTH catch and effort information by year available in the SPC observer dataset.
Year
Sets
BTH Catch
Records
1995
469
3
1996
485
4
1997
621
9
1998
581
38
1999
456
39
2000
507
61
2001
634
62
2002
1 576
136
2003
1 536
87
2004
1 428
86
2005
1 834
247
2006
2 497
876
2007
1 960
698
2008
1 540
111
2009
1 581
150
2010
1 284
23
2011
1 346
63
2012
1 566
187
2013
3 328
131
2014
2 887
101
The SPC observer dataset is distributed with low coverage over a wide area from 1993-2015. Detailed
analysis of thresher shark data in the SPC observer set was conducted by Clarke et al. (2011) and Rice et
al. (2015) but it should be noted that most of those analyses were conducted for Alopias spp (see
section 1). The spatial distribution of the SPC observer dataset is shown in Figure 2.
2.3 US observer data
Data from the US longline observer programme were prepared by NOAA on 11 March 2015 and sent by
post to the ABNJ TCSB in the Federated States of Micronesia. When the ABNJ TCSB began using the data
for this study in March 2016 it was discovered that all Hawaiian longline fleet data for 2002 were missing
from the provided dataset. Upon request, the missing 2002 data were provided by NOAA via a secure
download facility on 24 March 2016. Table 2 shows the number of sets observed, total catch, and the
number of BTH caught per year, for the observed sets in the Hawaii-permitted and American Samoa-
permitted longline fleets.
Pacific-wide sustainability risk assessment of bigeye thresher shark 13
Table 2: Number of sets observed, total number of fish (etc.) caught, and BTH caught by year in the observed
sets of the two fleets covered by the US observer programme and used in this study.
Year
Sets
Total Catch
Records
BTH Catch
Records
Sets
Total Catch
Records
BTH Catch
Records
Hawaii-permitted Longline Fishery
American Samoa-permitted Longline
Fishery
1995
519
26,422
75
0
0
0
1996
587
28,560
208
0
0
0
1997
443
30,507
140
0
0
0
1998
556
31,511
229
0
0
0
1999
421
24,794
83
0
0
0
2000
1,370
69,393
399
0
0
0
2001
2,699
132,214
692
0
0
0
2002
3,296
152,505
1,271
0
0
0
2003
3,078
160,255
765
0
0
0
2004
3,855
186,788
1,789
0
0
0
2005
5,829
274,322
1,158
0
0
0
2006
4,120
180,912
1,521
235
27,100
20
2007
4,762
223,752
1,293
327
40,497
19
2008
4,968
226,722
1,075
266
29,254
19
2009
4,683
199,899
1,660
237
26,167
24
2010
4,958
246,262
1,381
890
100,052
61
2011
4,572
236,003
1,319
1,017
90,357
67
2012
4,639
224,117
1,708
592
57,427
28
2013
4,389
262,919
1,645
584
44,863
49
2014
4,857
279,463
3,828
515
40,115
43
Total
64,601
3,197,320
22,239
4,663
455,832
330
Length and sex data may exist in the US observer dataset but were not included in the subset provided
for this study. Regarding fate and condition classification, the US observer programme only records
shark condition at retrieval as alive or dead, and at release as alive, dead or kept. This simplified
distinguishing between BTH which did and did not survive until release.
As for the SPC observer data, a number of filters were considered to clean and format the US observer
data (see section 2.3). Of these, six filters were applied with the following results:
Removed due to missing lat/long information (9 sets and 1 BTH);
Removed due to missing hooks fished values (6 sets and no BTH);
Removed due to missing hooks between floats (22 sets and 8 BTH);
Removed due to too many or too few hooks (293 sets and 17 BTH);
Removed due to too many or too few hooks between baskets (186 sets and 9 BTH); and
Removed due to being outside the spatial boundaries of the assessment (551 sets and 11 BTH) (see
section 3.1 for the spatial range criteria applied).
In total 1,067 sets were removed from the analysis, containing 46 BTH, leaving 69,264 sets and 22,523
BTH.
14 Pacific-wide sustainability risk assessment of bigeye thresher shark
The US observer dataset is a rich source of BTH data with considerably more records for this species
than the SPC dataset (22,523 BTH in 69,264 sets versus 3,112 BTH in 28,116 sets). The spatial
distribution of the US observer dataset is shown in Figure 2.
2.4 Japanese observer data
Japan’s longline observer program has been operating since 2007 but has only been fully implemented
since 2011. A data confidentiality agreement was negotiated between the Japan Fisheries Agency, NIWA
and the ABNJ (Common Oceans) Tuna Project on 24 March 2016. Data were provided using a secure
internet file sharing system on the same day and re-provided on 25 March 2016 to correct minor
formatting errors. The number of sets observed, total number of thresher sharks caught and the number
of BTH caught per year for the observed Japanese longline sets as received are shown in Table 3.
Table 3: Number of sets observed, total number of threshers caught, and BTH caught by year in the observed
sets of the Japanese longline fleet as provided by Japan. Note that Japan did not provide catch records for
species other than thresher sharks (bigeye, pelagic, common and unknown).
Year
Sets
Total Catch of Threshers
Catch of BTH
2007
13
4
4
2008
143
27
20
2009
89
4
2
2010
162
183
28
2011
638
275
152
2012
908
357
57
2013
1,756
972
376
2014
1,877
788
513
2015
1,371
355
171
Total
6,957
2965
1323
Length data were provided for 949 BTH and sex data for 939 BTH. These data have not yet been
formatted for use. Fate and condition data were not provided.
Filters were considered and applied as for the other observer data (see section 2.3). Of these, six filters
were applied with the following results:
Removed due to missing lat/long information (317 sets and 28 BTH);
Removed due to missing hooks fished values (1 set and 3 BTH);
Removed due to missing hooks between floats (33 sets and 20 BTH);
Removed due to being outside the spatial boundaries of the assessment (218 sets and 6 BTH) (see
section 3.1 for the spatial range criteria applied).
In total 569 sets were removed from the analysis, containing 57 BTH, leaving 6,405 sets and 1,266 BTH.
The Japan observer dataset contains 1,266 BTH from 6,405 sets. The number of BTH per set in the Japan
observer dataset (0.20) is intermediate between that of the SPC observer dataset (0.11) and the US
observer dataset (0.33). The spatial distribution of the Japanese observer dataset is shown in Figure 2.
Pacific-wide sustainability risk assessment of bigeye thresher shark 15
2.5 Composite dataset
A composite dataset composed of the SPC, US and Japanese observer data consisting of 104,320 sets
and 26,917 BTH was compiled on 25 March 2016. The distribution of BTH captures by 5x5 grid and
source dataset is shown in Figure 2. The annual observed effort and annual observed catch by source
dataset are shown in Figure 3.
Fields such as the number of hooks between floats (HBF), bait_type, hook_type and wire_trace that
were recorded for some sets were retained for analyses. HBF was used as a proxy for the fishing depth
of pelagic longline sets. Information on the time of set start and hauling start was used to estimate
fishing duration at night (number of hours fishing in dark conditions) for each set. This was done by
relating the reported setting and hauling times with the expected sunrise and sunset times at each
location and date.
A standardised measure of SST was assigned to each set, corresponding to monthly SSTs averaged over
2x2 degree cells from 1995 to 2014, available from NOAA Extended Reconstructed Sea Surface
Temperature (ERSST) database (http://www1.ncdc.noaa.gov/pub/data/cmb/ersst/v4/netcdf/). Other,
finer scale datasets were sought but could not be accessed in a workable format within the timeframe of
this study.
16 Pacific-wide sustainability risk assessment of bigeye thresher shark
Figure 2: Distribution of bigeye thresher caught in observed sets in the Pacific, 1995-2014. Catches from the SPC dataset are in blue, the US dataset in red
and the Japanese dataset in black. The size of the circle is proportional to log(catch) as shown in the legend. The grey-shaded portion in each grid square
represents the proportion of sets with positive catches of BTH. Catches from grids where fewer than three vessels caught BTH are not shown. The numbers
in parentheses are numbers of BTH caught.
Pacific-wide sustainability risk assessment of bigeye thresher shark 17
Figure 3: Total observed effort (in million hooks) by data source (top panel) and total number of BTH observed
by data source (bottom panel), 1995-2014.
18 Pacific-wide sustainability risk assessment of bigeye thresher shark
3 APPROACH AND METHODS
3.1 Analytical approach
The analytical framework is risk-based and spatially-explicit. Sustainability status
S
is assessed
relative to current impacts from fisheries (or relative fishing mortality F) and a maximum impact
sustainable threshold (MIST) limit reference point (LRP):
LRP
F
MIST
Impact ≈=S
Uncertainty in all parameters is quantified and propagated through the assessment framework. In
this context, sustainability risk
R
is the probability
p
, given the uncertainty, that the total impact
exceeds the MIST:
MIST]Impact p[ >=R
The assessment is conducted over a spatial grid of 5 by 5 degree latitude and longitude cells (section
3.2). Fishing impact is estimated as the average of fishing mortality
i
F
weighted by species relative
abundance
i
n
in each cell:
∑
∑
=
ii
iii
n
nF
Impact
Cell-specific
i
F
is calculated as the product of fishing effort
E
and catchability
q
distinguished
among (and summed across) fishery groups
j
:
∑
=
jjjii qEF ,
where
j
q
expresses the fraction of the total population in each cell that is available for capture by
each unit of effort, adjusted for capture efficiency in fishery group
j
.
Effort differentiation into fishery groups serves to handle the effects of different fishing operations
and operational practices on total impact. Impacts are assumed to be cumulative across fishery
groups and over the spatial domain of the assessment. As a result, sustainability risk, fishing impact
and uncertainty can be disaggregated in space and among fishery sectors.
MIST is the sustainable reference threshold for the species. The MIST is defined based on population
productivity inferred from life history data. Life history parameters are used to estimate a maximum
intrinsic population growth rate r, with uncertainty. In turn, r is used to derive sustainable impact
thresholds similar to the fishing mortality-based sustainability reference points (Fcrash, Fmsm, F lim )
described by Zhou et al. (2011).
The assessment is implemented in a flexible framework allowing incremental improvements and
fine-tuning as data are augmented and/or better information becomes available.
Pacific-wide sustainability risk assessment of bigeye thresher shark 19
A summary of data inputs, analytical methods and key parameters is presented in Figure 4. Details
on all components are presented in the following sections.
Figure 4: Conceptual representation of data inputs, analytical methods and key parameters used in Pacific-
wide spatially-explicit sustainability assessment of bigeye thresher shark. BDM = Bayesian state-space
biomass dynamics model. The dashed outline box represents analytical methods applied to an area subset
of the available data.
3.2 Spatial and temporal domains of the assessment
The spatial domain of the assessment was defined as the region between 38°N and 42°S latitude and
120°E and 70°W (290°E on map) longitude. The latitudinal range is based on published information
on the geographic distribution of bigeye thresher in the Pacific Ocean (Compagno 2001, Matsunaga
& Yokawa 2013). The longitudinal range is arbitrarily defined, with the eastern limit set to
encompass the full eastern extent of the Pacific (i.e., area offshore of the boundary between Peru
and Chile) and the western limit set near the Makassar Strait between Borneo and Sulawesi.
The assessment is conducted over a spatial grid of 5 by 5 degree latitude and longitude cells,
corresponding to the spatial resolution of the catch and effort data available for assessment. Three
area subsets were distinguished for analyses within the spatial domain of the assessment (Figure 5):
1) Assessment Area - corresponding to all grid cells in which at least one specimen of bigeye
thresher was caught between 2000 and 2014 (n=219 cells);
2) Core Area – corresponding to those grid cells that together contributed 95% of bigeye
thresher captures between 2000 and 2014 (n=62 cells).
3) Calibration Area – subset of grid cells from the Core Area (above) corresponding to the area
covered by the US Hawaii observer data (n=33 cells).
20 Pacific-wide sustainability risk assessment of bigeye thresher shark
Figure 5: Spatial domain of the assessment as defined in 5x5 degrees of latitude and longitude grid cells,
showing the three area subsets considered for analyses: Assessment Area (cells with numbers); Core Area
(shaded grey cells), and Calibration Area (cells with red borders). Cell numbers were assigned sequentially
from west to east and from south to north and are used to identify each cell in the datasets.
The timeframe of the assessment was set to include all commercial effort (logsheet) and observer
data from 1995 to 2014 in preliminary analyses. The start of this period corresponds to the full-scale
implementation of the SPC and US observer programmes. Species distribution was estimated using
the composite observer dataset including data from 2000 to 2014 (section 3.4). The start year of
2000 reflects the small amount of observer data in previous years (Figure 3). The catchability (q)
parameter calibration was performed using observer and commercial effort data for the period
1995-2014. A longer time period was considered in this process to better inform the catch series and
abundance index required by the calibration. Impact was estimated using the total commercial
pelagic longline fishing effort from the last fifteen years (2000-2014).
3.3 Catch groups and fishery groups definition
Fishery groups were defined as the combination of catch groups and fishing season (Jan-Mar, Apr-
Jun, Jul-Sep and Oct-Dec). Catch groups were determined by performing clustering analyses on
logsheet data using the “k-means” algorithm (see Hoyle et al. (2015) for details). Logsheet data
(rather than observer data) were used as they contain complete and reliable information on catch
composition by species for the main target species.
Catch data for albacore tuna, southern bluefin and yellowfin tuna, bigeye tuna, broadbill swordfish
and striped marlin were clustered over two periods (1995
−
2004 and 2005
−
2014) to account for
potential changes in fishing operations over time. Clustering was conducted on species composition
aggregated by year, month and 5x5 degree cell strata. The optimal number of clusters was
determined based on the maximum reduction of mean square error (Figure 6).
For both time periods, the analyses produced four clusters corresponding to a predominance of BET,
ALB, YFT or SWO in the catch, as well as an additional cluster (‘others’) in which none of the main
Pacific-wide sustainability risk assessment of bigeye thresher shark 21
five target species (above) were caught. The five clusters were used to distinguish catch groups in
the assessment.
Commercial effort (logsheet) data were categorised into fishery groups for impact estimation using
catch group and fishing season information. Each group is assumed to represent different
operational characteristics of the effort, as this is likely to affect capture efficiency for bigeye
thresher.
Figure 6: Diagnostics from kmeans cluster analysis showing the optimal number of targeting strategies based
on the species composition of the longline catch for 1995-2004 (left) and 2005-2015 (right).
For spatial and temporal standardisations, each observer set was assigned to a catch group defined
using the aggregated logsheet data. The catch group (which may or may not represent actual
targeting strategies) was assigned based on set location (5x5 grid cell) and time of year (year/month)
information, under the assumption that fishing activities predominantly catching ALB, BET, YFT and
SWO would be separated in space and/or time. This approach of assigning catch groups was used
because information on targeting strategies among observer programmes is inconsistent and often
unreliable. Operational characteristics of the effort reported in the observer data (e.g., HBF) can be
used to infer targeting strategies however, to ensure consistency among datasets and to avoid
double counting of information, these variables were separately included in standardization
procedures, along with the catch groups inferred based on logsheet data.
Variations in the number of hooks between floats (HBF) and fishing duration at night among catch
groups are shown in Figure 7. Sets predominantly catching BET generally fished deeper (HBFs mostly
ranging between 20 and 30) and operated during daylight hours, right before sunset. Sets mainly
catching SWO were mostly shallow and fished during the night. Other catch groups (YFT and ALB)
covered a broad range of HBF values (with some differences among datasets) and mainly fished
during daylight hours.
Agreement between catch groups inferred from cluster analyses and recorded target species was
assessed using the Japanese observer data (not including SBT effort). Recorded target species in the
Japanese observer data are believed to be representative of targeting strategies (Y. Semba, AFFRC,
pers. comm.). Proportions of matching sets (i.e. agreement between inferred catch groups vs
recorded target species) were 62% for ALB (inferred catch group), 59% for YFT, 94% for BET and only
5% for SWO.
22 Pacific-wide sustainability risk assessment of bigeye thresher shark
Figure 7: Total observed effort (no. of sets) as related to (a) hooks between floats (HBF) and (b) nighttime
fishing duration (hours in darkness) among catch groups (YFT, ALB, BET and SWO) and observer datasets
(US, JP and SPC) in the Assessment Area, 2000-2014.
(a)
(b)
Pacific-wide sustainability risk assessment of bigeye thresher shark 23
3.4 Species distribution estimation
3.4.1 Approach and input data
Standardisation analyses performed on observer catch and effort data were used to infer the spatial
distribution of bigeye thresher. The composite observer dataset for the period 2000-2014 was used.
Data from 1995-1999 were excluded owing to comparatively limited spatial and numerical coverage.
Two standardisation models were applied for comparison: a zero-inflated negative binomial model
(ZINB) (Zuur et al. 2009) and a geo-statistical delta-generalised linear mixed model (delta-GLMM)
(Thorson et al. 2015). Both were used to standardise catch rates of bigeye thresher in 5x5 degree
cells. The standardised catch rates or relative densities are assumed to be representative of spatial
abundance distribution for the species.
Data from all observed longline sets were included in the spatial standardisations (i.e., no
representative ‘fishery subset’ was defined for the species). Outputs from both models as well as
strengths and limitations are compared and discussed in the context of spatially-explicit
sustainability risk assessment for pelagic shark species.
3.4.2 ZINB standardisation
ZINB models serve to handle overdispersed count data with excessive number of zeros (Zuur et al.
2009). The relationship between the response variable (in this case, the number of bigeye thresher
caught per set) and a set of explanatory variables is modelled as a mixture of an encounter
probability (binomial process) and a negative binomial count process (that allows for overdispersion
and zero occurrences).
The estimation of spatial effects in each grid cell requires a large number of coefficients to be
estimated. To reduce the number of parameters and improve estimation, the fitting of the ZINB
model was restricted to observer catch and effort data from the Core Area (Figure 5, section 3.2).
This was required to ensure successful model convergence. Likewise, convergence problems caused
by the estimation of a large number of coefficients precluded the inclusion of vessel effects in the
ZINB model. The implication of this is that the abundance outside the Core Area is assumed to be
very low so that its contribution to the overall fishing impact on the whole population is negligible.
Explanatory variables considered in spatial standardisations are listed in Table 4. A number of
variables including bait_type, hook_type, wire_trace, sst and night_fishing were included in
preliminary analyses but excluded from the final models due to missing or ambiguous values
(wire_trace and hook_type); too many values (too many coefficients to be estimated and no clear
basis for grouping) (bait_type); confounding effects with other covariates (night_fishing) and
dubious relationships to the response variable (sst). Other variables were offered sequentially,
producing a series of nested models. The same sets of variables were offered simultaneously to both
the zero and count components of ZINB models. Likelihood ratio tests (performed using function
lrtest in R package lmtest (R core development team 2016)) were used to assess the effect of each
additional variable on model fit and explanatory power. Alternative models were also compared
using AIC (Akaike Information Criterion).
24 Pacific-wide sustainability risk assessment of bigeye thresher shark
Table 4: Summary of explanatory variables offered to ZINB models for spatial standardisation of catch rates
of bigeye thresher in observed pelagic longline fisheries in the Pacific Ocean. Continuous variables were
modelled as natural splines with 3 degrees of freedom.
Variable
Type
Description
year
Categorical
Calendar year (2000-2014)
cell
Categorical
5x5 degree grid cells in the Core Area
month
Continuous
Calendar month (1-12)
catch group
Categorical
Species catch composition
log(effort)
Offset
No. of hooks per set
HBF
Continuous
Hooks between floats
bait_type
Categorical
Types of bait used
hook_type
Categorical
Types of hooks used
wire_trace
Categorical
Presence/Absence (retention effect)
night_fishing
Continuous
Fishing duration at night (hours)
SST
Continuous
Sea surface temperature
Spatial indices of relative abundance were derived as the predicted catch rate (no. of bigeye
thresher caught per 1000 hooks) for each grid cell in the Core Area, with other covariates fixed to a
reference value corresponding to the coefficient calculated for the intercept term (categorical
variables) or the median observed value multiplied by the coefficient (continuous variables).
Model fit was assessed using a number of diagnostics plots, including observed versus fitted catch
rates, plots of Pearson residuals versus fitted values and Pearson residuals by year and grid cell.
