Download Census of seafloor sediments in world`s ocean basins

Submitted Manuscript: Confidential
22 March 2015
Census of seafloor sediments in world’s ocean basins
Adriana Dutkiewicz1*, R. Dietmar Müller1, Simon O‟Callaghan2, Hjörtur Jónasson1
1
EarthByte Group, School of Geosciences, University of Sydney, Sydney NSW 2006, Australia.
2
NICTA, Australian Technology Park, Eveleigh NSW 2015, Australia.
*Correspondence to: [email protected].
Knowing the patterns of distribution of sediments in the global ocean is critical for understanding
biogeochemical cycles and how deep sea deposits respond to environmental change at the sea
surface. We present the first digital map of seafloor lithologies based on original descriptions of
nearly 14,500 samples and show that their distribution is complex with major differences in
global abundances and regional deviations from earlier maps. By coupling our map to
oceanographic datasets we show that the occurrence of pelagic oozes is strongly linked to sea
surface parameters. Diatom oozes are not a reliable proxy for surface productivity and their
accumulation is instead strongly dependent on surface temperature, salinity, dissolved oxygen
and nutrients and limited completion from biogenous and detrital components. Understanding
the link between the seafloor and the sea surface will lead to more reliable reconstructions and
predictions of climate change and its impact on the ocean environment.
The first digital map of seafloor sediments provides a new resource for the exploration of global
relationships between oceanographic parameters and the geological record.
Modern oceanic sediments cover 70% of our planet‟s surface, forming the substrate for
the largest ecosystem on Earth and its largest carbon reservoir. The composition and distribution
of sediments in the world‟s oceans underpins our understanding of global biogeochemical cycles,
the occurrence of metal deposits, sediment transport mechanisms, the behavior of deep ocean
currents, reconstruction of past environments and the response of the deep ocean to global
warming. Although there are many versions of this map (1-5) they all show strikingly similar
distributions of clays, calcareous and siliceous oozes with large areas of the ocean basins draped
in either pelagic red clay or lithogenous sediments (Fig. S1). Despite the vast acquisition of new
data, this hand-drawn map has changed very little since its first inception (6). Here we present
the first digital map of recent sediments of the oceans based on carefully-selected descriptions of
surface sediment samples contained in cruise reports from the 1950s until the present-day. The
coupling of the sediment map to key oceanographic parameters provides new insights into the
processes governing the distribution of sediments in the world‟s oceans and highlights several
key discrepancies in the earlier maps.
Our map was created mostly using surface sample locations and descriptions obtained
through the Index to Marine and Lacustrine Geological Samples (IMLGS) (7, 8). The IMLGS
contains data on over 200,000 marine sediment samples, the vast bulk of which post-dates
creation of the commonly used Deck41 dataset (9) and 1983, the date of the last incarnation of
the global map of oceanic sediments (5). We selected 14,399 data points using strict qualitycontrol criteria [Fig. 1] (8).
There are many schemes by which marine sediments may be classified (10) resulting in at
least 80 different sediment types. The classification scheme that we have used here is
deliberately generalized in order to successfully depict the main types of sediments found in
global oceans and to overcome the shortcomings of inconsistent, poorly-defined and obsolete
classification schemes and terminologies that feature in the vast number of cruise reports. Our
goal was to adhere to the classification scheme currently used by the International Ocean
Discovery Program (11), focusing on the descriptive aspect of the sediment rather than its
genetic implications. We have used the following classes [Fig. 1] (8): gravel, sand, silt, clay,
calcareous ooze, radiolarian ooze, diatom ooze, sponge spicules, mixed calcareous-siliceous
ooze, shells and coral fragments, fine-grained calcareous sediment (not ooze), siliceous mud, and
volcaniclastics.
The map was created using a support vector machine (SVM) (12) classifier to predict the
lithology in unobserved regions (8). The SVM is a non-parametric model, which adapts in
complexity as new data are added. To reduce the risk of over-fitting to the measurements at the
expense of the model's ability to generalize into areas outside of the sampled regions, a crossvalidation approach was employed to train the classifier. This approach maximizes the model‟s
accuracy on observations that are withheld from the training set. For prediction, a one-againstone method (13) was used to address the problem of modeling multiple classes with a bilinear
classifier. Classes were weighted inversely proportional to their number of recorded instances to
account for the unbalanced nature of the data. Deep sea lithologies that collectively comprise
over 70% of seafloor sediment have been predicted with very high accuracy (up to 80%) [Figs
S2, S3].
