Radiolarians (holoplanktonic protozoa) preserved in marine sediments are
commonly used as palaeoclimate proxies for reconstructing past Southern
Ocean environments. Generating reconstructions of past climate based on
microfossil abundances, such as radiolarians, requires a spatially and
environmentally comprehensive reference dataset of modern census counts. The
Southern Ocean Radiolarian (SO-RAD) dataset includes census counts for 238
radiolarian taxa from 228 surface sediment samples located in the Atlantic,
Indian, and southwest Pacific sectors of the Southern Ocean. This compilation
is the largest radiolarian census dataset derived from surface sediment
samples in the Southern Ocean. The SO-RAD dataset may be used as a reference
dataset for palaeoceanographic reconstructions, or for studying modern
radiolarian biogeography and species diversity. As well as describing the
data collection and collation, we include recommendations and guidelines for
cleaning and subsetting the data for users unfamiliar with the procedures
typically used by the radiolarian community. The SO-RAD dataset is available to download from
The Southern Ocean is an important part of the global climate system as a major hub of oceanic circulation and nutrient redistribution. Future changes to its physical, chemical, and biological properties will therefore have global climate implications. Palaeoceanographic reconstructions of the Southern Ocean reveal its response to past climate and how it might respond to future climate change. Marine microfossils are commonly used as climate proxies to obtain such reconstructions (Yasuhara et al., 2020). Plankton distribution and abundance are related to the oceanic conditions prevalent in the water masses where these organisms live. After death, their skeletons settle to the ocean floor, becoming part of the sedimentary record. In the Southern Ocean, two main siliceous microfossil groups, diatoms and radiolarians, have been used to determine past environmental conditions such as ocean temperature (Cortese and Abelmann, 2002; Esper and Gersonde, 2014b; Panitz et al., 2015) and sea-ice coverage (Bianchi and Gersonde, 2004; Crosta et al., 2004; Gersonde et al., 2005; Esper and Gersonde, 2014a; Ferry et al., 2015).
Radiolarians are unicellular, eukaryotic marine microzooplankton, generally
ranging in size from 30–300
Siliceous microfossils, particularly diatoms, are useful for paleoclimate reconstructions at high latitudes and in deep-sea regions where carbonate microfossils are generally not well preserved (Crosta et al., 2005; Gersonde et al., 2005). After diatoms, radiolarians are the second largest source of biogenic silica in marine sediments (Anderson, 2019). Fossil radiolarians have long been used as a palaeo-proxy for Southern Ocean sea-surface temperature (Lozano and Hays, 1976; Cortese and Abelmann, 2002; Cortese et al., 2007; Rogers and De Deckker, 2011; Panitz et al., 2015) and, more recently, have been used for reconstructing subsurface ocean temperature (Hernández-Almeida et al., 2020; Civel-Mazens et al., 2021a, b). Such palaeoceanographic reconstructions rely on a spatially and environmentally comprehensive radiolarian reference dataset.
In this paper we present, for the first time, the Southern Ocean Radiolarian (SO-RAD) dataset consisting of census data for 238 radiolarian taxa from 228 surface sediment samples. We illustrate the spatial and environmental coverage of the dataset, briefly describe the distribution and abundance of taxa, and discuss how the dataset might be cleaned and/or split into subsets prior to analysis. The SO-RAD dataset can be used as a reference dataset for palaeoceanographic reconstructions and to explore new parameters for estimation, as stand-alone data to assist studies in the biogeography and diversity of modern Southern Ocean radiolarian species, or for the development of new statistical methods in microfossil analysis.
