Testing pre-trained models for earthquake detection at Raoul Island volcano, Aotearoa New Zealand

Pre-trained models offer a means to deploy machine learning earthquake detection at volcanoes with sparse training data. Here I test this approach with multiple potential training datasets on seismic data recorded at Raoul Island volcano during the 2006 eruption. I compared earthquake detections rates from the EQTransformer model with the classic STA/LTA technique. EQTransformer performs well with all training datasets during intense volcano-seismic activity, with overcounts avoided by combining the results from each dataset. However, I highlight the need for a New Zealand volcano-focused training dataset that can help detect low-frequency volcano seismicity.