Supplementary Material

Global Carbon Cycling on a Heterogeneous Seafloor

Paul V. R. Snelgrove,Department of Ocean Sciences and Biology Department, Memorial University of Newfoundland, St. John's NL A1C 5S7 Canada.

Karline Soetaert, Estuarine and Delta Systems. Netherlands Institute of Sea Research and Utrecht University. The Netherlands.

Martin Solan, Ocean and Earth Science, National Oceanography Centre Southampton, University of Southampton, Waterfront Campus, European Way, Southampton, UK. SO14 3ZH.

Simon Thrush, Institute of Marine Science, The University of Auckland, Auckland, 1142, New Zealand.

Chih-Lin Wei, Institute of Oceanography, National Taiwan University, Taipei 106, Taiwan.

Roberto Danovaro,Dept. Life and Environmental Sciences, Polytechnic University of Marche, Italy and Stazione Zoologica Anton Dohrn, Naples, Italy.

Robinson W. Fulweiler, Departments of Earth and Environment and Biology, Boston University, Boston, MA. USA.

Hiroshi Kitazato, Tokyo University of Marine Science and Technology, 4-5-7 Konan, Minato-ku, Tokyo 108-8477, Japan.

Baban Ingole, National Institute of Oceanography, Dona Paula, Goa, India.

Alf Norkko,Tvärminne Zoological Station, University of Helsinki. Hanko, Finland.

R. John Parkes,School of Earth and Ocean Sciences, Cardiff University, Cardiff, CF10 3AT, UK.

Nils Volkenborn, School of Marine and Atmospheric Sciences, Stony Brook University, Stony Brook, NY. USA.

Corresponding author:Dr. Paul Snelgrove, , 709-864-3440

We based our characterization of benthic conditions on a total of 24 global-coverage seafloor environmental datasets (described below). Among those, we derived bathymetric data from the Cell/pixel-registered ETOPO1 bedrock global relief model ( reducing the grid resolution from 1 minute to 5 minutes by averaging. We calculated terrain characteristics including slope and aspect of terrain in radian degree, tpi (mean of the absolute differences between the value of a cell and the value of its 8 surrounding cells), tri (difference between the value of a cell and the mean value of its 8 surrounding cells) and roughness (difference between the maximum and the minimum value of a cell and its 8 surrounding cells) from the global relief model. Decadal (1955-2012) seafloor mean temperature and salinity (1/4 degree resolution), as well as oxygen, silicate, phosphate and nitrate concentration (1 degree grid) were downloaded from World Ocean Atlas 2013 v2 ( The global climatological monthly mean SeaWiFS (1998-2002) and MODIS (form 2003-2014) Level-3 chlorophyll a (chl mean) concentration and Level-4 Vertical General Production (vgpm mean) data[1] (5 minutes grids) were downloaded from We calculated export POC flux (lutz_poc) at the seafloor using an empirical equation[2] based on the mean and seasonality (svi=sd/mean) of primary production, as well as the global mean export depth below the euphotic layer over the 16-year period (1998 to 2014). We calculated euphotic zone depth using a Case I model from mean surface chlorophyll concentrations[3], and export depth by subtracting the euphotic depth from water depth. The projected seafloor living standing stocks (1 degree grids, including interpolated values over data sparse regions) including abundance and biomass (biom) of prokaryotes (bact), meiofauna (meio), macrofauna (macro), and megafuna (mega) were derived from published estimates[4]. The projected and observed biomass were cross-validated by Wei et al. [4], in which empirical models independently explained 63 to 88% of the observationacross different benthic size classes. Mean multivariate environmental data were calculated from all pixels within a circular area of 110 km radius (approximately 1 degree at the equator) surrounding the geographic coordinates of sedimentary community oxygen consumption (SCOC) observations from multiple sources, many from our own labs.

We used a binary decision tree method, Random Forest (RF)[5] to fit logarithm transformed SCOC with the 24 environmental factors listed above. In the RF, we randomly selected an arbitrary 2/3 of the SCOC data to construct the model, retaining the other 1/3 of the data to validate the model accuracy and calculate explained variance (R2). We repeated this resampling (with replacement) procedure 5,000 times and calculated the average R2 (=0.8). In other words, among the 5000 iterations, the models constructed by training the dataset (2/3 of total observations) on average independently explained 80% of the total variability of the validating dataset (1/3 of total). This explanatory power gives us high confidence on the predictability of the model. In order to extrapolate SCOC to un-sampled areas, we calculated the average predicted SCOC by inserting the multi-layer seafloor environmental data into each of the 5000 models to calculate the mean. The resulting global map was fully interpolated, based on 1075 SCOC observations.

We calculated our geochemically based flux as follows:

According to [2], the POC deposition flux to the seafloor can be represented by:

Lutz_poc=npp∙p_ratio

p_ratio =prd∙e^(-1/rld∙ze )+prr

Here nppdenotes mean net primary production, as estimated from remote sensing; p_ratio denotes the fraction of the net production deposited on the sediment surface. This fractionconsists of two parts, one that decays with export depth zeaccording to an exponential coefficient 1/rld (units of [m-1]), and another,prr, that remains constant; prd, prr and rld are functions of the strength of seasonal variation in net primary production (sd/mean) as in [2], zedenotes the depth of the seafloor below the euphotic zone. Based on the similarity of the formula describing the decaying fraction with the analytical solution of a reactive-sinking model, the coefficient of exponential decay 1/rld can be considered equivalent to k0/w, wherek0 isthe first-order decay coefficient of the labile fraction (unit of [/d]) and wis a constant sinking speed (unit of [m/d]). A measure of the decay rate of POC deposited on the sediment (ks) is then:

ks=(prd∙e^(-k0/w∙ze ))/p_ratio∙k0

Assuming w = 150 m d-1, k0can be estimated from the estimates of rld[2]; this yields mean decay rates that approximate the biological estimate of lability.

1. Behrenfeld, M.J. and Falkowski, P.G. (1997) Photosynthetic rates derived from satellite‐based chlorophyll concentration. Limnology and oceanography 42 (1), 1-20.

2. Lutz, M. et al. (2002) Regional variability in the vertical flux of particulate organic carbon in the ocean interior. Global Biogeochemical Cycles 16 (3).

3. Morel, A. and Berthon, J.F. (1989) Surface pigments, algal biomass profiles, and potential production of the euphotic layer: Relationships reinvestigated in view of remote-sensing applications. Limnology and Oceanography 34, 1545-1562.

4. Wei, C.L. et al. (2010) Global patterns and predictions of seafloor biomass using Random Forests. PLoS ONE 5 (12), e15323.

5. Breiman, L. (2001) Random forests. Machine learning 45 (1), 5-32.