Spatially predicted concentrations of particulate organic carbon in surface (0-10cm) sediments of parts of the north-west European continental shelf (1996-2015), supplement to: Diesing, Markus; Kröger, Silke; Parker, Ruth; Jenkins, Chris; Mason, Claire; Weston, Keith (2017): Predicting the standing stock of organic carbon in surface sediments of the North–West European continental shelf. Biogeochemistry

Abstract

The dataset contains spatially predicted concentrations of particulate organic carbon (POC), expressed as percent of dry weight, in the top 10cm of the sediment column in parts of the north-west European continental shelf (North Sea, English Channel, Celtic Sea). The results are presented as a geo-referenced floating point TIFF-file with a spatial resolution of 500m and ETRS89 Lambert Azimuthal Equal-Area projection as spatial reference.Measurements of sediment organic carbon in marine sediments around England and Wales collected between 1996 and 2015 (http://dx.doi.org/10.14466/CEFASDATAHUB.32) were initially transformed using an arcsine transformation (Sokal and Rohlf 1981). Subsequently, we spatially predicted transformed POC concentrations using a Random Forest (Breiman 2001) regression model based on those point samples and a number of important and uncorrelated predictor variables. Two thirds of the samples were used to train the model. Predictors were selected from an array of variables available with full coverage of the study area including bathymetry, Euclidean distance to the nearest shoreline, geographic position (eastings, northings), sediment composition (mud, sand and gravel fraction), earth observation data (chlorophyll-a, depth of the euphotic zone and suspended particulate matter concentrations) from the Moderate Resolution Imaging Spectroradiometer (MODIS), hydrodynamic model data (depth averaged mean and peak current speed, peak wave orbital velocity and peak wave-current shear stress), water-column bottom salinity (annual average and range), water-column bottom temperature (annual average and range) and stratification (thermal and salinity). We used a variable selection wrapper algorithm (Kursa and Rudnicki 2010) to identify important predictor variables and subsequently reduced the set of variables to those that were uncorrelated (|r| < 0.5). The selected predictor variables, in decreasing order of importance, were: mud fraction of the surficial seabed (Stephens and Diesing 2015); annual average bottom water temperature (Berx and Hughes 2009); Euclidean distance to coast; longitudinal position (eastings); gravel fraction (Stephens and Diesing 2015) of the surficial seabed; and peak wave orbital velocity (Aldridge et al. 2015; Bricheno et al. 2015) at the seabed. Model validation statistics, derived from the remaining one third of the samples not used for model training, showed that the final model was not over-fitted and explained 77.5% of the variance in the response variable. Lastly, the spatially predicted transformed POC concentrations were back-transformed to POC concentrations.