AEBR 257 Continuous Plankton Recorder Sampling Southern Ocean 2002 2018 3347
AEBR 257 Continuous Plankton Recorder Sampling Southern Ocean 2002 2018 3347
K.V. Robinson,
M. Decima,
M.H. Pinkerton,
R. Stewart
April 2021
Requests for further copies should be directed to:
Email: [email protected]
Telephone: 0800 00 83 33
Facsimile: 04-894 0300
This publication is also available on the Ministry for Primary Industries websites at:
http://www.mpi.govt.nz/news-and-resources/publications
http://fs.fish.govt.nz go to Document library/Research reports
EXECUTIVE SUMMARY 1
1. INTRODUCTION 3
1.1 Background 3
1.2 Overall objective 4
1.3 Specific objectives 4
2. METHODS 4
2.1 CPR methodology 4
2.2 Study area 5
2.3 CPR deployments in Ross Sea region 6
2.4 Post-voyage laboratory analysis 11
2.5 SO-CPR data extract used in the present study 11
2.6 Phytoplankton Colour Index 12
2.7 Taxonomic grouping of data 13
2.8 Latitudinal variation in zooplankton abundance in the Ross Sea region 15
2.9 Spatial distribution of zooplankton abundance 15
2.10 Trends in zooplankton abundance 15
2.11 Model-based analysis of zooplankton abundance 15
2.12 Model-based analysis of zooplankton communities 18
2.13 Average Copepod Community Size (ACCS) 19
2.14 Microplastics 19
2.15 Data management and integration 20
3. RESULTS 20
3.1 Latitudinal variation in zooplankton abundance in the Ross Sea region 20
3.2 Spatial distribution of zooplankton abundance 23
3.3 Trends in zooplankton abundance 26
3.4 Model-based analysis of zooplankton abundance 28
3.5 Model-based analysis of zooplankton communities 37
3.6 Average Copepod Community Size (ACCS) 45
3.7 Microplastics 47
4. DISCUSSION 48
4.1 CPR sampling and analysis 48
4.2 Patterns in zooplankton abundances by broad taxonomic group 48
4.3 Zooplankton communities 49
4.4 Trends and long-term change 50
4.5 Average Copepod Community Size (ACCS) index 51
4.6 Comparison between Ross Sea and the rest of the Southern Ocean 52
4.7 Future effects of climate change 52
4.8 Northern hemisphere comparisons 53
4.9 Final comments 54
5. MANAGEMENT IMPLICATIONS 54
6. ACKNOWLEDGMENTS 55
7. REFERENCES 55
10. APPENDIX 3: Mid-range lengths of copepod species used to calculate the Average
Copepod Community Size index 78
EXECUTIVE SUMMARY
Robinson, K.V.; Decima, M.; Pinkerton, M.H.; Stewart. R. (2021). Continuous Plankton
Recorder sampling of zooplankton in the Southern Ocean 2002–2018: spatial, seasonal, and long-
term changes.
This report marks the end of the second phase of the New Zealand Continuous Plankton Recorder (CPR)
survey between New Zealand and the Ross Sea. The first phase was 2008–2013 (ZBD2008-15), the
second phase was 2013–2018 (ZBD2013-03), and the third phase started in 2019 (ZBD2018-01).
Through the continued collaboration with Sanford Limited, the FV San Aotea II was used to deploy the
CPR on transits between New Zealand and the Ross Sea (Antarctica) to gather zooplankton samples for
analysis. The overall objective of the project was to use these CPR samples to understand spatial,
seasonal, and long-term changes to zooplankton in New Zealand’s southern EEZ and on the transit to
the Ross Sea.
Zooplankton are small invertebrates living in the water column and are a key part of the food chain,
linking changes in primary production with consumers (such as fish, seabirds, and mammals). It is well
established that zooplankton can be effective indicators of ecosystem health and can be used for
monitoring, anticipating, and managing responses to ecosystem change, including regime shift. CPR
research in the Southern Ocean started in 1991, and over the last decade New Zealand has become the
third largest contributor of data. During the last 5 years of this project, 37 individual tows of the CPR
from the FV San Aotea II were carried out, covering a total of 11 407 nautical miles (> 20 000 km) of
towed distance. In total, 600 000 individual zooplankton specimens were counted and identified.
Zooplankton data from the circumpolar Southern Ocean CPR project were analysed in relation to
environmental measurements over the period 1997–2018, covering phytoplankton biomass, primary
productivity, water temperature, mixed layer depth, sea ice, and activity of ocean fronts. Statistical
models were used to determine relationships between key taxonomic groups of zooplankton,
zooplankton communities, and environmental conditions.
Key conclusions of this analysis:
1. Concentrations of chlorophyll-a (chl-a) in the Southern Ocean generally increased between
1997 and 2018, especially around the southern limit of the Antarctic Circumpolar Current,
around the Argentine continental shelf, and in the Subtropical Front across New Zealand. In
contrast, there were decreasing trends in chl-a over some of the Antarctic continental shelf and
in the larger Ross Sea region.
2. For 6 of the main taxonomic groups of zooplankton, statistical models could reproduce broad
spatial patterns of zooplankton abundances measured by the CPR survey. The models
developed can potentially be used in conjunction with Earth-system models to project the
effects of climate change on zooplankton in the Southern Ocean.
3. Abundances of zooplankton in general were higher in the Ross Sea region than in the Southern
Ocean as whole, by a factor of about 2 on average but varying among zooplankton taxa.
Abundances of Foraminifera and Fritillaria spp. were more than 3 times higher in the Ross Sea
region than elsewhere in the Southern Ocean. These differences are not due to the Ross Sea
region having unusual environmental conditions and are likely to be ecologically-driven.
4. Changes to environmental conditions over the last 20 years (1997–2018) are likely to lead to
increases in the abundance of most broad taxonomic groups of zooplankton in the Southern
Ocean, with copepods, Foraminifera, and Fritillaria spp. especially likely to benefit.
1.1 Background
Zooplankton 1 are a key link in marine ecosystems, connecting oceanographic/climate drivers and higher
trophic level consumers including fish, seabirds and marine mammals (Bathmann et al. 2000, Dubischar
& Bathmann 1997). Zooplankton can be useful indicators of the state of the marine environment
(Atkinson 1998, Smetacek et al. 2004, Mackas & Beaugrand 2010) and can act as early warning
indicators of critical shifts in ecosystem function (Beaugrand 2005, Batten et al. 2016). For example,
variations in plankton abundance in the Atlantic have been linked to climatic variability, including the
North Atlantic Oscillation, Gulf Stream Index, and changing sea-surface temperatures (Reid et al. 1998,
Beare & McKenzie 1999). Zooplankton spatial analyses are also useful in the context of bio-
regionalisation and spatial management including in the context of designing and evaluating Marine
Protected Areas (Raymond & Constable 2006, Sharp et al. 2007).
Despite their fundamental importance, information on zooplankton is typically poor because the
community is diverse (hundreds of species with many developmental stages), highly variable (both in
space and time – Fasham & Pugh 1977; Mackas et al. 1985; Atkinson et al. 2008), and approaches to
observation are limited. Satellite data provide year-round data (e.g., sea-surface temperature and ocean
primary productivity) but cannot observe zooplankton. Both station-based and underway sampling
methods for zooplankton abundances (e.g., Burkill & Gallienne 2001, Davis et al. 2004, Hosie et al.
2003, Pollard et al. 2002, Postel et al. 2000) continue to rely on deployment of gear from vessels which
limits the scale and frequency of field measurements of zooplankton.
The Continuous Plankton Recorder (CPR) is a high-speed plankton sampler originally designed to be
towed behind merchant ships as they steam at cruising speed along regular shipping routes (Warner &
Hays 1994). In the North Sea and Northeast Atlantic for example, CPRs have been deployed from such
vessels for over 75 years. Such long-term marine environmental datasets are extremely rare, and the
CPR time series is globally one of the longest running monitoring programmes in the world (Hays et
al. 1993). The SCAR 2 Southern Ocean CPR survey (SO-CPR) was established in 1991 to map the
spatial and temporal patterns of zooplankton in the Southern Ocean (Hosie et al. 2003). SO-CPR is one
of nine separate CPR surveys that are now contributing data to the Global Alliance of CPR Surveys
(GACS) to understand the effects of environmental changes on plankton biodiversity, and to monitor
ecosystem health at a global level. Up until 2008 however, there were few CPR tows in the Ross Sea
region 3. New Zealand has special interest in this sector of the Southern Ocean and so, with funding
from MFish (now MPI and Fisheries New Zealand) and with the support and cooperation of Sanford
Limited, this important information gap began to be filled with the instigation of a biannual CPR survey
between New Zealand and the Ross Sea.
This report presents the combined data from the Phase 1 (2008–2013) and the Phase 2 (2013–2018)
Ross Sea region CPR projects, along with data obtained from SCAR SO-CPR Survey database. Phase
3 (2018–2023) of New Zealand SO-CPR research commenced in 2019.
1
Zooplankton are planktonic animals which drift in oceans, seas, and bodies of fresh water. In the ocean,
zooplankton range in size from nanozooplankton (2–20 µm), microzooplankton (20–200 µm), mesozooplankton
(200 µm–2 mm), and microzooplankton (> 2 mm). Zooplankton can be heterotrophic (feed on other organisms)
or detritivorous (feed on non-living detrital material) plankton.
2
Scientific Committee on Antarctic Research, https://www.scar.org/
3
Ross Sea region is often defined as lying between 160° E and 150° W, and south of 60° S.
1. Conduct annual return transits with the CPR to the Ross Sea during the austral summer in
collaboration with Sanford limited, and contribute the data collected to the SCAR Southern Ocean
CPR Survey (SO-CPR).
2. Combine the data collected during phase 1 (years 1–5) with the data collected under this project to:
i. Determine zooplankton and phytoplankton trends in the full dataset: and
ii. Compare results with datasets available through the SO-CPR Survey.
5. Describe broad similarities and differences observed between NZ, and other southern hemisphere
data with patterns of trends observed in the northern hemisphere.
2. METHODS
The CPR is a robust plankton sampler that was designed in the 1920s by Sir Alister Hardy in order to
address questions regarding plankton distribution patchiness (Reid et al 2003). The design of the unit
that was deployed for both Phase 1 (2008–13) and Phase 2 (2013–18) MPI projects is based on the
original model, with some minor modifications to allow for the more extreme sea conditions in the
Southern Ocean (Hosie et al. 2003). The unit is towed horizontally at a depth of approximately 10 m
and approximately 100 m astern of the ship as it travels at cruising speed (about 10 kn.). During a tow
water enters a small aperture at the front of the CPR body where it passes through a 270 µm filtering
collecting silk (Figure 2-1). As the CPR moves through the water, an external impeller advances the
silk at a rate of ~ 1 cm per 1 nautical mile through the water irrespective of the speed of the ship over
the ground. (The speed of the ship through the water and over the ground are different because of ocean
currents and sea-state. For example, greater vertical undulation because of large swell results in shorter
distance towed over the ground, whereas a following sea can lead to longer distance towed over the
ground.) The filtering silk then gets covered by a ‘covering’ silk, with the plankton sandwiched between
the two layers. Both silks are then rolled into a storage tank, where a supply of buffered 37%
formaldehyde preserves the sample. The maximum towing distance for a silk is about 450 nautical miles
through the water.
The sampling aperture of the CPR is 1.27 cm2 which is most effective for capturing mesozooplankton
(0.2–2 mm size). Larger organisms are unlikely to be sampled well by the CPR, particularly those such
as Antarctic krill that form swarms and have strong predator evasion responses. This size selectivity
results in under-estimates of the abundance of krill compared with large net systems such as the
Rectangular Midwater Trawl, or when compared with biological hydro-acoustic estimates. Nonetheless,
krill of all species are collected by the CPR and their smaller developmental stages (larvae) are
sometimes abundant in samples. Parallel analyses have shown that the CPR results are comparable with
plankton net systems fitted with the same mesh and sampled within a similar depth range (Hunt & Hosie
2003). However, gelatinous zooplankton, e.g., salps and jellyfish, are poorly sampled by the CPR
because their large size and soft bodies can be damaged when trapped between the silks.