The Southern Ocean Radiolarian (SO-RAD) dataset includes census counts for 238 taxa/taxonomic groups from 228 surface sediment samples located in the Atlantic, Indian, and southwest Pacific sectors of the Southern Ocean (Fig. 1). This compilation is the largest dataset available for radiolarian-based palaeoenvironmental reconstructions constrained to the Southern Ocean. Sector-specific datasets have been previously published as follows: Atlantic sector (Abelmann et al., 1999; Cortese and Abelmann, 2002), southwest Pacific sector (Cortese and Prebble, 2015), and Indian sector (Rogers and De Deckker, 2007). These existing data, after a recount of the Indian sector sites to ensure taxonomic consistency, have been compiled as one dataset, which additionally includes new census data from 26 previously unpublished sites in the Ross Sea and the Indian sector of the Southern Ocean. Sediment coring in the Southern Ocean, particularly in areas close to the ice edge where sea-ice coverage varies from season to season, is expensive and difficult to plan. The SO-RAD dataset is, thus, the result of cooperative international partnerships based on decades of oceanographic voyages undertaken by scientists from many institutions and nations. In a compilation such as the SO-RAD dataset, sample locations are not chosen according to uniform spatial coverage criteria, but rather dictated by the availability of, and access to, radiolarian-bearing surface sediment samples.
Surface sediment sites included in the SO-RAD database shown with
In bringing together census data for this global dataset from various sources, the following data limitations and handling procedures are detailed.
Census data from the Atlantic sector include radiolarian counts from 64 sites. Data from 44 sites were published by Abelmann et al. (1999), and 20 sites were subsequently added and published in Cortese and Abelmann (2002). Sites from Abelmann et al. (1999) and Cortese and Abelmann (2002) do not include the full counts of all taxa observed at each site; rather they include only a selection of the taxa observed. Taxa were excluded from these sites prior to their original publication based on criteria outlined in Abelmann et al. (1999) such as their relative abundances (i.e. excluding rare taxa, with the threshold set at 2 %), depth preferences, and clarity of identification. These sites have been included in the dataset, despite containing counts for only a percentage of the taxa observed, because they contain valuable information about species useful for palaeo-reconstructions and for single-species studies.
Radiolarian census data from 87 sites located in the southwest Pacific sector published by Cortese and Prebble (2015) contained full census counts and required no further treatment.
The Indian Ocean sector includes census data from 41 sites first published by Rogers and De Deckker (2007). The slides from these sites were recounted to ensure consistency with the taxonomy of the SO-RAD compilation presented here (described in further detail in Sect. 2.1.3). Also located in the Indian Ocean sector are census data from four sites from Abelmann et al. (1999) and six from Cortese and Abelmann (2002).
We also include radiolarian census data from 26 previously unpublished
sites. These include counts from 14 archive samples from sites in the Ross
Sea, retrieved during multiple expeditions by the National Institute of
Water and Atmospheric Research (NIWA) and its predecessor New Zealand
Oceanographic Institute; five sites near the Sabrina Coast, East Antarctica
(three retrieved during the 2017 RV
In general, radiolarian slides were prepared using a similar method as
described in Cortese and Prebble (2015). Sediment samples were dried,
weighed, placed in beakers with 100 mL of distilled water, and heated to
60–70
Note that the specific chemical separation steps may differ between authors,
as may the mesh size used. A mesh size of 40–45
Radiolarian census counts were determined using transmitted-light microscopy
at 100–400
Species identification is mainly based on the taxonomic concepts of
Petrushevskaya (1967, 1971), Nigrini and Moore (1979), and Boltovskoy (1998).
This was supplemented by consulting Lazarus et al. (2015), the online
dataset at
A total of 238 radiolarian taxa have been recognised. Radiolarian taxa identified in SO-RAD are from the super-class Polycystinea, which is divided into three orders, Nassellaria, Spumellaria, and Collodaria. When present, phaeodarian species were identified, counted, and recorded. The binomial names for each of the taxa included in the SO-RAD dataset are listed in Sect. S2.