3.4.3 Delta-GLMM standardisation
The delta-GLMM model developed by Thorson et al (2015) allows for extrapolation to nearby cells
(i.e., density estimation in cells with no observations) by assuming spatially correlated spatial
variation. Similar to the ZINB, the delta-GLMM includes a binomial process that models the
probability of encounter (i.e., proportion of sets that catch bigeye thresher) and a count process
(positive catch rates) that follows a gamma distribution. Additional complexity relates to the
integration and differentiation of fixed and random effects.
Random spatial variation and spatiotemporal variation are approximated using Gaussian Markov
random fields over a number of ‘knots’. The location of each knot is determined by applying the k-
means clustering algorithm to the positional information in the available data (i.e., latitude and
longitude data from all sets converted to eastings and northings). This results in a distribution of
‘knots’ with density proportional to sampling intensity (or in this case, fishing intensity as related to
observer coverage). The knots define the model’s ‘predictive framework’ and allow for piecewise-
constant random fields approximation. This approach has a number of computational advantages
and assumes that density at any location is equal to the density value estimated at the nearest knot.
The number of knots can be specified within the model framework, allowing control over the
accuracy of random effects estimation. This can also be used to achieve a balance of accuracy and
computational speed (Thorson et al. 2015). Both the encounter probability and catch process are
modelled using a link function and a combination of linear predictors including the random fields.
Fixed effects are estimated using maximum marginal likelihood (approximated using the Laplace
approximation), while integrating across all random effects. The model is implemented in template
model builder (Kristensen et al. 2014).
Pacific-wide sustainability risk assessment of bigeye thresher shark 25
For application to bigeye thresher, year was included as a fixed effect and vessel was included as a
random effect in all models. Other variables considered and included as potential linear predictors
were fishery groups, HBF and month (see Table 4, section 3.4.2 for details). The number of knots was
fixed at 1000 in all runs. The estimation of spatial abundance indices (number of bigeye thresher
caught per 1000 hooks) involved a two-step process: 1) fine-scale extrapolation; and 2) density
estimation at the spatial scale of the assessment (5x5 degree cells).
The Assessment Area (Figure 5, section 3.2) was subdivided into a fine-scale (10x10 km cells)
extrapolation grid. Density extrapolation was restricted to cells with observations (i.e., in which
there was a recorded longline set start position) and to cells with no observations but a recorded
longline set start position within a maximum distance of 50 km. The resulting predictive framework
was composed of 296 045 square grids of 100 km2 each and an extrapolation layer of 1000 knots.
Relative abundance at the scale of 5x5 degree cells was calculated as the average density estimated
in 10x10km cells in the predictive framework. Three separate delta-GLMM models were fitted and
compared: 1) a spatial model (assuming constant spatial variation over time); 2) a spatiotemporal
model (allowing spatial variation to differ among years); and 3) a core vessels model (like the spatial
model in 1) but including only vessels that caught at least one specimen of bigeye thresher).
Spatial correlation was assessed using geometric anisotropy plots. Estimated vessel effects on
encounter probability and positive catch rates were plotted (with 95% confidence intervals) and
differentiated among contributing observer datasets.
3.4.4 Uncertainty estimation
Uncertainty in species distribution inferred from the final ZINB model was estimated using a
bootstrap (resampling) procedure that resampled data from all sets within each grid cell (with
replacement) and refitted the standardisation model to predict spatial indices (300 iterations).
Uncertainty in species distribution inferred from the delta-GLMM model was reported as the
marginal standard deviations estimated for the spatial effects and spatiotemporal effects on
encounter probabilities and positive catch rates. Details on the computation of marginal standard
deviation for random fields are available in Thorson et al. (2015). However, uncertainty estimation
and summarization for the delta-GLMM model still require further research (Thorson et al. 2015).
Additional complications also arise when extrapolating spatial effects to obtain spatial indices on 5x5
cells. For these reasons, uncertainty for the spatial indices inferred from the delta-GLMM model is
not formally quantified.
3.4.5 Key assumptions
The estimation of a species distribution layer using available data from observed fishing events
assumes that the aggregated data from observer programmes from 2000 to 2014 are representative
of the species distribution in the Pacific. The estimated spatial distribution for bigeye thresher is
assumed to have remained constant over the timeframe of the assessment (2000-2014; see Section
5.2 for discussion of this assumption).
The delta-GLMM model applied in this study was designed to estimate population abundance from
survey (fishery-independent) data and area-swept by trawl gear. Its application to estimate spatial
indices of abundance for bigeye thresher using fishery-dependent catch and effort data from pelagic
longlines assumes that all observed longline sets have a comparable area of impact. Constant gear-
affected area has been assumed in the catchability studies for passive fishing methods including
longline by Zhou et al. (2014).
26 Pacific-wide sustainability risk assessment of bigeye thresher shark
3.5 Catchability estimation
3.5.1 Approach and input data
The approach to catchability estimation was developed based on the assumption that the available
data were insufficient to estimate absolute catchability, but could be used to calibrate a relative
catchability parameter for use in relative impact estimation. Plausible values for the population
catchability scalar
q
were derived in a calibration exercise using available life history information for
bigeye thresher and a representative subset of the observer data within a subsection of the
Assessment Area (the Calibration Area (AΩ) - see Figure 5, section 3.2). The rationale for using the
Calibration Area is that this data subset (US Hawaii longline fishery) is likely to provide more credible
estimates of catch history and standardised CPUE index which are required for catchability
calibration. The Calibration Area accounted for 82% of all captures in the observer datasets and is
assumed to be representative of population dynamics for the species.
The calibration fits a Bayesian state-space biomass dynamics model (BDM, Edwards 2016) to an
index of relative abundance with year effects (CPUEΩ) (section 3.5.3) and a catch series (CΩ) (section
3.5.2) (Figure 4). The model assumes a uniform prior on log(K) (the biomass at unexploited
equilibrium), with prior bounds defined based on expert knowledge on bigeye thresher maximal
density in hot spot areas (and a range of sensitivities based on blue shark Prionace glauca
assessment values) (section 3.5.4); and an informed prior on r (the maximum intrinsic population
growth rate) estimated using life history data (section 3.7). The BDM estimates a distribution of
posterior samples for q, which is taken to represent the range of plausible values of
q
Ω for the
species in the Calibration Area, with uncertainty. This catchability scalar
q
Ω is then adjusted by
fishery group (catch group and fishing seasons) and scaled to the spatial resolution (5x5 degree cells)
used to estimate fishing impact in the assessment.
3.5.2 Catch history
A catch history (CΩ) for bigeye thresher in the Calibration Area AΩ was constructed by scaling the
number of observed captures by the ratio of total effort to total observed effort. Data from all
observer sets in the Calibration Area for the period 1995-2014 and commercial effort (logsheet) data
aggregated in 5x5 degree cells for the period 1952-2014 (which covers the time span of extracted
logsheet data), were used.
Catch estimation was stratified by year, by year and fishery group, or by year and season (Jan−Mar,
Apr−Jun, Jul-Sep, and Oct−Dec). The number of observed captures was multiplied by the ratio of the
total number of hooks (logsheet data) and the number of observed hooks within each stratum,
summed over all strata to obtain the annual catch from 1995 to 2014. Historical (pre-1995) catches
were calculated by scaling the average observed catch for the period 1995−2014, by the ratio of the
annual (logsheet) effort to the average annual observed effort (1995-2014) in each year from 1952 to
1994. The catch history calculated for the pre-1995 period is highly uncertain and is provided only as
an indication (i.e., only the 1995-2014 catch history is included in BDM runs for
q
Ω calibration).
3.5.3 Abundance index
Year effects standardisations of observer catch and effort (CPUE) data were used to estimate annual
indices of relative abundance (CPUEΩ) for bigeye thresher in the Calibration Area AΩ.
Standardisations were performed by fitting a ZINB model to the US Hawaii observer data in AΩ from
1995 to 2014. These data accounted for the majority (82%) of observed BTH captures in the
composite observer dataset (see sections 2.3 and 2.5) and provided a relatively long and spatially
consistent time series of catch and effort information over a region with generally high observer
Pacific-wide sustainability risk assessment of bigeye thresher shark 27
coverage (10% or higher since 2000). Pre-2000 data were included to estimate a more informative
index of abundance for the BDM process, but were characterized by comparatively limited observer
coverage.
Explanatory variables included in year effects standardisations were month, HBF, catch group, effort
(log no. of hooks) and subarea. Variables were offered sequentially and nested models were
compared using the likelihood ratio test and AIC.
Subarea was included to account for spatial effects on a coarser scale than the 5x5 degree cells used
to estimate species relative densities (section 3.4) and fishing impact (section 3.6). This was done to
ensure that spatial effects on annual indices of relative abundance are estimated at a scale that
reflects differences in fishing intensity (as opposed to an arbitrarily defined geometric grid). The data
were partitioned into 12 knots (subareas) by applying the k-means clustering algorithm (similar to
that used in the geostatistical delta-GLMM model (see section 3.4.3)) The k-means clustering
algorithm was applied to the positional information in the data from all sets in the calibration Area
(i.e., latitude and longitude data from all sets converted to eastings and northings) to determine the
location of each subarea. The number of knots was based on the maximum reduction of mean
square error from the clustering (as shown in section 3.3).
Because the aim of this analysis was to derive an annual CPUE index for use in
q
Ω calibration, re-
scaling was required to ensure that CPUE indices reflect average catch rates of BTH in the Calibration
Area (as opposed to within a specific subarea) (Appendix A). To this end, the following procedure
was carried out:
The annual CPUE index for a “reference” subarea was predicted using the ZINB standardization
model (fitted with subarea covariate) and fixing the value of all covariates (intercept term for
categorical variables including subarea and median value for continuous variables). A ‘non-spatial’
model (ZINB model without spatial effects) was fitted to estimate the effort-weighted average
annual CPUE over all subareas (Appendix A). Annual indices predicted by the reference ZINB model
for the reference subarea were then scaled to have the same mean as the annual CPUE predicted by
the ‘non-spatial’ model:
∑
∑−
Ω
=
y
y
ref
y
yspatialnon
y
ref
y
CPUE
CPUE
CPUECPUE
Where
is the index for the Calibration Area (AΩ) in year y;
is the CPUE
index from the non-spatial model in year y; and
is the annual CPUE index from the
reference ZINB model (fitted with subarea covariate).
Sensitivity testing of year effects standardisation was performed by fitting a number of geostatistical
delta-GLMM models (n=4) and a delta lognormal model to the same dataset and using the same
explanatory variables as the final ZINB model.
3.5.4 BDM calibration
The index of relative abundance (CPUEΩ) (section 3.5.3) and catch history (CΩ) (section 3.5.2) for
bigeye thresher in the Calibration Area AΩ were inputted into the BDM to estimate a range of
plausible values for
Ω
q
.
A detailed description of the BDM model is presented in Appendix B. The model describes changes in
biomass in response to a particular harvest regime and according to the generalised (hybrid)
28 Pacific-wide sustainability risk assessment of bigeye thresher shark
production function described by McAllister et al. (2000). The catchability scalar relates the
abundance index and estimated biomass trajectory and is calculated as a set of most likely values
relative to the values of other parameters, assuming a uniform prior on the natural scale.
For
q
calibration runs, the shape parameter value is arbitrarily fixed at 0.4 (
K4.0=
ϕ
) and the
observation error (
o
σ
) was fixed at 0.2. BDMs were fitted to the catch history (CΩ) and abundance
index (CPUE Ω) for bigeye thresher in the Calibration Area, and to an informed prior on the maximum
intrinsic population growth rate r for the species (lognormal with mean 0.03 and standard deviation
0.02) (section 3.7) (see Figure 4 for conceptual representation).
The population was unlikely to be in an unfished equilibrium state at the start of our time series in
1995 (i.e., initial depletion or initial stock status relative to the unfished biomass
u
<1). Because the
initial depletion state could not be estimated by the model, a set of values
u
were randomly sampled
from three normal distributions with means 0.3 (low initial status), 0.5 (medium initial status) and
0.7 (high initial status) and a standard deviation of 0.05. Each was sampled 300 times, for a total
sample of 900
u
values ranging from 0.15 to 0.84 (Figure 8). A BDM was fitted to each
u
to obtain
1000 posterior samples of
q
(total 3000 samples across the three u assumptions). The combined
samples constitute the plausible range of
q
across the three initial stock status scenarios.
A uniform prior was assumed for K (unfished biomass at equilibrium) in log space (which serves to
give lower probabilities to higher K values). Prior bounds were defined to constrain the estimates to
biologically plausible values. The lower bound for K was set at 30 000 sharks, based on the estimated
maximum annual catch in the Calibration Area. The upper bound was determined based on expert
knowledge and advice on plausible values of bigeye thresher density in fishery hotspots. The base
case scenario used an upper bound of 2 million sharks, which is the biomass (K) value corresponding
to a ≤ 5% chance of encountering more than one shark per km2 in hotspots cells (i.e., cells 436, 437,
and 438 - see Figure 11). This value was derived based on the combined surface area of the three
cells (approximately 900 000 km2) and the fact that approximately 50% of the estimated density for
the Calibration Area was located in those cells (see spatial standardisation results, section 4.1). As
sensitivities, alternative values for the upper bound were considered based on estimates of K from
the most recent assessment of blue shark in the North Pacific region (WCPFC 2014), however
recognising that blue shark is a more productive species with potentially higher abundance.
Alternative upper bound values were derived using two approaches: (1) by applying the ratio of the
upper to lower bound of K for blue shark (= 200) to the maximum observed catch of bigeye thresher
in the Calibration Area (30 000 sharks) (= 6 million upper bound); and (2) by using the best estimate
of K for blue sharks in the North Pacific region (i.e., average of the median estimates from the four
main reference models = 1.04 million tonnes (see Table 8 of WCPFC (2014)), corresponding to
approximately 37 million sharks, assuming an average weight of 28 kg for blue shark caught in the
longline fishery (Nakano 1994)). This number was scaled to the Calibration Area for bigeye thresher
(using the results of the spatial standardisation in Section 3.4), to yield an upper bound of 16 million
bigeye thresher.
This multiple scenario approach served to constrain the uncertainty in the estimated parameters,
and ensure that q estimates derived from the calibration process reflect a range of biologically
plausible values. This was required as the lack of contrast in the abundance index did not permit the
BDM to estimate the upper range of the unfished biomass at equilibrium K.
The state-space estimation procedure implemented in BDM allows for the inclusion of process error.
The value inputted for the process error standard deviation (0.05) was based on recommendations
by McAllister (2013). Process error allows the model to account for inter-annual variability in stock
Pacific-wide sustainability risk assessment of bigeye thresher shark 29
biomass caused by temporal changes in biological processes that are not observed or modelled
(Edwards 2016). In this case, this includes potential immigration/emigration of bigeye thresher
to/from the Calibration Area. The effect of process error inclusion on q estimation was tested and
demonstrated in sensitivity analyses.
Both process error and K upper bound (prior) sensitivities were conducted using initial depletion
state (u) samples drawn from the medium (0.5) initial status distribution (n=300).
Figure 8: Initial population status (depletion) level u was sampled from three normal distributions
(n=300 from each), with means of 0.3, 0.5, and 0.7 (vertical dashed lines), respectively, and a standard
deviation of 0.05.
3.5.5 Post-capture survival
The catch history of bigeye thresher in the Calibration Area (see Section 3.5.2) was assumed to be
known without error, and to represent mortality induced by the fishery (i.e., no post-capture
survival). However, fate and condition data can be used to distinguish between bigeye thresher
which were and were not alive upon release (see datasets section of this report). In the SPC observer
data, the calculated proportion of bigeye thresher released alive showed very large inter-annual
variability with an overall average of 30% between 1995 to 2014 (Figure 9). In the US observer data,
the calculated proportion of bigeye thresher released alive was relatively stable between 2004 and
2014 and averaged 70%. Yet the post-release survival or actual mortality rate for the species remains
unknown. To account for the occurrence of live releases and potential survival, q calibration runs
were carried out which incorporated a range of post-capture survival values.
30 Pacific-wide sustainability risk assessment of bigeye thresher shark
This required the catch vector (or annual harvest rate Ht) in the Calibration Area to be adjusted by
applying an assumed post-capture survival rate s:
( )
sHH
t
t
−= 1
'
where H’ is the annual harvest rate (ratio of catch over abundance) adjusted for post-capture
survival. In each BDM run (and across the range of initial population status scenarios described
above), s was randomly drawn from a uniform distribution with bounds [0.3, 0.7]. This range was
loosely based on the calculated proportion of BTH released alive in the SPC and US observer datasets
(Figure 9).
A similar adjustment of posterior estimates for q in the Calibration Area (qΩ) was necessary to
propagate post-capture survival assumptions into fishing mortality estimation:
( )
sqq −= ΩΩ 1
'
Figure 9: Annual proportion of BTH released alive for US observer data 2003–2014 and SPC observer data
1995–2014
3.5.6 Spatial scaling and adjustments by fishery groups
The catchability scalar was adjusted by fishery groups to account for differences in operational
practices and associated capture efficiency for bigeye thresher among fishery sectors.
Fishery group-specific catchability
j
q
was estimated as:
jj
qfq =
Pacific-wide sustainability risk assessment of bigeye thresher shark 31
where
q
is the average population catchability (qΩ) scaled to the spatial resolution of the
assessment (5x5 degree cells) and
j
f
is the adjustment factor for fishery group
j
(see Appendix C for
complete derivation).
The adjustment factor
j
f
was calculated as the predicted catch rate for each fishery group relative
to a reference group (defined as the ‘BET’ catch group and ‘Jan-Mar’ season) using the final (spatial
standardisation) ZINB model fitted to all observer data within the Core Area. Since month was
modelled as a continuous variable, seasonal predictions were based on the intermediate month
within each season (i.e., February for Jan-Mar).
Uncertainty in
j
f
can be estimated using a bootstrap procedure (similar to that used for the spatial
abundance indices) but this was done in this assessment.
3.5.7 Key assumptions
Our
q
estimation method assumes that the Calibration Area and US Hawaii observer data are
representative of population dynamics for bigeye thresher sharks at the scale of the Pacific Ocean.
This means we assume that on average the fishing power of longline sets on bigeye thresher is the
same across the Pacific region, but differences in relative catchability (catch group and seasonal
effects) and population density explain the differences in catch rates. This is unlikely to be the case
but was a necessary assumption in the absence of informative data indicating otherwise. The
assumptions made on the initial population status (low, median, and high) are arbitrary and
intended to improve estimation and ensure realistic outcomes in q estimation. Indirectly, initial stock
status assumptions also served to deal with uncertainty in post-capture survival of bigeye thresher in
pelagic longline fisheries (e.g., a high post-capture survival scenario can be expected to result in a
higher initial stock status, and vice-versa). As in most age-structured stock assessment models,
values of q are assumed to remain constant over the time frame of the assessment (2000-2014).
3.6 Impact estimation (fishing mortality)
Impact was estimated relative to the total (commercial) pelagic longline effort available in the CES
Longline Logsheet dataset, from 2000 to 2014.
Spatially-explicit impact is the average annual fishing mortality in 5x5 degree cells calculated using
commercial effort data (split by fishery groups), species relative density and fishery group
catchability. We assumed cumulative fishing mortality as contributed from different fishery groups
in each cell, and cumulative impact over the spatial domain of the assessment.
Fishing mortality in each cell was calculated as the product of effort and fishery group catchability
and contrasted across a range of scenarios (i.e., the three initial population status assumptions used
to calibrate the catchability scalar and with and without taking into account the occurrence of post-
capture survival). Since the catchability parameter was calibrated using a plausible range of K values
with upper bound constrained based on expert opinion (or information from another species in
sensitivity analyses), estimates of fishing mortality represent the plausible range of fishing mortality
for the species over the timeframe of the assessment, as constrained by the available data and
expert knowledge.
Impacts were estimated for the Core Area (using species relative density estimates derived from the
ZINB model) and the Assessment Area (using density estimates from the delta-GLMM model).
Uncertainty in species distribution information is incorporated in impact estimation by resampling
density indices from bootstrapped estimates.