The Ross Sea region is defined in this report as the area between New Zealand and the Ross Sea proper,
lying between 160° E and 150° W, and south of 60° S. The Ross Sea is wholly contained within
FAO/CCAMLR statistical area 88. This region was visited annually in the austral summer by the FV
San Aotea II from 2008–2018 and by RV Tangaroa in 2006, 2008, 2010, and 2013. The 2013 voyage
of the RV Tangaroa generated samples from both within the Ross Sea region as well as the East
Antarctic region by crossing the 160° E line. The sampling area of the FV San Aotea II extended from
just south of the port of Timaru in the South Island of New Zealand, to the Ross Sea, Antarctica (Figure
2-2). The exact route, especially in the south of the region, varied among years because this depended
on the fishing locations chosen by the master of the fishing vessel. The route in higher latitudes was
also affected by the distribution and concentration of sea ice.
CPR data from the Ross Sea region are compared in this report to those collected from the rest of the
Southern Ocean. Most CPR data in the Southern Ocean are from the East Antarctic region, defined as
between 60° E and 160° E, and south of the Subtropical Front, and largely contained within
FAO/CCAMLR statistical area 58 (between longitudes 80° and 150° E).
FV San Aotea II
Sampling in this project was conducted through a collaboration with Sanford Limited, which annually
sends vessels to the Ross Sea to harvest Antarctic toothfish (Dissostichus mawsoni). The CPR was
towed by the FV San Aotea II between New Zealand and the Ross Sea annually in the austral summer
between 2008 and 2013 (Phase 1 survey) and 2013 and 2018 (Phase 2 survey). Training was provided
to the crew on deployment/recovery methodology, and one person per voyage was trained as the CPR
“handler”. The CPR handler and the assisting FV San Aotea II crew are responsible for (1) loading
internal cassettes with new silks into the main body of the CPR, deploying the CPR at regular intervals,
(2) retrieving the CPR with the sampled silks, (3) labelling and preserving the silks, and (4) recording
appropriate data. Global Positioning System (GPS) time and position data were recorded hourly (where
possible) in a logbook and later matched with the plankton data. The sampled silks were preserved in
Table 2-1: Summary of FV San Aotea II CPR deployments as part of the Phase 1 project 2008–2013.
Transit: S=southward (New Zealand to Ross Sea); N=northward (Ross Sea to New Zealand). A
‘segment’ is a 5-cm section of silk representing an approximate 5 nm analysed section of the tow.
Number Total
Number of Distance zooplankton
Season Transit Start date End date of tows segments towed (nm) count
RV Tangaroa sampling
The CPR was also deployed during five RV Tangaroa voyages (2002, 2006, 2008, 2010, 2013) under
separate funding: (1) TAN0202, JARE43 (06/02/2002–05/03/2002); (2) TAN0602 (28/01/2006–
17/03/2006); (3) TAN0802, IPY-CAML, International Polar Year – Census of Antarctic Marine Life
(26/01/2008–21/03/2008); (4) TAN1002, Tangaroa Whale Research voyage (31/01/2010–16/03/2010);
and (5) TAN1302, RV Tangaroa Mertz Polyna Voyage (01/02/2013–14/03/2013). The locations of
CPR tows from the RV Tangaroa varied significantly among the voyages and were often distant from
the usual transit routes of the FV San Aotea II. Consequently, the CPR data from the RV Tangaroa
voyages are not as useful for assessing variability in zooplankton over time as those from the FV San
Aotea II voyages. In total, 48 individual tows of the CPR from the RV Tangaroa were carried out (Table
2-3), comprising 722 segments, with a total of 10 846 nautical miles of towed distance. Zooplankton
In the laboratory after the voyage, each pair of silks from the Ross Sea CPR tows were unrolled and cut
into segments representing ~5 nautical miles of towed distance. The last segment at the end of a tow
was sometimes shorter than 5 nautical miles and was either included with the penultimate segment if
less than 2.5 nautical miles or retained as a short segment if more than 2.5 nautical miles. All
zooplankton were rinsed off each 5 nautical mile silk segment and counted and identified under a
dissecting microscope using the SO-CPR Survey method (Hosie et al. 2003). This method was
developed from the northern hemisphere CPR Survey’s method to be a more conservative and thorough
approach, and to eliminate the need for highly specialised equipment.
When zooplankton abundances in the SO-CPR samples were very high, subsampling using a Motoda
plankton splitter was carried out (Motoda 1959). The efficiency of the plankton splitter was tested to
ensure that each subsample was an accurate representative of the whole sample. Zooplankton were
identified to the lowest taxonomic level possible, ideally to species.
Antarctic krill (Euphausia superba) and other euphausiids were identified to developmental stage where
possible, as were copepods (adults and copepodite stages) (McLeod et al. 2010). Zooplankton that are
easily damaged, and notably gelatinous and soft bodied species, were identified to a coarser taxonomic
resolution only. Larvaceans (solitary, free-swimming tunicates) are very abundant in CPR samples, and,
although the body of the larvacean is often damaged or missing, the tails are easy to identify to genus
level and count. Other harder body zooplankton such as foraminiferans can usually be identified to an
acceptable taxonomic level and counted, although damaged individuals can only be identified to a
coarser taxonomic resolution than intact specimens.
For the analysis given in this report, zooplankton data were extracted from the SO-CPR database as
current on 28 August 2018. The total number of SO-CPR observations across the Southern Ocean at
the time of the extraction is shown in Figure 2-5. It is clear that the majority of the SO-CPR samples
Figure 2-1: Number of CPR samples by area. The colours show the number of CPR segments in 81 km
pixels over the 29 years of the SO-CPR survey. Black lines are nominal positions of (from north):
Subantarctic Front (SAF, solid line), Polar Front (PF, solid line), average maximum northern
extent of seasonal sea ice (“Ice”, dashed line), and the Southern boundary of the Antarctic
Circumpolar Current (SB-ACC, solid line, Orsi et al. 1995).
Based on the colour of the CPR silks, a Phytoplankton Colour Index (PCI) score was assigned to each
5 nautical mile segment using a Pantone colour charts (www.pantone.com) with scores of 0 = no colour,
1 = very pale green, 2 = pale green, and 3 = green. Pantone colour codes corresponding to these indices
depend on the dominant phytoplankton group and other coloured material in the water. The PCI is
indicative of the chlorophyll a (chl-a) concentration based on the Marine Biological Association
(MBA), CPR Survey method (Reid et al. 1998). Training in the use of the PCI was provided by staff at
the Australian Antarctic Division (AAD), Tasmania, as well as senior analysts at the MBA, CPR Survey
in Plymouth, England.
Spatial and temporal patterns in PCI were not analysed in this study as satellite ocean colour
observations provide a more reliable and more powerful method of exploring changes in upper ocean
phytoplankton abundance in the Southern Ocean. The coarse categorisation of PCI and the lack of
quantitative values is limiting. PCI values from the global CPR surveys were compared with satellite
data for the Southern Ocean and for the Ross Sea region alone to “ground truth” the satellite data, albeit
at the coarse 4-category level. Monthly satellite measurements of chl-a at the same point as the CPR
sample were compared, and these were logged before comparison because chl-a is close to being log-
Figure 2-2: Phytoplankton Colour Index (PCI) showing comparison with (natural log of) satellite chl-a
concentration (mg m-3) on y-axis. PCI values are dithered by a random amount to show point
densities (colours).
CPR zooplankton samples were identified to 293 taxonomic categories in the SO-CPR database. For
the purposes of the present study, these data were further grouped into broad and fine taxonomic groups
(see Appendix 1). The broad grouping (Table 2-4) captured coarse patterns in the data and was similar
to that used in previous CPR analysis (e.g., Robinson et al. 2014).
The fine grouping of 60 taxonomic groups retained as much fine-scale taxonomic data as possible. This
double taxonomic grouping (broad and fine) was necessary because a large proportion of the
zooplankton were unidentified or unidentifiable at a fine taxonomic level, but the fine-scale data
nevertheless included important information on subtle changes in zooplankton community composition.
The fine-scale grouping was determined on a group-by-group basis considering the ambiguity caused
by unidentified specimens and included 35 separate species or genuses of calanoid copepods (see
Appendix 1). Care was taken to avoid spurious effects due to the large proportion of unidentified
copepods in the samples on the fine-scale analysis. Specifically, individuals unidentified to genus or
species (37% of all copepods) were excluded from the fine-scale analysis. The excluded copepod
samples will be one of: (1) very rare species that are not generally identified in the CPR analysis; (2)
early developmental stages (approximately C3–4 4 stages or earlier); and (3) damaged specimens. For
the 2018 data from the Ross Sea, the laboratory analysis separately recorded numbers of calanoid
copepods that could not be identified because of damage, and numbers that could not be identified for
other reasons. Within this test set of data, only 70 of 4323 unidentified calanoid copepods were damaged
or “very rare” (1.6%) with the vast majority being unidentifiable because of being early developmental
4Copepods develop in stages from eggs, to nauplii, to copepodites, and then to adults. The nauplii stages are called N1–N2,
and the copepodite stages C1–C5, where the number indicates the growing developmental life stage. “C3” is hence a
copepodite stage with 4 pairs of legs, and “C5” is the final stage before the adult.
Table 2-1: “Broad” taxonomic groupings of zooplankton taxa from the CPR used in this study.
7 Fritillaria spp. Two abundant larvacean genera. These are both solitary, free-
8 Oikopleura spp. swimming filter feeding tunicates.
For CPR data collected along a regular north-south transect line (such as the FV San Aotea II CPR
data), the zooplankton abundances were simply shown as a stacked bar chart against latitude. Missing
data (e.g., because of failure of the CPR unit or operational issues with deployment) were identified.
This analysis is relevant to specific objective 3.
To examine large-scale spatial patterns in zooplankton abundance the SO-CPR data were divided into
large spatial blocks, following the approach previously used for summarising CPR data (e.g., Beaugrand
et al. 2003; Beaugrand 2005; McLeod et al. 2010). Blocks of 81 km x 81 km were used so that each
block contained ~9 segments of 5 nm each time a CPR tow crossed it. This block size was chosen to be
small enough to give reasonable spatial resolution, but large enough that each contained a reasonable
amount of data. This analysis is relevant to specific objective 3.
To investigate long-term trends in zooplankton abundances in the raw CPR data (objective 3), counts
at the broad taxonomic level were grouped into spatial blocks of 300 km x 300 km. Larger blocks were
used for the trend analysis than for the preceding spatial summary because these contain more data and
time-series must have relatively complete coverage in time for trend analysis to be useful.
Linear trends were determined using the Mann-Kendall Z statistic (Mann 1945; Kendall 1975). The
magnitude of trends was estimated using Sen’s slope (Sen 1968). Trends were identified as significant
if the p-value was less than 5% and the Sen’s slope was not equal to zero (Hipel & McLeod 1994,
Scarsbrook 2006). Such a non-parametric approach for detecting linear trends is preferred over
parametric methods (such as linear regression) because the form of the relationships is not known, and
the Sen’s slope estimator reduces the effect of outliers (Hastie et al. 2001). Data were de-seasonalised
prior to analysis by removing the climatological monthly mean from the measurements to create
monthly anomalies.