Our digital map (Fig. 2) reveals that the pattern of distribution of different lithologies is
more complex with significant regional deviations from earlier maps (1-5) [Fig. S1]. As with
older maps, clay dominates the Pacific Ocean, calcareous oozes the Atlantic and Indian oceans
and diatom ooze forms a belt in the Southern Ocean. However, these occur in drastically
different proportions globally (Table S1) with coverage by calcareous sediment and clay each
having increased by about 30% while that of diatom and radiolarian oozes having decreased by
about 25% and 60%, respectively. Rather than forming a belt in the equatorial Pacific extending
to 30°S along the west coast of South America (Fig. 1) radiolarian oozes occur as isolated
pockets around the equatorial Pacific in association with patches of mixed oozes and as a
component of diatom ooze within the Peru Basin (Fig. 3). This is also the case in the Atlantic and
Indian oceans where radiolarian oozes are mixed with calcareous and diatom oozes. Patches of
radiolarian ooze are, however, common in the Southern Ocean, which is not apparent on preexisting maps. There are numerous large areas of diatom ooze within predominant clay lithology
in the northern Pacific and central Indian oceans. The circum-Antarctic belt of diatom ooze is
discontinuous on our map with a major interruption in the Drake Passage where sedimentation is
dominated by a large body of sand (Fig. 3). Sponge spicules form a significant component of
seafloor sediment in parts of the Australian-Antarctic Basin where they co-occur with diatom
and radiolarian oozes. Compared to earlier maps clay occupies a considerably larger area around
eastern and western South America and is significantly more abundant in the Indian Ocean
where its southerly extent from the Bengal Fan is interrupted only by the 90°E and Broken
ridges. Clay is dominant within the South Australia Basin (Fig. 3) but does not spread into the
Southern Ocean as shown on older maps (Fig. S1).
Marine planktonic organisms play a critical role in the global cycling of silica and carbon
and in the biological pump of CO2 (14). However, many of the mechanisms that are thought to
control the geologic accumulation of biogenic carbonate and silica as are very difficult to
quantify, even on a local scale (e.g., 14, 15). Our digital map of seafloor sediments provides a
missing link for constraining global relationships between the seafloor and the sea surface for
which comprehensive datasets exist (8). For example, the bulk of diatom oozes occurs at seafloor
depths of 3300-4800 m, below surface water that has very restricted and low temperature (0.95.7 °C), consistent with but slightly narrower than the 0.8-8 °C range required for optimal diatom
growth in the Southern Ocean (16), low and narrow salinity range (33.8-34.0 PSS) a relatively
high and narrow range of dissolved oxygen (7.1-7.8 ml/L). The summer productivity is modest
(230-840 mgC/m2/day in the northern hemisphere and 175-260 mgC/m2/day in the southern
hemisphere) based on satellite-derived surface chlorophyll-a concentrations, and in the Southern
Ocean may be limited by iron and light (8, 17). Diatom oozes are associated with the highest and
narrowest ranges of surface nutrients, especially growth-limiting silicate (18), of all lithology
classes including siliceous radiolarian oozes (Figs S4, S5; Table S2).
Recent computations of global distributions of phytoplankton species from chlorophyll-a
satellite data calibrated with situ measurements of diagnostic pigments (19) show that diatoms
are a major contributor to primary productivity in the Southern Ocean, peaking in numbers in the
austral summer (20, 21). However, even with these vastly improved maps of diatom abundances
that capture seasonal blooms, we fail to find a strong link between diatom chlorophyll
concentration and diatom ooze occurrence (Figs 3, S6 and S7). Diatom ooze association with
high diatom chlorophyll concentration in the north Weddell Sea and around Prydz Bay is an
exception, not the rule. Diatom ooze is most common below waters with very low diatom
chlorophyll concentration forming prominent zones between 50 to 60°S in the AustralianAntarctic Basin and the Bellinghausen Basin (Fig. 3). These large-scale patterns cannot be easily
explained by post-depositional sediment redistribution (22) and contradict the widely held view,
based on hydrographic and sediment trap data, that biogenic opal accumulation in the Southern
Ocean is linked to high surface productivity (e.g., 23, 24). As biogenic silica preservation
efficiency is no longer considered critical for diatom accumulation in the Southern Ocean (2325), we propose that the accumulation of diatom oozes is strongly dependent on limited
completion from biogenous and lithogenous components. Diatoms are conspicuously absent
below the prominent 40°S diatom chlorophyll concentration belt (Fig. 3). The belt coincides with
a marked northward increase in salinity, temperature and dissolved oxygen and a decrease in
dissolved silicate (Fig S7) conducive for the proliferation of biocalcareous organisms and the
subsequent accumulation of calcareous oozes on the seafloor above the carbonate compensation
depth (15) (Figs 3, S4, S5). Likewise, diatoms oozes are absent below high diatom chlorophyll
areas near continents where their presence in the sediment is diluted by terrigenous input.
Therefore, we conclude that diatom ooze is not a reliable proxy for surface productivity,
but it is a good indicator of sea surface oceanographic variables (26), including temperature (27,
28). It will accumulate under restricted conditions even when surface productivity is low
provided that there are no dominant competing components, such as nanoplankton (19, 20) and
diatom-grazing radiolarians (4), and that diatom frustules are not significantly dissolved in the
water column or on the seafloor (14, 25, 29). Our seafloor lithology map is a new digital, openaccess resource that provides a basis for elucidating global relationships between the sedimentary
record and a variety of oceanographic parameters, providing additional constraints for models of
paleoproductivity and global biogeochemical cycles.