32 Pacific-wide sustainability risk assessment of bigeye thresher shark
3.7 Population productivity and MIST estimation
3.7.1 Maximum intrinsic growth rate r
The life history module (LHM) for BDM developed by Edwards (2016) was used to estimate a
distribution for the maximum intrinsic population growth rate
r
for bigeye thresher. The model
implements Monte Carlo sampling of life history parameter distributions, with iterated solving of the
Euler-Lotka equation (McAllister et al. 2001). The Euler-Lotka equation defines maximum intrinsic
growth r as the net balance of survivorship s and unconstrained fecundity f, integrated over all age
classes a:
safae-ar
∞
a=0 =1
sa=e-aM
fa=α mawa
Survivorship (s) is a function of the natural mortality
M
, assumed constant across ages. Fecundity (f)
is the product of female maturity m, weight w and the maximum recruits per spawner α (in the
absence of density dependent effects). The relevant functional forms are the maturity-at-age ma,
length-at-age la (modelled as per von Bertalanffy growth), weight-at-age wa and recruits per spawner
α:
ma=
1+exp
((
a50-a
)
)
-1
la=l
1-exp
-k
(
a-t0
)
wa=ala
b
α
=
4h
ρ
(
1-h
)
Recruits per spawner is related to steepness h and the female spawning biomass per recruit ρ,
assuming a Beverton-Holt stock recruitment relationship. Current LHM parameterisation requires a
value of h to be specified.
The model incorporates uncertainty in all parameters, which can be fixed on input (for details
see https://github.com/cttedwards/lhm). Maximum age Amax is treated as a single asymptotic value
in the model. Steepness is modelled as a bounded beta distribution and all other life history
parameters are modelled as random lognormal variables. Life history data used to estimate a
distribution for r are summarized in Table 5. Parameter values calculated for females of bigeye
thresher were used whenever possible. Parameters that were poorly-informed or unobserved (i.e.,
those relating to maturation and recruitment) were given a higher cv (0.2) in the estimation process,
and others that were estimated based on observations with sample sizes >100 specimens (growth
and longevity) were given a cv of 0.10. We assumed that females have a litter size of two (Chen et al.
1997) and an annual reproductive cycle.
Pacific-wide sustainability risk assessment of bigeye thresher shark 33
The maximum observed age for female bigeye thresher in the Atlantic was 22 yr (Fernandez-
Carvalho et al. 2011) and the maximum observed age in the Pacific was 21 yr (Liu et al. 1998). True
longevity in an unfished population probably exceeds both these values, so we used the larger value
in the Euler-Lotka equation. Natural mortality estimates were available from Smith et al. (2008)
(M=0.223) and Chen and Yuan (2006) (M=0.147). Additional M estimates were derived using four
empirical equations summarised in Tsai et al. (2010), including the Hoenig (1983) and Campana et al.
(2001) approximations based on maximum age; and the Jensen (1996) approximations based on age
at maturity and the growth parameter of the von Bertalanffy equation. The value in the table
represents the mean value for M (and calculated cv) obtained using the four empirical relationships.
A number of sensitivities were performed on selected input parameters, including Amax, h, M and
parameters of the maturity ogive. A thousand (x1000) iterations were performed in each run.
3.7.2 Maximum Impact Sustainable Threshold (MIST)
The MIST was set at 1.0r = Fcrash (the instantaneous fishing mortality rate corresponding to the
minimum unsustainable instantaneous fishing mortality rate) (Zhou et al. 2011). The MIST was used
to compute sustainability status and sustainability risk for the species in the Pacific.
3.7.3 Key assumptions
The intrinsic growth rate r is assumed to represent population productivity (and thus resilience and
recovery potential) for bigeye thresher. Productivity is assumed to have remained constant over the
spatial domain of the assessment, from 2000 to 2014. This implies a stable environment and stable
state (equilibrium) population dynamics for the species.
Table 5: Input life history information used to develop a prior for the maximum intrinsic population growth
rate of bigeye thresher in the Pacific. Maturation, Growth and Recruitment parameters are based on
available information for females only.
Process
Parameter
Value
cv
Reference(s)
Longevity
Amax (yr)
22
Fernandez-Carvalho et al. 2011
Maturation
A50 (yr)
13.4
0.20
Liu et al. 1998
delta δ
0.6
0.20
estimated
Growth
Linf (cm, PCL)
224.6
0.10
Liu et al. 1998
k
0.092
0.10
Liu et al. 1998
t0
-4.21
0.10
Liu et al. 1998
a
6.87x10
-5
0.10
Liu et al. 1998
b
2.769
0.10
Liu et al. 1998
Recruitment
α
2
Liu et al. 1998
h
0.30
0.20
estimated
Mortality
M
0.171
0.17
See text
34 Pacific-wide sustainability risk assessment of bigeye thresher shark
3.8 Sustainability risk calculations
Sustainability status was determined relative to fishing impact from pelagic longline fisheries in the
Pacific over the period 2000-2014, and computed relative to a MIST=1.0r= Fcrash. A sustainability risk
metric, corresponding to the ratio of total impact to the species MIST, was computed and compared
between impact estimated at the scale of the Core Area (ZINB model species distribution) and
impact estimated at the scale of the Assessment Area (delta-GLMM model species distribution).
The probability that current impacts exceed the MIST (Pr(Impact/MIST>1)) was calculated by re-
sampling across the uncertainty range estimated for all parameters. Additional sustainability risk
thresholds were defined a posteriori based on the distribution of annual sustainability status and
uncertainty.
4 ASSESSMENT RESULTS
Characteristics of the observer data (and observer data coverage) relevant to the assessment are
presented in Appendix D. Owing to the complexity and multiple dimensions of this assessment, the
following sections were structured to describe the main results of key components, namely species
distribution, catchability, fishing impact and sustainability status and risk. Details on multiple model
fitting, assumptions, sensitivities and comparisons, are presented in the appendices and referred to
as appropriate in the text.
4.1 Species distribution
The estimated spatial distribution for bigeye thresher using the final ZINB model (Core Area
distribution) and the spatial delta-GLMM model (Assessment Area distribution) are mapped in Figure
10. Abundance ‘hot spots’ and predicted densities (spatial indices of relative abundance) in Core
Area cells were similar between the two models (Figure 11). Highest densities occurred between
latitude 5°N and 15°N and, according to the extrapolated distribution from the delta-GLMM model,
spanned a broad longitudinal range from 150°E to approximately 140°W (220°E on map).
The three delta-GLMM models predicted similar, highest densities within three adjacent cells (cell ID
438, 437 and 436 – see Figure 5), which demonstrates the smoothing effect of the spatial correlation
(Figures 10, 11). ZINB outputs were patchier and predicted a higher relative density in cell ID 438 and
a lower density in cell ID 436. Encounter probabilities were highly variable among cells and ranged
from 2% to 72% within the Core Area. Predicted catch rates per cell (no. of captures per 1000 hooks)
were generally less than 2.
The final ZINB model included cell_ID, year, month, catch group and HBF as significant covariates
(Appendix E). Effort (no. of hooks) was included as an offset term to predict relative densities as the
number of captures per 1000 hooks in 5x5 degree cells. Relationships between explanatory variables
and encounter probabilities and catch rates were generally weak, suggesting that spatial effects
explained most of the variation in catch rates of bigeye thresher within the Core Area.
Pacific-wide sustainability risk assessment of bigeye thresher shark 35
Figure 10: Spatial distribution of bigeye thresher density over the spatial domain of the assessment, as
estimated using the final ZINB model (Core Area distribution, top) and the spatial delta-GLMM model
(Assessment Area distribution, bottom). Spatial indices of relative abundance are in relative log density in
cells centered on a 5x5 degree grid.
36 Pacific-wide sustainability risk assessment of bigeye thresher shark
Figure 11: Predicted indices of relative abundance for bigeye thresher (no. captures per 1000 hooks) in 5x5
degree cells in the Core Area of the Assessment Area, as estimated using the final ZINB model (leffort, 95%
confidence interval plotted) and three geostatistical delta-GLMM models (spatial, spatiotemporal, and core
(nonzero) vessels; uncertainty is not estimated (see Section 3.4.4)).
The delta-GLMM models were fitted to the same covariates as the final ZINB model plus random
vessel effects. These models permitted extrapolation of abundance up to 50 km beyond the area of
our observations within the Assessment Area. Predicted densities in 5x5 degree cells were generally
similar between the spatial, spatiotemporal and core vessel models (Figure 11). Spatial relative
abundance as estimated on the (finer scale) extrapolation grid (10x10km cells) used in model fitting
is shown for the spatial model in Figure 12 (year effects held constant) and annually for the
Pacific-wide sustainability risk assessment of bigeye thresher shark 37
spatiotemporal model in Figure 13. Annual indices suggest the spatial distribution of bigeye thresher
in the Assessment Area has remained relatively constant between 2000 and 2014 (Figure 13). The
spatiotemporal model predicted increased abundance in some ‘hot spot’ areas in recent years.
The Core Area (ZINB) model assumed that the population distribution for bigeye thresher was
restricted to Core Area cells in the Pacific, while the delta-GLMM model served to extrapolate
abundance over a broader region (Assessment Area plus 50 km distance from recorded
observations). Outputs from the delta-GLMM model suggested that between 20% and 44% of the
bigeye thresher population was distributed outside the Core Area cells in the Pacific, depending on
whether densities in 5x5 degree cells were averaged (44%) or summed (20%) over the fine scale
(10x10 km) grids of the predictive framework. This underlined the influence of data re-scaling
procedures when predicting spatial densities from a finer-resolution extrapolation grid to a coarser
grid used for assessment (in this case, determined by the spatial resolution of the commercial effort
(logsheet) data). The average re-scaling procedure was used and retained for further analyses in this
assessment.
Figure 12: Fine-scale relative density estimates (log scale) for bigeye thresher over the Assessment Area,
2000-2014, as predicted from the spatial delta-GLMM model. Data are plotted in 10x10 km grid cells used to
standardise and extrapolate catch rates in all GLMM models.
38 Pacific-wide sustainability risk assessment of bigeye thresher shark
Figure 13: Fine-scale, annual relative density estimates (log scale) for bigeye thresher in the Assessment
Area, from 2000 (top-left) to 2014 (bottom-right), as predicted from the spatial-temporal delta-GLMM
model. Annual data are plotted in 10x10 km grid cells used to standardise and extrapolate catch rates in all
GLMM models.
4.2 Catchability
The base case scenario for catchability estimation consisted of a 2 million upper prior bound on K,
three initial population status assumptions (i.e., initial biomass level relative to the unfished biomass
at equilibrium) and the assumption of 100% capture mortality.
The distribution of estimated values for the population catchability scalar qΩ across base case
scenarios is shown in Figure 14. Values ranged from 0.27 to 0.96 (10-6) at low (0.3) initial status
(median 0.5x10-5); from 0.19 to 0.77 (10-5) assuming a medium (0.5) initial status (median 0.34 x10-5);
and from 0.14 to 0.66 (10-5) (median 0.26 x10-5) in the high (0.7) initial status scenario (see BDM runs
1-3 in Table F1, Appendix F). A lower initial population status resulted in higher catchability values,
and vice versa.
Inter-annual variability in fishery group catchability (qΩ adjusted for spatial variation in effort
distribution by season and catch group) is illustrated in Figure 15. These values are based on the
medium (0.5) initial stock status assumption. From 2011 to 2014, higher and lower catchability
Pacific-wide sustainability risk assessment of bigeye thresher shark 39
characterised the Apr-Jun/Oct-Dec and Jan-Mar/Jul-Sep seasons, respectively. Effort mainly catching
SWO had comparatively lower (and a narrower range of) catchability. Higher catchability
characterised the fleet that mainly caught bigeye tuna (BET) in each year. Other catch groups
showed intermediate and generally variable catchability values. A summary of inter-annual
variability in fishery groups catchability by year, catch group and season, for the period 2000-2014, is
presented in Tables F2 and F3 of Appendix F.
Figure 14: Estimated distributions of the catchability scalar
Ω
q
for each initial stock status (biomass level
relative to the unfished biomass at equilibrium) assumption (low=0.3; medium=0.5 and high=0.7) considered
in BDM calibration (BDM run 1-3 in Table F1, Appendix F).
Assuming a range of post-capture survival rates produced median catchability estimates that were
32-35% lower than the 100% capture mortality scenarios (run 1s-3s in Table F1, Appendix F).
Increasing the upper bound value of the K prior likewise resulted in lower catchability estimates
(runs 2a and 2b in Table F1, Appendix F). The calibration model was sensitive to the inclusion of
process error, which permitted fitting the index of relative abundance without over-estimating the
unobserved biomass state (Appendix F).
40 Pacific-wide sustainability risk assessment of bigeye thresher shark
(a)
(b)
Figure 15: Distributions of estimated catchability q by year and fishery groups disaggregated by (a) fishing
season and (b) catch group. Values are based on the distribution for
Ω
q
derived assuming a medium (0.5)
initial biomass level for the stock.
The catch history (assuming 100% capture mortality) and standardised CPUE index of abundance
used in BDM calibrations are shown in Figure 16 and Figure 17, respectively. Annual CPUE indices are
tabulated (with 95% confidence intervals) in Table 6. Time series of estimated depletion and
predicted abundance indices for the base-case scenarios are given in Appendix F (Figure F1). The
CPUE index of abundance was not informative of stock status as the model was able to fit the trend
in the CPUE indices by varying initial status assumptions. This is because the standardised CPUE
index lacked contrast and was inconsistent with the derived catch series (i.e., the presence of trends
in some parts of the CPUE series, including a substantial decline from 1996 to 2000 and a slight
increase through the early 2000s, were not well explained by the catch history). This did not permit
the biomass dynamic model to infer stock status by interpreting changes in relative abundance in
Pacific-wide sustainability risk assessment of bigeye thresher shark 41
response to the harvest. Re-constructed catches between 2000 and 2014 were similar among
stratifications but differed in some years prior to 2000 (Figure 16). This may or may not be related to
comparatively limited observer coverage over the 1995-2000 period. The complete catch series and
detailed information on observer and commercial effort data used to derive catch estimates for BTH
in the Calibration Area are given in Appendix G.
Figure 16: Estimated catch (1995-2014) and approximated catch history (1952-1994) for BTH within the
Calibration Area. Catch estimates for 1995
−
2014 are compared among stratification methods. The catch
history for 1952
−
1994 is shown for comparison but was not used for analyses.
The standardised CPUE indices of abundance predicted using different models showed a consistent
trend over time. Explanatory variables included in the final ZINB (and other) models were year,
month, catch group, subarea, HBF and effort (log no. of hooks). Subarea had a large influence in
determining the annual trend (the trend in annual indices changed very little when other variables
were offered). Important fluctuations and higher variability early in the time series probably reflect
comparatively limited effort and/or observer coverage. Changes in fishing patterns over time were
evidenced in coefficient-distribution influence plots (Bentley et al. 2012) for explanatory variables in
the delta lognormal model (Figure 18). Numbers of shallow sets (low HBFs) were higher in 1995-
1997 and decreased thereafter. Effort mainly catching ALB was reduced in 2002-2014 compared to
1995-2001. The slight increasing trend from 2000 onwards can be partly explained by an increase in
observed effort in subareas that have lower catch rates during this period (i.e., in the
standardisation model, catch rates are dependent on year, subarea, and other factors in a
multiplicative manner. Thus, everything else being equal, subareas with lower effects probably
served to push the standardised year effects higher, and vice versa (see Figure 18)).
The nominal CPUE in 2014 was more than twice that in 2013. The reason for this is not clear,
although further investigation suggested most of the increase in catch occurred in one cell (cell_ID
436), where a gradual increase in effort has also occurred over the last five years. In addition, over a
third of catch in 2014 was taken in May, which had the highest SST among the last 8 years. A
sensitivity model that removed the 2014 data did not change the overall abundance trend (Appendix
H), consistent with the results of a similar sensitivity test on these data (Young et al. 2016). Detailed
results, model diagnostics and additional sensitivities performed in the development of a CPUE index
of abundance for bigeye thresher in the Calibration Area are presented in Appendix H.
42 Pacific-wide sustainability risk assessment of bigeye thresher shark
Figure 17: Annual CPUE indices of relative abundance (mean and 95% CI) for bigeye thresher in the
Calibration Area, 1995-2014, predicted using (a) the final ZINB model and the delta-GLMM model fitted with
1000 knots; and b) the final ZINB model, the same ZINB model excluding year 2014, and a delta lognormal
GLM model. All models were fitted to US Hawaii observer data from the Calibration Area. Grey dots in (b)
are nominal (unstandardised) CPUE. All indices were normalised to the mean of each series to allow
comparison.
Pacific-wide sustainability risk assessment of bigeye thresher shark 43
Table 6: Final standardised, and re-scaled (see Appendix A) CPUE index of abundance (number of
captures per 1000 hooks with confidence intervals) for bigeye thresher in the Calibration Area (AΩ)
estimated using the US Hawaii observer data, 1995
−
2014. The indices from other models (see Figure 16) are
not used in the catchability calibration analysis and therefore are not presented here.
95%CI
Year
CPUEΩ
lower
upper
1995
0.1664
0.0596
0.1247
1996
0.3260
0.1196
0.2490
1997
0.2585
0.0979
0.1949
1998
0.2356
0.0855
0.1610
1999
0.1561
0.0176
0.1171
2000
0.1364
0.0447
0.0904
2001
0.1390
0.0560
0.0904
2002
0.1633
0.0708
0.1037
2003
0.1421
0.0595
0.0874
2004
0.1833
0.0784
0.1064
2005
0.1660
0.0704
0.0989
2006
0.2055
0.0881
0.1242
2007
0.1562
0.0669
0.0946
2008
0.1805
0.0753
0.1111
2009
0.1994
0.0813
0.1309
2010
0.1776
0.0738
0.1093
2011
0.1623
0.0690
0.1020
2012
0.1995
0.0855
0.1238
2013
0.2050
0.0845
0.1322
2014
0.3027
0.1304
0.1867
44 Pacific-wide sustainability risk assessment of bigeye thresher shark
(a)
(b)
(c)
(d)
Figure 18: Influence plots (Bentley et al. 2012) for explanatory variables of the lognormal component of the
delta-lognormal model fitted to US Hawaii BTH catch and effort observer data, 1995–2014. (a) month, (b)
subarea, (c) catch group, and (d) HBF. Influence plots are used to visualize the effect of explanatory
variables on annual CPUE indices. Each plot shows the relative effects by levels of the explanatory variable
(top panel), the relative distribution of the variable by year (bottom left panel) and the calculated influence
of the variable on the unstandardized CPUE by year (bottom right panel).
Pacific-wide sustainability risk assessment of bigeye thresher shark 45
4.3 Fishing impacts
The breakdown of commercial effort data (no. of hooks) by year and annual proportions of total
effort located in the Core Area and corresponding to the different catch groups and seasons used to
distinguish fishery groups, are shown in Table 7.
Estimates of annual fishing mortality in 5x5 degree cells are mapped for the recent period (2011-
2014) in Figure 19. Time series of annual impact for the Core Area (ZINB species distribution) and the
Assessment Area (delta-GLMM species distribution) are presented in Figure 20. Trends were similar
under both distribution assumptions. Higher and lower impact was estimated for the lower (0.3) and
higher (0.7) initial population status, respectively. Median fishing impacts were lower than the mean
intrinsic population growth rate r for the species if the initial population status was assumed to be at
or above 0.5, and higher than r if a lower initial status (0.3) was assumed (see Figure 24, section 4.4
for detailed results on the r prior distribution for bigeye thresher). In all scenarios, variability in
annual estimates generally overlapped with the uncertainty range for r. (Figure 20).
Impact was lowest in 2001 and highest in 2012 (Figure 20). In 2001, the fishing mortality F ranged
0.010-0.044 among cells in the Core Area and 0.009-0.034 in the Assessment Area. In 2012, F ranged
0.018-0.078 in the Core Area and 0.021-0.085 in the Assessment Area. Higher impact in 2012-2013 in
the Assessment Area suggests an increase in fishing effort outside the Core Area in recent years.
Over the recent period (2011-2014), median impact (assuming a 0.5 initial stock status) was 0.030
(95% quantile range 0.015-0.073) in the Assessment Area, and 0.029 (95% quantile range 0.015-
0.066) in the Core Area. Over the longer period (2000-2014), median impact was 0.023 (0.011-0.059)
in the Assessment Area and 0.026 (0.013-0.061) in the Core Area (Table 8).