To spatially categorise the dataset, each site was assigned to one of four zones: the Antarctic Zone (AZ; south of the Antarctic Polar Front), Polar Frontal Zone (PFZ; between the Polar Front and the Subantarctic Front), Subantarctic Zone (SAZ; between the Subantarctic Front and the Subtropical Front), or Subtropical Zone (STZ; north of the Subtropical Front), based on frontal boundaries delineated by Orsi et al. (1995). Sites were also assigned to one of three sectors, the Atlantic, Indian, or southwest Pacific, based on the longitudinal boundaries used by the World Ocean Circulation Experiment (WOCE Data Products Committee, 2002).
To illustrate the environmental coverage of the dataset, austral summer
(January–March) observations of temperature (
The SO-RAD dataset consists of 228 rows (plus header rows) and 251 columns (Table 1).
Each row contains data for one site from which a surface sediment sample was
obtained. Columns 1–13 contain information relating to the site or sample,
while columns 14–251 are the radiolarian taxa observed (one taxon per
column). The SO-RAD dataset is available to download from
All figures and summary statistics in this paper were generated using R
statistical software (R core team, 2020) and the packages
The SO-RAD dataset is the most comprehensive Southern Ocean radiolarian surface sediment dataset available. However, the spatial coverage is still uneven with, for example, more Antarctic Zone sites in the Atlantic sector than in the Indian and southwest Pacific sectors and more Subtropical Zone sites in the southwest Pacific sector (Table 2). The latter sector has better overall coverage than the Atlantic and Indian sectors; nevertheless there is a paucity of sites from the Pacific Ocean east of the Ross Sea. Sites represent locations covering a large water depth gradient, from the Antarctic continental shelf (235 m b.s.l.) to the abyssal plain (5883 m b.s.l.).
Description of variables in the SO-RAD dataset.
Number of sites located in each zone and sector of the Southern Ocean.
Sites included in the SO-RAD dataset are representative of a broad range of
environmental conditions. Temperature values range from
Locations of sites (black dots) and the six environmental variables at 100 m water depth used to demonstrate the environmental coverage of the SO-RAD dataset (data source: Locarnini et al., 2018; Garcia et al., 2019a, b; Zweng et al., 2019).
Distribution of sites across observed environmental variables at 100 m water depth (Locarnini et al., 2018; Garcia et al., 2019a, b; Zweng et al., 2019) identified by sector – Atlantic Ocean sector (ATL, green), Indian Ocean sector (IND, orange), and southwest Pacific sector (SWP, purple).
The SO-RAD dataset documents 238 identified taxa characterised by subspecies/species level identifications through to individuals identified only as belonging to the Nassellaria or Spumellaria orders.
Radiolarian species (i.e. not including taxa of genus level and above or
taxa that had not been formally described) with the highest average relative
abundances were
Other well-distributed and abundant species were
Distribution maps for the nine most abundant and widely distributed radiolarian taxa. Ocean front boundaries from Orsi et al. (1995).
Thirty-eight taxa were found at more than 100 sites.
There were 71 taxa (30 % of the total number of taxa) in the SO-RAD
dataset considered to be “rare”; i.e. they had a maximum relative abundance
While the SO-RAD dataset is predominantly a Southern Ocean dataset, there
are several sites located considerably north of the Subtropical Front (the
northern boundary of the Southern Ocean). One of the anticipated uses of the
SO-RAD dataset is as a reference dataset for radiolarian-based
palaeoenvironmental reconstructions from sediment cores retrieved from the
Southern Ocean and subtropical Southern Hemisphere. One of the key
requirements of a reference dataset is that it should cover a wide range of
environmental conditions. It has been demonstrated elsewhere that
radiolarian assemblages from temperate latitudes, such as offshore New
Zealand, show large and distinct changes over glacial and interglacial
periods (e.g., Lüer et al., 2009). The broad coverage of the SO-RAD
modern reference dataset facilitates identification of future faunal shifts and
imparts robustness to the environmental variable estimates derived from
them. In a broader sense, the large spatial and environmental coverage of
this dataset can aid in refining palaeoenvironmental reconstructions
using radiolarian assemblages from periods that were substantially cooler or
warmer than the present. The SO-RAD database is a valuable resource for
exploring how physical, biological, and chemical characteristics of the
Southern Ocean influence modern radiolarian biogeography, ecology, and
species diversity. This ecological information and species diversity may
provide key insights into the past palaeoceanography. The SO-RAD dataset also
holds potential for the exploration of sea-ice coverage and concentration as a
causal mechanism underlying the abundance patterns for some of the species
occurring in the seasonally ice-covered zone. It may be possible to
establish surface-dwelling radiolarian species as sea-ice proxies, as
routinely done for diatom species, such as
When constructing any reference dataset based on surface sediment samples, there are possible sources of error and assumptions that apply. Reproducibility of species abundance data can be affected by differing taxonomic concepts held by the analysts. We have minimised this potential source of error by including counts by radiolarists who regularly collaborate on taxonomic concepts. Furthermore, the vast majority of the samples included here have been counted by only one of us (Giuseppe Cortese). As more samples are added to the SO-RAD dataset, only samples for which quality control of the census data is possible will be included in the database in an effort to minimise this source of error.