References and notes
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
E. J. Barron, J. M. Whitman, in The Sea, C. Emiliani, Ed. (Wiley Interscience, New
York, 1981), vol. 7, pp. 689-733.
W. H. Berger, Biogenous deep sea sediments: production, preservation and interpretation.
Chemical Oceanography 5, 265-372 (1976).
T. A. Davies, D. S. Gorsline, in Chemical Oceanography, J. P. Riley, R. Chester, Eds.
(Academic Press, London, 1976), vol. 5, pp. 1-80.
H. Hüneke, T. Mulder, Deep-Sea Sediments. (Elsevier, 2011), vol. 63.
A. P. Trujillo, H. V. Thurman, Essentials of Oceanography. (Prentice Hall, ed. 11,
2014).
W. H. Berger, in The Geology of Continental Margins, C. A. Burk, C. D. Drake, Eds.
(Springer-Verlag, New York, 1974), pp. 213-241.
Curators of Marine and Lacustrine Geological Samples Consortium: Index to Marine and
Lacustrine Geological Samples (IMLGS). National Geophysical Data Center, NOAA.
doi:10.7289/V5H41PB8 [accessed August-November 2014].
Materials and methods are available as supplementary materials on Science Online.
S. Bershad, M. Weiss, "Deck41 surficial seafloor sediment description database,"
(National Geophysics Data Center NOAA, 1974). doi: doi:10.7289/V5VD6WCZ
[accessed July 2014]
J. Kennett, Marine Geology. (Prentice Hall, Englewood Cliffs, New Jersey, 1982).
J. Mazzullo, A. Meyer, R. Kidd, in The ODP Shipboard Scientist's Handbook, P. D.
Rabinowitz, A. W. Meyer, L. E. Garrison, Eds. (Texas A & M University, 1990), pp.
125-130.
C. Cortes, V. Vapnik, Support-vector networks. Machine Learning 20, 273-297 (1995).
C. M. Bishop, Pattern Recognition and Machine Learning. (Springer New York, 2006),
vol. 4.
O. Ragueneau, P. Tréguer, A. Leynaert, R. F. Anderson, M. A. Brzezinski, D. J.
DeMaster, R. C. Dugdale, J. Dymond, G. Fischer, R. Francois, A review of the Si cycle
in the modern ocean: recent progress and missing gaps in the application of biogenic opal
as a paleoproductivity proxy. Global and Planetary Change 26, 317-365 (2000).
W. S. Broecker, A need to improve reconstructions of the fluctuations in the calcite
compensation depth over the course of the Cenozoic. Paleoceanography 23, (2008).
A. Neori, O. Holm-Hansen, Effect of temperature on rate of photosynthesis in Antarctic
phytoplankton. Polar Biology 1, 33-38 (1982).
P. Claquin, V. Martin‐ Jézéquel, J. C. Kromkamp, M. J. W. Veldhuis, G. W. Kraay,
Uncoupling of silicon compared with carbon and nitrogen metabolisms and the role of
the cell cycle in continuous cultures of Thalassiosira pseudonana (Bacillariophyceae)
under light, nitrogen, and phosphorus control. Journal of Phycology 38, 922-930 (2002).
V. Martin-Jézéquel, M. Hildebrand, M. A. Brzezinski, Silicon metabolism in diatoms:
implications for growth. Journal of Phycology 36, 821-840 (2000).
T. Hirata, N. J. Hardman-Mountford, R. J. W. Brewin, J. Aiken, R. Barlow, K. Suzuki, T.
Isada, E. Howell, T. Hashioka, M. Noguchi-Aita, Synoptic relationships between surface
Chlorophyll-a and diagnostic pigments specific to phytoplankton functional types.
Biogeosciences 8, 311-327 (2011).
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
J. Uitz, H. Claustre, B. Gentili, D. Stramski, Phytoplankton class‐ specific primary
production in the world's oceans: seasonal and interannual variability from satellite
observations. Global Biogeochemical Cycles 24, (2010).
M. A. Soppa, T. Hirata, B. Silva, T. Dinter, I. Peeken, S. Wiegmann, A. Bracher, Global
retrieval of diatom abundance based on phytoplankton pigments and satellite data.
Remote Sensing 6, 10089-10106 (2014).
L. Dezileau, G. Bareille, J. L. Reyss, F. Lemoine, Evidence for strong sediment
redistribution by bottom currents along the southeast Indian ridge. Deep Sea Research
Part I: Oceanographic Research Papers 47, 1899-1936 (2000).
