Characterizing Lava Lake Processes by Using a Convolutional Neural Network and Support Vector Machine to Detect Strombolian Eruptions in Infrared Images at Mount Erebus, Antarctica

Strombolian eruptions offer insight into the subsurface processes of lava lakes and their magma conduits. Understanding the characteristics of the lava lake and conduit is essential for better eruption forecasting. Our team has previously demonstrated that machine learning, specifically neural networks, is a tool that can be used in conjunction with existing methods to characterize the lava lakes eruptive behavior (Dye and Morra 2020). In this research, we utilize findings from our previous and ongoing research in machine learning methods for detecting Strombolian eruptions to discover insights into the processes driving the convection of the lava lake, and the magma chamber and conduit flow. For this research, we used two machine learning methods to detect both large Strombolian eruptions and small bubbles. For large Strombolian eruptions, we developed a convolutional neural network specifically for identifying Strombolian eruptions in infrared images. We utilized Support Vector Machine as a second machine learning method tasked with detecting frequent small bubbles which, due to their size, are difficult to detect using other methods such as luminosity, gas chromatography, and seismic cross-correlation. The combination of these methods has proven to be an efficient supplemental method to manually detecting eruptions in various datasets.