Global seamount detection using machine learning

Volcanic in origin, seamounts are widely scattered throughout the world’s ocean basins. Research ships with single beam and multibeam echo-sounders have discovered many small seamounts and some large ones while satellite altimetry has found many large seamounts and some small ones. The exact number of seamounts in the world’s ocean basins is, however, unknown. Here we use machine learning in an attempt to locate all seamounts, to estimate their height, area and volume, and to speculate on their origin. We use the seamounts that have already found along ship track from single beam. echo-sounder data and from the satellite-derived vertical gravity gradient as a ‘ground truth’ data set and a GEBCO 2025 grid which combines shipboard single beam and multibeam (swath) bathymetry data with predicted bathymetry from satellite altimeter data in regions of sparse ship tracks as an ‘input’ data set. The two data sets are used to determine the 6 main attributes (channels) of seamounts, 4 of which refer to their directional slopes, and then are used to train the model. Finally, we use the model together with morphological based post-processing algorithms to update the global seamount census and assess their precision, recall and F1-score. Results show that machine learning predicts up to factors of 1.8–2.7 more seamounts than previous ground truth studies, 28,434 of which rise 1 km and higher above the surrounding depths of the ocean floor.