MULTISEISMO: A Multimodal Seismic Dataset and Model for Cross-Modal Seismic Understanding

A new multimodal seismic dataset named MultiSeismo has been introduced, comprising over 16,000 seismic events recorded over 13 years from 2010 to 2023. This dataset integrates waveform recordings, intensity maps, population exposure visualizations, and textual descriptions in a standardized JSON format, addressing the lack of comprehensive datasets in seismology that combine various data modalities.

The development of MultiSeismo and the accompanying multimodal instruction set, MISCE, is significant as it enhances the ability to train and evaluate generalist multimodal models in seismology, potentially leading to improved understanding and prediction of earthquake phenomena.