If post-capture survival was taken into account, median estimates of fishing impact were generally
below the mean intrinsic population growth rate, except in 2012 and 2013 when median impacts
exceeded r under the assumption of a low (0.3) initial stock status (Figure 21). Over the recent
period (2011-2014), median impact (again assuming a 0.5 initial stock status) was 0.021 (95%
quantile range 0.007-0.063) in the Assessment Area, and 0.019 (95% quantile range 0.006-0.063) in
the Core Area. Over 2000-2014, median impact was 0.016 (0.005-0.051) in the Assessment Area
relative to 0.018 (0.006-0.057) in the Core Area (Table 9). In all cases, differences between the Core
Area and Assessment Area were very small.
Fishing impact calculated using the range of catchability values derived in sensitivity testingof BDM
calibration runs, are shown in Appendix I. Catchability estimates for higher upper bounds on K
produced lower median fishing impacts below the mean intrinsic population growth rate r for the
species. Minimizing process error likewise resulted in a reduction of median of fishing impacts to a
level below the population growth rate r, whereas doubling the process error resulted in higher
impacts above r. These findings and related assumptions are pondered and deliberated in the
discussion.
46 Pacific-wide sustainability risk assessment of bigeye thresher shark
(a)
(b)
Figure 19: Estimates of fishing mortality for BTH in 5x5 degree cells for (a) the Core Area (species relative
abundance in each cell predicted from the ZINB model) and (b) the Assessment Area (species relative
abundance predicted using the delta-GLMM model). The assumed initial stock status was 0.5. Blue is lower F
and red is higher F (log scale).
Pacific-wide sustainability risk assessment of bigeye thresher shark 47
Table 7: Summary of commercial (pelagic longline) effort information used to estimate fishing impact for bigeye thresher in the Pacific Ocean. Total number of
hooks by year and percentages (%) of total effort within the Core Area, and corresponding to each of the catch groups and seasons used to differentiate fishery groups.
Total hooks
Core Area
Catch group (%)
Season (%)
Year
(millions)
(%)
YFT
ALB
BET
SWO
Others
Jan–Mar
Apr–Jun
Jul–Sep
Oct–Dec
2000
507
25.69
26.95
22.88
48.35
1.82
<0.01
25.18
23.63
27.14
24.05
2001
574
24.90
26.14
23.86
48.91
1.08
<0.01
27.13
23.31
25.38
24.19
2002
681
26.63
19.15
27.68
52.40
0.77
<0.01
24.53
21.69
27.91
25.86
2003
711
32.34
23.27
26.11
49.84
0.78
<0.01
23.47
24.80
26.13
25.60
2004
681
26.00
16.18
27.80
55.22
0.79
<0.01
25.02
22.47
26.44
26.07
2005
614
30.39
15.61
33.72
49.98
0.69
<0.01
25.66
23.40
26.70
24.23
2006
607
32.26
15.94
36.01
47.35
0.71
<0.01
22.71
25.56
26.88
24.85
2007
599
31.19
19.89
32.49
46.02
1.59
<0.01
23.74
23.96
27.30
25.00
2008
610
30.55
16.75
33.41
48.53
1.26
0.04
23.55
25.64
25.82
24.99
2009
647
27.22
19.73
38.88
39.53
1.86
0.01
23.11
24.99
27.31
24.59
2010
676
28.84
18.62
41.45
38.17
1.75
0.02
22.60
24.63
27.45
25.32
2011
730
30.69
23.05
36.62
38.43
1.85
0.05
22.79
24.50
25.44
27.28
2012
776
28.19
15.10
43.20
40.16
1.53
0.01
24.24
25.73
24.96
25.08
2013
729
31.04
18.74
45.37
35.14
0.74
0.00
24.57
28.02
26.31
21.11
2014
649
32.13
23.36
40.91
34.77
0.94
0.02
22.34
26.33
26.53
24.80
48 Pacific-wide sustainability risk assessment of bigeye thresher shark
(a)
(b)
Figure 20: Annual impact (median values and 95% quantile range) estimated for (a) the Core Area and (b)
the Assessment Area, using catchability estimates derived assuming 100% post-capture mortality and three
initial population status assumptions (low (0.3), medium (0.5), and high (0.7)) (BDM runs 1, 2, 3 from Table
F1, Appendix F). The dashed line is the mean value for the estimated r prior (0.03), with 95% quantile range
grey band (see section 4.4 for details).
(a)
(b)
Figure 21: Annual impact (median values and 95% quantile range) estimated for (a) the Core Area and (b)
the Assessment Area, using catchability estimates adjusted for the occurrence of post-capture survival
(random occurrence between 0.3 and 0.7) and assuming three initial population status assumptions (low
(0.3), medium (0.5), and high (0.7)) (BDM run 1s, 2s, 3s from Table F1, Appendix F). The dashed line is the
mean value for the estimated r prior (0.03), with 95% quantile range (grey band) (see section 4.4 for details).
Pacific-wide sustainability risk assessment of bigeye thresher shark 49
Table 8: Total impact (median F and 95% quantile range among cells) for the fifteen year period (2000-
2014) and the recent period (2011-2014) in the Core Area and the Assessment Area, estimated assuming
100% post-capture mortality and three initial population status assumption (low (0.3), medium (0.5) and
high (0.7)).
Impact
Impact
Impact
Low (0.3)
Medium (0.5)
High (0.7)
Core Area
2000-2014
0.038 (0.02-0.076) 0.026 (0.013-0.061) 0.019 (0.01-0.05)
2011-2014
0.042 (0.023-0.084) 0.029 (0.015-0.066) 0.022 (0.011-0.055)
Assessment
Area
2000-2014
0.034 (0.016-0.078) 0.023 (0.011-0.059) 0.018 (0.008-0.049)
2011-2014
0.044 (0.021-0.096) 0.03 (0.015-0.073) 0.023 (0.011-0.061)
Table 9: Total impact (median F and 95% quantile range among cells) for the fifteen year period (2000-
2014) and the recent period (2011-2014) in the Core Area and the Assessment Area, estimated assuming the
occurrence of post-capture survival (random occurrence between 0.3 and 0.7) and three initial population
status assumptions (low (0.3), medium (0.5) and high (0.7)) (BDM run 1s, 2s, 3s from Table F1, Appendix F).
Impact
Impact
Impact
Low (0.3)
Medium (0.5)
High (0.7)
Core Area
2000-2014
0.025 (0.009-0.066)
0.018 (0.006-0.057)
0.014 (0.004-0.048)
2011-2014
0.028 (0.01-0.073)
0.019 (0.006-0.063
0.015 (0.005-0.052)
Assessment
Area
2000-2014
0.024 (0.008-0.065)
0.016 (0.005-0.051)
0.012 (0.004-0.045)
2011-2014
0.03 (0.011-0.081)
0.021 (0.007-0.063)
0.015 (0.005-0.055)
Fishery group contributions to annual impact over the recent period (2011-2014) indicated a lower
fishing mortality in Jul-Sep of each year in both the Core Area and Assessment Area, and higher
fishing mortality in Apr-Jun (Figure 22). In 2011, a higher F was associated with the Oct-Dec season in
the Core Area. Fishing mortality was highest in the BET catch group in all years, while the SWO fleet
contributed minimal impact (Figure 23). Higher impact in 2012-2013 in the Assessment Area was
linked to comparatively higher F contributions from the YFT fleet and the BET fleet during the Jan-
Mar season.
50 Pacific-wide sustainability risk assessment of bigeye thresher shark
(a)
(b)
Figure 22: Annual fishing mortality (2011
−
2014) as contributed from different fishery groups (here
distinguished by season) over (a) the Core Area and (b) the Assessment Area. Values are based on the medium
(0.5) initial biomass level assumption.
(a)
(b)
Figure 23: Annual fishing mortality (2011
−
2014) as contributed from different fishery groups (here
distinguished by catch group) over (a) the Core Area and (b) the Assessment Area. Values are based on the
medium (0.5) initial biomass level assumption.
4.4 Sustainability risk
The estimated distribution of the maximum intrinsic population growth rate r for bigeye thresher is
shown in Figure 24. The distribution had a median 0.028 and cv 0.53 (Table 10). Results of sensitivity
analyses indicated that the r estimation method used this study was most sensitive to recruitment
assumptions relating to steepness (and associated maximum recruits per spawner). Decreasing the
maximum annual number of recruits per spawner to less than 2 (in this case, 1.7 and 0.9) reduced
median r values to 0.02 and 0.01, respectively, with increased variability (calculated cv=0.95 for r
distribution estimated assuming α=0.9) (Table 10).
Pacific-wide sustainability risk assessment of bigeye thresher shark 51
Figure 24: Estimated distribution of the maximum intrinsic population growth rate r for bigeye thresher,
using Monte Carlo sampling of life history parameters distributions with iterated solving of the Euler-Lotka
equation.
Table 10: Results of sensitivity analyses on maximum intrinsic growth r estimation for bigeye thresher using
the life history module (LHM). Different models were fitted by varying input values for M, δ, A50, h and
Amax.
Model
mean
median
sd
cv
Reference
Base model
0.031
0.028
0.017
0.528
Table 5, section 3.7
M=0.223 (cv=0.17)
0.033
0.031
0.019
0.571
Smith et al. 2008
M=0.147 (cv=0.17)
0.031
0.029
0.017
0.543
Chen and Yuan 2006
δ=0.3 (cv=0.20)
0.030
0.028
0.017
0.545
δ=1.2 (cv=0.20)
0.032
0.030
0.018
0.563
δ=1.7 (cv=0.20)
0.034
0.031
0.019
0.564
A50=12.3 (cv=0.20)
0.034
0.031
0.019
0.566
A50=12.3, δ=1.2 (cv=0.20)
0.033
0.031
0.019
0.558
h=0.28 (cv=0.20) (α=1.7)
0.026
0.023
0.016
0.620
h=0.25 (cv=0.20) (α=0.88)
0.016
0.012
0.016
0.945
Amax=21 (F)
0.033
0.030
0.019
0.567
Liu et al. 1998 (estimated for
largest observed size)
Amax=20 (M)
0.033
0.031
0.019
0.557
Liu et al. 1998 (estimated for
largest observed size)
Amax=40
0.028
0.025
0.016
0.585
52 Pacific-wide sustainability risk assessment of bigeye thresher shark
The estimated r distribution is taken to represent the maximum impact sustainable threshold (MIST)
for the species, corresponding to the minimum unsustainable instantaneous fishing mortality rate (r
= MIST =Fcrash ) for the stock.
Assuming 100% capture mortality, sustainability risk (ratio of impact to MIST) ranged from 0.6 to 1.2
in the Core Area (median values among initial population status assumptions) and between 0.6 and
1.1 in the Assessment Area, over the entire (2000-2014) assessment period (Figure 25, Table 11). For
the recent period (2011-2014), sustainability risk ranged from 0.7 to 1.3 in the Core Area and from
0.7 to 1.4 in the Assessment Area. Sustainability risk was highest (and generally close to 1) under the
assumption of a lower initial stock status for the species in the Pacific, and lower if initial status was
assumed to be high.
Accounting for post-capture survival reduced sustainability risk in all scenarios (range 0.4-0.8 for
2000-2014 in both the Core Area and Assessment Area) (Figure 25, Table 12).
There was considerable uncertainty about sustainability risk. The upper range estimates (95%
quantile) were above 1 in all scenarios, indicating the possibility that total impact from pelagic
longline fisheries in the Pacific exceeded the minimum unsustainable fishing mortality rate for the
species. Assuming 100% capture mortality, the calculated probability that the annual impact
exceeded the MIST, given the uncertainty, averaged 0.4 and ranged 0.1-0.7 among years and
distribution scenarios (Table 13). If post-capture survival was taken into account, the probability that
the annual impact exceeded the MIST ranged from 0.1 to0.4 and averaged 0.2 (Table 13).
Sustainability risk was generally stable among years and showed no directional trend over time,
aside from a small increase in risk ratio over the recent period in the Assessment Area (Figure 25b).
The probability that total impact exceeded the MIST was higher in the Assessment Area relative to
the Core Area in 2012 and 2013. In other years, sustainability risk probabilities were higher in the
Core Area (Table 13).
Table 11 : Sustainability risk (ratio of impact to MIST with values >1 considered to be unsustainable) (median
values and 95% quantile range) for bigeye thresher in the Pacific, as estimated for the Core Area and the
Assessment Area assuming 100% capture mortality in impact estimation and three initial population status
assumptions (low (0.3), medium (0.5), and high (0.7)). Results are contrasted for the fifteen year period (2000-
2014) and the recent period (2011-2014).
Impact/MIST Impact/MIST Impact/MIST
Low (0.3) Medium (0.5) High (0.7)
Core Area
2000-2014 1.218 (0.527-3.05) 0.834 (0.349-2.357) 0.624 (0.252-1.883)
2011-2014 1.334 (0.597-3.253) 0.922 (0.385-2.537) 0.701 (0.292-2.052)
Assessment
Area
2000-2014 1.076 (0.434-3.002) 0.746 (0.291-2.239) 0.567 (0.216-1.875)
2011-2014 1.387 (0.576-3.657) 0.959 (0.386-2.775) 0.733 (0.283-2.325)
Pacific-wide sustainability risk assessment of bigeye thresher shark 53
Table 12: Sustainability risk (ratio of impact to MIST with values >1 considered to be unsustainable) (median
values and 95% quantile range) for bigeye thresher in the Pacific, as estimated for the Core Area and the
Assessment Area assuming the occurrence of post-capture survival (random occurrence between 30% and
70%) in impact estimation and three initial population status assumptions (low (0.3), medium (0.5), and high
(0.7)). Results are contrasted for the fifteen year period (2000-2014) and the recent period (2011-2014).
Impact/MIST Impact/MIST Impact/MIST
Low (0.3) Medium (0.5) High (0.7)
Core Area
2000-2014 0.815 (0.247-2.54) 0.563 (0.164-2.154) 0.438 (0.119-1.764)
2011-2014 0.902 (0.281-2.794) 0.619 (0.184-2.399) 0.483 (0.134-1.961)
Assessment
Area
2000-2014 0.755 (0.23-2.426) 0.519 (0.148-1.89) 0.379 (0.11-1.62)
2011-2014 0.974 (0.302-3.051) 0.677 (0.193-2.428) 0.488 (0.142-2.065)
54 Pacific-wide sustainability risk assessment of bigeye thresher shark
(a)
(b)
Figure 25: Annual sustainability risk for bigeye thresher in the Pacific, 2000-2014, as estimated for (a) the
Core Area and (b) the Assessment Area. Sustainability risk and uncertainty are distinguished based on the
occurrence (green) or absence (red) of post-capture survival. Sustainability risk (x-axis) is the ratio of total
impact (combined values across three initial population status assumptions) to the maximum impact
sustainable threshold (MIST) (here corresponding to the maximum intrinsic rate of population increase r). A
ratio above one indicates potentially unsustainable fishing impact.
Pacific-wide sustainability risk assessment of bigeye thresher shark 55
Table 13: Sustainability risk probabilities (Pr(Impact/MIST)>0.5 and Pr(Impact/MIST)>1.0) for bigeye thresher in the Pacific, 2000-2014, assuming 100% capture
mortality (left) and the occurrence of post-capture survival (right) over the Core Area and the Assessment Area (combined values across three initial population status
assumptions). MIST = r = Fcrash
Absence of post-capture survival (100% capture mortality) Occurrence of post-capture survival (random between 30% and 70%)
Core Area Assessment Area Core Area Assessment Area
Year Pr(Impact/MIST) > 0.5 Pr(Impact/MIST) > 1
Pr(Impact/MIST) > 0.5 Pr(Impact/MIST) > 1 Pr(Impact/MIST)
> 0.5
Pr(Impact/MIST)
> 1
Pr(Impact/MIST)
> 0.5
Pr(Impact/MIST)
> 1
2000 0.756 0.295
0.645 0.188 0.510 0.163 0.405 0.108
2001 0.684 0.226
0.538 0.129 0.435 0.126 0.320 0.062
2002 0.818 0.372
0.673 0.216 0.558 0.218 0.429 0.117
2003 0.905 0.521
0.853 0.413 0.673 0.308 0.616 0.248
2004 0.803 0.359
0.689 0.228 0.556 0.197 0.442 0.124
2005 0.926 0.565
0.837 0.392 0.706 0.333 0.593 0.224
2006 0.834 0.405
0.668 0.224 0.597 0.229 0.437 0.114
2007 0.870 0.463
0.796 0.347 0.644 0.269 0.558 0.191
2008 0.822 0.375
0.779 0.323 0.572 0.211 0.537 0.175
2009 0.776 0.319
0.820 0.356 0.513 0.175 0.571 0.214
2010 0.799 0.338
0.740 0.285 0.549 0.193 0.499 0.158
2011 0.849 0.414
0.832 0.379 0.598 0.236 0.581 0.222
2012 0.936 0.586
0.965 0.674 0.727 0.369 0.790 0.434
2013 0.897 0.501
0.943 0.614 0.660 0.298 0.743 0.392
2014 0.836 0.411
0.806 0.353 0.584 0.233 0.560 0.204
56 Pacific-wide sustainability risk assessment of bigeye thresher shark
5 DISCUSSION AND RECOMMENDATIONS
The assessment of fishing effects on pelagic sharks is complicated by limited data, inaccurate and
incomplete catch records, and changes in fishing practices, reporting requirements and observer
coverage over time. In thresher sharks, additional complexity arises from species identification
problems leading to the grouping of three species (bigeye, common and pelagic thresher) into a
single gender/species complex (Alopias or Thresher spp.) in commercial catch reports (logsheet
data). Age data and information on movements and migratory patterns are also limited. As a result,
population trends for thresher sharks are largely unavailable and highly uncertain (Young et al. 2016)
and the present data situation limits the use of age-structured stock assessment models to infer
biomass trajectories and evaluate stock status relative to reliable estimates of population
abundance.
In this study, we developed and implemented a spatially-explicit and quantitative framework for
sustainability risk evaluation of bigeye thresher shark. This approach served to integrate available
data for the species in the Pacific into a comprehensive framework permitting quantification of
relative fishing impact (with uncertainty) in relation to the species’ ability to withstand fishing
pressure, as determined from life history productivity parameters. This represents an improvement
over traditional low-information assessment methods, including stand-alone trends in relative
abundance inferred from catch and effort data in relevant fisheries and indicator-based analyses for
monitoring changes in distribution, median size, sex ratio and catch composition by species (Clarke
et al. 2011, Clarke 2011, Francis et al. 2014, Rice et al. 2015). It is also an improvement over semi-
quantitative risk assessments that serve to evaluate and rank species vulnerability to exploitation,
but do not quantify impacts or fishing-induced mortality (Cortés et al. 2008, 2010, 2012).
The available data for bigeye thresher in the Pacific were used to infer species distribution and to
calibrate a range of plausible catchability values for use in spatially-explicit fishing mortality
estimation. This served to shift the assessment focus from poorly informed abundance trends to
quantitative impact estimation and mapping over space and across fishery groups. Quantifying and
incorporating uncertainty in the estimation of both impact and population productivity, allowed
estimation of sustainability risk as a probability with measured uncertainty.
The catchability calibration was performed using a Bayesian state-space surplus production model
applied to a subset of the observer data assumed to represent population dynamics for bigeye
thresher in the Pacific (i.e., the US Hawaii observer dataset). The constructed catch series and index
of relative abundance derived from these data lacked sufficient contrast to allow the model to
define an upper bound for the biomass estimates. The catchability estimates were therefore
constrained by fixing the upper bound value for the prior over K, the unfished biomass at
equilibrium, based on expert advice on maximal plausible bigeye thresher density in fishery
hotspots. Structural uncertainty (resulting from limited information provided by the CPUE indices of
abundance and the lack of accurate and reliable catch estimates) was too vast to use the biomass
dynamics model directly to evaluate stock status for the species at the scale of the Pacific Ocean.
Instead, spatially-explicit fishing mortality was estimated using a plausible range of catchability
values (with uncertainty), scaled by fishery groups and weighted by species relative density. This
approach assumed that the species range generally overlaps with the spatial extent of pelagic
longline fisheries. Fishing impact and sustainability risk ratios were compared between scenarios
assuming 100% capture mortality and a range of post-capture survival rates to account for the
occurrence of live release. The outcomes represent the probability distribution of fishing impact and
sustainability risk from pelagic longline fisheries for bigeye thresher in the Pacific, as distinguished
over space and across fishery sectors, over the period 2000-2014.