“Block-based” abundance and trend analysis (see previous sections) is limited by the coverage of the
CPR survey and by the inherent variability of zooplankton in time and space. The most appropriate way
of overcoming complex, patchy, and time-varying ecological information is to use data-mining
approaches (Hochachka et al. 2007). In the present study, statistical models were used to investigate
spatial and temporal patterns in the abundances of the broad groups of zooplankton. This analysis was
needed to address specific objectives 4 and 5.
The multivariate statistical technique of Boosted Regression Trees (BRT: Friedman 2001; Ridgeway
2006; De'ath 2007; Elith et al. 2008) was used. BRT has been used to relate sparse measurements of
species abundance to environmental characteristics to investigate patterns of occurrence over large
spatial scales (103–105 km) and over extended periods of time (months–decades) (Guisan &
Zimmermann 2000; Leathwick et al. 2006; Elith et al. 2007, 2008; Pinkerton et al. 2010b). The
technique used in the present study broadly followed the methodology given by Pinkerton et al. (2010b)
for Oithona similis. However, the analytical approach of Pinkerton et al. (2010b) was extended by
exploring the effect of using different time scales of environmental variation in the analysis. Five
different models were used, representing a range of environmental integrations, from annual averages
to actual data at the time and point of capture (Table 2-5). Environmental data used in the present study
are shown in Appendix 2.
combined Combination of {cumulative, annual, clim, None directly, but all above indirectly.
climAnom, actual} models as Equation 2.
It is important to use an appropriate number of trees to avoid overfitting BRT models (Elith et al. 2007,
2008). Using too few trees gives models with poor performance (i.e., ones that explain only a low
proportions of the total variance), whereas too many trees gives models with little generality
(“overfitted” models). To explore the appropriate number of trees for each BRT model, K-fold analysis
was used where data were withheld in K=10 folds, respecting CPR tows, as recommended for CPR data
previously (Pinkerton et al. 2010b). Within-bag 5 and out-of-bag comparisons were also calculated;
these are also sometimes effective at giving insights into the appropriate number of BRT trees to use.
Different BRT models were combined taking into account the relative uncertainties in estimates from
different models (Equation 1). Weighting individual model outputs by the inverse of their variance
provides the best ensemble estimate. Combining different regression tree models with different sets of
predictor variables is consistent with the approaches of Random Forest (Breiman 2001) and Gradient
Forests (Ellis et al. 2012; Stephenson et al. 2018).
5
BRT models use “bagging” where only a randomly selected subset of data is used to fit each model tree. When
the model starts to become overfitted (too many trees), there will be an increasing difference between the
performance of the model based on the “within-bag” data (that used to fit the model) and the “out-of-bag” data
(not used to fit the model).
∑σ zi
2
i
zbest = i =1
n
Equation 1
∑σ
i =1
1
2
Where zbest is the best ensemble estimate, generated from n separate models, where zi is the estimate
from model i, which has variance σ2.
The various BRT models were applied to predict zooplankton abundance for a particular month and
year (Table 2-6). Environmental data used for predictions are shown in Appendix 2. Predictions of
abundance were made for a standard segment length of 5 nm. Predictions will vary by time of day
because Ecoszen is used in the BRT models so three predictions representing three times of day were
averaged. For each different location and month, the value of Ecoszen used in the prediction (Epredict)
was calculated as Equation 2, where α={0.17, 0.5, 0.83}. Here, Emin and Emax are the minimum and
maximum values of Ecoszen over the middle day-of-month at a given location.
= α Emin + (1 − α ) Emax
E predict Equation 2
Table 2-3: Data used in different types of BRT models for prediction in year y and month m. The
oceanographic environmental layers used (chl, sst, sstgrad, SeaIce, vgpm, mld0p030, mld0p125) are
explained in Appendix 2 and text.
Differences in zooplankton abundances between regions may arise from differences in environmental
or ecological conditions between regions, or for reasons unrelated to environmental conditions (for
example, ‘upstream’ conditions may vary). To investigate whether and why zooplankton abundances
differed between the Ross Sea region and the rest of the Southern Ocean the following analyses were
carried out. (1) The relationship between environmental conditions and zooplankton abundance was
modelled twice, first for the Southern Ocean and second for just the Ross Sea region. (2) Using the
whole Southern Ocean model, estimates of zooplankton abundances in the Ross Sea region to those
outside it were compared. (3) Again, using the whole Southern Ocean model, model residuals
In line with specific objectives 4 and 5, the fine taxonomic group data were used to describe zooplankton
assemblages or communities as a whole (Beaugrand 2005, Beaugrand et al. 2003, Hunt & Hosie 2005,
2006a, b). There are many alternative approaches to describing variation in multivariate communities
(e.g., Sørensen 1948, Bray & Curtis 1957, Legendre & Legendre 1998, Legendre & Gallagher 2001,
Clarke et al. 2006, De Caceres et al. 2013), but no established best practice for zooplankton community
analysis. In the present study, fine-scale SO-CPR zooplankton data from the whole Southern Ocean
were clustered along a gradient of community dissimilarity using a method that broadly followed
Beaugrand (2005) and Beaugrand et al. (2003).
There are a number of different dissimilarity measures available that emphasize different aspects of
community composition, but no standard measure for zooplankton communities. Some common
community dissimilarity measures emphasize presence/absence of species (Sørensen metric; Sørensen
1948), whereas some studies highlight the value of considering relative abundance of different species
(e.g., Clarke et al. 2006, Sharp et al. 2007). In the present analysis, dissimilarity between zooplankton
communities at location X and Y (DXY) was calculated as the product of the Euclidian distance metric
in log-space (because counts are close to log-normally distributed) for the broad group data (counts of
taxa i at site X and Y being xi and yi respectively), and the Sørensen distance metric for the fine grouped
data (Equation 3). In this equation, Nxy is the number of taxa present both at site X and at site Y, NX is
the number of taxa present at site X, and NY is the number of taxa present at site Y. Other dissimilarity
metrics were tried and gave similar results.
2 2 N XY
DXY ∑ ( ln xi − ln yi ) × 1 −
= Equation 3
i N X + NY
Agglomerative hierarchical cluster analysis was used to progressively group samples along a gradient
of increasing dissimilarity. Clusters were considered to be significantly different if their group-to-group
dissimilarities were greater than a given [‘threshold’ level. Determining the appropriate threshold was
important because it determined how many significantly different clusters were identified. To determine
the threshold dissimilarity, within-group dissimilarities (pairwise averaging) from 20 neighbouring
CPR segments were calculated. This can be thought of as the ‘sampling uncertainty’ because of the
high spatial autocorrelation between CPR segments (Pinkerton et al. 2010b). Clusters were assessed as
being significantly different when their dissimilarities were more than twice the sampling uncertainty.
The analysis was repeated with other sizes of groups (10 and 30) which had little effect.
Next, BRT modelling (see previous section) was applied to explore the relationships between the
dissimilarities of each community cluster and environmental variables. Environmental data used in the
present study are shown in Appendix 3. As before, a number of different models and combinations of
environmental data were used (Table 2-5). The final models were chosen from a combination of the
cumulative, annual, clim, and climAnom models. Combining models with different independent
variables is akin to the Random Forest approach (Breiman 2001), where the ensemble approach can
provide greater robustness and performance. This approach is preferred over multivariate modelling
approaches for multinomial responses (e.g., Generalised Dissimilarity Analysis; Ferrier et al. 2007)
because it allows us to consider explicitly the dissimilarity of a point from different cluster centres. The
zooplankton community and environmental conditions within a given modelled cluster were
The BRT models were used to predict zooplankton community clusters for each month where CPR data
were available (October-March) in each year (1998–2018) which allows changes in the areas predicted
to be occupied by the communities over time to be investigated. Mann Kendall trend analysis using Sen
slopes was applied as explained in Section 2.10. Areas were first de-seasonalised by subtracting the
monthly mean over the whole time series.
The Average Copepod Community Size (ACCS) metric (Equation 4) is the mean adult length of all
copepods in a sample, based on individual representative lengths (see Appendix 2) and can be used to
monitor changes in the dominance of large or small species in the community. The ACCS was
calculated using the mid-range length of adult female copepods (L) as an index of the relative size of a
species (i), multiplied by the number of individuals of that species (X). This is calculated for all copepod
species in a sample, summed for all species, and then divided by total number of all copepods (N) to
produce a single ACCS value (Beaugrand et al. 2003, Batten et al. 2016).
Equation 4
The ACCS was calculated for all samples in the SO-CPR dataset, covering the whole Southern Ocean
between 1991 and 2018. Only the copepod groups from the fine-scale CPR data were used for this
analysis. The mid-range lengths for adult females used were those from Beaugrand et al. (2003), with
the exception of a few species for which mid-range lengths were recalculated because these tend to be
different in the Southern Ocean compared with the global average (Appendix 3).
The ACCS metric will increase if the number of larger individuals relative to smaller species increases.
Conversely the ACCS index will decrease if there is a shift in dominance to smaller copepod species.
In the northern hemisphere, changes in the ACCS have been found to be indicative of biogeographical
changes in the copepod community due to oceanic temperature changes (Beaugrand et al. 2003,
Beaugrand et al. 2009). For example, the body size of copepods is related to water temperature, with
warmer water species tending to be smaller than those in polar oceans (Lopez & Anadon 2008). This
analysis is hence relevant to specific objective 5.
2.14 Microplastics
Microplastics were not a specific objective of this study, but there is growing interest in these particles
in the Southern Ocean as elsewhere in the world. During the zooplankton analysis process, preliminary
observations were made on samples that showed evidence of non-organic particles; items that were
believed to be microplastics, and other materials. These particles were recorded as ‘presence’ only, and
not counted. Categories of particles included single stranded filaments, flakes, or beads. The colour of
particle was also noted. These observations were not initially recorded in the 2008–09 season but were
initiated the following season and these data are available only for the 2009–2018 period. Brief results
are given in section 3.7 and Grover-Johnson (2018) gives further analysis of the data collected during
Phase 1 and Phase 2 of the MPI project.
The raw zooplankton count data generated in this study and from other CPR work in the Southern Ocean
carried out by NIWA (e.g., from Tangaroa voyages) were forwarded to the marine section of the SCAR
Antarctic Biodiversity Information Facility (ANTABIF) where they can be used for the new SCAR Life
Sciences Programme State of the Antarctic Environment (AntEco) and the Antarctic Thresholds -
Ecosystem Resilience and Adaptations (AnT-ERA). The New Zealand SO-CPR data were also
transmitted to the global CPR database managed by the Global Alliance of CPR Surveys (GACS),
which produces an annual state of the environment report on ocean plankton. Data are available for use
in the Southern Ocean Observing System (SOOS), Global Ocean Observing System (GOOS), and are
available for use by the Commission for the Conservation of Antarctic Marine Living Resources
(CCAMLR). Data will be held by the Fisheries New Zealand Research Data Management team as well
as the SCAR SO-CPR Survey database hosted by the Australian Antarctic Data Centre:
https://data.aad.gov.au/aadc/cpr/.
3. RESULTS
The broad (ten most abundant) zooplankton taxonomic groups enumerated on each transect of each
season sampled by F.V. San Aotea II during Phase 2 (seasons 2013–14 to 2017–18) are shown in Figure
3-1. Overall, including all years and water types, the most numerous zooplankton groups were Oithona
spp., Copepoda, Fritillaria spp., Foraminifera, pteropods, and Euphausiidae (Figure 3-1).
Zooplankton abundances were patchy, with the most consistent peak observed between latitudes 60°–
65° S, roughly coincident with the Polar Front Zone, and most noticeable during season 2013–14
(Figure 3-1a, b), 2014–15 southward transect (Figure 3-1c), and season 2017–18 (Figure 3-1i, j). The
southward transects (first one of the season) do not typically sample the area south of 65° S due to sea
ice, whereas the northward transects occur late enough in the season to provide data on the zooplankton
community closer to the Ross Sea shelf, between 70° and 78° S. Although the composition of the
community in this area does vary by year, the presence of Fritillaria spp. (2014–15, 2015–16, and
2017–18; Figure 3-1d, f, & j), and pteropods (2014–15, 2015–16; Figure 3-1d, f) were notable in some
years.