D. M. Nelson, R. F. Anderson, R. T. Barber, M. A. Brzezinski, K. O. Buesseler, Z.
Chase, R. W. Collier, M.-L. Dickson, R. François, M. R. Hiscock, Vertical budgets for
organic carbon and biogenic silica in the Pacific sector of the Southern Ocean, 1996–
1998. Deep Sea Research Part II: Topical Studies in Oceanography 49, 1645-1674
(2002).
P. Pondaven, O. Ragueneau, P. Tréguer, A. Hauvespre, L. Dezileau, J. L. Reyss,
Resolving the „opal paradox‟in the Southern Ocean. Nature 405, 168-172 (2000).
D. M. Nelson, P. Tréguer, M. A. Brzezinski, A. Leynaert, B. Quéguiner, Production and
dissolution of biogenic silica in the ocean: revised global estimates, comparison with
regional data and relationship to biogenic sedimentation. Global Biogeochemical Cycles
9, 359-372 (1995).
W. L. Cunningham, A. Leventer, Diatom assemblages in surface sediments of the Ross
Sea: relationship to present oceanographic conditions. Antarctic Science 10, 134-146
(1998).
O. E. Romero, L. K. Armand, X. Crosta, J. J. Pichon, The biogeography of major diatom
taxa in Southern Ocean surface sediments: 3. Tropical/Subtropical species.
Palaeogeography, Palaeoclimatology, Palaeoecology 223, 49-65 (2005).
U. Zielinski, R. Gersonde, Diatom distribution in Southern Ocean surface sediments
(Atlantic sector): implications for paleoenvironmental reconstructions. Palaeogeography,
Palaeoclimatology, Palaeoecology 129, 213-250 (1997).
P. J. Tréguer, C. L. De La Rocha, The world ocean silica cycle. Annual Review of Marine
Science 5, 477-501 (2013).
A. H. Orsi, T. Whitworth, W. D. Nowlin, On the meridional extent and fronts of the
Antarctic Circumpolar Current. Deep Sea Research Part I: Oceanographic Research
Papers 42, 641-673 (1995).
Acknowledgements
This work would not have been possible without the enormous effort of the huge number of
participants involved in oceanographic cruises whose sample descriptions have enabled us to
make this map. We are grateful to the curators of the IMLGS at NOAA for creating and
maintaining an open-access global database of marine geological samples. Many thanks to
Mariana Soppa for making the diatom chlorophyll dataset available to us and to Clark Alexander
and Charlotte Sjunneskog for access to sample descriptions not available via the IMLGS. We
extend our gratitude to Maria Seton and Paul Wessel for their reviews of an earlier version of the
manuscript. This research was supported by the Science Industry Endowment Fund (RP 04-174)
Big Data Knowledge Discovery project. The seafloor lithology grid can be downloaded via the
supplementary materials and at: ftp:// XXXXX.
Fig. 1. Seafloor sediment sample locations. Lithology-coded sample locations of surface
sediments (n =14399) used to create the digital map of seafloor sediments in world‟s ocean
basins (Fig. 2). The lithologies are based on descriptions of surface or near-surface samples that
were collected using coring, drilling or grabbing methods and whose descriptions were
unambiguous and could be verified using original cruise reports and core logs. Mollweide
projection.
Fig. 2. Digital map of major lithologies of seafloor sediments in world’s ocean basins.
Legend is the same as in Fig. 1. Percentage estimates of lithologies in major oceans are given in
Table S1. Mollweide projection. The gridded data are available for download from: XXXX.
Fig. 3. Biosiliceous oozes versus diatom chlorophyll concentration in the Southern Ocean.
Outlines of areas of biosiliceous sediments superimposed on austral summer average of diatom
chlorophyll concentrations (mg/m3) for the period 2003-2013 from the model of Soppa et al.
(21). Seafloor regions where we map diatom oozes, radiolarian oozes and sponge spicules are
outlined in white, light grey and dark grey, respectively. Polar Frontal Zone (PFZ) bounded by
the Polar Front to the south and by the Subantarctic Front to the north, both outlined in yellow,
after (30). Stereographic projection with 10° latitude gridlines. Color scale has been adjusted to
highlight subtle variations in diatom chlorophyll concentration; maximum values (dark red)
reach about 18 mg/m3.
Supplementary materials
www.sciencemag.org/content/XXXXX
Materials and methods
Figs S1-S7
Tables S1-S2
References (31-44)
Supplementary Materials for
Census of Seafloor Sediments in World‟s Ocean Basins
Adriana Dutkiewicz, R. Dietmar Müller, Simon O‟Callaghan, Hjörtur Jónasson
correspondence to: [email protected]
This PDF file includes:
Materials and Methods
Figs. S1 to S7
Tables S1 to S2
Materials and Methods
Sample selection criteria and lithology classes
We selected 14,399 data points from the total number available through the Index to Marine and Lacustrine
Geological Samples [IMLGS] (7) using strict quality-control criteria. For areas with poor sample coverage, our
dataset was supplemented by descriptions acquired from non-participating agencies such as Geoscience Australia
(31). Trawl samples were not used as they represent mixed material whose exact location is unknown. We did not
use dredge samples as these usually sample rock rather than overlying soft sediment and represent a sampling bias.