Pacific-wide sustainability risk assessment of bigeye thresher shark 57
Different methods have been developed to estimate fishing mortality in data limited situations.
Methods based on age- or length-cohort analyses are usually data intensive and require samples
distributed across life stages. Tsai et al. (2010) used a Virtual Population Analysis approach to
estimate fishing mortality for pelagic thresher shark in the north-western Pacific based on catch-at-
age data inferred from size distributions of landed (market) fish. This approach assumes that the size
structure of targeted and landed sharks is representative of the whole-population. For bigeye
thresher in the Pacific, selectivity patterns in pelagic longline fisheries are likely to be confounded
with spatial differences in life stage and gender distribution (Clarke 2011a, 2011b, Matsunaga &
Yokawa 2013). Other ‘swept area’ methods have been applied to estimate fishing mortality for non-
target species in the North Sea and Australia based on the overlap of fishing effort and species range
(Pope et al. 2000; Zhou et al. 2008, 2009). This approach is difficult to apply to passive fishing
methods such as pelagic longlines for which the effective area of fishing (and related area of impact)
cannot be easily quantified.
Spatially-explicit fishing impacts for bigeye thresher in the Pacific were estimated using information
on species distribution and fishery catchability derived from observer data, and aggregated fishing
effort data in pelagic longline fisheries differentiated by catch group and season. Median fishing
mortality ranged from 0.01 to0.04 across years and scenarios representing different distribution,
initial population status and post-capture survival assumptions. Highest impact overlapped with the
region of higher relative abundance for the species, corresponding to a narrow latitudinal-
longitudinal band between 10-15°N and 150°E-220°E. Fishing operations targeting bigeye tuna and
operating during the April-June season had the highest impact over the recent period (2011-2014).
The ability to disaggregate impact spatially and among fishery groups is an important outcome of the
risk assessment framework implemented in this study, allowing consideration of more focused
management options.
5.1 Fishery groups
Fishery groups corresponding to different catch groups and seasons were used to differentiate
impact contributed by different fishery sectors. The definition of fishery groups in this case was
limited by the coarse-scale resolution of the data provided for the assessment (i.e., catch and effort
data for the main target species aggregated in 5x5 degree cells). Differences in fleet composition and
fishing practices at finer operational scales (i.e., vessel and/or trip) were neither available nor
considered in the assessment. The pelagic longline fleet in the WCPO comprises a mix of vessels that
specifically target sharks, vessels engaged in ‘mixed targeting’ and vessels that target tuna and other
non-shark species and report sharks solely as bycatch (Young et al. 2016). Such distinctions will
affect fishing mortality estimation for bigeye thresher in the Pacific. Random vessel effects were
considered when predicting spatial indices of relative abundance for the species using the delta-
GLMM model, and in year-effects standardisations of CPUE data for catchability estimation in the
Calibration Area. A lack of finer operational level information to better distinguish targeting
behaviour and operational practice and more accurately define fishery groups and associated
catchability was an important limitation of this study. A more detailed definition of fishery groups
would improve the accuracy of fishing impact estimation for bigeye thresher in the Pacific.
58 Pacific-wide sustainability risk assessment of bigeye thresher shark
5.2 Species distribution
Spatial abundance distribution of bigeye thresher was estimated using standardisation models
applied to observer data. Despite operating on different spatial scales, the ZINB model and
geostatistical delta-GLMM showed consistent results (within overlapping areas). Both models
predicted highest densities between 10–15°N and 150°E–220°E. According to Matsunaga & Yokawa
(2013), the area between 10–15°N and 180°E –210°E is the parturition and nursery ground for
bigeye thresher in the Pacific. The ZINB model estimated the abundance distribution within 63 grid
cells assumed to represent the “Core” area of distribution and occurrence. Although the Core area
accounted for a large proportion of the observed catch, it is unlikely to represent the full distribution
range for the species.
A possible limitation in species distribution modelling was that the observer coverage has been low
and disproportionate among regions (Clarke et al. 2011a). In the WCPO, annual average Regional
Observer Programme (ROP) coverage in longline fisheries from 2005–2008 was <1% and has
remained below 2% since 2009 (Clarke et al. 2011a, Clarke, 2013). = If a substantial portion of the
bigeye thresher population is distributed outside the range of the observer data, restricting the
analysis to the Core Area (as done in the ZINB model) could potentially bias estimates of
sustainability risk upward or downward, depending on the level of fishing impact in the un-observed
fraction of the population. Lower estimated densities for bigeye thresher south of the equator, may
be an artefact of unbalanced sampling linked to comparatively poorer observer coverage in the SPC
dataset in the southern regions. The delta-GLMM model served to relax the assumption that the
species range overlapped with the fishery areas. This represents a significant improvement over
other standardisation methods for predicting more realistic spatial indices of relative abundance and
minimising bias in sustainability risk evaluation resulting from inadequate population distribution
assumptions. The present results indicated that between 22% and 40% of the bigeye thresher
population in the Pacific may be distributed outside the Core Area (see section 4.1). This constitutes
an important finding that requires further investigation, ideally with the inclusion of environmental
covariates. The geostatistical delta-GLMM model framework applied in this study (Thorson et al.
2015) allows for the inclusion of environmental covariates, which can serve to extrapolate further
beyond the area of observations. This was outside the scope of this study, but would be a valuable
exercise in similar, future assessments.
The delta-GLMM model has been demonstrated to improve precision of abundance indices based on
survey data, by incorporating spatial correlation in distribution estimation (Thorson et al. 2015). The
assumption that population spatial densities vary in a smooth fashion is biologically appealing.
However, its application to fishery dependent data may require further evaluation, as the potential
correlation between sampling intensity and underlying abundance as a result of targeting behaviour
may increase bias (Thorson et al. 2015). It is not known whether geostatistical approaches applied to
fishery-dependent data can produce more precise estimates of abundance than conventional
standardisation models. Herein, the delta-GLMM applied to bigeye thresher had a number of
advantages compared to the ZINB model. Firstly, it allowed abundance distribution to be estimated
on a finer spatial scale without estimating an excessive number of parameters. Secondly, it allowed
temporal changes in abundance distribution to be estimated more efficiently (we have assumed that
the distribution of bigeye thresher remained constant over the assessment period to simplify the
analysis and because only minimal spatial-temporal variation was found). Thirdly, vessel effects can
be accounted for (the ZINB model could not standardise for vessel effects owing to the large number
of vessels and hence parameters). Finally, and most importantly, the delta-GLMM model permitted
the expansion of the spatial scale of the assessment (from the Core Area to the Assessment Area)
and the detection of spatial changes in fishing intensity contributing to increased fishing mortality in
recent years.
Pacific-wide sustainability risk assessment of bigeye thresher shark 59
A potential issue encountered here was that set locations in the SPC and Japan observer data were
provided at a lower resolution, which may have affected the estimation and extrapolation processes
when deriving spatial indices of abundance. Uncertainty estimation and propagation from the delta-
GLMM model were not performed in this study owing to time and data accessibility constraints, but
could be attempted and tested in future work.
Likewise, inclusion of environmental covariates and improved parameterisation of the ZINB and
delta-GLMM models (i.e., to deal with issues such as discontinuity between December/January
months as a result of modelling month as a continuous variable) have not been fully explored in this
analysis and would be useful in future work.
5.3 Catchability
Methods for estimating catchability include experiments, census, or other sampling methods that
compare catch rates with abundance. Model-based approaches are also common, yet can
underestimate uncertainty due to process error (Ward 2007). Herein, catchability estimation was
performed by means of a calibration exercise implemented in a Bayesian state-space biomass
dynamics model (Edwards 2016, McAllister et al. 2001). This approach relied on the assumption that
the available data were insufficient to estimate absolute catchability, but could be used to calibrate
a relative catchability parameter for use in spatially-explicit impact estimation. The resulting
catchability estimates were influenced by strong assumptions on stock productivity imposed by the
BDM (surplus production) calibration model (i.e., feasible population trajectories determined by the
logistic growth curve and controlled by the intrinsic growth rate parameter), and were constrained
by the prior bound assumptions on the unfished biomass at equilibrium K.
The calibration permitted us to account for observation and process errors in relative catchability
estimation, and to incorporate uncertainty in the intrinsic growth rate and initial population status.
It did not, however, deal with structural uncertainty (i.e., the difference between the model world
and the real world). Alternative models such as the delay-difference model that might better mirror
the population dynamics of bigeye thresher should be explored for their applicability in deriving
feasible population trajectories and relative and absolute catchability.
Sensitivity runs on the BDM calibration suggested the estimate was not sensitive to the shape of the
production curve but was sensitive to the inclusion of process error (Appendix F). Within the
Bayesian state-space estimation framework implemented in BDM, process error consists of a time-
dependent, multiplicative error term that accounts for inter-annual variability in stock biomass
caused by temporal changes in biological processes that are not observed or modelled. As such,
process error inclusion provided the model with some ‘space’ to deal with inconsistencies in the
data, by assuming there are factors other than those considered (in this case, a constructed catch
series and CPUE index of abundance for an area subset), that are likely to affect biomass. We
considered this to be a reasonable and defensible assumption based on: 1) the difference in scale
between the calibration dataset (US Hawaii observer data in the Calibration Area) and the spatial
domain of the assessment (Pacific-wide); 2) the lack of a proper and reliable catch series for bigeye
thresher (which is unlikely to be obtained for the whole Pacific pelagic longline fishery owing to the
historical lack of shark reporting in logsheet data as well as the current requirement to report
thresher sharks only at the genus level (Clarke 2011a, Young et al. 2016)); 3) limited knowledge of
population structure, movements and migration patterns for bigeye thresher in the Pacific; 4)
uncertainty regarding the length of the reproductive cycle for the species; and 5) a high degree of
uncertainty in the index of relative abundance used in the calibration.
60 Pacific-wide sustainability risk assessment of bigeye thresher shark
The standardised indices suggested that relative abundance in the Calibration Area has been stable
over time, yet trends in reported catch rates in fisheries data may not represent changes in relative
abundance for thresher sharks (Young et al. 2016). Temporal changes in observer coverage and
population movements such as significant emigration or immigration of bigeye thresher from and
into the area, may cause hyperstability in the abundance indices despite high fishing pressure. There
is little information on the migratory patterns of bigeye thresher in the Pacific, although Matsunaga
and Yokawa (2013) suggested seasonal migrations in latitude, based on latitudinal segregation by
gender and life stage. Outputs from the spatiotemporal delta-GLMM model indicated higher local
densities in the Calibration Area in recent years, but no obvious decrease in density elsewhere in the
Assessment Area. If hyperstability in the Calibration Area is supported by immigration from other
areas, we might expect to see declines in catch rates in other areas sampled in the SPC or Japan
observer data. Such declines were not identified in the raw (unstandardized) data (see Appendix J)
but might still be present although obscured by seasonal effects and a lower sampling coverage. The
observer programme for the Hawaii-based pelagic longline was initiated in 1994, with observer
coverage ranging between 3% and 10% in 1994-2000, to a minimum of 20% beginning in 2001. The
deep-set fishery targeting tuna is currently observed at a minimum of 20% coverage and the
shallow-set fishery targeting swordfish has 100% observer coverage (Young et al. 2016).
Minimizing the process error in the BDM calibration resulted in lower catchability estimates and thus
lower fishing mortality and sustainability risk for the species. Increasing the process error to 0.10 had
the opposite effect. In the absence of process error, the model was forced to increase biomass
estimates in order to fit the relative abundance index, thus producing lower q values. Process error
reduction equates to assuming that the constructed catch series and CPUE index of abundance for
bigeye thresher in the Calibration Area are the only factors determining population abundance for
the species in the Pacific, and also implies that the underlying biomass dynamic equation is adequate
to describe the population trend in the presence of fishing with no margin of error. This is unlikely to
be the case, based on the reasons outlined above. The assessment outcomes therefore, were very
much dependent on the assumption that factors other than those provided by the available data are
affecting the population dynamics and catchability of the species at the scale of the Pacific. We have
included the influence of such factors as a 0.05 process error variance in the calibration model to
estimate catchability. The choice of process error can be contentious and the value of 0.05 was
loosely based on the range recommendation from McAllister (2013). Moderate to long-lived species
that are not expected to have much inter-annual recruitment variability could be expected to have
very low process error variance (M. McAllister, pers. comm.). Model selection techniques such as
those based on Bayesian factor can be used to discriminate the process error variance, but often
don’t work well when the data are not informative (for the same reason estimating the process error
variance when there are relatively uninformative catch and abundance index data should be
avoided, M. McAllister, pers. comm.).
The calibration process was also sensitive to the definition of prior bounds over K, which served to
constrain the catchability estimates to a range of biologically plausible values. The choice of prior
bounds remains subjective to some extent. The base case scenario (upper bound of 2 million) was
determined based on expert advice on maximum density values for the species in fishery hotspots.
The derivation of this upper bound assumed that relative densities per location were invariant at
different abundance levels (i.e., it is possible that when the population is unfished, densities
elsewhere will increase more than in the hotspot; and if the hotspot is a “preferred habitat” for
breeding or feeding, density is likely to be hyperstable, remaining high as abundance declines
elsewhere). Scenarios with wider bounds derived from blue shark assessments were also
considered, which resulted in much lower catchability estimates for bigeye thresher. But since blue
shark is a more productive and abundant species, the use of these bounds probably underestimated
Pacific-wide sustainability risk assessment of bigeye thresher shark 61
fishing impacts and sustainability risk for bigeye thresher. Longer time series of catch and CPUE data
and further work to validate and incorporate expert advice and opinion, would serve to improve
catchability estimation in pelagic sharks and enhance biological realism in fishing impact estimation.
The population status (current biomass relative to the unfished equilibrium state) of bigeye thresher
in the Pacific at the onset of the assessment (mid-1990s) is unknown. A range of initial population
status was assumed in catchability estimation and these scenarios were treated as equally likely.
Studies suggest that most pelagic sharks including bigeye thresher are highly vulnerable to pelagic
longline fisheries. The stock assessment of common thresher shark along the west coast of North
America indicated a stock status less than 30% of unfished level in the mid-1990s (the stock
subsequently recovered following management interventions, with stock status estimated at
approximately 94% of unfished level in 2014 (see Teo et al. 2015)). Assessment outcomes derived
assuming a low initial population status for bigeye thresher therefore, may represent the most
probable as well as precautionary scenario.
Relative catchability scaled to the 5 x 5 grid cells was assumed to be constant, meaning that the
potential effects of environmental fluctuations on catchability were ignored. Catchability is also
inversely proportional to the stock’s area (Paloheimo and Dickie 1964) however, differences in
effective population area among cells were not considered in this study (and assumed to be
equivalent). In multi-species fisheries, catchability is known to be influenced by target species, gear
configuration, and skipper experience (Polacheck 1991). Fishery group specific catchability was
incorporated in the assessment but was assumed to be constant for each fishery group over the
spatial domain of the assessment. Catchability may vary temporally as a result of fishing gear
improvements and Ward (2007) described methods for estimating changes in relative catchability
for a number of factors relating to target species in longline fisheries. As mentioned above, finer-
resolution catch effort data and detailed information on fishing practices and how these may change
over time are required to refine the estimation of fishery group catchability and quantify temporal
variations, and this would improve the estimation of fishing impact.
5.4 Post-capture survival
The assessment has investigated incorporating the occurrence of post-capture survival of bigeye
thresher in the estimation of fishing impact and sustainability risk. While the inclusion of post
capture survival resulted in lower risk estimates, we note that the ability of the assessment to
quantify the effect of post-capture survival was limited by the available data and the calibration
model and related assumptions, which required a two-step adjustment procedure of harvest rates
(constructed catch series) and posterior catchability estimates. The range of post-capture survival
rates considered in the assessment (0.3-0.7) was loosely based on fate and condition data available
from the SPC and US observer datasets. The true survival rate of bigeye thresher after release is
unknown and is likely to depend on the type of fishery operation and practices. In longline fisheries
bigeye thresher sharks are often hooked by the tail and “die soon afterward" (IOTC 2015; Gallagher
et al. 2014). Carvalho (2014) found that bigeye thresher was among the shark species with the
highest hooking mortality rates in the Portuguese pelagic longline fleet. The potential occurrence of
temporal changes in post-capture survival is also unknown. Assuming that 100% of captures result in
mortality therefore, represents a precautionary approach. On the other hand, better quantifying
uncertainty in the catch history resulting from capture mortality being potentially less than 100%
would help distinguish a more probable range of initial population status (i.e., high post-capture
survival and a constant environment would support a higher initial status assumption, and vice-
versa).
62 Pacific-wide sustainability risk assessment of bigeye thresher shark
5.5 Maximum impact sustainable threshold (MIST)
The MIST was defined based on population productivity parameters and is assumed to represent the
population’s ability to withstand fishing pressure. A single MIST value was applied in the assessment,
meaning that stock productivity and environmental conditions were assumed to have remained
constant over time (2000-2014). Consistent with the estimation of fishing mortality using a range of
scenarios, the MIST was defined as the maximum intrinsic population growth rate r for the species,
corresponding to the minimum unsustainable fishing mortality rate (Fcrash ) limit reference point
(Zhou et al. 2011). Risk-based fishing mortality limit reference points were recommended by Clarke
and Hoyle (2014) for data limited assessment of elasmobranchs.
The estimation of the maximum intrinsic growth rate r was performed using Monte Carlo sampling
of life history parameter distributions, with iterated solving of the Euler-Lotka equation. This method
has the advantage of incorporating uncertainty in all parameters. In this case, a higher uncertainty
(cv=0.20) was assigned to reproduction (maturation) and recruitment parameters and a lower
uncertainty (cv=0.10) was assigned to growth parameters. The rationale for this was that growth
parameters were obtained from direct observations (with sample sizes ≥100 specimens) while
reproduction and recruitment parameters were inferred from small sample sizes and in some cases
without observations (i.e., we assumed that females produce two pups per year although the exact
duration and frequency of the reproductive cycle is unknown). Uncertainty in natural mortality
(cv=0.17) was calculated across a range of values derived from multiple life history invariant
estimators. The principal limitations of the method are that density-dependent processes are
ignored and age-based processes are averaged across cohorts. Fecundity (as the product of the
number of recruits per spawner and the spawners biomass) was allowed to increase with age (as per
the maturity ogive and length-weight relationship) but survivorship was assumed constant across
age groups. Assumed density-independence implies that life history data for bigeye thresher in the
Pacific represent the maximum demographic values that would be achieved under ideal
environmental conditions (i.e., unlimited resources in the absence of fishing). As this is not the case,
the maximum growth rate r estimated in this study is probably an underestimate (Cortés 2016).
The mean value of the estimated r distribution for bigeye thresher (r=0.03) was higher than
previously reported for the species on the basis of demographic analyses using age-structured life
tables and Leslie matrices (median 0.01, 95% CI -0.006-0.025) (Cortés et al. 2002, 2012). Such low r
values were obtained assuming a slightly lower maximum age (Amax = 20 yr) and age-specific
mortality ranging from M=0.288 (at age-0) and from M=0.236 to M=0.094 in ages 1 to 20 (E. Cortés,
pers. comm.). Sensitivity testing revealed that the model-based approach implemented in our study
was highly sensitive to input values for recruitment parameters. Decreasing steepness to 0.25
and/or the maximum recruit per spawner to 0.88, produced an r distribution with mean=0.016,
which is within the range estimated by Cortés et al. (2002, 2012), however with increased
uncertainty (cv=0.95) and some estimates being less than zero (which is theoretically impossible
since we are estimating the maximum intrinsic growth r).
In a recent paper looking at the efficiency of multiple methods for maximum r estimation in shark
populations, Cortés (2016 and pers. comm.) obtained deterministic values for bigeye thresher that
ranged from 0.010 to 0.049, depending on the method. The paper recommended the Euler-Lotka
equation for estimation of the maximum intrinsic growth r for different degrees of data availability,
and to provide sensible advice for conservation and management in data-limited situations (Cortés
2016). Thus, the estimated r distribution for bigeye thresher in this study is assumed to constitute a
reliable and precautionary measure of population productivity for the species, given the available
data and uncertainty.
Pacific-wide sustainability risk assessment of bigeye thresher shark 63
Future work in terms of r estimation and MIST determination for bigeye thresher should focus on
environmental effects characterisation.