Based on the latitudinal plots from the FV San Aotea II transects only (Figure 3-1), the relative
abundances of different groups varied between years, yet no broad zooplankton taxon was
overwhelmingly more abundant during any particular sampling season, in contrast to the bloom of
Fritillaria spp. observed during the 2009–10 season in Phase 1 (Robinson et al. 2014). Although these
larvaceans were always present in significant numbers in the samples, order-of-magnitude higher
abundances during any of the seasons sampled during Phase 1 were not observed.
c d
Figure 3-1: Abundance of zooplankton taxa in broad taxonomic groups, plotted by year and transect
(S=southward, N=northward). a) 2013–14 S, b) 2013–14 N, c) 2014–15 S, d) 2014–15 N. Note that
zero total abundance indicates no data. “nm” = nautical mile. Missing data (e.g., because of failure
of the CPR unit or operational issues with deployment) are shown grey. Dashed lines indicate
boundaries between oceanographic zones (see section 2.9, Figure 2-3, Figure 2-4). (Continued on
next page)
g h
Figure 3-1: continued. Abundance of zooplankton taxa in broad taxonomic groups, plotted by year and
transect (S=southward, N=northward). e) 2015–16 S, f) 2015–16 N, g) 2016–17 S, h) 2016–17 N.
Note that zero total abundance indicates no data. “nm” = nautical mile. Missing data (e.g., because
of failure of the CPR unit or operational issues with deployment) are shown grey. Dashed lines
indicate boundaries between oceanographic zones (see section 2.9, Figure 2-3, Figure 2-4).
(Concluded on next page)
Figure 3-1: continued and concluded. Abundance of zooplankton taxa in broad taxonomic groups, plotted
by year and transect (S=southward, N=northward). i) 2017–18 S, j) 2017–18 N. Note that zero total
abundance indicates no data. “nm” = nautical mile. Missing data (e.g., because of failure of the CPR unit
or operational issues with deployment) are shown grey. Dashed lines indicate boundaries between
oceanographic zones (see section 2.9, Figure 2-3, Figure 2-4).
Spatial patterns of total zooplankton abundance were driven largely by variations in copepod and
Oithona spp. numbers which were numerically dominant (Figure 3-2). Amphipoda, Chaetognatha, and
Ostracoda were most abundant south of the southern limit of the Antarctic Circumpolar Current (ACC)
near the Antarctic continent. Euphausiidae abundances were higher to the north, especially around the
south of New Zealand. The other groups (copepods, Oithona spp., Foraminifera, Fritillaria spp.,
Oikopleura spp., pteropods) generally had higher abundances in the frontal area separating Subantarctic
waters from the sea ice zone around the Antarctic continent, though substantial variability and
heterogeneity in the spatial distributions is clear in Figure 3-2. Some “hotspots” in pteropods abundance
are seen over the Ross Sea shelf and offshore of Prydz Bay. Abundances of several groups (copepods,
Oithona spp., Foraminifera, Fritillaria spp.) and total abundance of zooplankton were substantially
higher in the Ross Sea region compared with the rest of the Southern Ocean (Figure 3-2).
c d
e f
Figure 3-2: Long-term average total abundance and abundances of the broad groups. a) total abundance,
b) Amphipoda, c) Chaetognatha, d) Copepoda, e) Euphausiidae, f) Foraminifera. Missing data are
shown as white. (Continued on next page)
i j
Figure 3-3: continued. Long-term average total abundance and abundances of the broad groups. g)
Fritillaria spp., h) Oikopleura spp., i) Oithona spp., j) Ostracoda, k) Pteropods. Missing data are
shown as white.
Relatively few significant long-term trends in abundances (Figure 3-3) were found from the raw CPR
data, which was likely because of the lack of consistent CPR sampling in any given location even at
this relatively-coarse spatial resolution (300-km blocks). For example, in the case of total abundance,
isolated locations with high positive trends (+20%/y) and high negative trends (-12%/y) were found,
but a predominance of no significant trend. There were also no widespread, spatially coherent patterns
in long-term trends for the other taxonomic group.
a b
c d
e f
Figure 3-4: Long-term trends in total abundance and broad groups. a) total abundance, b) Amphipoda, c)
Chaetognatha, d) Copepoda, e) Euphausiidae, f) Foraminifera. Fewer than 2 measurements at
different times at a given pixel are shown as grey. Trends which are not significant at the 0.05 level
are shown as white. (Continued on next page)
i j
Figure 3-3: concluded. Long-term trends in total abundance and broad groups. g) Fritillaria spp., h)
Oikopleura spp., i) Oithona spp., j) Ostracoda, k) Pteropods. Fewer than 2 measurements at
different times at a given pixel are shown as grey. Trends which are not significant at the 0.05
level are shown as white.
Table 3-1: Performances of fitted models for all broad taxonomic groups from each Boosted Regression
Tree (BRT) model. The proportion of variation explained (R2) and coefficients of variation (CV =
root mean square differences as a proportion of the mean) are shown.
Total abundance 0.43 0.38 0.44 0.36 0.40 0.39 0.44 0.37 0.43 0.37 0.54 0.34
Oithona spp. 0.44 0.82 0.46 0.77 0.43 0.83 0.45 0.78 0.42 0.80 0.54 0.74
Copepoda 0.41 0.73 0.44 0.67 0.42 0.72 0.44 0.69 0.42 0.70 0.53 0.65
Euphausiidae 0.26 0.77 0.27 0.77 0.25 0.78 0.27 0.76 0.25 0.77 0.32 0.75
Foraminifera 0.31 3.42 0.32 3.96 0.30 3.55 0.34 3.64 0.27 3.83 0.46 3.12
Fritillaria spp. 0.40 1.82 0.38 2.09 0.33 1.95 0.36 1.96 0.31 2.04 0.52 1.71
Pteropods 0.35 0.66 0.36 0.67 0.32 0.68 0.40 0.63 0.35 0.66 0.47 0.61
Table 3-2: Dependencies on sampling and environmental variables of the combined Boosted Regression
Tree (BRT) model for each univariate model. Values shown are the proportions of variance
explained by each of the variables (%). Ecoszen represents the time of day of sampling in terms of
the cosine of the solar zenith angle. Environmental variables include all components (i.e., month,
climatological month, climatological year average, and annual anomaly). mld includes both
mld0p030 and mld0p125 variables.
Total abundance 5.1 8.1 3.3 4.9 14.8 8.2 3.4 6.4 54.2
Oithona spp. 5.3 1.5 4.4 3.5 20.9 10.4 3.1 5.1 54.1
Copepoda 4.5 15.9 2.2 4.0 12.7 6.2 2.4 5.0 53.0
Euphausiidae 1.5 15.0 0.1 2.4 4.6 2.9 1.7 3.9 32.2
Foraminifera 4.4 0.4 1.2 5.9 10.1 13.9 4.2 6.2 46.2
Fritillaria spp. 5.6 0.9 1.7 7.6 11.7 12.5 6.2 5.9 52.2
Pteropods 2.9 7.4 1.3 8.4 10.0 7.2 4.6 5.1 46.8
The predicted abundances of the separate broad groups of zooplankton (Oithona spp., Copepoda,
Euphausiidae, Foraminifera, Fritillaria spp., and pteropods) (Figure 3-6) show differences in their
distributions. Copepoda in general and Oithona spp. were predicted to have a broad latitudinal
distribution with maximum abundances to the south of the Polar Front. Predicted maximum abundances
of Foraminifera occupied a narrower latitudinal range which was further north than that of copepods.
Fritillaria spp. also have a narrow latitudinal range of high abundances which is located further south
than that of copepods. The predicted abundances of Euphausiidae were higher to the north of the
Southern Ocean, and lower further south. This group included all euphausiids and not just Antarctic
krill (Euphausia superba).
c d
e f
Figure 3-5: Total abundance of Southern Ocean zooplankton for February based on separate BRT models
and oceanographic environmental information. a: cumulative; b: annual c: clim; d: climAnom; e:
actual; f: combined. Different models cover different total areas because predictions were not made
when the environmental conditions were outside the range of the training data.
c d
e f
Figure 3-6: Total zooplankton abundances (log scale) predicted using the combined BRT model and
oceanographic environmental information for months: a: October; b: November; c: December; d:
January; e: February; f: March. Areas with environmental conditions outside the range of the
training data were excluded from the prediction.
c d
e f
Figure 3-7: Abundances of broad zooplankton taxonomic groups predicted using the combined BRT model
and oceanographic environmental information (long-term average). Areas with environmental
conditions outside the range of the training data were excluded from the prediction. a: Copepoda;
b: Euphausiidae; c: Foraminifera; d: Fritillaria spp.; e: Oithona spp.; f: Pteropods.
Table 3-3: Comparison between zooplankton abundances (numbers per 5 nm segment) in the Ross Sea
region (RSR) and the rest of the Southern Ocean. Information from the combined BRT model
developed on all Southern Ocean data is shown in the left part of the table (“Southern Ocean
model”). Information on the Ross Sea region based on the BRT combined model fitted to only Ross
Sea region data is shown on the right (“Ross Sea model”). A two-sided t-test (N=120) was used to
assess whether zooplankton abundances in the Ross Sea region were significantly different to those
in the rest of the Southern Ocean, evaluated over the months October-March and years 1998–2018:
* p ≤ 0.05, ** p ≤ 0.01, *** p ≤ 0.001. Non-significant differences are shaded grey.
Total abundance 67.7 57.7 4.29 0.000*** 0.9 136.3 11.65 0.000*** 2.0
Oithona spp. 15.2 11.3 4.38 0.000*** 0.7 23.9 5.31 0.000*** 1.6
Copepoda 23.5 20.5 3.54 0.001*** 0.9 40.2 10.77 0.000*** 1.7
Euphausiidae 1.3 1.4 1.59 0.115 1.0 2.1 21.78 0.000*** 1.6
Foraminifera 2.8 2.8 0.09 0.931 1.0 9.1 11.38 0.000*** 3.2
Fritillaria spp. 1.4 1.3 0.69 0.493 0.9 4.9 8.01 0.000*** 3.5
Pteropods 0.5 0.4 4.88 0.000*** 0.8 0.5 1.43 0.155 1.1
Across the Southern Ocean, significant, positive long-term trends in total zooplankton abundance, and
for most broad taxonomic groups (except pteropods; Table 3-4), were found. Based on the circumpolar
modelling, similar significant, positive long-term trends in abundance were also predicted for the Ross
Sea region. The mean rate of increase was 0.62% per year in the period 1997–2018 (range +0.19%/y to
+0.83%/y), or a mean 12% increase over the modelled 20-year period. The models developed
specifically for the Ross Sea region predicted higher rates of increase for the same groups compared
with the Southern Ocean models (mean +0.91%/y, range +0.56%/y to +1.52%/y).
Spatial plots of statistically significant long-term trends are shown in Figure 3-7 for the abundances of
broad taxonomic groups of zooplankton, and in Figure 3-8 for total zooplankton abundance. The models
predict that total zooplankton abundance is increasing in the zone closest to the Antarctic continent in
the sea ice zone south of the Polar Front, especially in the East Antarctic, Indian, and Atlantic sectors.
