Sediment Maps for the Continental Shelf of the US Gulf of America and South Atlantic Bight Using Compositional Kriging

Abstract

We produced maps of sediment fractions for the US Gulf of America (formerly Gulf of Mexico) and South Atlantic Bight at 1 km² spatial resolution using compositional kriging. Quantitative tools were used to identify the optimal pixel size of the output map, which was produced using compositional kriging of log-ratio transformed variables. Input data were extracted from the databases usSEABED and dbSEABED, and were in the form of 167,854 sediment samples with the percentage composition of sand, mud and gravel. Sediments for the Gulf of America were mostly muddy (35% median, while sand and gravel took 20% and 0%) and for the South Atlantic Bight were mostly sandy (86%, with sand and gravel fractions having 0% of the median). Gravel was always the least common fraction. Anisotropy (variable spatial continuity in different directions) was negligible in the Gulf of America but relevant in the South Atlantic Bight. Sediment data were uncorrelated with bathymetry in both regions. Spatial resolution for the output maps was identified as 1 km² based on quantitative analyses. Interpolated maps were computed using compositional kriging on log-ratio transformed variables. The standard deviation of the estimator based on the kriging variance was 0.12 for gravel, 0.18 for sand and 0.06 for mud in the Gulf and 0.14 for gravel, 0.17 for sand and 0.001 for mud in the Atlantic. Compositional kriging is the method that provides the best accuracy in terms of mean absolute error. Interpolation of raw variables provides the best accuracy according to root mean square error, but handling of fractions individually is statistically inappropriate for this type of data.