5.6 Sustainability risk
We used a scenario-based approach to evaluate sustainability risk for bigeye thresher in the Pacific,
with scenarios ranging from more to less precautionary and representing different species
distribution, initial population status, maximum density and post-capture survival assumptions. This
approach served to cope with high uncertainty in population status, movements and biology, as well
as inherent caveats in the available datasets.
Sustainability risk outcomes differed among scenarios and were notably sensitive to post-capture
survival, initial status and process error assumptions. The base case scenario, developed by fixing the
process error standard deviation at 0.05 and assuming a medium (mean 0.5) initial stock status with
100% capture mortality over the broader (Assessment Area) species distribution, produced a median
sustainability risk of 0.75 (range 0.29-2.24 among years) over the period 2000-2014. Including a
range of post-capture survival values (with uncertainty) reduced the median sustainability risk to
0.52 (range 0.15-1.89 among years) over the same period.
These results indicate that total impacts from pelagic longline fisheries in the Pacific since 2000 are
generally sustainable, but have exceeded the minimum unsustainable fishing mortality rate for
bigeye thresher in some years. The calculated probability that fishing impacts exceed the species
MIST, given the uncertainty, averaged 0.34 (range 0.13-0.67 among years) for base case scenarios
assuming 100% capture mortality, and 0.20 (range 0.06-0.43 among years) when accounting for the
potential occurrence of post-capture survival. We note that initial population status assumptions
were combined and treated as equally likely in sustainability risk probability calculations.
Uncertainty in key components of the risk assessment, namely species distribution, catchability and
life history traits, was included in impact and population growth rate estimation and propagated to
the evaluation of sustainability risk for the species across the Pacific. This represents a strength of
the quantitative risk assessment framework, permitting integration, characterisation (and
disaggregation) of uncertainty associated with the various datasets.
Impact was uncertain in the southwest Pacific. Effort levels were high in this area, but predicted
densities were low and highly uncertain, possibly as a result of limited observer coverage. This
resulted in asymmetric distribution of bigeye thresher around the equator in the southwest (see
Figure 10), which may not be biologically realistic. If densities (used to scale the F) are over- or
underestimated, this area may contribute higher or lower sustainability risk for the species in the
Pacific than currently estimated.
The risk-based, spatially-explicit and quantitative approach, served to integrate available data for
bigeye thresher into a comprehensive framework that permitted quantification of fishing impact in
relation to population resilience estimated from life history productivity parameters. Areas with
higher impact, and fishery sectors (catch group and seasons) contributing higher impact, were
identified. Fishing mortality was highest in April–June, and in the BET catch group in all years, while
the SWO fleet contributed minimal impact. Highest impact overlapped with the region of higher
relative abundance for the species, corresponding to a narrow latitudinal-longitudinal band between
10-15°N and 150°E-220°E. These results will assist with the development of more focused
management options for the species in the Pacific. Elasmobranchs are among the species most
vulnerable to overfishing, consequently their management should be precautionary (Zhou 2008).
The scenario-based approach implemented in this study, including the assumption of 100% capture
64 Pacific-wide sustainability risk assessment of bigeye thresher shark
mortality and estimation of sustainability risk relative to a higher mean population growth rate r
than previously reported for the species (yet potentially representing an under-estimate if density-
dependent mechanisms are affecting population dynamics for the species), contributed to ensure
that the results of the assessment include the most precautionary outcomes.
The conclusions that can be drawn from this assessment relate to the sustainable (or unsustainable)
character of current impact levels relative to population productivity, not population abundance. A
sustainability risk assessment is appropriate when data are insufficient to inform population
abundance estimation, as is the case for bigeye thresher in the Pacific. The spatially-explicit
approach shifts the assessment focus from poorly-informed abundance estimates to spatially-
explicit estimates of fishing mortality inferred using available information on species distribution and
on the occurrence, intensity and efficiency of fishing activities. The main strengths of this approach
include data integration, quantitative impact and population productivity estimation with
uncertainty, and the mapping of fishing impact in space and among fishery sectors.
5.7 Recommendations for future developments and implementations
The following extensions and developments will assist with improving sustainability status
evaluation (and minimising uncertainty) for bigeye thresher and other pelagic sharks in the Pacific
and elsewhere:
1. Weighting of the different datasets for proportional representation (i.e., differences in observer
coverage among areas) in spatial density estimation.
2. Further testing of the delta-GLMM model for species distribution estimation using fisheries-
dependent data, including simulation testing to evaluate model performance against variable
targeting behaviour and correlations among fishing events, and the inclusion of environmental
covariates to extrapolate species range beyond the fishery areas.
3. Sourcing of fine-scale resolution environmental covariates (e.g., sea surface temperature (SST),
ocean currents, wind and moon phases data) for inclusion in spatial and year effects
standardisations.
4. Exploration and testing of alternative methods for relative catchability estimation, including
simulation testing in data-rich fisheries and comparisons with catchability estimates derived
from full stock assessments.
5. Refinement of fishery groups definition to better account for differences in operational
practices affecting catchability for bigeye thresher in pelagic longline fisheries in the Pacific.
6. Estimate and incorporate uncertainty in catchability adjustment factors by fishery groups.
7. Evaluation of temporal changes in fishing patterns and practices that might cause changes in
catchability for bigeye thresher in pelagic longline fisheries over time.
8. Use available length data from recent years (i.e., last five years of observer datasets) to
investigate potential changes in population productivity (using life-history invariant methods on
median length).
Pacific-wide sustainability risk assessment of bigeye thresher shark 65
9. Seek further advice on initial status, biomass at unfished equilibrium and post-capture survival
assumptions, as this will serve to weight alternative scenarios and improve the accuracy of
impact and risk estimation.
Ultimately, there is a need for data enhancement for the species at the scale of the Pacific, including
tagging and tracking studies of neonates, juveniles and females, and sexing and ageing of bycatch
samples both spatially and seasonally; as this will improve the understanding of population structure,
movements, productivity and abundance mechanisms for the species. Longer CPUE time series based
on consistent reporting schedules will also serve to improve the accuracy of species distribution and
catchability estimation.
6 ACKNOWLEDGMENTS
We thank the following people and organisations for providing data or permission to access their
data:
Fisheries observers for collecting important data onboard fishing vessels
John Hampton, Peter Williams, Neville Smith, Manu Schneiter and Bruno Deprez (Pacific Community
(SPC))
Keith Bigelow, Daniel Luers, John D Kelly and Eric Forney (United States National Oceanic and
Atmospheric Administration)
Yujiro Akatsuka, Hiroaki Okamoto and Yasuko Semba (Japan Fisheries Agency and National Research
Institute of Far Seas Fisheries)
Alexandre Aires-da-Silva and Nick Vogel (IATTC)
James Larcombe (Australia Department of Agriculture and Water Resource)
Georgia Langdon & Ben Ponia (Cook Islands, Ministry of Marine Resources)
Eugene Pangelinan & Bradley Philip (Federated States of Micronesia, National Oceanic Resource
Management Authority)
Anare Raiwalui (Fiji Ministry of Fisheries and Forests)
Cedric Ponsonnet (French Polynesia Marine Resources and Mining Department)
Glen Joseph & Berry Muller (Marshall Islands Marine Resources Authority)
Regis Etaix-Bonneux (New Caledonia, Department of Maritime Affairs)
John Annala (New Zealand Ministry of Primary Industries)
Joyce Samuelu Ah-leong (Samoa Ministry of Agriculture and Fisheries)
Sylvester Diake (Solomon Islands Ministry of Fisheries and Marine Resources)
Vilimo Fakalolo (Tonga Fisheries Department)
William Naviti (Vanuatu Fisheries Department)
We wish to thank and acknowledge Dr. Charles Edwards for science advice, comments and revision
on an earlier version of this report.
We also thank Yasuko Semba (Japan Fisheries Agency and National Research Institute of Far Seas
Fisheries) and Felipe Carvalho and Donald Kobayashi (United States National Oceanic and
Atmospheric Administration) for useful comments and suggestions on the final report.
This study was funded through an ABNJ (Common Oceans) Tuna Project supported WCPFC contract
to NIWA. Additional funding for methods development was provided under NIWA core funding.
66 Pacific-wide sustainability risk assessment of bigeye thresher shark
7 REFERENCES
Amorim, A. et al. 2009. Alopias superciliosus. The IUCN Red List of Threatened Species 2009:
e.T161696A5482468. http://dx.doi.org/10.2305/IUCN.UK.2009-2.RLTS.T161696A5482468.en
Bentley, N., Kendrick, T. H., Starr, P. J., Breen, P. A. (2011) Influence plots and metrics: tools for
better understanding fisheries catch-per-unit-effort standardisations. ICES Journal of Marine
Science, 69: 84-88.
Braccini, J. M., Gillanders, B. M., & Walker, T. I. 2006. Hierarchical approach to the assessment of
fishing effects on non-target chondrichthyans: case study of Squalus megalops in southeastern
Australia. Canadian Journal of Fisheries and Aquatic Sciences, 63, 2456-2466.
Carvalho, J.F.D. (2014). Population dynamics and fisheries assessment of the bigeye thresher
(Alopias superciliosus) in the Atlantic: a comparison between North Atlantic and South Atlantic
stocks. PhD Thesis.
Chen, C.-T., Liu, K.-M., Chang, Y.-C. 1997. Reproductive biology of the bigeye thresher shark, Alopias
superciliosus (Lowe, 1839) (Chondrichthyes: Alopiidae), in the northwestern Pacific. Ichthyological
Research 44, 227-235.
Chen, P., Yuan, W. 2006. Demographic analysis based on the growth parameter of sharks. Fisheries
Research 78, 374-379.
Clarke, S. 2009. An alternative estimate of catches of five species of sharks in the Western and
Central Pacific Ocean based on shark fin trade data. Western and Central Pacific Fisheries
Commission, Scientific Committee Paper SC5/EB-WP-02.
Clarke, S. 2011. A status snapshot of key shark species in the Western and Central Pacific and
potential management options. WCPFC-SC7-2011/EB-WP-04.
Clarke, S., S. Harley, S. Hoyle and J. Rice. 2011a. An indicator-based analysis of key shark species
based on data held by SPC-OFP. WCPFC-SC7-2011/EB-WP-01. Accessed online
at https://www.wcpfc.int/node/2766
Clarke, S., Yokawa, K., Matsunaga, H. and Nakano, H. 2011b. Analysis of North Pacific shark data
from Japanese commercial longline and research/training vessel records. WCPFC-SC7-2011/EB-WP-
02.
Clarke S., M. Sato, C. Small, B. Sullivan, Y. Inoue and D. Ochi. 2014a. Bycatch in longline fisheries for
tuna and tuna-like species: A global review of status and mitigation measures. FAO Fisheries
Technical Paper 588. Rome. 199 p.
Clarke, S., L. Manarangi-Trott and S. Brouwer. 2014b. Issues for t-RFMOs in relation to the listing of
shark and ray species by the Convention on International Trade in Endangered Species (CITES).
WCPFC-SC10-2014/EP-IP-05. Accessed online at http://www.wcpfc.int/system/files/EB-IP-
05%20(CITES%20%20Sharks).pdf
Clarke, S., R. Coelho, M. Francis, M. Kai, S. Kohin, K.M. Liu, C. Simpfendorfer, J. Tovar-Avila, C. Rigby
and J. Smart. 2015. Report of the Pacific Shark Life History Expert Panel Workshop, 28-30 April 2015.
WCPFC-SC11-2015/EB-IP-13.
Pacific-wide sustainability risk assessment of bigeye thresher shark 67
Clarke S. and Hoyle, S. 2014. Development of limit reference points for elasmobranchs. WCPFC-
SC10-2014/ MI-WP-07. 43 pp.
Compagno, L.J.V. 2001. Sharks of the world. An annotated and illustrated catalogue of shark species
known to date. Volume 2. Bullhead, mackerel and carpet sharks (Heterodontiformes, Lamniformes
and Orectolobiformes). FAO, Rome.
Compagno, L., Dando, M. and Fowler, S. 2005. A field guide to the sharks of the world. Harper Collins
Publishers, London, UK. 368 pp.
Cortés, E. 2002. Incorporating uncertainty into demographic modeling: application to shark
populations and their conservation. Conservation Biology 16, 1048-1062.
Cortés, E. 2008. Comparative life history and demography of pelagic sharks. In: Sharks of the Open
Ocean: Biology, Fisheries and Conservation. (Eds. M.D. Camhi, E.K. Pikitch, E.A. Babcock), Blackwell
Publishing Ltd, pp. 309-320.
Cortés, E. 2016. Perspectives on the intrinsic rate of population growth. Methods in Ecology and
Evolution. doi: 10.1111/2041-210X.12592
Cortés, E., Arocha, F., Beerkircher, L., et al. 2010. Ecological risk assessment of pelagic sharks caught
in Atlantic pelagic longline fisheries. Aquatic Living Resources 23, 25-34.
Cortés, E., Domingo, A., Miller, P., et al. 2012. Expanded ecological risk assessment of pelagic sharks
caught in Atlantic pelagic longline fisheries. SCRS/2012/167, 1-56.
Edwards, C.T.T. 2016. BDM: Bayesian Biomass Dynamic Model,
URL https://github.com/cttedwards/bdm.git
Fernandez-Carvalho, J., R. Coelho, J. Mejuto, E. Cortés, A. Domingo, K. Yokawa, K.M. Liu, B. García-
Cortés, R. Forsellado, S. Ohshimo, A. Ramos-Cartelle, W.P. Tsai, M.N. Santos. 2015. Pan-Atlantic
distribution patterns and reproductive biology of the bigeye thresher, Alopias superciliosus. Reviews
in Fish Biology and Fisheries 25: 511-568.
Gallagher, A. J., Kyne, P. M., & Hammerschlag, N. 2012. Ecological risk assessment and its application
to elasmobranch conservation and management. Journal of Fish Biology 80, 1727-1748.
Gallagher, A. J., Orbesen, E.S., Hammerschlag, N., Serafy, J.E. (In press). Vulnerability of oceanic
sharks as pelagic longline bycatch. Global Ecology and Conservation.
Griffiths, S. P., Brewer, D. T., Heales, D. S., Milton, D. A. & Stobutzki, I. C. 2006. Validating ecological
risk assessments for fisheries: assessing the impacts of turtle excluder devices on elasmobranch by-
catch populations in an Australian trawl fishery. Marine and Freshwater Research 57, 395–401.
Hoyle, S.D., Okamoto, H., Yeh, Y.-M., Kin, Z.G., Lee, S.I., Sharma, R. 2015. IOTC-CPUEWS02: Report of
the 2nd CPUE Workshop on Longline Fisheries, 30 April- 2 May 2015. Indian Ocean Tuna Commission;
2015.
ICCAT (International Convention for the Conservation of Atlantic Tunas). 2012. 2012 Shortfin Mako
Stock Assessment and Ecological Risk Assessment Meeting. Olhão, Portugal – June 11-18, 2012.
Accessed online at https://www.iccat.int/Documents/Meetings/Docs/2012_SHK_ASS_ENG.pdf
68 Pacific-wide sustainability risk assessment of bigeye thresher shark
ICCAT (International Convention for the Conservation of Atlantic Tunas). 2015. SCRS Executive
Summary Sharks. Summary of ecological risk assessment results. Accessed online
at https://www.iccat.int/Documents/SCRS/ExecSum/SHK_EN.pdf
IOTC (Indian Ocean Tuna Commission). 2012. Ecological Risk Assessment. IOTC Paper 2012-SC15-
INF10 (rev 1). Accessed online
at http://iotc.org/sites/default/files/documents/proceedings/2012/sc/IOTC-2012-SC15-
INF10%20Rev_1.pdf.
Kristensen, K., Thygesen, U. H., Andersen, K. H., and Beyer, J. E. 2014. Estimating spatio-temporal
dynamics of size-structured populations. Canadian Journal of Fisheries and Aquatic Sciences, 71,
326-336.
Lawson, T. 2011. Estimation of catch rates and catches of key shark species in tuna fisheries of the
Western and Central Pacific Ocean using observer data. WCPFC-SC7-2011/EB-IP-02.
Liu, K.-M., Chiang, P.-J., Chen, C.-T. 1998. Age and growth estimates of the bigeye thresher shark,
Alopias superciliosus, in northeastern Taiwan waters. Fishery Bulletin 96, 482-491.
Liu, K.M., Chang, Y.T., Ni, I.H. and Jin, C.B. 2006. Spawning per recruit analysis of the pelagic thresher
shark, Alopias pelagicus, in the eastern Taiwan waters. Fisheries Research 82, 52-64.
Matsunaga, H. and K. Yokawa. 2013. Distribution and ecology of bigeye thresher Alopias
superciliosus in the Pacific Ocean. Fisheries Science 79, 737-748.
McAllister, M.K. 2013. A generalized Bayesian surplus production stock assessment software (BSP2).
ICCAT SCRS/13/100.
McAllister, M. K., Pikitch E. K., and Babcock E. A. 2001. Using demographic methods to construct
Bayesian priors for the intrinsic rate of increase in the Schaefer model and implications for stock
rebuilding. Canadian Journal of Fisheries and Aquatic Sciences 58, 1871-1890.
McAllister, M.K., Babcock, E.A., Pikitch, E.K. and Prager, M.H., 2000. Application of a non-equilibrium
generalized production model to South and North Atlantic swordfish: combining Bayesian and
demographic methods for parameter estimation. Col. Vol. Sci. Pap. ICCAT, 51(5), pp.1523-1550.
Musyl, M.K., Brill, R.W., Curran, D.S., et al. 2011. Postrelease survival, vertical and horizontal
movements, and thermal habitats of five species of pelagic sharks in the central Pacific Ocean.
Fishery Bulletin 109, 341-368.
Nakano, H (1994). Age, reproduction and Migration of blue shark in the North Pacific Ocean. Bull.
Nat. Res. Inst. Far Seas Fish. No 31.
Nakano, H., Matsunaga, H., Okamoto, H., Okazaki, M. 2003. Acoustic tracking of bigeye thresher
shark Alopias superciliosus in the eastern Pacific Ocean. Marine Ecology Progress Series 265, 255-
261.
NMFS-NOAA. 2015. Status review report: common thresher (Alopias vulpinus) and bigeye thresher
(Alopias superciliosus) sharks. 199 pp.
Pacific-wide sustainability risk assessment of bigeye thresher shark 69
Polacheck, T. (1991). Measures of effort in tuna longline fisheries: changes at the operational level.
Fisheries Research 12:75–87.
Rice, J., L. Tremblay-Boyer, R. Scott, S. Hare and A. Tidd. 2015. Analysis of stock status and related
indicators for key shark species of the Western Central Pacific Fisheries Commission. WCPFC-SC11-
2015/EB-WP-04-Rev 1. Accessed online at https://www.wcpfc.int/system/files/EB-WP-
04%20shark%20indicators%20Rev%201.pdf
Rice, J., Harvey, S. 2012. Stock assessment of oceanic whitetip sharks in the western and central
Pacific Ocean. WCPFC-SC8-2012/SA-WP-06 Rev 1.
Rice, J., Harvey, S. 2013. Updated stock assessment of silky sharks in the western and central Pacific
Ocean. WCPFC-SC9-2013/ SA-WP-03.
Rice, J., Harvey, S., Kai, M. 2013. Stock assessment of blue shark in the North Pacific Ocean using
stock synthesis. WCPFC-SC10-2014/ SA-WP-08.
Rice, J., L. Tremblay-Boyer, R. Scott, S. Hare and A. Tidd. 2015. Analysis of stock status and related
indicators for key shark species of the Western Central Pacific Fisheries Commission Rev 1. WCPFC-
SC11-2015/EB-WP-04. Accessed online at https://www.wcpfc.int/node/21719
Smith, S.E., David, W.A., Christina, S. (2008). Intrinsic rates of increase in pelagic Elasmobranchs. In
Sharks of the Open Ocean: Biology, Fisheries and Conservation, 288-97. Oxford: Blackwell Science,
2008.
Stobutzki, I. C., Miller, M. J., Heales, D. S., & Brewer, D. T. 2002. Sustainability of elasmobranchs
caught as bycatch in a tropical prawn (shrimp) trawl fishery. Fishery Bulletin 100, 800-821.