Decreases in total abundance are predicted around the Polar Front and Subantarctic Front. Similar
spatial patterns in long-term trends are seen for Copepoda and Oithona spp. Temporal trends in
euphausiids are smaller, with trends being positive, and higher trends in the area of the Polar Front and
Table 3-4: Trends in abundances of zooplankton taxa (total and in broad taxonomic groups) from the
combined BRT models of the Southern Ocean and Ross Sea. The de-seasonalised Mann-Kendall
statistic (Z) and probability (p) are shown. Significance of Mann-Kendall trend: * p ≤ 0.05,
** p ≤ 0.01, *** p ≤ 0.001, with non-significant trends shaded grey. The Sen’s slope trend in the
area is shown as a proportion of the mean area.
Sen slope
Taxa or index Z p
(%/y)
Southern Ocean (Southern Ocean combined model)
Total abundance 7.46 0.000*** +0.64
Oithona spp. 6.09 0.000*** +0.59
Copepoda 8.18 0.000*** +0.65
Euphausiidae 4.40 0.000*** +0.19
Foraminifera 8.19 0.000*** +0.83
Fritillaria spp. 7.12 0.000*** +0.82
Pteropods 1.38 0.169 +0.13
Ross Sea (Southern Ocean combined model)
Total abundance 3.02 0.002** +0.38
Oithona spp. 1.09 0.275 +0.15
Copepoda 4.00 0.000*** +0.40
Euphausiidae 5.55 0.000*** +0.40
Foraminifera 4.22 0.000*** +0.62
Fritillaria spp. 2.62 0.009** +0.42
Pteropods -0.65 0.515 -0.05
Ross Sea (Ross Sea combined model)
Total abundance 3.57 0.000*** +0.56
Oithona spp. 4.72 0.000*** +1.16
Copepoda 4.50 0.000*** +0.76
Euphausiidae 6.01 0.000*** +0.59
Foraminifera 4.57 0.000*** +1.52
Fritillaria spp. 3.45 0.001*** +0.87
Pteropods -5.31 0.000*** -0.63
c d
e f
Figure 3-8: Predicted long-term trends in abundances of broad zooplankton taxonomic groups predicted using the
combined BRT model and oceanographic environmental information. The coloured data shown are the Sen’s
slope trends (average change per year for the period 1998–2018) where these trends were significant at better
than the 5% level assessed using the de-seasonalised Mann-Kendall test. a: Oithona spp.; b: Copepoda; c:
Euphausiidae; d: Foraminifera; e: Fritillaria spp.; f: pteropods.
Figure 3-10: Dendrogram of agglomerative hierarchical clustering applied to the whole SO-CPR dataset as
described in the text. Between-cluster dissimilarity is shown on the x-axis (note that the axis is
reversed to show more similar clusters on the right). Significant differences are those greater than
twice the mean dissimilarity in the 20-segment subsets (shown as the vertical red line at
dissimilarity of 1.25). Significant clusters (= communities) are shown as different colours on the
furthest right bar. The largest clusters are numbered. The three coloured bars (see colour scales)
show the distribution of data by three variables: season, day, and latitude. “Season” here is the
later year of an Antarctic summer (e.g., 2000 for December 1999 and January 2000). “Day” is the
number of days after 1 January within a season.
The optimum number of trees to use to avoid overfitting based on the K-fold analysis was 161 (range
of 57–699), with best R2 values of 0.18 (0.06–0.28). Hence, 150 trees were used for all BRT models of
cluster distances to obtain models with a high degree of generality. The separate BRT models explained
0.38 of the variance (range of 0.28–0.45), whereas the combined model explained on average 0.50 of
the variance (Table 3-5). The climAnom model performed best across all clusters and the actual model
the worst, though the difference in performance across the models was small. The CVs were relatively
high (mean of 0.48 for the combined model), but all models gave significant reductions in variance than
a null model at better than the 0.05% level for all clusters (F>1.34, p < 0.0005, for N=47 574). In terms
of dependencies on environmental variables, across all clusters, mixed layer depth was the most
important variable, explaining 12.5% of the total variance in cluster distances (Table 3-5). The other
variables in decreasing order of overall importance were SST, primary productivity (vgpm), time of day
(solar zenith angle, Ecoszen), chl-a concentration (chl), month, SST gradient (indicative of frontal
activity), and sea ice concentration (seaIce). The fact that month explained a low proportion of variance
(4.2% overall) means that cluster or community membership is primarily determined from the
oceanographic conditions rather than the date, which gives more likelihood that these models can be
used to understand the effects of oceanographic change on zooplankton communities.
After BRT modelling of cluster dissimilarities, Clusters 6, 22, and 24 had few or no members and were
excluded from further consideration. One cause of these changes is that the initial zooplankton
clustering was based on pooled data (20 CPR data points to establish cluster significance), whereas the
BRT modelling was based on individual CPR data points (not pooled), and this affected dissimilarity
based on the fine-scale Sørensen distance metric. Comparing results from different block sizes indicated
that this is a relatively minor effect, however. The second and more important factor is that the original
clustering was based exclusively on CPR data (i.e., zooplankton counts), and these clusters overlapped
considerably in environmental space. In other words, zooplankton communities in a given cluster
substantially shared the same environmental conditions as different clusters. The BRT analysis
attempted to separate out these different zooplankton communities based on environmental conditions.
Because many points in different clusters were closer in environmental space than those in the same
cluster, this environmental-based analysis led to changes in cluster memberships. Three Clusters (6, 22,
and 24) did not occupy distinct environmental space so using environmental conditions to separate
clusters meant that these clusters could not be resolved and were subsumed into other clusters.
Two analyses were used to show that, despite these changes, the remaining eight clusters (Clusters 1,
4, 9, 10, 37, 39, 40, and 43) remained significantly different from each other when measured by the
original cluster distances and, crucially, retained the characteristics (in terms of the zooplankton
communities) as the original clusters. First, within cluster dissimilarities were always significantly
smaller than between cluster dissimilarities (one-sided t-test, p<0.01). Second, analysis showed that
after the BRT analysis, clusters had the same key characteristics as in the original clustering. Cluster
(zooplankton community) characteristics (Table 3-6) did not change significantly after the BRT
modelling from the original clustering. The post-BRT cluster characteristics explained 90% of the
variance in the original cluster characteristics for the broad taxonomic groups, 77% of the variance for
the fine taxonomic groups, and 76% of the variance for the environmental data across all environmental
variables.
Standardised difference
Cluster Log abundance (broad taxa) Probability of presence (fine taxa) (environmental)
1 Foraminifera +0.86 *** Rhincalanus gigas +0.41 *** mld0p125_mP +0.82 ***
1 Oithona +0.85 *** Ctenocalanus citer +0.15 *** mld0p125_mPClim +0.80 ***
1 Total abundance +0.72 *** Clausocalanus sp. +0.08 *** mld0p030_mP +0.76 ***
1 Fritillaria sp. +0.69 *** Ostracoda indet -0.09 *** mld0p030_mPClim +0.75 ***
1 Copepoda +0.46 *** Neocalanus tonsus -0.09 *** sst_mPClim -0.61 ***
1 Ostracoda -0.23 *** Metridia sp. -0.10 *** sst_mP -0.61 ***
4 Copepoda +1.02 *** Calanus simillimus +0.42 *** SeaIce_cum -0.37 ***
4 Total abundance +0.93 *** Clausocalanus sp. +0.31 *** SeaIce_anMean -0.37 ***
4 Oithona +0.77 *** Rhincalanus gigas +0.28 *** chl_mPClim -0.41 ***
4 Pteropods +0.68 *** Ostracoda indet +0.22 *** chl_cum -0.42 ***
4 Ostracoda +0.58 *** Ctenocalanus citer +0.20 *** vgpm_anMean -0.48 ***
4 Amphipoda +0.46 *** Calanoides acutus +0.15 *** vgpm_cum -0.57 ***
9 Copepoda +0.06 *** Oithona sp. +0.06 *** mld0p030_anMean -0.35 ***
9 Oithona +0.04 *** Metridia sp. +0.05 *** mld0p030_cum -0.35 ***
9 Total abundance -0.05 *** Calanoides acutus +0.03 *** mld0p125_anMean -0.35 ***
9 Pteropods -0.09 *** Neocalanus tonsus +0.03 *** mld0p125_cum -0.37 ***
9 Fritillaria sp. -0.14 *** Ctenocalanus citer -0.05 *** mld0p125_mPClim -0.46 ***
9 Foraminifera -0.50 *** Rhincalanus gigas -0.08 *** mld0p125_mP -0.47 ***
10 Foraminifera +1.15 *** Metridia sp. +0.18 *** mld0p125_mP +1.08 ***
10 Euphausiidae +0.31 *** Clausocalanus sp. +0.14 *** mld0p125_mPClim +1.08 ***
10 Chaetognatha +0.19 *** Thalicidae +0.12 *** mld0p030_mP +0.94 ***
10 Pteropods -0.17 *** Ostracoda indet +0.12 *** mld0p030_mPClim +0.92 ***
10 Fritillaria sp. -0.46 *** Calanoides acutus -0.14 *** sst_cum +0.81 ***
10 Oithona -0.55 *** Rhincalanus gigas -0.26 *** sst_anMean +0.78 ***
Standardised difference
Cluster Log abundance (broad taxa) Probability of presence (fine taxa) (environmental)
37 Pteropods -0.40 *** Neocalanus tonsus +0.36 *** sst_anMean +1.58 ***
37 Copepoda -0.41 *** Calanoides acutus -0.12 *** sst_cum +1.58 ***
37 Foraminifera -0.47 *** Ctenocalanus citer -0.14 *** vgpm_cum +1.47 ***
37 Fritillaria sp. -0.67 *** Oithona sp. -0.21 *** sst_mP +1.38 ***
37 Total abundance -0.76 *** Rhincalanus gigas -0.32 *** sst_mPClim +1.37 ***
37 Oithona -1.14 *** Calanus simillimus -0.40 *** vgpm_anMean +1.34 ***
39 Euphausiidae -0.29 *** Ctenocalanus citer -0.14 *** sst_mP +0.67 ***
39 Fritillaria sp. -0.31 *** Metridia sp. -0.14 *** sst_mPClim +0.67 ***
39 Ostracoda -0.33 *** Calanoides acutus -0.14 *** sst_anMean +0.62 ***
39 Oithona -0.75 *** Calanus simillimus -0.22 *** sst_cum +0.61 ***
39 Total abundance -0.77 *** Clausocalanus sp. -0.28 *** vgpm_cum +0.51 ***
39 Copepoda -0.93 *** Rhincalanus gigas -0.29 *** vgpm_anMean +0.46 ***
40 Oikopleura sp. -0.29 *** Metridia sp. -0.11 *** SeaIce_cum +1.24 ***
40 Ostracoda -0.31 *** Ostracoda indet -0.12 *** SeaIce_anMean +1.22 ***
40 Foraminifera -0.48 *** Ctenocalanus citer -0.18 *** chl_cum +1.06 ***
40 Oithona -0.74 *** Rhincalanus gigas -0.27 *** chl_mPClim +0.99 ***
40 Total abundance -0.88 *** Clausocalanus sp. -0.33 *** SeaIce_mPClim +0.86 ***
40 Copepoda -0.96 *** Calanus simillimus -0.35 *** SeaIce_mP +0.81 ***
43 Foraminifera -0.66 *** Ostracoda indet -0.16 *** SeaIce_cum +2.75 ***
43 Oikopleura sp. -0.67 *** Ctenocalanus citer -0.28 *** SeaIce_mPClim +2.72 ***
43 Fritillaria sp. -0.71 *** Rhincalanus gigas -0.33 *** SeaIce_anMean +2.69 ***
43 Copepoda -1.61 *** Clausocalanus sp. -0.42 *** SeaIce_mP +2.41 ***
43 Oithona -1.70 *** Calanus simillimus -0.45 *** sstgrad_cum -1.91 ***
43 Total abundance -1.99 *** Oithona sp. -0.54 *** sstgrad_anMean -1.98 ***
Cluster characteristics are shown in Table 3-6. Clusters 1 and 4 had high total zooplankton abundance.