The species-level taxonomy of all microfossil groups has been based on phenotype/morphology, and not on molecular biology. Identification can be difficult when there are only minor differences in morphology between species, or where there are fragments of an individual, or additionally, where a juvenile specimen has been encountered. In such instances individual radiolarians have been assigned to higher-rank taxonomic categories, i.e. genus or family rather than species or subspecies.
When using surface sediment samples as a modern analogue record, it is assumed that the fossil assemblages found within those sediments are representative of modern assemblages that live in the water column above. This is problematic for two reasons. The first is that the coring device can affect the quality/age of the surface sediment sample. For example, using a multi-corer allows for easier observation, and intact retrieval of, the surface sediment–bottom water interface (Barnett et al., 1986). With other coring devices it is not possible to observe if the uppermost sediment layer is the true surface sediment sample, or if the top of the coring device was buried beneath the surface. This means the “core top” sample may be from several centimetres below the surface and is not the most recently deposited sediment. Some of these sediment samples have been radiocarbon dated using planktic foraminifera in the same samples (Cortese and Prebble, 2015), but this is not possible with all samples, particularly those that have no carbonate present. Secondly, even when the sediment sample is truly from the surface, the transportation or reworking of sediment, or preferential dissolution of species, means that the assemblages observed in the sediment may not be truly representative of the assemblages that lived in the water column above.
Where possible, we have presented census counts in their complete form to cater for a variety of analyses as well as the development of novel techniques for radiolarian data analysis. We have opted not to present a “reconstruction ready” dataset as we believe the full dataset is useful for more than just palaeoenvironmental reconstructions. However, for most analyses the dataset would not be used in its raw form. Here we offer practical suggestions for preparing and producing a subset from this dataset based on practices utilised by radiolarists in the past.
Users of the dataset may wish to extract a subset of sites from the SO-RAD dataset for quality control reasons. For example, placing a lower bound on the total number of individuals counted per site may depend on which species are of interest to the user. In general, lower total counts are sufficient when only the dominant species will be used in the analysis, while higher total counts are needed when a user is interested in species comprising as little as 1 % of the population. For example, there is a 30 % probability of failing to observe a species that comprises 1 % of the population when counting 100 individuals but, by increasing the count to 300 individuals, this probability is reduced to less than 1 % (Fatela and Taborda, 2002).
Sites in the Atlantic sector, as well as some in the Indian sector, did not have full census counts available, and therefore a percentage of the assemblage is recorded as “Other radiolarians”. Rules for the exclusion of species at these sites can be found in Abelmann et al. (1999).
The coring apparatus used to retrieve a core can affect the quality of the sample collected, as well as its age. In most studies, the latter is usually assumed to be modern, but this can only be confirmed by radiocarbon dating of each surface sediment sample. Of the sites included in the SO-RAD dataset, the only radiocarbon-dated samples are those reported in Cortese and Prebble (2015). Where known, the apparatus used to retrieve the core has been listed in the “Sample method” column to facilitate the extraction of a data subset based on this variable.