Thorson, J. T., Shelton, A. O., Ward, E. J., & Skaug, H. J. 2015. Geostatistical delta-generalized linear
mixed models improve precision for estimated abundance indices for West Coast groundfishes. ICES
Journal of Marine Science: Journal du Conseil, doi: 10.1093/icesjms/fsu243.
Thorson, J.T., Ward, E.J., 2014. Accounting for vessel effects when standardizing catch rates from
cooperative surveys. Fish. Res. 155, 168–176. doi:10.1016/j.fishres.2014.02.036
Thorson, J.T., Skaug, H.J., Kristensen, K., Shelton, A.O., Ward, E.J., Harms, J.H., Benante, J.A., 2014.
The importance of spatial models for estimating the strength of density dependence. Ecology 96,
1202–1212. doi:10.1890/14-0739.1. URL: http://www.esajournals.org/doi/abs/10.1890/14-0739.1
Trejo, T. 2005. Global Phylogeography of Thresher Sharks (Alopias spp.) Inferred from Mitochondrial
DNA Control Region Sequences.” Thesis, California State University, Monterey
Bay. http://digital.mlml.calstate.edu/islandora/object/islandora%3A1682/datastream/OBJ/downloa
d/Global_phylogeography_of_thresher_sharks__Alopias_spp___inferred_from_mitochondrial_DNA
_control_region_sequences.pdf
Teo, S.L.H., E. Garcia Rodriguez and O. Sosa-Nishizaki. 2016. Status of Common Thresher Sharks,
Alopias vulpinus, along the West Coast of North America. NOAA Technical Memorandum (NOAA-TM-
NMFS-SWFSC-557), Southwest Fisheries Science Center, La Jolla, California. 198 pp.
70 Pacific-wide sustainability risk assessment of bigeye thresher shark
Tsai, W.P., Liu, K.M. and Joung, S.J. 2010. Demographic analysis of the pelagic thresher shark, Alopias
pelagicus, in the north-western Pacific using a stochastic stage-based model. Marine and Freshwater
Research 61: 1056-1066.
WCPFC (Western and Central Pacific Fisheries Commission). 2006. WCPFC Second Scientific
Committee (Manila, Philippines), Summary Report. Summary of ecological risk assessment results.
Accessed online at https://www.wcpfc.int/meetings/2nd-regular-session
WCPFC (Western and Central Pacific Fisheries Commission). 2012. Process for designating WCPFC
key shark species for data provision and assessment. Accessed online
at https://www.wcpfc.int/system/files/Key-Doc-SC-08-Process-Designation-Key-WCPFC-Shark-
Species.pdf
WCPFC (Western and Central Pacific Fisheries Commission). 2014. Stock Assessment and Future
Projections of Blue Shark in the North Pacific Ocean. Accessed online
at https://www.wcpfc.int/node/19204
Ward, P. (2006). Preliminary Estimates of Historical Variations in the Fishing Power and Catchability
of Pelagic Longline Fishing Gear. WCPFC-SC3-ME SWG/WP-7. Accessed online at
https://www.wcpfc.int/system/files/SC2_FT_WP1.pdf
Weng, K.C. and B.A. Block. 2004. Diel vertical migration of the bigeye thresher shark (Alopias
superciliosus), a species possessing orbital retia mirabilia. Fishery Bulletin 102: 221-229.
Young, C.N., J. Carlson, M. Hutchinson, D. Kobayashi, C. McCandless, M.H. Miller, S. Teo and T.
Warren. 2016. Status review report: common thresher shark (Alopias vulpinus) and bigeye thresher
shark (Alopias superciliosus). Final Report to National Marine Fisheries Service, Office of Protected
Resources. March 2016. 199 pp.
Zhou S. and Griffiths S.P. 2008. Sustainability assessment for fishing effects (SAFE): a new
quantitative ecological risk assessment method and its application to elasmobranch bycatch in an
Australian trawl fishery. Fisheries Research 91, 56-68.
Zhou, S., Smith, A. D., & Fuller, M. 2011. Quantitative ecological risk assessment for fishing effects on
diverse data-poor non-target species in a multi-sector and multi-gear fishery. Fisheries Research 112,
168-178.
Zhou, S., Klaer, N.L., Daley, R.M., Zhu, Z., Fuller, M., Smith, A.D.M. 2014. Modelling multiple fishing
gear efficiencies and abundance for aggregated populations using fishery or survey data. ICES
Journal of Marine Science: Journal du Conseil. doi:10.1093/icesjms/fsu068.
Zuur, A. F., Ieno, E. N., Walker, N. J., Saveliev, A. A., and Smith, G. M. 2009. Mixed effects models and
extensions in ecology with R; ed. Gail M, Krickeberg K, Samet JM, Tsiatis A, Wong W. New York, NY:
Spring Science and Business Media.
Pacific-wide sustainability risk assessment of bigeye thresher shark 71
Appendix A – CPUEΩ estimation and derivation
In this section we derive (the CPUE index for the Calibration Area AΩ) from (CPUE in
subarea a within AΩ).
is used to calibrate a range of plausible values for using BDM. is the predicted
catch rate for subarea a from the final “leffort” ZINB standardisation model (see section .
Firstly,
∑
∑
=
Ω
Ωa
a
a
aa
n
n
qEEq
Where is the catchability scalar over AΩ and q is the average catchability in each subarea a.
and are the total effort and abundance in subarea a, respectively, and is total abundance in
AΩ.
Assuming that CPUE index is proportional to abundance implies that:
=
=
Therefore,
∑∑
∑
Ω
Ω
Ω
Ω=
=
aaa
a
a
a
aa
CPUEE
En
n
qE
E
n
CPUE 1
72 Pacific-wide sustainability risk assessment of bigeye thresher shark
Appendix B – BDM description
The Biomass dynamic model (BDM) developed by Edwards (2016) implements the Fletcher-Schaefer
hybrid model proposed by McAllister et al. (2000) in a state-space modelling framework that
describes changes in stock depletion in response to fishing,:
00
µε
=x
0
for =t
(1)
)
2
1
exp(
2
12
1
1
11 ptt
x
ttt H
x
rxxx
σε
ϕ
−
−
−+= −
−
−−
ϕ
<>
t
xt and
0
for
(2)
( )
( )
)
2
1
exp(1
2
12
1
1
111 ptt
n
tttt Hxxgr
xx
σεϕ
−
−−+= −
−
−−−
ϕ
≥>
t
xt and 0for
(3)
where
t
x
is the depletion in year
t
(abundance as a percent of unfished equilibrium abundance);
u
is the initial biomass;
ϕ
is the depletion level at which Maximum Sustainable Yield occurs,
which is controlled by a shape parameter
n
, and
1
1
1
−
=
n
n
ϕ
(4)
1
1
−
=
−
n
n
g
n
n
(5)
r
is the intrinsic growth rate.
t
H
is the harvest rate in year t, and
K
C
H
t
t
=
(6)
where
t
C
is the catch is year t and
K
is the unfished equilibrium abundance,
t
ε
is the process error
in year
t
following a normal distribution:
( )
2
,0 normal~
p
t
σε
(7)
p
σ
is the standard deviation for the process error. The expected abundance index in year
t
,
t
I
ˆ
is
calculated as,
)
2
1
exp(
ˆ
2
ottt
qKxI
σξ
−=
(8)
Pacific-wide sustainability risk assessment of bigeye thresher shark 73
Where
q
is the catchability coefficient and
t
ξ
is the observation error in year
t
, and
),0( normal~
2
o
t
σς
(9)
Where
o
σ
is the standard deviation for observation errors.
The hybrid model allows
K5.0<
ϕ
whilst maintaining an ecologically consistent interpretation of
r
.
Using a Bayesian framework, BDM estimates the marginal posterior distribution of underlying
parameters including
K
,
r
, and
q
, by incorporating time series of catches and observed abundance
indices.
74 Pacific-wide sustainability risk assessment of bigeye thresher shark
Appendix C – q adjustment by fishery groups and spatial scaling
In this section, we derive
j
q
, the catchability for fishery group
j
at the level of 5x5 degree cells used
in the assessment. Firstly,
jj
qfq =
(1)
where
q
is the average catchability on the grid cell (constant across spatial domain) and
j
f
is the
adjustment factor for fishery group
j
, calculated as the predicted CPUE for each fishery group
(averaged over space and time) relative to a reference fishery group (i.e., catch group of “BET” in
February).
To obtain the qj, we first write the fishing mortality in the Calibration Area,
Ω
F
, as
∑
ΩΩ =
ji ij
EqF ,
(2)
where
Ω
q
is the catchability over AΩ,
ji
E,
is the fishing effort for fishery group
j
in grid cell
i
. Using
a spatially-explicit approach:
( )
∑∑
=
Ω
Ωi j jij
iEq
n
n
F,
(3)
Where
i
n
is the abundance (relative density) in cell
i
and
Ω
n
is the total relative abundance in the
Calibration Area AΩ
Combining (1), (2), and (3) we obtain:
( )
j
ijji
j
i
ji ij
jf
E
f
n
n
Eq
q
∑∑
∑
=
Ω
Ω
,
,
(4)
Pacific-wide sustainability risk assessment of bigeye thresher shark 75
Appendix D – Observer data characterisation
Only a few bigeye thresher captures were recorded outside the Core Area of the Assessment Area
from 2000-2014 (Figure D1-a). Observer coverage (number of observed hooks) was also
comparatively lower outside the Core Area, but higher in 2013-2014 (Figure D1-c). The number of
captures peaked in March, April and May. Higher catches were also observed in June and in
November-December (Figure D1-b).
Figure D1: Proportion of observed bigeye thresher (BTH) captures between the Core Area and non-core
areas of the Assessment Area by (a) year and (b) month (2000-2014); and (c) proportion of observed hooks
(1995-2014).
The number of captures per set was highly skewed towards zero and 1. Of all the observed sets in
the Assessment Area, 86.4% did not catch bigeye thresher. Of the remaining 13.6% of sets that
reported a positive catch, 70% caught one specimen, 26% caught between 2 and 9, and less than 2%
caught ≥10 (Figure D2-a). The maximum observed number of captures per set was 94 specimens.
Catch per set was not related to the number of hooks (Figure D2-b), suggesting that the number of
captures per set could be an adequate measure of catch-per-unit-effort (CPUE) for bigeye thresher.
(a)
(b)
(c)
76 Pacific-wide sustainability risk assessment of bigeye thresher shark
Fifty percent of all observed captures within the Core Area from 2000 to 2014 were located within
five (5x5 degree) cells and 80% were located within 17 cells (Figure D3).
Figure D2: Numbers of bigeye thresher (BTH) captures in observed pelagic longline sets that reported a catch
of BTH, 2000-2014. (a) Frequency distribution of BTH captures per set among observer datasets. (b)
Variation in the number of observed hooks per set relative to the number of BTH captures in each set.
Figure D3: Proportions of the total observed catch of bigeye thresher (BTH) by 5x5 degree latitude/longitude
grid cells of the Core Area, 2000
−
2014. Grid cells are ordered by catch. Cell numbers correspond with cell IDs
as mapped in Figure 1. Vertical lines indicate cumulative catch proportions of 50% and 80% respectively.
(a)
(b)
Pacific-wide sustainability risk assessment of bigeye thresher shark 77
Appendix E – Spatial standardisations
ZINB models
Results of likelihood ratio tests for nested ZINB models are presented in Table E1. Predicted
densities in 5x5 degree cells of the Core Area are compared among the different models in Figure E1.
Spatial indices were similar between models, suggesting most explanatory variables had little effects
on spatial relative abundance at the level of 5x5 degree cells.
Relationships between explanatory variables and encounter probabilities and catch rates were
generally weak (Figure E2). The model including SST had the lowest AIC (80 363) but predicted a
monotonic relationship between catch rates and SST without changing the indices. The hbf model
including cell_id, year, month, catch group and HBF gave the next best fit (AIC 80 540). Effort (log no.
of hooks) was included as an offset term to predict relative densities as the number of captures per
1000 hooks in 5x5 degree cells. This model (‘leffort’) was selected as the final ZINB model and used
to predict spatial abundance indices for bigeye thresher within the Core Area.
Model fits are shown in Figure E3. Residual patterns were investigated further using a simulation
exercise. The simulated catch in each observed set (including zeros and counts) were generated
from the predicted distribution of BTH catch from the final ZINB model, and the model was then
fitted to the simulated catch using the same set of covariates. The residuals patterns from the
simulated data were consistent with that in the original data (Figure E3-b).
Fitted encounter probabilities matched the observed proportions of zero sets (Figure E3-a). The
predicted number of sets that caught two or more BTH were higher than observed values (Figure E3-
b), indicating that sets with occasionally high capture rates were not well fitted by the model. At the
level of 5x5 degree cells however, predicted encounter probabilities and catch rates were consistent
with observed values (Figure E3-c and E3-d).
Table E1: Summary of the ZINB models fitted to observer data 2000
−
2014 in Core Area. The same variables
are included in both components of the model. The likelihood ratio test is performed to nested models
sequentially (e.g. between model “year” and model “id”). Model “leffort” is the final model. “df”, degree of
freedom; “Pr(>Chisq)”, P value from the likelihood ratio test.
Model
Variables
df
AIC
Pr(>Chisq)
id
Cell_id
125
81885
-
year
Cell_id+year
153
81277 <2.2E-16
month
Cell_id+year+ns(month,3)
159
80666
<2.2E-16
kmeans
Cell_id+year+ns(month,3)+catch_group
167
80635 1.45E-07
hbf
Cell_id+year+ns(month,3)+ catch_group+ns(hbf,3)
173 80540 <2.2E-16
sst
Cell_id+year+ns(month,3)+ catch_group+ns(hbf,3) + ns(sst,3)
179 80363 <2.2E-16
leffort
Cell_id+year+ns(month,3)+ catch_group+ns(hbf,3) + offset (leffort)
173
80575
-
78 Pacific-wide sustainability risk assessment of bigeye thresher shark
Figure E1: Predicted indices of relative abundance for BTH in 5x5 degree cells of the Core Area, as estimated
using a series of ZINB models fitted to different combinations of explanatory variables (see Table 1 for each
model specifications).
Pacific-wide sustainability risk assessment of bigeye thresher shark 79
(a)
(b)
(c)
(d)
Figure E2: Relationships between encounter probabilities (in logit space) and BTH catch rates (in log space) in
Core Area cells (2000
−
2014) as related to (a) catch group; (b) HBF (shallow sets (HBF <15) and deep sets (HBF
≥15)); (c) fishing duration at night (hours in darkness); and (d) wire trace (“U” represents unknown).
Residuals diagnostics for the final ZINB model are presented in Figure E4. Predicted catch rates by
explanatory variables are shown in Figure E5. Highest catch rates occurred between March and May
and again in November-December, and were associated with catch group ‘BET’.
80 Pacific-wide sustainability risk assessment of bigeye thresher shark
(a)
(b)
(c)
(d)
Figure E3: Predicted vs observed encounter probabilities and numbers of BTH per set for the final ZINB model
“leffort” fitted to observer data in the Core Area, 2000
−
2014. (a) overall proportion of sets with zero catch
(the bars on the left) and overall proportion of sets with positive catch); (b) distribution of catch per set; (c)
encounter probabilities in each grid cell; (d) catch per set for in each grid cell.
Pacific-wide sustainability risk assessment of bigeye thresher shark 81
(a)
(b)
Figure E4: Residuals diagnostics for (a) the final ZINB (leffort) model fitted to Core Area observer data for the
period 2000-2014; and (b) the same model fitted to simulated BTH catch data based on the predicted catch
distribution derived from the final model. Pearson Residuals vs fitted values (top-left panel in both (a) and
(b)); Observed catch vs fitted catch (top-right); Pearson Residuals by 5x5 degree cell (bottom-left); Pearson
Residuals by year (bottom-right).
82 Pacific-wide sustainability risk assessment of bigeye thresher shark
Figure E5: Predicted CPUE (number of BTH per 1000 hooks) by covariate for the final ZINB model “leffort”
fitted to observer data 2000
−
2014 in Core Area.
Geostatistical delta-GLMM models
Estimated variance parameters for the three fitted delta-GLMM models are presented in Table E2.
Spatial variability in encounter probabilities was greater than the spatial variability in positive catch
rates. In the spatial model, density was correlated over a longer distance on the longitudinal axis
(East-West plane) than on the North-South axis (Figure E6).
Pacific-wide sustainability risk assessment of bigeye thresher shark 83
Table E2: Estimated variance parameters for the three geostatistical delta-GLMM models (spatial,
spatiotemporal and core vessels) fitted to observer data from the Assessment Area, 2000
−
2014. 5
Random fields (marginal SD)
Vessel effects
Model
)
(p
ε
σ
)
(r
ε
σ
)( p
ϖ
σ
)
(r
ϖ
σ
)
(p
v
σ
)(r
v
σ
spatial
-
-
1.16
0.42
0.81
0.25
spatiotemporal
0.70
0.34
0.76
0.25
0.71
0.22
core vessels
-
-
1.05
0.43
0.49
0.24
Figure E6:Ellipses representing estimates of geometric anisotropy (where spatial correlation will have
dropped to 10%) for encounter probabilities and positive catch rates of bigeye thresher estimated using the
spatial delta-GLMM model fitted to the observer data from the Assessment Area, 2000
−
2014.
Random vessel effects are plotted for each model in Figure E7. The dataset included a total of 849
vessels, of which 314 did not catch BTH (mostly SPC vessels). Random effects were estimated to be
nil (zero) for these vessels (Figure E7). Excluding zero catch vessels (core vessels model) did not
affect the estimated variability in catch rates but reduced the spatial variability in encounter
probabilities (Table E2).
5
)( p
ε
σ
and
)(r
ε
σ
are the marginal standard deviations for the spatiotemporal random fields effects on
encounter probabilities and positive catch rates, respectively.
)( p
ϖ
σ
and
)(r
ϖ
σ
are for spatial effects; and
)( p
v
σ
and
)(r
v
σ
are for random vessel effects.
84 Pacific-wide sustainability risk assessment of bigeye thresher shark
Figure E7: Estimated vessel effects (with 95% confidence intervals) on encounter probabilities (top) and
positive catch rates (bottom), as estimated in (a) the spatial delta-GLMM model; and (b) the core vessels
delta-GLMM model. Effects are distinguished by observer dataset: “Red” is US observer data, “Green” is
“SPC” data, “Blue” is for Japanese data.
(a)
(b)
Pacific-wide sustainability risk assessment of bigeye thresher shark 85
Appendix F – q calibration Results and sensitivities
Time series of abundance and predicted abundance indices associated with the range of
Ω
q
values
used in the assessment are shown in Figure F1. BDM-estimated depletion showed a similar trend
among initial biomass level assumptions (Figure F1−a). Predicted abundance indices were consistent
with the fitted index (Figure F1-b).
Posterior estimates for
K
with alternative prior bounds are shown in Figure F2. The use of bounds
based on blue shark assessments (WCPFC 2016) resulted a much wider range of
K
being retained,
but with decreasing probabilities for higher values (Figure F2-a). The mode of the posterior
distributions was similar among the three assumed upper bound values. Posterior estimates for r
were similar for these runs (Figure F2-b) and retained a range of values of r similar to the
constructed prior (shown in Figure 23, section 4.4) (Figure F2).
86 Pacific-wide sustainability risk assessment of bigeye thresher shark
(a)
(b)
Figure F1: Estimated time series of (a) depletion and (b) predicted (ribbon) vs fitted (line and dots)
abundance indices for the three initial stock status assumptions (BDM runs 1, 2, and 3 in Table F1). The
ribbon represents the 95 confidence intervals for each scenario.
Pacific-wide sustainability risk assessment of bigeye thresher shark 87
(a)
(b)
Figure F2: (a) Posterior estimates of K obtained in BDM calibration and (b) posterior estimates of the
maximum intrinsic population growth rate r estimated assuming a medium initial stock status (0.5) and
three different upper bound values for the prior over K (2 million (base case), 6 million and 16 million) (BDM
runs 2, 2a, and 2b in Table F1 ).