Cluster 1 was characterised by high Foraminifera, Oithona spp., and Fritillaria spp. abundances, high
probability of presence of Rhincalanus gigas, and deep mixed layer depths. Cluster 4 was characterised
by high copepod (and total zooplankton) abundance, high probability of presence of Calanus simillimus,
and low sea ice concentrations. Cluster 9, which covered the largest area, was characterised by
“average” conditions in most variables apart from low foraminifera abundance and shallow mixed layer
depths. Cluster 10 had high abundances of foraminifera, euphausiids, and Metridia spp., warm water,
and a deep mixed layer. Cluster 37 had low abundances of Oithona spp., low overall zooplankton
abundances, high probability of presence of Neocalanus tonsus, and very warm productive waters.
Cluster 39 had low zooplankton abundances (especially low copepod abundances), and warm and
productive water. Cluster 40 was also characterised by low total zooplankton abundance, low copepod
abundance, and had high sea ice and high chl-a concentration. Cluster 43 had the lowest overall
zooplankton abundances (especially for Oithona spp. and calanoid copepods) and the greatest effect of
sea ice. A brief title defines each cluster: Cluster 1, “Polar Front-Rhincalanus gigas zone”; Cluster 4,
“Polar Front-Calanus simillimus zone”; Cluster 9, “Open ocean zone”; Cluster 10, “Polar Front-
foraminifera zone”; Cluster 37, “warm Subantarctic zone”; Cluster 39, “cool Subantarctic zone”;
Cluster 40, “seasonal sea ice zone”; Cluster 43, “extreme sea ice zone”.
c d
e f
Figure 3-12: Zooplankton communities based on predicted cluster distances using the combined BRT model
and oceanographic environmental information for climatological months. a: October; b:
November; c: December; d: January; e: February; f: March. Colours show the different
zooplankton communities (clusters). Areas with environmental conditions outside the range of the
training data were excluded from the prediction.
Sen’s slope
Cluster Description Area (%) Z p
(% per y)
Southern Ocean
1 PF-Rhincalanus gigas zone 17.7 6.36 0.000*** +2.65
4 PF-Calanus simillimus zone 5.7 3.84 0.000*** +1.49
9 Open ocean zone 37.6 -2.02 0.043* -0.39
10 PF-Foraminifera zone 1.1 1.52 0.129 +0.73
37 Warm Subantarctic zone 22.1 0.76 0.447 +0.08
39 Cool Subantarctic zone 10.2 -3.71 0.000*** -0.72
40 Seasonal sea ice zone 4.7 -0.90 0.368 -0.29
43 Extreme sea ice zone 0.9 -5.56 0.000*** -7.04
Ross Sea
1 PF-Rhincalanus gigas zone 12.9 4.53 0.000*** +2.54
4 PF-Calanus simillimus zone 2.0 0.80 0.423 +0.68
9 Open ocean zone 33.9 -1.44 0.148 -0.42
10 PF-foraminifera zone 1.1 0.46 0.642 +0.41
37 Warm Subantarctic zone 29.4 3.38 0.001*** +0.76
39 Cool Subantarctic zone 13.6 -1.54 0.124 -0.49
40 Seasonal sea ice zone 6.6 -2.37 0.018* -1.08
43 Extreme sea ice zone 0.4 -3.88 0.000*** -3.08
a b
Figure 3-13: Average copepod community size (ACCS) index. a: Spatial average ACCS from the raw CPR
data. b: Predicted using the combined BRT model and oceanographic environmental information
(6-month, long term average). Areas with environmental conditions outside the range of the
training data were excluded from the prediction.
c d
e f
Figure 3-14: Predicted average copepod community size (ACCS) index by month. a: October; b:
November; c: December; d: January; e: February; f: March. Areas with environmental conditions
outside the range of the training data were excluded from the prediction.
Figure 3-15: Long-term trends in Average Copepod Community Size (ACCS) index. Sen’s slope trends
(average change per year for the period 1998–2018) a: Long-term trends in ACCS from the raw
CPR data. White indicates no significant trend; grey indicates not enough data for trend analysis.
b: Trends in ACCS predicted using the combined BRT model and oceanographic environmental
information. White indicates no significant trend. “Significant” means at better than the 0.05
confidence level assessed using the de-seasonalised Mann-Kendall test. Note that the colour scales
are different (a: -0.10 to +0.10 mm/y; b: -0.02 to +0.02 mm/y).
3.7 Microplastics
Microplastics, primarily in the form of threads, were present in 1597 of the 4159 samples analysed
between 2009 and 2018. The percentage of samples with microplastics varied little among the different
water masses, between ~25 to 40% containing at least one thread (Figure 3-15). Variability among years
was higher, with ~5 to 85% of the samples collected during a given season containing microfibres and
higher percentages in the most recent two seasons (2016–2017 and 2017–2018).
Figure 3-16: a: percentage of samples with microplastics by water type, and b: percentage of samples with
microplastics by season.
Different types of threads, as indicated by thread colour (called ‘thread diversity’), present in different
water masses and seasons were investigated. This metric varied little between water masses (Figure
3-16a) or season (Figure 3-16b). There did seem to be a prevalence of blue and red threads, with higher
Figure 3-17: a: thread ‘diversity’ (types of different coloured threads) by water mass; b: thread ‘diversity’
by season, and c: colour of threads found throughout the whole time series.
4. DISCUSSION
There are well-known advantages and disadvantages to sampling zooplankton communities by the CPR
rather than using vertical, depth-resolved, or oblique net hauls from research ships (e.g., Reid et al.
2003, Hosie et al. 2003, Pinkerton et al. 2010b). The main disadvantages include surface-only sampling,
more damaged specimens, greater ability to avoid capture, and poorer retention of small zooplankton
species for the CPR versus research nets. However, for long-term monitoring of zooplankton over large
areas, it is generally accepted that the advantage of sampling consistently and cost-effectively over large
areas and over extended periods of time (decades) outweighs the disadvantages of CPR sampling.
This report has used a dual broad- and fine-scale grouping of SO-CPR data for analysis. These
groupings were chosen to maximise the useful information content of the dataset whilst being minimally
affected by uncertainties caused by the different levels of taxonomic identification possible in different
samples. Retaining unidentified taxa in the broad grouping enabled the best analysis of spatial, seasonal,
and long-term changes in Southern Ocean zooplankton generally. Excluding poorly-identified
specimens and using the finely-grouped data allowed more detailed examination of the composition of
the zooplankton community.
These analyses of the raw CPR data in the Southern Ocean confirmed the expectation that zooplankton
abundances had high spatial and temporal variability at the broad taxonomic level. This variability
meant that trend analysis of the raw CPR data, even at a coarse spatial scale of 300 km, had limited
power to identify spatially-coherent, long-term changes. Using a data-mining analysis consistent with
best practice (e.g., Hochachka et al. 2007), the underlying spatial and temporal patterns in the CPR data
for many of the broad groups was revealed. The BRT models effectively allowed regionally-restricted
CPR sampling to be extrapolated to the whole Southern Ocean using satellite observations and data-
assimilating hindcasts of ocean primary productivity, phytoplankton biomass, surface water
temperature, mixed layer depth, sea ice, and frontal activity.
No standard method is available for zooplankton community analysis and the approach developed in
the present study is a step forward in this kind of analysis. Our method appropriately captured the
combination of broad-scale and fine-scale variability in zooplankton communities from the CPR data
and robustly obtained the number of significantly-different clusters.
Using cluster analysis, 11 “clusters” in the CPR data were identified as distinct zooplankton
communities in the Southern Ocean. Of these, 8 could be resolved using environmental properties: two
communities in the Subantarctic water (Cluster 37, warm Subantarctic zone and Cluster 39, cool
Subantarctic zone); the largest open ocean zone (Cluster 9); three frontal zone communities (Cluster 1,
Polar Front-Rhincalanus gigas zone; Cluster 4, Polar Front-Calanus simillimus zone; and Cluster 10:
Polar Front-foraminifera zone); and two se -ice communities (Cluster 40, seasonal sea ice zone; Cluster
43, extreme sea ice).
Cluster 40 is the least well resolved cluster based on environmental properties, and in March becomes
split latitudinally making its interpretation problematic. Otherwise, the positions and characteristics of
The CPR data in broad taxonomic groups were analysed by oceanographic zone in the Ross Sea region
and in large (300 km) spatial blocks. From analysis of the “raw” CPR data, only a few significant trends
were found, and the magnitudes of these significant trends tended to be very high (>10% y-1), especially
for Fritillaria spp. These results contrasted with those from the model-based analysis; the BRT models
showed significant trends in the abundances of almost all broad zooplankton groups, but with very low
rates of change (less than 1% y-1).
The fact that month explained a relatively low proportion of variance in the abundances of the broad
taxonomic groups (4.0% overall) means that predicted values based on the BRT models are primarily
determined from the oceanographic conditions rather than the date per se. This result gave more
likelihood that the BRT model-based analysis can be used to understand the effects of oceanographic
change on zooplankton communities. The fact that the same types of changes were predicted by
different BRT models (cumulative, annual, etc.) based on different sets of environmental variables also
supported the conclusion that the predictions of long-term changes from the BRT modelling were robust
to the particular type of model developed.
What seemed to be occurring is that the environmental conditions have generally changed in a way
likely to be beneficial (in terms of leading to higher abundances) for most broad taxonomic groups of
zooplankton (Oithona spp. and calanoid copepods, euphausiids, foraminifera, Fritillaria spp.) over the
last 20 years (1997–2018). The environmental changes were likely to have been especially beneficial
in the area between the Subantarctic Front and the southern boundary of the ACC, though the changes
were “patchy” geographically. Positive trends in zooplankton abundances tended to follow increased
ocean productivity (higher chl), warmer water (higher sst), and deeper mixed layers (higher mld0p125).
In contrast, the environmental conditions for pteropods had worsened over the last 20 years over the
Ross Sea shelf especially, and this corresponded to lower phytoplankton biomass (chl) in the region of
decrease.
Superimposed on these environmentally-driven changes in zooplankton communities was a very large
amount of spatial and short-term variability which was not captured by the models. This intrinsic
variability in zooplankton abundances and the limited nature of the sampling prevents the underlying
changes to zooplankton abundances being apparent in simple analysis of the CPR data alone. Hence,
occasional, spatially-incoherent, high-magnitude, significant trends were identified from the “raw” CPR
data, but the large-area, relatively small-magnitude, environmentally-driven trends that are likely to be
present were not resolved. As the time series of observation increases, the power to observe these long-
term changes above the short-term fluctuations will improve.
Corresponding to the change in the abundances of broad taxonomic groups of zooplankton, the analysis
showed that the environmental properties have been changing in a way this is likely to lead to significant
long-term changes in the spatial extent of many zooplankton communities over time. Based on the
environmental change and model relationships between the environment and communities, two zones
in the vicinity of the Polar Front (Cluster 1, Polar Front-Rhincalanus gigas zone; Cluster 4, Polar Front-
Calanus simillimus zone) are predicted to have increased in area between 1997 and 2018. In contrast,
Cluster 39 (cool Subantarctic zone) and southernmost community (Cluster 43) are predicted to have
BRT modelling was used to investigate changes in the average size of copepods in the Southern Ocean.
An updated average size for groups of zooplankton (at the genus level or above) in the Southern Ocean
was used (Appendix 3). Based on the BRT model, this analysis suggested that the ACCS is decreasing
at high latitudes in the Southern Ocean (approximately south of the southern limit of the ACC). These
high latitude areas showed no trends, or slightly cooling trends, in SST. In contrast, north of the southern
limit of the ACC, surface waters were warming strongly in most areas. The warming of the surface
ocean in the Southern Ocean seems to be associated with a shift to larger zooplankton and, in contrast,
cooling with a shift to smaller zooplankton.
