Downloaded from geology.gsapubs.org on August 20, 2015 Census of seafloor sediments in the world’s ocean Adriana Dutkiewicz1, R. Dietmar Müller1, Simon O’Callaghan2, and Hjörtur Jónasson1 1EarthByte Group, School of Geosciences, University of Sydney, Sydney, NSW 2006, Australia 2National ICT Australia (NICTA), Australian Technology Park, Eveleigh, NSW 2015, Australia ABSTRACT and to overcome the shortcomings of inconsis- Knowing the patterns of distribution of sediments in the global ocean is critical for under- tent, poorly defined, and obsolete classification standing biogeochemical cycles and how deep-sea deposits respond to environmental change schemes and terminologies that are detailed in at the sea surface. We present the first digital map of seafloor lithologies based on descrip- the majority of cruise reports. Our goal is to ad- tions of nearly 14,500 samples from original cruise reports, interpolated using a support vec- here to the classification scheme currently used tor machine algorithm. We show that sediment distribution is more complex, with significant by the International Ocean Discovery Program deviations from earlier hand-drawn maps, and that major lithologies occur in drastically (Mazzullo et al., 1990), focusing on the descrip- different proportions globally. By coupling our digital map to oceanographic data sets, we tive aspect of the sediment rather than its genetic find that the global occurrence of biogenic oozes is strongly linked to specific ranges in sea- implications. As a result, we identify the follow- surface parameters. In particular, by using recent computations of diatom distributions from ing major classes of marine sediment (Fig. 1): pigment-calibrated chlorophyll-a satellite data, we show that, contrary to a widely held view, gravel, sand, silt, clay, calcareous ooze, radiolar- diatom oozes are not a reliable proxy for surface productivity. Their global accumulation is ian ooze, diatom ooze, sponge spicules, mixed instead strongly dependent on low surface temperature (0.9–5.7 °C) and salinity (33.8–34.0 calcareous-siliceous ooze, shells and coral frag- PSS, Practical Salinity Scale 1978) and high concentrations of nutrients. Under these condi- ments, fine-grained calcareous sediment (not tions, diatom oozes will accumulate on the seafloor regardless of surface productivity as long ooze), siliceous mud, and volcaniclastics (see as there is limited competition from biogenous and detrital components, and diatom frustules the Data Repository). are not significantly dissolved prior to preservation. Quantifying the link between the seafloor The map is created using a support vector and the sea surface through the use of large digital data sets will ultimately lead to more robust machine (SVM) (Cortes and Vapnik, 1995) reconstructions and predictions of climate change and its impact on the ocean environment. classifier to predict the lithology in unobserved regions (see the Data Repository). The SVM is INTRODUCTION map of recent sediments of the oceans based on a nonparametric model that adapts in complex- Modern oceanic sediments cover 70% of the carefully selected descriptions of surface sedi- ity as new data are added. To reduce the risk of planet’s surface, forming the substrate for the ment samples contained in cruise reports from overfitting to the measurements at the expense largest ecosystem on Earth and its largest carbon recent expeditions and as long ago as the 1950s. of the model’s ability to generalize into areas reservoir. The composition and distribution of The coupling of the sediment map to key ocean- outside of the sampled regions, a cross-valida- sediments in the world’s oceans underpins our ographic parameters provides new insights into tion approach was employed to train the classi- understanding of global biogeochemical cycles, the processes governing the distribution of sedi- fier. This approach maximizes the model’s ac- the occurrence of metal deposits, sediment trans- ments in the world’s oceans and highlights sev- curacy on observations that are withheld from port mechanisms, the behavior of deep-ocean eral key discrepancies in the earlier maps. the training set. For prediction, a one-against- currents, reconstruction of past environments, one method (Bishop, 2006) was used to address and the response of the deep ocean to global METHODOLOGY the problem of modeling multiple classes with warming. A comprehensive map of ocean sedi- Our map was created mostly using surface a bilinear classifier. Classes were weighted ments can help greatly in planning oceanographic sample locations and descriptions obtained inversely proportional to their number of re- expeditions, submarine search and recovery op- through the Index to Marine and Lacustrine corded instances to account for the unbalanced erations, and the assessment of geohazards and Geological Samples (IMLGS) (Curators of Ma- nature of the data. Deep-sea lithologies that col- potential sites for the