Palaeoenvironmental reconstructions require a reference dataset that provides appropriate spatial and environmental coverage. Reference sites may be included based on regional needs, e.g., by sub-setting the data based on latitudinal and/or longitudinal boundaries, based on sector or oceanographic zone, or based on certain values of the environmental variable of interest.
Here we outline some practices to consider when refining the SO-RAD dataset for use as a reference dataset in palaeoceanographic reconstructions. These include selecting only taxa that can be clearly identified and with clear taxonomic descriptions, selecting appropriate taxa for the depth level of the reconstruction, grouping species/subspecies with similar environmental affinities into higher-order groupings, and removing rare species.
If the focus of a study is on environmental reconstructions, the signal-to-noise ratio in radiolarian census data can be increased by removing broad, often higher-order, counting categories having no clear taxonomic boundaries/definitions, and hence questionable environmental significance. The inclusion of unidentifiable individuals in higher-order taxonomic groupings in the SO-RAD dataset ensures the correct calculation of species relative abundances; however these taxa are not useful for analyses that rely on genus- or species-level data and might be removed.
Broad, higher-order categories that can be taken into consideration for
removal from the SO-RAD dataset prior to analysis include the indeterminate
Nassellaria and Spumellaria taxonomic groups, along with taxa including, but
not necessarily limited to, Litheliidae/Pyloniidae,
Some Phaeodarea taxa have been included in the census counts. Studies
focusing only on polycystine radiolarians should remove the phaeodarians
Radiolarian reference datasets often contain taxonomic groupings made up of species/subspecies grouped at the genus level, or of unrelated species with the same environmental affinities that may be routinely confused for one another during the counting procedure (e.g., Cortese and Prebble, 2015; Matsuzaki and Itaki, 2017). Grouping species is especially relevant for less abundant taxa, and for those with a particularly complex taxonomy. Grouping species and/or subspecies that have similar distributions according to water column and modern sediment sampling can increase the statistical weight (and thus their chances of inclusion in a working reference dataset) of rare species that are known to carry a strong environmental signal. For example, Collosphaeridae is a taxonomic group that can be composed of several to tens of species that are most abundant at low latitudes (Cortese and Prebble, 2015; Biard et al., 2016). At high latitudes, their presence alone, even in very small numbers, may be indicative of warmer-than-usual conditions. It is, therefore, useful to group these rare species so that their signal is not lost when removing rare species prior to analysis. Grouping species, while obviously leading to the loss of species-level information, also increases inter-comparability of results between different researchers/teams, as higher-level categories are generally more accommodating and less prone to identification mistakes based on fine skeletal differences. Recommended groupings are listed in Sect. S3 and are adapted from Cortese and Prebble (2015).
Previous radiolarian-based palaeoenvironmental reconstructions and
biogeographical studies have removed rare taxa using one, or both, of the
following methods: (1) removing species observed at fewer than a certain
number of sites (e.g., Abelmann et al., 1999; Boltovskoy and Correa, 2016)
and/or (2) removing species based on their low abundances (e.g.
As radiolarians are present throughout the water column, a surface sediment sample contains a collection of taxa that, while living, do not inhabit the same depth but eventually end up in the same surface sediment sample. It is possible to limit the reference dataset to taxa that are known to live at the depth for which a specific environmental variable is being reconstructed. Examples of studies where the authors have used a subset of taxa based on their vertical distributions include Abelmann et al. (1999), Cortese and Abelmann (2002), and Matsuzaki and Itaki (2017).
There are noticeable gaps in the spatial coverage of the SO-RAD dataset, in particular the area between the Ross Sea and the Antarctic Peninsula, comprising a vast part of the high-latitude south Pacific Ocean, and between Australia and the Antarctic continent in the Indian Ocean. The intention of the authors is that the SO-RAD dataset will constantly evolve as more surface sediment samples, collected during upcoming oceanographic expeditions, become available for census counts.