A caveat to these results arises from the factors involved in calculating the ACCS. Because the average
female adult size is used and not a measurement of each specimen within the sample which could
include other developmental stages, the estimate will be more ‘true’ if the real catch comprised more
adults. More importantly, the catchability of a zooplankter by the CPR varies with size, with small
organisms like Oithona spp. likely to be under-sampled because of net efficiencies, and larger copepods
and euphausiids under-sampled because larger organisms tend to live deeper in the water column, with
more pronounced diel vertical migration (De Robertis 2002; Steinberg et al. 2002). If communities are
shifting significantly into the size-spectral area that is under-sampled by the CPR (either too large or
too small), this effect could reduce the reliability of trends in the ACCS. In addition, because most of
the sampling occurs within a restricted time-frame (November-February), ocean-warming induced
changes in phenology could result in our CPR sampling missing (or under-sampling) the abundance of
species which have pronounced seasonal peaks, e.g., Calanoides acutus (Atkinson 1998).
Notwithstanding these caveats, these results are contrary to the expectation that warming oceans would
see a shift to warm-water species that tend to be smaller (Lopez & Anadon 2008). The rationale for this
premise is that tropical species are typically very small, temperate species tend to be bigger, and the
largest species are usually found in sub-polar/polar waters or in deeper cooler waters. However, iron-
limited Subantarctic waters also tend to have phytoplankton communities dominated by small cells,
which is similar to tropical systems, and thus have the same effect of reducing zooplankton species size
(Bradford-Grieve et al. 2003). The near surface Antarctic waters, notably in higher latitudes, are
typically characterised by large species such as Calanoides acutus, Calanus propinquus, Rhincalanus
gigas, and several species of Paraeuchaeta. However, the Southern Ocean also has quite a broad size
range of numerous abundant small species of 1–2 mm common in the CPR samples, e.g., species of
Clausocalanus, Ctenocalanus, and then there is the numerically most abundant species, the
cosmopolitan Oithona spp. at < 1 mm. Given that the BRT model predicts that total zooplankton
abundances are generally increasing, especially just to the north of the seasonal limit of sea ice (Figure
3-8), it is likely that decreases in ACCS in these high latitude areas are driven by the increasing
abundance of small zooplankton taxa. Of the broad groups, Oithona spp. shows the greatest increasing
trend at these high latitudes and is likely responsible for some or much of the decreases in ACCS.
Areas of increasing trends in ACCS are mainly located between the Subantarctic Front and the Polar
Front and correspond with increases in the abundance of Foraminifera (Figure 3-8). Foraminifera
abundances were not used to calculate the ACCS, which is restricted to abundances of copepods. The
hypothesis from this analysis therefore is that large copepod species are increasing in abundance faster
than smaller copepod species in the Subantarctic Front and Polar Front areas and that this is ecologically
associated with increasing abundances of Foraminifera in these areas. It is possible that increasing
abundances of Foraminifera in these areas are fueling (i.e., as prey) increases in the abundance of large,
carnivorous zooplankton species, or that the underlying ecological drivers increasing copepod
abundances are also positively affecting Foraminifera, perhaps linked to increases in productivity or
more palatable phytoplankton.
There is interest in considering whether zooplankton abundances are different in the Ross Sea region
than in the rest of the Southern Ocean. Some studies have suggested that mesozooplankton biomass in
the Ross Sea is unusually low (Tagliabue & Arrigo 2003), which was not confirmed by later analysis
(e.g., Stevens et al. 2014, Pinkerton et al. 2010a, b, Robinson et al. 2014). In this study, spatial statistical
models were used to explore large-area differences in abundances between the regions and whether this
could be due to differences in environmental conditions between the regions or different ecological
characteristics.
Overall, analysis presented in this report showed that zooplankton abundances were higher in the Ross
Sea region than the Southern Ocean as whole, by a factor of about 2 on average, but this factor varied
between zooplankton taxa. Foraminifera and Fritillaria spp. were more than 3 times as abundant in the
Ross Sea region than elsewhere in the Southern Ocean. The reasons for these differences in abundances
are not known, but they are not related to the environmental conditions considered here (primary
productivity, phytoplankton biomass, SST, mixed layer depth, sea ice, or frontal activity). The spatial
differences in zooplankton could be related to factors including (but not restricted to): (1) different types
of phytoplankton in the Ross Sea region; (2) different microbial communities; and (3) different grazing
pressure by higher trophic level consumers. Seeking large-area information on these factors would be
a useful next step from this study and is underway elsewhere (e.g., under the Ross-RAMP project 6).
Interestingly, long-term trends in the abundances of zooplankton groups in the Ross Sea region were
very similar irrespective of whether a Southern Ocean or specific Ross Sea region model was used
(Table 3-4) indicating the changes to the environmental envelopes of zooplankton are likely to be
changing at similar rates inside and outside the Ross Sea region.
The data and results developed in this report pave the way to explore the effects of climate change on
Southern Ocean zooplankton populations. A large global effort is focussed on forecasting the effects of
increasing carbon dioxide levels in the atmosphere on atmospheric and oceanographic conditions. For
example, the CMIP5 7 and CMIP6 programmes bring together more than 14 Earth-system models from
leading research groups around the world in a way in which the results can be examined and re-used
(Bopp et al. 2013). Previous work has examined which of the many Earth system models work best in
the New Zealand and in the Ross Sea regions by comparing model results with observational data from
satellites (Rickard & Behrens 2016).
Less developed are methods to use these predicted changes in physical and chemical characteristics to
explore the potential effects on living organisms. Zooplankton, although a key part of ocean ecosystems
worldwide, are particularly problematic because they are ecologically diverse (hundreds of taxa),
poorly-observed, and spatially and temporally variable. The relationships developed in the present study
provide a step forward by enabling the forecasting of the effect of climate change on key taxonomic
groups of zooplankton and zooplankton communities. Although carrying out this analysis is beyond the
scope of this report, this analysis will continue under funding from the MBIE-Endeavour RAMP 8
programme and potentially the Antarctic Science Platform.
6
Ross-RAMP (Ross Sea region Research and Monitoring Programme): MBIE Endeavour Fund programme,
2017–2023. Pinkerton PI (NIWA).
7
Coupled Model Intercomparison Project - https://cmip.llnl.gov/.
8
Ross Sea Research and Monitoring programme associated with the Ross Sea region MPA.
A formal comparison between Northern Hemisphere and Southern Hemisphere CPR datasets was not
possible because of methodological differences in the way CPR zooplankton samples were analysed,
which means that treating the two datasets in the same way would be inappropriate (Beaugrand et al.
2001, 2003). Also, the CPR survey in the North Atlantic started in 1948, so there is less ability to track
long-term (decadal) changes in zooplankton in the Southern Ocean because the SO-CPR survey started
in 1991.
Trends in zooplankton in the North Atlantic show communities shifting polewards, accompanying
warming ocean temperatures (Beaugrand et al. 2002), with once-tropical species now found in more
temperate latitudes, and temperate species shifting to polar waters (Beaugrand et al. 2002, Beaugrand
et al. 2001, Beaugrand et al. 2003). These changes in the North Atlantic are not necessarily forced by
the local phytoplankton communities, because these seem to be responding to decadal, rather than
longer-term, forcing (Harris et al. 2014, Harris et al. 2015). The changes would tend to lead to a
decreasing ACCS index in the North Atlantic over time.
The zooplankton community (cluster) analysis of Southern Ocean zooplankton data showed a similar
pattern of temperate zooplankton communities (especially in the Polar Front region) expanding
polewards and the high-latitude polar zooplankton communities (Antarctic “sea ice” zooplankton
communities, Cluster 43) being squeezed out. The decline in abundance of pteropods over the Ross Sea
shelf is also notable (Figure 3-7). Analysis of the ACCS patterns in the SO-CPR data also suggested
more smaller copepods compared with more sizeable polar species at high latitudes. There were
corresponding negative (decreasing) long-term trends in ACCS in the Southern Ocean at high latitudes.
What is different between hemispheres is that the negative trends in ACCS in the high-latitude Southern
Ocean correspond to increases in SST (warming) rather than cooling in the North Atlantic. Also, the
drivers of the changes in zooplankton communities between the two hemispheres may be different; in
the Southern Ocean, the increases in abundances of small copepods are tentatively interpreted as
resulting from improved food-conditions (given positive relationships with chl-a) rather than the longer-
term (potentially physical) forcing in the North Atlantic.
It is important to discuss both the different latitudinal range sampled as part of each of these
programmes, and the inherent differences in the oceanography of the southern versus the northern
hemisphere. The North Atlantic CPR coverage spans a slightly different latitudinal range, with the bulk
of measurements carried out between 40°–65° N (Beaugrand & Ibanez 2002), i.e., slightly more
temperate than the SO-CPR sampling. The oceanography of the North Atlantic basin is also quite
different from the Southern Ocean, with large portions of land that prohibit a circumpolar water
circulation pattern being established. This means the copepod communities sampled as part of the North
Atlantic programme inhabit large areas of somewhat homogenous oceanic conditions, e.g., large areas
of Subtropical waters not structured by the Subtropical Front, or polar waters without the strong
Antarctic Polar Front. These fundamental oceanographic differences between hemispheres mean that
the differences in zooplankton abundance patterns might not be so surprising. Furthermore, there have
been differences in ocean heat absorption between hemispheres. Since 2006, an estimated 60–90% of
global ocean heat content change associated with global warming has occurred in the Southern Ocean
(Roemmich et al. 2015, Llovel & Terray 2016, Sallée 2018), suggesting that changes in the Southern
Hemisphere would be expected to be more dramatic than in the Northern Hemisphere.
The time series analysed here is presently too short to determine these long-term changes with certainty,
but the BRT model results suggest that the large-scale pattern of Southern Hemisphere warming is likely
the cause in changing communities, with higher total abundances and relative increases in key
taxonomic groups of zooplankton which are not paralleled by changes in the North Atlantic time series.
Longer time series and enhanced spatial resolution will allow us to confirm these patterns.
This report has explored the relation between zooplankton abundance, zooplankton communities, and
environmental conditions in the Southern Ocean, with a focus on the Ross Sea region. It seems that
changes to the environmental conditions have already occurred and are likely to continue to cause long-
term change in Southern Ocean zooplankton. The effect of these changes on higher trophic level species
like fish, seabirds, and marine mammals is not known. The data and analyses presented here are suitable
to underpin multi-trophic studies to see if changes to zooplankton can help explain changes to larval
fishes, mesopelagics, seabirds, and marine mammals across the Southern Ocean. Continued collection
of zooplankton data in the future will provide an ability to confirm empirically whether these patterns
and trends are genuine, and how zooplankton taxonomic groups and communities change in the future.
5. MANAGEMENT IMPLICATIONS
1. Concentrations of chl-a in the Southern Ocean (likely indicative of changes to oceanic primary
production) are generally increasing, especially around the southern limit of the Antarctic
Circumpolar Current (particularly in East Antarctica), around the Argentine continental shelf,
and in the Subtropical Front around New Zealand. In the Ross Sea region (especially over the
continental shelf), however, chl-a concentrations show significant negative long-term trends
(decreases).
2. Abundances of zooplankton are higher in the Ross Sea region than in the Southern Ocean as
whole, by a factor of about 2 on average but varying between zooplankton taxa. Abundances
of Foraminifera and Fritillaria spp. were the most different in the Ross Sea region than
elsewhere. These differences are not due to the Ross Sea region having unusual environmental
conditions and are more likely to be ecologically-driven.
