Automated geostatistical seafloor classification—Principles, parameters, feature vectors, and discrimination criteria

Abstract

A geostatistical method for automated seafloor classification is developed and applied to bathymetrie data for a 150 × 100 km area at 26 °N on the western flank of the Mid-Atlantic Ridge. The objective of seafloor classification is to characterize seafloor properties quantitatively, and to use such spatial characteristics to distinguish roughness provinces, and geologic and morphologic units automatically. The method presented here is based on the calculation of directional variograms as spatial structure functions. Parameters determined from filtered variogram functions are used to compose feature vectors, which are shown to be characteristic of morphologic prototypes and surface roughness types, and therefore facilitate a classification. Discrimination criteria include spacing and strike of abyssal hill terrain, smoothness resulting from sediment cover, and parameters related to complexity and morphological significance of abyssal hills and their slopes. Complications of automating the process concern robustness of parameter estimation, optimal window size, and subselection of data. By moving the classification operation through the study area and color-coding property classes, seafloor classification maps are obtained. The concepts of characteristic parameters, feature vectors and discrimination criteria are illustrated in applications to bathymetric data from the western flank of the Mid-Atlantic Ridge. Resultant classification maps are presented, with classes including roughness provinces and morphologic units.