Neuro-fuzzy Classification of Submarine Lava Flow Morphology

Abstract

This paper presents a novel approach to semi-automated classification of volcanic morphology on the seafloor using high-resolution multibeam sonar bathymetry and side-scan sonar backscatter imagery. The classification methodology combines a fuzzy inference system and neural network theory in an adaptive neuro-fuzzy inference system (ANFIS) and is capable of rapidly classifying submarine lava morphology based on bathymetry-derived surface geometry and backscatter-derived attributes of acoustics and texture. The system has been applied to a study area on a seafloor spreading ridge, the Galapagos Spreading Center (GSC), in order to quantify the distribution and relative abundance of lava flow types, which can be used to indicate variations in eruption and emplacement dynamics. A detailed assessment shows the classification has an overall accuracy of almost 90 percent with a kappa coefficient of 0.84. The neuro-fuzzy method described here is shown to be an efficient and reliable tool for classification of submarine lava morphology.