3. Predictions of abundances and distributions of key zooplankton groups based on environmental
conditions (phytoplankton biomass, primary productivity, water temperature, mixed layer
depth, sea ice, or frontal activity) are generally good at representing average spatial patterns in
occurrence of particular broad groups of zooplankton taxa. The models developed in this study
can be used in conjunction with Earth-system models to forecast the effects of climate change
on zooplankton in the Southern Ocean. This has relevance to anticipating, understanding, and
managing change in Antarctic ecosystems because variability in zooplankton will affect other
parts of the marine food-web, including fishes, seabirds, and mammals.
4. Environmental conditions for most broad taxonomic groups of zooplankton in the Southern
Ocean have generally improved over the last 20 years (1997–2018) and these environmental
changes are likely to drive increases in abundances of most zooplankton groups in the future.
5. In contrast, the environmental conditions for pteropods (small marine snails living in the water
column) seems to have worsened over the last 20 years over the Ross Sea shelf especially. The
negative pressure on pteropod abundance may have implications for the Ross Sea shelf
ecosystem where pteropods are a major contributor to the flow of organic matter to the seabed.
6. There were significant increases in the Average Copepod Community Size (ACCS) metric in
the Ross Sea region during the sampling period, suggesting a shift towards larger copepod
species. Ocean warming in the Southern Ocean generally seems to be associated with a shift to
larger zooplankton, and cooling with a shift to smaller cold-water copepods such as Oithona
spp. This finding is contrary to the prevailing hypothesis that ocean warming favours smaller
copepods.
7. Superimposed on these predicted changes to zooplankton due to trends in environmental
conditions were shorter-term and localised fluctuations in zooplankton abundances. These
6. ACKNOWLEDGMENTS
This work was completed under Ministry for Primary Industries project ZBD2013-03 Objective 10.
Additional resources for CPR sample and data analysis were provided by NIWA Strategic Science
Investment Fund (SIFF) Coasts & Oceans, Programme 4: Structure and function of marine ecosystems,
and MBIE Endeavour Fund programme Ross-RAMP (C01X1710). The work would not have been
possible without the assistance of Sanford Limited, Dave Bilton, Monique Messina, and the officers
and crew of FV San Aotea II. We acknowledge the scientists, officers, and crew of RV Tangaroa for
CPR data collection on voyages between 2006 and 2013. CPR equipment and training were provided
by Graham Hosie and John Kitchener from the SO-CPR survey (Australian Antarctic Division, Hobart,
Australia) and The Marine Biological Association CPR Survey (formerly SAHFOS), Plymouth, UK.
We thank Ashley Rowden (NIWA) for statistical discussions and for review of this report which helped
improve it. We also thank Barb Hayden (NIWA) for review.
7. REFERENCES
Ainley, D.; Ballard, G.; Ackley, S.; 10 other co-authors. (2007). Paradigm lost, or is top-down forcing
no longer significant in the Antarctic marine ecosystem? Antarctic Science 19(3): 283–290.
Atkinson, A. (1998). Life cycle strategies of epipelagic copepods in the Southern Ocean. Journal of
Marine Systems 15: 289–311. doi: 10.1016/s0924-7963(97)00081-x.
Atkinson, A.; Siegel, V.; Pakhomov, E.A.; Rothery P.; Loeb V.; Ross R.M.; Quetin L.B.; Schmidt, K.;
Fretwell, P.; Murphy, E.J.; Tarling, G.A.; Fleming, A.H. (2008). Oceanic circumpolar habitats
of Antarctic krill. Marine Ecology Progress Series 362: 1–23.
Bathmann, U.; Priddle, J.; Treguer, P.; Lucas, M.I.; Parslow, J.; Hall, J. (2000). Plankton ecology and
biogeochemistry in the Southern Ocean: a first review of Southern Ocean JGOFS. In: Hanson,
R.; Ducklow, H.W.; Field, J.G. (Eds.), The Changing Ocean Carbon Cycle: A Mid-Term
Synthesis of the JGOFS. Cambridge University Press, UK.
Batten, S.D.; Chiba, S.; Edwards, M.; Hall, J.; Hosie, G.; Melrose, C.; Muxagata, E.; Richardson, A. J.;
Verheye, H.M.; Helaouet, P.; Rochester, W.; Stevens, D. (2016). Chapter 5.3: The Status of
Zooplankton Populations. In: UNESCO IOC and UNEP (2016). The Open Ocean: Status and
Trends. United Nations Environment Programme, Nairobi, pp. 154–165.
Table 8-1. Groupings of the standard CPR identification groups used in the analysis. Two levels of grouping
were used: “broad” and “fine”. In a small number of fine groups, subgroups were also used; these
subgroups are shown separated with ‘|’. In these cases, counts were partitioned pro rata amongst subgroups
on a sample-by-sample basis unless none of the subgroup counts were greater than zero, in which case the
average proportions in the tow were used. Broad and fine groups marked “other” were excluded from
analysis.
chl mg m-3 Near surface concentration of chl a estimated Figure 9-1 Pinkerton et al. (2018); Hooker et
from the default (case 1) processing of MODIS- al. (1992); Murphy et al. (2001);
Aqua (version R2018.0) and SeaWiFS (version NASA (2018a, b)
R2018.0) satellite sensors, blended using overlap
period 2002-2010. The data set spans the period
September 1997 to August 2018.
sst °C Sea Surface Temperature (SST). The dataset used Figure 9-2 Reynolds et al. (2002)
is the Optimum Interpolation Sea Surface
Temperature (OISST), version 2, based on
satellite measurements by the Advanced Very
High Resolution Radiometer series, operated by
NOAA (National Oceanic and Atmospheric
Administration of the USA) The data set spans
the period 1981 to August 2018.
sstgrad °C km-1 Spatial gradient of SST obtained from a 20 pixel Figure 9-3 NIWA unpublished data
(180 km) smoothed version of OISST, extended
beyond the analysis domain to avoid edge effects.
SeaIce % Sea ice concentration data set derived using Figure 9-4 Cavalieri et al. (1990), updated
measurements from the Scanning Multichannel 2007; https://nsidc.org/
Microwave Radiometer (SMMR) on the Nimbus-
7 satellite and from the Special Sensor
Microwave/Imager (SSM/I) sensors on the
Defense Meteorological Satellite Program's
(DMSP) -F8, -F11, and -F13 satellites.
Measurements from the Special Sensor
Microwave Imager/Sounder (SSMIS) aboard
DMSP-F17 are also included. The data set has
been generated using the Advanced Microwave
Scanning Radiometer - Earth Observing System
(AMSR-E) Bootstrap Algorithm with daily
varying tie-points.
vgpm mgC m-2 d-1 Net primary productivity from the Vertically Figure 9-5 Behrenfeld & Falkowski (1997);
Generalized Production Model (VGPM). Based O’Reilly & Sherman (2016)
on www.science.oregonstate.edu/
ocean.productivity/
mld0p030 m Mixed layer depth calculated from numerical Figure 9-6 Unpublished data:
oceanographic model data using a potential orca.science.oregonstate.edu
density difference of 0.030 kg m-3 from the 1080.by.2160.
surface. Models used are: (1) hycom: from day monthly.hdf.mld.hycom.php
265 (2008) to present; (2) fnmoc: from day 169
(2005) to present; (3) soda: from day 249 (1997)
to end of 2004; (4) tops: from day 001 (2005) to
225 (2010).
mld0p125 m As above but using a potential density difference Figure 9-7 Unpublished data:
of 0.125 kg m-3 orca.science.oregonstate.edu
1080.by.2160.
monthly.hdf.mld.hycom.php
As context to the analysis of the CPR zooplankton data, long-term trends in environmental properties
in the Southern Ocean were investigated by applying Mann Kendall (Mann 1945, Kendall 1975) and
Sen’s slope analysis to each pixel in the dataset. Statistical significance was assessed using unadjusted
Mann-Kendall p values; trends were identified as significant when the p-value was less than 5%. The
magnitude of any statistically significant trends was estimated using the Sen’s slope (Sen 1968). The
Mann-Kendall trend is preferred over linear regression analysis because it is non-parametric
(distribution free) and does not require an assumption that the data are normally distributed. The
insensitivity of the Sen’s slope to outliers means that it is generally the preferred non-parametric method
for estimating a linear trend (Hipel & McLeod 1994).
Example environmental data are shown below: chl (Figure 9-1), sst (Figure 9-2), sstgrad (Figure 9-3),
SeaIce (Figure 9-4), vgpm (Figure 9-5), mld0p030 (Figure 9-6), mld0p125 (Figure 9-7).
c d
Figure 9-1: Chl-a concentration (chl): Example of chl-a environmental data used in this study. a: Long-
term average (CU, “cumulative”); b: Annual average (anMean); c: Monthly climatology (MC), February;
d: Monthly average (MO), February 2011; e: Monthly anomaly (MAnom), February 2011.
c d
Figure 9-2: Sea surface temperature (sst): Example of sst environmental data used in this study. a: Long-
term average (CU, “cumulative”); b: Annual average (anMean); c: Monthly climatology (MC), February;
d: Monthly average (MO), February 2011; e: Monthly anomaly (MAnom), February 2011.
c d
Figure 9-3: Sea surface temperature spatial gradient (sstgrad): Example of sstgrad environmental data used
in this study. a: Long-term average (CU, “cumulative”); b: Annual average (anMean); c: Monthly
climatology (MC), February; d: Monthly average (MO), February 2011; e: Monthly anomaly (MAnom),
February 2011.
c d
Figure 9-4: Sea ice concentration (SeaIce): Example of SeaIce environmental data used in this study. a:
Long-term average (CU, “cumulative”); b: Annual average (anMean); c: Monthly climatology (MC),
February; d: Monthly average (MO), February 2011; e: Monthly anomaly (MAnom), February 2011.
c d
Figure 9-5: Oceanic primary productivity by the vertically generalised production model (vgpm): Example
of vgpm environmental data used in this study. a: Long-term average (CU, “cumulative”); b: Annual
average (anMean); c: Monthly climatology (MC), February; d: Monthly average (MO), February 2011; e:
Monthly anomaly (MAnom), February 2011.
c d
Figure 9-6: Mixed layer depth with a 0.030 kg/m3 density different criterion (mld0p030): Example of
mld0p030 environmental data used in this study. a: Long-term average (CU, “cumulative”); b: Annual
average (anMean); c: Monthly climatology (MC), February; d: Monthly average (MO), February 2011; e:
Monthly anomaly (MAnom), February 2011.
c d
Figure 9-7: Mixed layer depth with a 0.125 kg/m3 density different criterion (mld0p125): Example of
mld0p125 environmental data used in this study. a: Long-term average (CU, “cumulative”); b: Annual
average (anMean); c: Monthly climatology (MC), February; d: Monthly average (MO), February 2011; e:
Monthly anomaly (MAnom), February 2011.
c d
Figure 9-8: Long-term trends (Sen’s slope values) in environmental properties of the Southern Ocean. a:
chl a concentration (chl, mg/m3/y); b: net primary production by the VGPM model (vgpm, mgC/m2/d/y); c:
sea-surface temperature (sst, °C per decade); d: seaice concentration (SeaIce, %/y). [Figure continued
below] Only trends assessed as being significant at better than 5% by the Mann-Kendall trend test are
shown coloured. Nominal positions of (from north) the Subantarctic Front, Polar Fronts, Southern
boundary of the Antarctic Circumpolar Current (Orsi et al. 1995), and the average maximum northern
extent of the sea ice are also shown.
Figure 9-8: continued and concluded. e: mixed layer depth (mld0p0030, m/y); f: mixed layer depth
(mld0p0125, m/y); g: sea surface temperature spatial gradient (sstgrad, m°C/km/y).
Table 10-1: Indicative mid-range length of adult females used to calculate the Average Copepod
Community Size index (ACCS) from Beaugrand et al. (2003) and with recalculated indicative values
appropriate for the Southern Ocean.