Only surface or near-surface (decimeters below surface) samples were used that were collected using coring, drilling
or grabbing methods. Although the IMLGS contains sediment classifications for many samples, these are generic
and ambiguous (e.g., “terrigenous mud or ooze”) preventing the sediment to be correctly classified. Consequently,
we only used samples whose descriptions could be verified using original cruise reports, cruise proceedings and core
logs. In addition, these descriptions had to be sufficiently detailed to allow classification into one of our designated
groupings. For example, if the sediment was simply described as mud or ooze in the original report without any
other information, it was not used. Although we have a relatively small number of points from the continental shelf,
we have mostly concentrated our efforts on the deep sea. Areas of the map with limited sample coverage (e.g., parts
of the Southern Ocean, the South Pacific, the Arctic Ocean and the East China Sea) indicate lack of sampling,
unavailability of cruise reports and hence lack of reliable sample descriptions. That the seafloor in the northern
hemisphere has been much more densely sampled than in the southern hemisphere is illustrated by the coordinates
of the sample locations in the IMLGS; 85% are from the northern hemisphere.
The lithology classes on our seafloor sediment map are defined below and include various
sediment types based on descriptions from cruise reports and logs:
Gravel: unconsolidated sediments coarser than sand; includes gravel, shelly gravel, sandy gravel, muddy gravel,
pebbles, cobbles, diamicton, erratics, rock fragments, loose conglomerate, ice-rafted exotics, till.
Sand: unconsolidated sediment dominated by sand-sized grains; includes quartz sand, detrital sand, shelly sand,
glauconitic sand, clastic sand, silty sand, muddy sand, bioclastic sand, clayey sand, clayey shelly sand.
Silt: unconsolidated sediment dominated by silt-sized grains; includes diatomaceous silt, clayey silt, sandy silt,
calcareous silt, muddy silt, lutaceous silt, shelly sandy silt.
Clay: unconsolidated sediment dominated by a clay-sized fraction and characterised by low carbonate and low
biogenic content; includes sediments variously described as lutite, red clay, brown clay, zeolitic clay, zeolitic red
clay, hemipelagic clay, pelagic clay, lutaceous clay, biosiliceous clay, radiolarian clay, diatomaceous clay,
foraminiferal clay, radiolarian clay, foram lutite, silty clay, palagonitic clay, ashy clay, manganese lutite.
Diatom ooze: unconsolidated sediment containing > 30% diatom skeletons, in which diatoms are the dominant
biogenic component; includes sediments variously described as diatom clay, clayey diatom ooze, diatomite, diatom
mud, muddy diatom ooze, diatom clay ooze, diatom silty clay, radiolarian diatom ooze, diatom marl.
Radiolarian ooze: unconsolidated sediment containing > 30% radiolarian skeletons, in which radiolarians are the
dominant biogenic component, includes radiolarite.
Sponge spicule ooze: unconsolidated sediments containing > 30% sponge spicules, includes spiculite.
Calcareous ooze: unconsolidated sediment containing > 30% calcareous skeletons (foraminifera, pterapods,
coccoliths); includes sediments described as chalk if biogenic content is > 30%, calcareous marly ooze,
foraminiferal marly ooze, foraminiferal chalk, chalk ooze, clayey nanno ooze, foraminiferal sand, foraminiferal marl
and calcilutite if foraminifera > 30%.
Mixed calcareous-siliceous ooze: unconsolidated sediment containing > 30% of approximately equal portions of
calcareous and siliceous microfossils. Proportions of these are often not specified in original reports or logs and
sediment is simply classified as calcareous-siliceous ooze. Biogenic material includes foraminifera, nannofossils,
discoasters, pteropods, diatoms, radiolarians, sponge spicules.
Shells and coral fragments: unconsolidated sediment dominated by biogenic fragments that are not oozes; includes
shell hash, coral fragments, bryozoan fragments, reef sand, skeletal debris, coral hash, calcareous sand with minor
terrigeous content, Halimeda fragments, coral hash, reef debris, limestone fragments, shells, mollusc fragments.
Transitional calcareous sediments: unconsolidated calcareous sediment containing < 30% identifiable biogenic
components; includes marl, foraminiferal marl, radiolarian marl, calcareous clay, calcareous mud, calcilutite, nanno
clay, foraminiferal clay/lutite, chalk if biogenic content is < 30%, mixtures of calcareous and siliceous clays,
mixtures of terrigenous and calcareous clay.
Transitional siliceous sediments: unconsolidated siliceous sediment containing no or low amounts of carbonate
including sediments described as mud, terrigenous mud, sandy mud, silty mud, sandy clay, sandy lutite, clayey mud,
diatom mud.
