Predicting the standing stock of organic carbon in surface sediments of the North–West European continental shelf

Shelf seas and their associated benthic habitats represent key systems in the global carbon cycle. However, the quantification of the related stocks and flows of carbon are often poorly constrained. To address benthic carbon storage in the North–West European continental shelf, we have spatially predicted the mass of particulate organic carbon (POC) stored in the top 10 cm of shelf sediments in parts of the North Sea, English Channel and Celtic Sea using a Random Forest model, POC measurements on surface sediments from those seas and relevant predictor variables. The presented model explains 78% of the variance in the data and we estimate that approximately 250 Mt of POC are stored in surficial sediments of the study area (633,000 km2). Upscaling to the North–West European continental shelf area (1,111,812 km2) yielded a range of 230–882 Mt of POC with the most likely estimate being on the order of 476 Mt. We demonstrate that the largest POC stocks are associated with coarse-grained sediments due to their wide-spread occurrence and high dry bulk densities. Our results also highlight the importance of coastal sediments for carbon storage and sequestration. Important predictors for POC include mud content in surficial sediments, annual average bottom temperature and distance to shoreline, with the latter possibly a proxy for terrestrial inputs. Now that key variables in determining the spatial distribution of POC have been identified, it is possible to predict future changes to the POC stock, with the presented maps providing an accurate baseline against which to assess predicted changes. Electronic supplementary material The online version of this article (doi:10.1007/s10533-017-0310-4) contains supplementary material, which is available to authorized users.


Estimation of the spatial extent of sediment types on the North-West European continental shelf
For the purpose of upscaling the results of the spatial predictions of particulate organic carbon (POC) stored in the top 10 cm of shelf sediments in the study area it was necessary to know the spatial extent of Folk (1954Folk ( , 1980) sediment types on the North-West (NW) European continental shelf. Although Stephens and Diesing (2015) recently presented a map of Folk sediment classes for a large part of the NW European continental shelf, significant parts of the NW European continental shelf are not covered by their map. These gaps cannot be closed with existing data products, e.g. EMODnet-Geology (http://www.emodnet.eu/geology) seabed substrate maps do not have the required thematic resolution. Hence, we have derived our own purpose-built Folk sediment map and describe the process and results in the following.
The long-term, collaborative project dbSEABED (see Goff et al., 2008;Jenkins, 2002) was employed to produce a map of sediment classes across the NW European Shelf area. dbSEABED is an information processing system that uses various mathematical methods to extract harmonised, quality controlled attributes from a large, structured archive of legacy-data analyses and descriptions of sediments. Analytical numerical data is simply extracted after quality assurance. The worddescriptive data is parsed and analysed as described in Jenkins (1997), again with quality checks, which have also included validations of the process. In cases where a water depth value was not available, that was attached using the SRTM 30-Plus global bathymetry (Becker et al., 2009). Note that skeletal biological components in the sediments, such as coral and shell, are counted in the gravel proportions.
A sub-set of 74,046 data points for the area ( Figure S1) was projected to ETRS89-LAEA coordinates. The median separation of data points is 1.4 km and the median distance from any ocean-area cell to nearest data is 20.0 km (mostly due to deep water sample spacings). Kilometre scale cell-wise values for the average and standard deviation of gravel, sand and mud contents were calculated for the data-bearing cells, then predicted for vacant cells within a 20 km radius (less as the shore is approached).
The prediction was carried out by local Inverse Distance Weighting (IDW; see Tomczak, 1998 for geographic distances) with a filter against data more than a factor of 2 different in water depths from the target cell's depth. This is important for mapping inshore as well as shelf-break and continental slope areas. Vacant parts of the map beyond the reach of this process were filled using a Random Forest (RF; Breiman, 2001) predictor, employing the variables water depth, bathymetric gradient and standard 2 deviation and distance to shoreline. After merging these, Folk codes (BGS variant) were computed from the gridded textural values. Uncertainties on the data were computed rigorously in the local IDW process, but were assigned to 100% of parameter range for the global RF process. Discrepancies due to 'compositional data' effects (Aitchison, 1986) were placed in the uncertainties.

Fig. S1
Locations of samples from dbSEABED utilised to predict sediment composition 3 Outputs of the spatial predictions were combined with the map of Stephens and Diesing (2015) to derive a map of Folk sediment classes for the NW European continental shelf ( Figure S2). The spatial extent of the different sediment classes was extracted and is detailed in Table S1.