Document Type


Date of Original Version



Grain size characteristics of pyroclastic deposits provide valuable information about source eruption energetics and depositional processes. Maximum size and sorting are often used to discriminate between fallout and sediment gravity flow processes during explosive eruptions. In the submarine environment the collection of such data in thick pyroclastic sequences is extremely challenging and potentially time consuming. A method has been developed to extract grain size information from stereo images collected by a remotely operated vehicle (ROV). In the summer of 2010 the ROV Hercules collected a suite of stereo images from a thick pumice sequence in the caldera walls of Kolumbo submarine volcano located about seven kilometers off the coast of Santorini, Greece. The highly stratified, pumice-rich deposit was likely created by the last explosive eruption of the volcano that took place in 1650 AD. Each image was taken from a distance of only a few meters from the outcrop in order to capture the outlines of individual clasts with relatively high resolution. Mosaics of individual images taken as the ROV transected approximately 150 meters of vertical outcrop were used to create large-scale vertical stratigraphic columns that proved useful for overall documentation of the eruption sequence and intracaldera correlations of distinct tephra units. Initial image processing techniques, including morphological operations, edge detection, shape and size estimation were implemented in MatLab and applied to a subset of individual images of the mosiacs. A large variety of algorithms were tested in order to best discriminate the outlines of individual pumices. This proved to be challenging owing to the close packing and overlapping of individual pumices. Preliminary success was achieved in discriminating the outlines of the large particles and measurements were carried out on the largest clasts present at different stratigraphic levels. In addition, semi-quantitative analysis of the size distribution could also be determined for individual images. Although a complete size distribution is not possible with this technique, information about the relative distribution of large and medium size clasts is likely to provide a reasonable proxy for the overall sorting of submarine deposits. Our preliminary work represents the first attempt to carry out an in situ granulometric analysis of a thick submarine pyroclastic sequence. This general technique is likely to be valuable in future studies of submarine explosive volcanism given the recent discoveries of extensive pumiceous deposits in many submarine calderas associated with subduction zone environments. AGU session number OS13A-1501