Volcaniclastics: any unconsolidated sediment in which volcaniclastic grains are dominant irrespective of grain-size;
includes ash, clayey ash, lapilli, glass, volcanic sand, volcanic mud, volcanic gravel, volcanic breccia, vitric silt, tuff,
palagonite, tephra, silty tephra, volcanogenic detritus.
Support Vector Machine Classifier
A Support Vector Machine (SVM) is a classifier that attempts to separate classes of data with a plane. The location
of the plane in the input space is selected based on a set of training examples,
, labeled
{
}. To query the label of a new point, the algorithm simply determines on which
with a binary variable,
[
] where
side of the plane the point sits,
represents the discrimant function associated with the
hyperplane
Here, ,
and denote the weight vector, the basis function and the bias term, respectively. In two dimensions,
this amounts to dividing the input space into two sections with a straight line. In higher dimensions, the division is
determined by a plane (or hyperplane in 4 or more dimensions). Typically, the original data are projected into a
higher dimensional feature space,
where
, using a kernel function
to produce more flexible
decision boundaries in the lower dimensional input space.
The weight vector
is optimised using a cost function:
∑
subject to the constraints:
(3)
and
Here, acts as a slack variable, which is required when the data is non-separable.
the model for misclassified points.
Finally, the weights
can be found from
controls the penalty incurred by
∑
where comes from reformulating the cost function through a Lagrange function and finding the Lagrange
multipliers via a dual optimization:
∑
∑
( )
subject to the constraint
for all
The computational cost of determining the dot
product
can be large for large . Fortunately, there exist kernel functions,
, that return the dot
product directly without the need to first calculate its factors.
The SVM in this paper was trained using five-fold cross-validation. The key parameters, and (a parameter of
the kernel function), were optimised to maximize the model‟s predictive power on observations that were held-out
during training. The final values employed for the results presented were
and
. A radial basis
function was used as the kernel to produce the results presented in this paper. In this case, refers to the function‟s
lengthscale parameter. A sensitivity analysis plot is included in Fig. S2 that illustrates the behavior of the classifier‟s
cross-validation accuracy as these key parameters were varied. The yellow circular marker indicates the final set of
parameters used by the model.
To handle the multi-label nature of the data, a one-versus-one approach was adopted. A separate SVM was trained
for each pair of labels resulting in
different classifers, where is the number of labels. During prediction,
each classifier was evaluated at the query location and the most frequently returned label was used as the model‟s
output.
The accuracy of the SVM classifier is illustrated in Figs S2 and S3. Deep-sea lithologies (diatom oozes, calcareous
oozes and clay) that collectively comprise over 70% of seafloor sediment (Table S1) have been predicted with very
high accuracy (up to 80%). Minor misclassifications of radiolarian ooze as clay, and that of transitional classes such
as mixed calcareous-siliceous ooze and fine-grained calcareous sediments as calcareous ooze are due to a
compositional overlap and physical proximity of these lithologies. This has not affected the veracity of the final
gridded map (Fig. 3). It is also common for a rare lithology (e.g., gravel and shells) to be misclassified as a common
lithology simply due to the large difference between their likelihoods of occurrence (Fig. S3).
Oceanographic Datasets
Datasets of sea-surface temperature, salinity, dissolved oxygen and dissolved inorganic nutrients (nitrate, phosphate,
silicate) (32-35) were accessed from the World Ocean Atlas 2009 at
http://www.nodc.noaa.gov/OC5/WOA09/woa09data.html between August 2014 and January 2015. The data used
represent objectively analyzed means at 1degree grid resolution.
The surface productivity dataset based on Sea-viewing Wide Field-of-view Sensor (SeaWiFS) r2010 reprocessing
results was obtained from
http://orca.science.oregonstate.edu/1080.by.2160.monthly.hdf.vgpm.s.chl.a.sst.php in January 2015. We calculated
summer month averages (periods of maximum phytoplankton productivity) for the Northern Hemisphere (June,
July, August) and Southern Hemisphere (December, January, February) for the period 1997-2009.
Bathymetry data were obtained in November 2014 from ETOPO1 Global Relief Model (36) at
http://www.ngdc.noaa.gov/mgg/global/global.html.
Although diatom blooms are enhanced by iron fertilization (37) certain open-water species have modified their
apparatus to require less iron (38) resulting in thicker and heavier frustules (39). However, we were unable to
explore the impact of iron concentrations on preserved lithologies as the data coverage is relatively poor and biased
towards the northern hemisphere (40). Large intra-basin variabilities that are known to exist (41) would invariably
lead to substantial gridding errors, particularly in the Southern Ocean high-nutrient low-chlorophyll (HNLC) region.
The Southern Ocean receives very little iron from eolian dust (42) and other sources (e.g., from continental margins)
have not been quantified.