Table F1 summarises the results of BDM sensitivities on the estimated range of plausible values for
Ω
q
. Increasing the upper bound of the prior over ln(K) to 6 million and 16 million produced a 45%
reduction and 65% reduction in median estimates of qΩ , respectively (BDM runs 2a and 2b in Table
F1). Estimates of K derived using upper bound values of 6 million and 16 million corresponded to a
median population density of over 1.4 and 2.2 sharks per km2 in fishery hotspots, respectively, much
higher than the median density of 0.8 per km2 corresponding to the base case scenario (upper
bound value of 2 million) (Table F1) .
The calibration was sensitive to the inclusion of process error (BDM runs 2c, 2d Table F1). Reducing
process error to 1% produced an 18% reduction in median estimates of qΩ (BDM run 2c, Table F1).
Increasing process error to 10% produced an increase in median estimates of qΩ of 15% (BDM run
2d, Table F1). The implications of process error assumptions are considered in the discussion
(section 5).
Assuming a lower prior distribution for the maximum intrinsic population growth rate r (with mean
0.009 and cv 0.1) resulted in a 12% decrease in median estimates of maximum
Ω
q
. A higher r (prior
mean 0.065 and cv 0.1) resulted in a 24% increase in median
Ω
q
. Catchability estimates were not
particularly sensitive to the shape of the production function: assuming
3.0=
ϕ
resulted in 5%
decrease in median
Ω
q
, while the
5.0=
ϕ
model resulted in 2% increase in median
Ω
q
.
Variability in q values adjusted for fishery groups (catch group and season) are shown in Tables F2
and Table F3, respectively.
88 Pacific-wide sustainability risk assessment of bigeye thresher shark
Table F1: Comparison of estimated range of
Ω
q
(median and 95% quantile range) for a set of BDM sensitivities representing different assumptions on initial stock status
(relative to the unfished biomass at equilibrium) (runs 1-3); whether post-capture survival was accounted for (1s, 2s, and 3s); K prior upper bound (2a, 2b); process
errors (2c, 2d), intrinsic growth rate prior mean (2e, 2f), and model configurations (shape of the production function) (2g, 2h).
Run
Initial status
(
u
)
Intrinsic growth
rate (
r
)
Depletion at MSY
(
ϕ
)
Process error
(
p
σ
)
Bounds of K prior
Post-capture
survival
accounted for
Ω
q
(
5
10
−
) number per Km2
(calibration area)
number per Km2
(hotspots)
1
Low
0.033
0.4
0.05
[3x104, 2x106]
No
0.5 (0.27-0.96)
0.15 (0.08-0.21)
0.94 (0.49-1.3)
2
Medium
0.033
0.4
0.05
[3x104, 2x106]
No
0.34 (0.19-0.77)
0.13 (0.06-0.21)
0.82 (0.37-1.28)
3
High
0.033
0.4
0.05
[3x104, 2x106]
No
0.26 (0.14-0.66)
0.12 (0.05-0.21)
0.78 (0.32-1.3)
1s
Low
0.033
0.4
0.05
[3x10
4
, 2x10
6
]
Yes
0.34 (0.13-0.84)
0.11 (0.04-0.2)
0.68 (0.25-1.28)
2s
medium
0.033
0.4
0.05
[3x104, 2x106]
Yes
0.23 (0.08-0.69)
0.09 (0.03-0.2)
0.59 (0.19-1.27)
3s
high
0.033
0.4
0.05
[3x104, 2x106]
Yes
0.17 (0.06-0.59)
0.09 (0.03-0.2)
0.57 (0.16-1.26)
2a
medium
0.033
0.4
0.05
[3x104, 6x106]
No
0.19 (0.07-0.68)
0.23 (0.07-0.6)
1.41 (0.42-3.76)
2b
medium
0.033
0.4
0.05
[3x104, 16x106]
No
0.12 (0.03-0.61)
0.36 (0.07-1.58)
2.23 (0.46-9.85)
2c
medium
0.033
0.4
0.01
[3x10
4
, 2x10
6
]
No
0.28 (0.18-0.58)
0.15 (0.08-0.21)
0.95 (0.49-1.31)
2d
medium
0.033
0.4
0.10
[3x104, 2x106]
No
0.39 (0.2-1.02)
0.12 (0.05-0.21)
0.76 (0.29-1.29)
2e
medium
0.009
0.4
0.05
[3x104, 2x106]
No
0.3 (0.19-0.63)
0.15 (0.07-0.21)
0.94 (0.46-1.29)
2f
medium
0.065
0.4
0.05
[3x104, 2x106]
No
0.42 (0.19-0.99)
0.1 (0.05-0.21)
0.63 (0.3-1.29)
2g
medium
0.033
0.3
0.05
[3x104, 2x106]
No
0.32 (0.19-0.7)
0.14 (0.07-0.21)
0.88 (0.42-1.29)
2h
medium
0.033
0.5
0.05
[3x10
4
, 2x10
6
]
No
0.35 (0.19-0.8)
0.13 (0.06-0.2)
0.79 (0.36-1.28)
Pacific-wide sustainability risk assessment of bigeye thresher shark 89
Table F2: Estimated range of fishery group catchability q (median and 95% quantile range) by catch group in
years 2000
−
2014. All values are based on the distribution of
Ω
q
derived assuming a medium (0.5) initial
biomass level (BDM run 2).
Year
YFT (
4
10
−
)
ALB (
4
10
−
)
BET(
4
10
−
)
SWO(
4
10
−
)
others (
4
10
−
)
2000
0.64 (0.22-2.24)
0.6 (0.2-2.11)
0.78 (0.26-2.72)
0.32 (0.11-1.13)
0.53 (0.18-1.86)
2001
0.51 (0.17-1.8)
0.48 (0.16-1.7)
0.62 (0.2-2.18)
0.26 (0.08-0.91)
0.42 (0.14-1.5)
2002
0.58 (0.19-2.09)
0.54 (0.18-1.97)
0.7 (0.23-2.54)
0.29 (0.1-1.05)
0.48 (0.16-1.74)
2003
0.45 (0.16-1.61)
0.43 (0.15-1.51)
0.55 (0.19-1.95)
0.23 (0.08-0.81)
0.38 (0.13-1.34)
2004
0.43 (0.14-1.6)
0.41 (0.13-1.5)
0.52 (0.17-1.94)
0.22 (0.07-0.8)
0.36 (0.12-1.33)
2005
0.72 (0.24-2.65)
0.67 (0.22-2.49)
0.87 (0.29-3.21)
0.36 (0.12-1.33)
0.59 (0.2-2.2)
2006
0.43 (0.14-1.6)
0.4 (0.13-1.51)
0.52 (0.17-1.94)
0.21 (0.07-0.8)
0.35 (0.12-1.33)
2007
0.69 (0.23-2.65)
0.65 (0.22-2.49)
0.84 (0.28-3.21)
0.35 (0.12-1.33)
0.58 (0.19-2.2)
2008
0.48 (0.16-1.74)
0.45 (0.15-1.64)
0.58 (0.2-2.11)
0.24 (0.08-0.87)
0.4 (0.14-1.44)
2009
0.63 (0.21-2.38)
0.59 (0.2-2.24)
0.76 (0.25-2.88)
0.32 (0.11-1.19)
0.52 (0.17-1.97)
2010
0.68 (0.23-2.55)
0.64 (0.21-2.4)
0.82 (0.27-3.1)
0.34 (0.11-1.28)
0.56 (0.19-2.12)
2011
0.54 (0.18-1.99)
0.51 (0.17-1.88)
0.65 (0.22-2.42)
0.27 (0.09-1)
0.45 (0.15-1.65)
2012
0.83 (0.28-3)
0.79 (0.27-2.82)
1.01 (0.34-3.64)
0.42 (0.14-1.51)
0.69 (0.24-2.49)
2013
0.75 (0.25-2.86)
0.71 (0.23-2.69)
0.91 (0.3-3.47)
0.38 (0.13-1.44)
0.62 (0.21-2.38)
2014
0.85 (0.29-3.23)
0.8 (0.27-3.04)
1.03 (0.35-3.92)
0.43 (0.14-1.63)
0.7 (0.24-2.68)
Table F3: Estimated range of fishery group catchability q (median and 95% quantile range) by season in years
2000
−
2014. All values are based on the distribution of
Ω
q
derived assuming a medium (0.5) initial biomass
level (BDM run 2).
Year
Jan–Mar(
4
10
−
)
Apr–Jun(
4
10
−
)
Jul–Sep(
4
10
−
)
Oct–Dec(
4
10−
)
2000
0.78 (0.26-2.72)
0.98 (0.33-3.44)
0.8 (0.27-2.79)
1.03 (0.35-3.59)
2001
0.62 (0.2-2.18)
0.78 (0.26-2.76)
0.63 (0.21-2.25)
0.81 (0.27-2.89)
2002
0.7 (0.23-2.54)
0.89 (0.3-3.21)
0.72 (0.24-2.61)
0.93 (0.31-3.35)
2003
0.55 (0.19-1.95)
0.7 (0.24-2.47)
0.57 (0.19-2.01)
0.73 (0.25-2.58)
2004
0.52 (0.17-1.94)
0.66 (0.22-2.45)
0.54 (0.18-2)
0.69 (0.23-2.56)
2005
0.87 (0.29-3.21)
1.1 (0.37-4.06)
0.89 (0.3-3.3)
1.15 (0.38-4.24)
2006
0.52 (0.17-1.94)
0.65 (0.22-2.45)
0.53 (0.18-2)
0.68 (0.23-2.56)
2007
0.84 (0.28-3.21)
1.06 (0.35-4.06)
0.87 (0.29-3.3)
1.11 (0.37-4.24)
2008
0.58 (0.2-2.11)
0.73 (0.25-2.67)
0.59 (0.2-2.17)
0.76 (0.26-2.79)
2009
0.76 (0.25-2.88)
0.97 (0.32-3.64)
0.79 (0.26-2.96)
1.01 (0.34-3.81)
2010
0.82 (0.27-3.1)
1.04 (0.35-3.92)
0.85 (0.28-3.19)
1.09 (0.36-4.09)
2011
0.65 (0.22-2.42)
0.83 (0.28-3.06)
0.67 (0.23-2.49)
0.86 (0.29-3.19)
2012
1.01 (0.34-3.64)
1.28 (0.44-4.6)
1.04 (0.35-3.74)
1.34 (0.45-4.81)
2013
0.91 (0.3-3.47)
1.15 (0.38-4.39)
0.94 (0.31-3.57)
1.21 (0.4-4.59)
2014
1.03 (0.35-3.92)
1.3 (0.44-4.96)
1.06 (0.36-4.04)
1.36 (0.46-5.18)
90 Pacific-wide sustainability risk assessment of bigeye thresher shark
Appendix G – Catch history
Table G1: Summary of observed number of hooks, observed BTH catch, total logsheet hooks (commercial
effort) and estimated total catch in the Calibration Area , 1995
−
2014. Total catch estimates were calculated
using the year-catch group stratification (assuming 100% capture mortality).
observed hooks
observed catch
Logsheet
Total scaled catch
Year
(million)
(number)
hooks (million)
(number)
1995
0.59
75
44.77
6629
1996
0.65
203
33.37
9901
1997
0.52
139
34.40
11579
1998
0.66
227
39.69
13428
1999
0.60
74
49.21
9057
2000
2.14
397
43.73
7388
2001
4.91
688
35.31
4628
2002
6.19
1250
46.01
8498
2003
6.16
759
63.34
7655
2004
7.58
1781
62.72
12575
2005
9.83
1138
68.09
8393
2006
7.52
1474
72.25
13352
2007
8.15
1274
62.30
8966
2008
8.54
1051
64.43
7720
2009
7.81
1618
52.85
11406
2010
8.37
1350
51.34
8340
2011
8.73
1311
68.61
9656
2012
8.89
1684
60.76
12550
2013
9.08
1634
61.98
11179
2014
9.85
3803
63.65
23705
Pacific-wide sustainability risk assessment of bigeye thresher shark 91
Appendix H – Year effects standardisations
ZINB model
Results of likelihood ratio tests for the nested ZINB models are presented in Table H1. The largest
improvement in AIC occurred when subarea was included. A comparison of predicted annual indices
for each of the fitted models is shown in Figure H1.
Model “leffort” is the final model selected to predict annual indices of abundance for bigeye thresher
in the Calibration Area. Diagnostics for the final model are shown in Figure H2.
Table H1: Summary of ZINB models fitted to US Hawaii BTH catch and effort observer data in the Calibration
Area, 1995
−
2014. df = degree of freedom; “Pr(>Chisq)”,P value from the likelihood ratio test.
Model Variables df AIC Pr(>Chisq)
year
year
41
82266
-
subarea
Year+subarea
63
71919
<2.2E-16
month year+subarea+ns(month,3) 69 71335 <2.2E-16
hbf Year+subarea+ns(month,3)+ns(hbf, 3) 75 71267 <3.2E-15
kmeans year+ subarea+ns(month,3) +ns(hbf, 3) + catch_group 81 71140 <2.2E-16
leffort
year+subarea+ns(month,3) +ns(hbf, 3) + catch_group + leffort
81
71147
-
Figure H1: Comparison of annual CPUE indices for BTH in the Calibration Area, as obtained from ZINB models
fitted to US Hawaii observer data 1995
−
2014, where variables were added to each model sequentially (see
Table H1). All indices were normalized by the mean of each series for comparison.
92 Pacific-wide sustainability risk assessment of bigeye thresher shark
Figure H2: Diagnostics of the final ZINB model “leffort” fitted to US Hawaii observer data 1995
−
2014 in the
Calibration Area: top-left is Pearson Residuals vs fitted , top-right, observed catch vs fitted catch; bottom left,
Pearson Residuals by region, bottom right, Pearson Residuals by year.
delta-GLMM model
Geostatistical delta-GLMM models were applied for comparison. For year effects estimation, the
area associated with each knot of the predictive grid was defined as the total areas of the grid cells
closest to that knot. The abundance index in a year is calculated by summing across the model
predicted density for all knots, where each is weighted by its area (Thorson et al. 2015).
Four models were fitted that differed in the number of knots. Spatial variability was estimated for
1000, 250, 50, and 10 knots, respectively.
Estimated variance for both encounter probabilities and positive catch rates increased significantly as
the number of knots fell below 50 (Table H2). The predicted annual indices were similar between the
four models, except for the model with the lowest number of knots (10) (Figure H3).
Pacific-wide sustainability risk assessment of bigeye thresher shark 93
Table H2: Estimates of variance parameters for the four delta-GLMM models fitted to estimate annual
indices of abundance for BTH using US Hawaii observer data from 1995−2014 in the Calibration Area.6
Random fields (marginal SD)
Vessel effects
Model
)
(p
ε
σ
)(r
ε
σ
)( p
ϖ
σ
)(r
ϖ
σ
)( p
v
σ
)
(r
v
σ
1000 knots
-
-
1.21
0.53
0.26
0.18
250 knots
-
-
1.21
0.53
0.26
0.18
50 knots
-
-
6.50
2.34
0.26
0.20
10 knots
-
-
6.51
2.36
0.26
0.20
Figure H3: Predicted annual indices (with confidence interval) for delta-glmm models fitted a predefined
number of knots equal to 1000, 200, 50, and 10, respectively. Indices were normalised by their mean to allow
comparisons.
6
)( p
ϖ
σ
and
)(r
ϖ
σ
for spatial random fields effects.
)( p
v
σ
and
)(r
v
σ
are for vessel random effects
94 Pacific-wide sustainability risk assessment of bigeye thresher shark
Variability for vessel effects is lower than estimated in similar models fitted to the composite
observer dataset (Appendix E). Spatial patterns of estimated density were not highly sensitive to the
number of knots defined to estimate spatial random effects, but using a very small number of knots
did not capture the variability in estimated densities in some hotspot areas (Figure H4).
(a)
(b)
(c)
(d)
Figure H4: Estimated density (log scale) extrapolated to the 10 km by 10 km square extrapolation grid used to
estimate spatial effects in the delta-GLMM models fitted to US Hawaii observer catch and effort data in the
Calibration Area, 1995-2014. (a) model with 1000 knots; (b) 250 knots; (c) 50 knots; and (d) 10 knots.
Pacific-wide sustainability risk assessment of bigeye thresher shark 95
Appendix I - Impact sensitivity
Impact sensitivity consisted of calculating total fishing impacts using catchability values derived from
BDM calibration runs performed assuming higher upper bounds for the prior over K (as per blue
shark assessment values) (BDM runs 2a & 2b - Table F1, Appendix F), and by varying process error
standard deviation from very low (0.01) to 0.1 (BDM run 2c, 2destimates (see Table F1 in Appendix
F). All such sensitivities were conducted assuming the medium (mean 0.5) initial stock status for the
species in the Pacific and 100% capture mortality.
Median impact calculated for the 6 million upper bound over K scenario was lower than 0.02 in both
the Core Area and in the Assessment Area, over both the full assessment period 2000-2014 and the
recent period 2011-2014 (Table I1). Median impact calculated for the 16 million upper bound over K
scenario was lower than 0.01 (Table I1). Annual impacts and variability within years were within the
range of estimated values for the maximum intrinsic population growth rate r for the species (mean
0.03 and cv 0.06, see section 4.4).
Median impact calculated for the minimum process error (0.01) scenario ranged from 0.019 to 0.025
(Table I2), and were below the mean r estimates for the species (Figure I2). Median impact calculated
for the higher (0.10) process error scenario ranged from 0.027 to 0.035 (Table I2), and were at or
above the mean intrinsic population growth rate.
(a)
(b)
Figure I1: Annual impact (median values and 95% quantile range) estimated for (a) the Core Area and (b) the
Assessment Area, using catchability estimates derived from three prior bounds scenarios over log(K) (upper
bound ranging from 2 x 106(base case scenario), 6 x 106, and 16 x 106 ) (runs 2, 2a, and 2b - Table F1,
Appendix F). Medium initial status (mean 0.5) and 100% capture mortality were assumed in each scenario.
The dashed line is the mean value for the estimated r prior (0.03), with cv 0.6 (shaded grey area) (see section
4.4 for details).
96 Pacific-wide sustainability risk assessment of bigeye thresher shark
(a)
(b)
Figure I2: Annual impact (median values and 95% quantile range) estimated for (a) the Core Area and (b) the
Assessment Area, using catchability estimates derived for three process error sigma values (0.01, 0.05 (base
case scenario), and 0.10) (runs 2, 2e, and 2f - Table F1, Appendix F). Medium initial status (mean 0.5) and
100% capture mortality were assumed in each scenario. The dashed line is the mean value for the estimated
r prior (0.03), with cv 0.6 (shaded grey area) (see section 4.4 for details).
Table I1: Total impact (median F and 95% quantile range among cells) for the fifteen year period (2000-2014)
and the recent period (2011-2014) in the Core Area and the Assessment Area, across log(K) prior upper
bound sensitivities (runs 2a and 2b - Table F1, Appendix F).
Impact
Impact
K prior bounds [30000, 6000000]
K prior bounds [30000, 16000000])
Core Area
2000-2014
0.015 (0.005-0.056) 0.008 (0.002-0.049)
2011-2014
0.017 (0.005-0.062) 0.009 (0.002-0.054)
Assessment
Area
2000-2014
0.013 (0.004-0.053) 0.009 (0.002-0.045)
2011-2014
0.017 (0.005-0.064) 0.011 (0.002-0.056)
Pacific-wide sustainability risk assessment of bigeye thresher shark 97
Table I2: Total impact (median F and 95% quantile range among cells) for the fifteen year period (2000-2014)
and the recent period (2011-2014) in the Core Area and the Assessment Area, across process error
sensitivities (runs 2c and 2d - Table F1, Appendix F).
Impact
Impact
Process error sigma 0.01
Process error sigma 0.10
Core Area
2000-2014
0.021 (0.012-0.045) 0.029 (0.013-0.085)
2011-2014
0.023 (0.014-0.049) 0.033 (0.015-0.093)
Assessment
Area
2000-2014
0.019 (0.01-0.045) 0.027 (0.012-0.077)
2011-2014
0.025 (0.013-0.056) 0.035 (0.015-0.097)
98 Pacific-wide sustainability risk assessment of bigeye thresher shark
Appendix J - Supporting information
This section illustrates available information on nominal (unstandardized) CPUE for bigeye thresher
in SPC and Japanese observer datasets (Figure J1).
Figure J1: Observed catch and unstandardised catch rate (number per 1000 hooks) for SPC observer data 1995–
2000 and Japan observer data 2007–2015.