disposal of nuclear waste. rine and Lacustrine Geological Samples Consor- lectively compose >70% of seafloor sediment Virtually every marine geology and ocean- tium, 2014). The IMLGS contains data on more have been predicted with very high accuracy (to ography textbook contains a global map of five than 200,000 marine sediment samples, the vast 80%) (Figs. DR2–DR4). or six dominant sediment types in the ocean bulk of which postdates creation of the com- basins. Although there are many versions of monly used Deck41 data set (Bershad and Weiss, RESULTS this map (Barron and Whitman, 1981; Berger, 1976) and the year (1983) of the last incarnation Our digital map (Fig. 2) reveals that the pat- 1976; Davies and Gorsline, 1976; Hüneke and of the global map of oceanic sediments (Trujillo tern of distribution of different lithologies is Mulder, 2011; Trujillo and Thurman, 2014), and Thurman, 2014). We selected 14,399 data more complex, with significant regional devia- they all show strikingly similar distributions of points (Fig. 1) using strict quality control criteria tions from earlier maps (Barron and Whitman, clays and calcareous and siliceous oozes, with (see the Data Repository). 1981; Berger, 1976; Davies and Gorsline, 1976; large areas of the ocean basins draped in either There are many marine sediment classifi- Hüneke and Mulder, 2011; Trujillo and Thur- pelagic red clay or lithogenous sediments (Fig. cation schemes (Kennett, 1982) resulting in at man, 2014) (Fig. DR1). The lithologies occur in DR1 in the GSA Data Repository1). Despite the least 80 different sediment types. The classifi- drastically different proportions globally (Table vast acquisition of new data, this hand-drawn cation scheme that we use here is deliberately DR1); coverage by calcareous sediment and map has changed very little since its inception generalized in order to successfully depict the clay each increased by ~30%, and that of dia- (Berger, 1974). Here we present the first digital main types of sediments found in global oceans tom and radiolarian oozes decreased by ~25% 1GSA Data Repository item 2015271, descriptions of sample selection criteria and lithology classes, support vector machine classifier, oceanographic datasets, Figures DR1–DR8, and Tables DR1 and DR2, is available online at www.geosociety .org /pubs/ft2015.htm, or on request from editing@geosociety .org or Documents Secretary, GSA, P.O. Box 9140, Boulder, CO 80301, USA. Gridded data are available at ftp://ftp.earthbyte.org/papers/Dutkie wicz_etal_seafloor_lithology/, and can be viewed on an interactive 3-D globe at http://portal .gplates .org /cesium /?view =seabed. GEOLOGY, September 2015; v. 43; no. 9; p. 795–798 | Data Repository item 2015271 | doi:10.1130/G36883.1 | Published online 5 August 2015 GEOLOGY© 2015 Geological | Volume Society 43 | ofNumber America. 9 Gold| www.gsapubs.org Open Access: This paper is published under the terms of the CC-BY license. 795 Downloaded from geology.gsapubs.org on August 20, 2015 80°N 70°N 60°N 50°N 40°N 30°N 20°N 10°N 0° 10°S 20°S 30°S 40°S 50°S 60°S 70°S 80°S Siliciclastic Volcaniclastic Transitional Gravel Ash and volcanic Fine−grained Siliceous Sand Silt Clay and coarser sand/gravel calcareous sediment mud Biogenic Mid-ocean Calcareous Radiolarian Diatom Sponge Mixed calcareous- Shells and coral ridge ooze ooze ooze spicules siliceous ooze fragments Figure 1. Seafloor sediment sample locations. Lithology-coded sample locations of surface sediments (n = 14,399) used to create the digital map of seafloor sediments in world’s ocean basins (Fig. 2). Mollweide projection. 80°N 70°N 60°N 50°N 40°N 30°N 20°N 10°N 0° 10°S 20°S 30°S 40°S 50°S 60°S 70°S 80°S Figure 2. Digital map of major lithologies of seafloor sediments in world’s ocean basins. Legend is the same as in Figure 1. More detailed views of major ocean basins and percentage estimates of lithologies are given in Figures DR4B–DR4E and Table DR1 (see footnote 1). Moll- weide projection. 796 www.gsapubs.org | Volume 43 | Number 9 | GEOLOGY Downloaded from geology.gsapubs.org on August 20, 2015 and 60%, respectively. Rather than forming a dissolution of biogenic SiO2 relative to typical that biogenic opal accumulation in the South- belt in the equatorial Pacific extending to 30°S seawater salinity (Roubeix et al., 2008). The ern Ocean is linked to high surface productivity along the west coast of South America (Fig. summer productivity is modest (230–840 mgC/ (e.g., Nelson et al., 2002; Pondaven et al., 2000). DR1), radiolarian oozes occur as isolated pock- m2/day in the Northern Hemisphere and 175– As biogenic silica preservation efficiency, esti- ets around the equatorial Pacific in association 260 mgC/m2/day in the Southern Hemisphere), mated to be a mere 1.2%–5.5% (Nelson et al., with patches of mixed oozes and as a compo- based on satellite-derived surface chlorophyll-a 2002), is no longer considered critical for diatom nent of diatom ooze within the Peru Basin (Fig.