Fig. S1. A typical example of a map showing the distribution of deep-sea sediments in the
world‟s oceans. Color scheme is the same as in Figs 1 and 2. Mollweide projection. Modified
after (5). This hand-drawn map has remained unchanged since 1983 and is based on the Deck41
data (9) supplemented by ODP data (Trujillo, A.P., pers. comm., October 2014).
Fig. S2. Confusion matrix for the digital sea-floor lithology map. The matrix illustrates the
support vector classifier's accuracy in predicting lithology in our dataset. Each row corresponds
to a distribution over the algorithm's prediction on a class label (lithology type) that was withheld
during the model's training phase. In the case where the classifier is 100% accurate at predicting
each lithology, the matrix would be a diagonal one. The more red the diagonal boxes of the
matrix, the higher the number of points for which the true lithology has been correctly predicted.
Off-diagonal values correspond to a number of points whose lithology has been incorrectly
predicted by the support vector classifier.
Fig. S3. Gridded seafloor lithology map with lithology-coded data points (n =14399)
superimposed. Circles with white outlines indicate a match between the gridded lithology and the
sample point. Minor mismatching is highlighted by colored sample points, which in most cases
are closely related to the gridded lithology (e.g., fine-grained calcareous sediment and calcareous
oozes).
Fig. S4. Box plots of lithology classes versus various oceanographic parameters. Lithology class
are linked to numbers as follows: 1 – gravel, 2 – sand, 3 – silt, 4 – clay, 5 – calcareous ooze, 6 –
radiolarian ooze, 7 – diatom ooze, 8 – sponge spicules, 9 – mixed ooze, 10 – shell and coral
fragments, 11 – ash and volcanic sand/gravel, 12 – siliceous mud, 13 – fine-grained calcareous
sediment. See section on oceanographic datasets for description and links to source material.
Fig. S5. Box plots of lithology classes versus various oceanographic parameters. Lithology class
are linked to numbers as in Fig. S4. See section on oceanographic datasets for description and
links to source material.
Fig. S6. Outlines of areas of biosiliceous sediments and lithology-coded surface sediment sample
locations superimposed on austral summer average of diatom chlorophyll concentrations
(mg/m3) for the period 2003-2013 from the model of Soppa et al. (43). Areas of diatom oozes,
radiolarian oozes and sponge spicules are outlined in white, light grey and dark grey,
respectively.
Fig. S7. Biosiliceous oozes superimposed on maps of A) annual sea-surface temperature, B)
annual sea-surface salinity, C) annual dissolved oxygen, D) annual sea-surface dissolved nitrate,
E) annual sea-surface dissolved silicate and F) annual sea-surface dissolved phosphate (32-35).
Stereographic projection, latitude in 10° intervals. Datasets downloaded from
http://www.nodc.noaa.gov/OC5/WOA09/woa09data.html.
Table S1. Distribution of main sediment types in major oceans (Arctic excluded due to poor coverage). Ocean area
calculations, average salinities, temperatures and carbonate ion concentrations for surface waters from Feely et al.
(44).
Ocean
Salinity
(g/kg)
Temp.
(°C)
Carbonate
Ion
(µmol kg1)
Clay
(%)
Calc.
ooze
(%)
Finegrained
calc.
sediment
(%)
27
19.56
222.7
27
36
North Atlantic 35.643
90°W to 20°E
± 1.37
± 7.3
± 34
65°N to 20°N
and
80°W to 10°E
20°N to 0°
36.022
22.54
236.6
34
39
21
South Atlantic
70°W to 20°E
± 0.79
± 3.7
± 22
0° to 40°S
34.051
20.81
208.1
62
16
5
North Pacific
120°E to
± 0.86
± 8.1
± 38
100°W
59.5°N to 20°N
and
120°E to 80°W
20°N to 0°N
35.287
23.29
223.4
43
29
15
South Pacific
120°E to 70°W
± 0.55
± 4.3
± 25
0° to 40°S
34.722
28.12
244.4
40
33
13
North Indian
20°E to 120°E
± 1.57
± 0.8
± 12
0° to 24.5°N
35.103
23.46
227.0
33
38
14
South Indian
20°E to 120°E
± 0.47
± 4.3
± 19
0° to 40°S
34.775
16.74
193.1
29
21
6
Southern
± 0.94
± 10.5
± 54
Ocean
South of 40°S
*Siliciclastic sediment (gravel, sand, silt, mud), shells, volcaniclastic sediment.
Rad.
ooze
(%)
Diatom
Ooze
(%)
Mixed
ooze
(%)
Other
(%)*
<1
<1
<1
8
<1
<1
1
4
3
4
1
9
1
1
3
9
1
1
5
8
2
3
3
7
3
19
2
20
Table S2. The range of variation (1QR) of key oceanographic parameters for major deep-sea lithologies. Based on box plots in Figs S4 and S5.