The SO-RAD dataset is available to download from
The SO-RAD dataset contains census data for 238 radiolarian taxa from 228 Southern Ocean sites. The dataset is the most comprehensive Southern Ocean radiolarian census dataset to date. We have provided a detailed description of the methods used to build the dataset, along with an overview of the spatial and taxonomic coverage of the data. The SO-RAD dataset may be used as a reference dataset for palaeoenvironmental reconstructions, or in studies of modern radiolarian biogeography, ecology, and species diversity.
KAL was responsible for the data curation (along with GC), formal analysis, investigation (along with GC and MCM), data visualisation, and preparation of the original draft. GC conceptualised and acquired funding for this study. Resources (access to samples) were provided by GC, HB, XC, AL, JR, and LKA. GC and LKA supervised this project. All authors contributed to the review and editing of the manuscript.
The contact author has declared that neither they nor their co-authors have any competing interests.
Publisher’s note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
A series of national and international core repositories provided the surface sample material used in this study. These include the core repository at the National Institute of Water and Atmospheric Research (NIWA, New Zealand); three of IODP’s (International Ocean Discovery Program) core repository centres: Kochi Core Center (jointly managed by Kochi University and JAMSTEC, the Japan Agency for Marine-Earth Science and Technology), the IODPWest Coast repository at Scripps Institution of Oceanography, and the IODP Gulf Coast Repository at Texas A&M University; the Oregon State University Marine Geology Repository located at the College of Earth, Ocean, and Atmospheric Sciences; the Woods Hole Oceanographic Institution core repository; the French EPOC and LSCE core repositories; the Alfred Wegener Institute for Polar and Marine Research (AWI) core facility in Bremerhaven; and the IfM/GEOMAR core repository in Kiel. The help of these repositories, and the funding agencies providing support to them, is hereby greatly appreciated. The authors wish to thank the crews and scientists that collected the sediment core material used in this project, as well as Sonja Penafiel-Bermudez (GNS, Lower Hutt), Ute Bock (AWI, Bremerhaven), and Natalie Kozlowski (Colgate University) for preparing the microscopic slides.
Samples not previously published were collected through the listed missions and supported by the following research support bodies.
For Sabrina Coast material, we thank the Marine National Facility (MNF) IN2017-V01 scientific party led by the chief scientists Leanne K. Armand and Philip E. O'Brien, CSIRO MNF support staff, and ASP crew members led by the captain Michael Watson on board the RV
Ross Sea material was supplied from the NIWA surface sediment and core repository and has been collected on numerous research voyages by NIWA and its predecessor NZOI since the 1950s. The samples have been collected by a range of different researchers on research voyages on a number of different vessels, funded by the New Zealand government through various funding programmes over the last 70 years.
We thank Alain Mazaud and Elisabeth Michel for providing access to cores MD11-3353 and MD12-3396CQ. We are also grateful to the crews and scientists on board R/V
We thank Demetrio Boltovskoy and the one anonymous reviewer for taking the time to provide comments that enabled us to improve and refine the paper.
This research was supported by the New Zealand Ministry of Business, Innovation and Employment through the Antarctic Science Platform (ANTA1801) and the Global Change through Time Programme (GCT SSIF, contract C05X1702) as well as through an IdEx (Initiative d’Excellence de l’Université de Bordeaux) visiting professorship grant (GC).
Kelly-Anne Lawler and Vikki Lowe are each supported by an Australian Research Training Program (RTP) scholarship.
Voyage IN2017-V01 was supported by the Australian government’s Australian Antarctic Science Grant Program (AAS no. 4333) and the Australian Research Council (grant no: DP170100557).
Voyage NBP1402 was supported by NSF Antarctic Integrated Systems Science Program (grant nos. 1143834, 1143836, 1143837, 1143843 and 1313826).
This paper was edited by Thomas Blunier and reviewed by Demetrio Boltovskoy and one anonymous referee.