Lithology
Clay
Calcareous
Radiolarian
Diatom ooze
Mixed ooze
ooze
ooze
3431-5198
2818-4286
4144-5126
3300-4791
3359-4504
Bathymetry (m)
7.2-25.9
13.2-26.9
1.7-27.2
0.9-5.7
10.5-27.2
Annual temperature at the
surface (°C)
33.786-35.212
34.349-35.679
33.84933.827-34.033
34.215-35.204
Annual salinity at the surface
34.306
(PSS)
4.736-6.872
4.653-6.098
4.654-7.660
7.057-7.780
4.635-6.480
Annual dissolved oxygen at
the surface (ml/L)
215-634
211-611
209-360
232-840
407-773
Surface productivity
northern hemisphere summer
(mgC/m2/day)
163-331
207-379
166-229
175-258
221-429
Surface productivity
southern hemisphere summer
(mgC/m2/day)
0.226-6.941
0.226-6.872
0.552-25.587
18.720-25.309
0.342-9.800
Annual nitrate at the surface
(mol/L)
0.159-0.772
0.154-0.655
0.272-1.756
1.424-1.760
0.234-0.957
Annual phosphate at the
surface (mol/L)
1.812-6.624
1.590-4.185
2.750-21.540
9.062-34.554
2.982-5.589
Annual silicate at the surface
(mol/L)
References
31.
32.
33.
34.
35.
36.
37.
38.
39.
40.
41.
42.
43.
44.
V. Passlow, J. Rogis, A. Hancock, M. Hemer, K. Glenn, A. Habib, "Final Report:
National Marine Sediments Database and Seafloor Characteristics Project. Record
2005/008," (Geoscience Australia, Canberra, 2005). Marine Science Database at
http://www.ga.gov.au/oracle/mars/ [accessed September-November 2014].
J. I. Antonov, D. Seidov, T. P. Boyer, R. A. Locarnini, A. V. Mishonov, H. E. Garcia, O.
K. Baranova, M. M. Zweng, D. R. Johnson, "World Ocean Atlas 2009 Volume 2:
Salinity," NOAA Atlas NESDIS 69 (Washington, D.C., 2010).
H. E. Garcia, R. A. Locarnini, T. P. Boyer, J. I. Antonov, O. K. Baranova, M. M. Zweng,
D. R. Johnson, "World Ocean Atlas 2009 Volume 3: Dissolved Oxygen, Apparent
Oxygen Utilization, and Oxygen Saturation.," NOAA Atlas NESDIS 70 (Washington,
D.C., 2010).
R. A. Locarnini, A. V. Mishonov, J. I. Antonov, T. P. Boyer, H. E. Garcia, O. K.
Baranova, M. M. Zweng, D. R. Johnson, "World Ocean Atlas 2009, Volume 1:
Temperature," NOAA Atlas NESDIS 68 (Washington, D.C., 2010).
H. E. Garcia, R. A. Locarnini, T. P. Boyer, J. I. Antonov, M. M. Zweng, O. K. Baranova,
D. R. Johnson, "World Ocean Atlas 2009, Volume 4: Nutrients (phosphate, nitrate, and
silicate)," NOAA Atlas NESDIS 71 (Washington, D.C., 2010).
C. Amante, B. W. Eakins, "ETOPO1 1 Arc-Minute Global Relief Model: Procedures,
Data Sources and Analysis," NOAA Technical Memorandum NESDIS NGDC-24
(National Geophysical Data Center, NOAA., 2009).
K. L. Denman, Climate change, ocean processes and ocean iron fertilization. Marine
Ecology Progress Series 364, 219-225 (2008).
R. F. Strzepek, P. J. Harrison, Photosynthetic architecture differs in coastal and oceanic
diatoms. Nature 431, 689-692 (2004).
D. A. Hutchins, K. W. Bruland, Iron-limited diatom growth and Si:N uptake ratios in a
coastal upwelling regime. Nature 393, 561-564 (1998).
J. K. Moore, O. Braucher, Sedimentary and mineral dust sources of dissolved iron to the
world ocean. Biogeosciences 5, (2008).
P. W. Boyd, M. J. Ellwood, The biogeochemical cycle of iron in the ocean. Nature
Geoscience 3, 675-682 (2010).
T. D. Jickells, Z. S. An, K. K. Andersen, A. R. Baker, G. Bergametti, N. Brooks, J. J.
Cao, P. W. Boyd, R. A. Duce, K. A. Hunter, Global iron connections between desert dust,
ocean biogeochemistry, and climate. Science 308, 67-71 (2005).
M. A. Soppa, T. Hirata, B. Silva, T. Dinter, I. Peeken, S. Wiegmann, A. Bracher, Global
retrieval of diatom abundance based on phytoplankton pigments and satellite data.
Remote Sensing 6, 10089-10106 (2014).
R. A. Feely, S. C. Doney, S. R. Cooley, Ocean acidification: present conditions and
future changes in a high-CO2 world. Oceanography 22, 37-47 (2009).