MDIO v1: Schematizing seismic data for AI and processing

The MDIO format is an open-source storage solution for seismic data, designed for processing, AI training and inference, data management, and visualization. Built on top of the open-source projects Zarr, Blosc, Xarray, and TensorStore, it enables efficient access to large datasets on both local and cloud storage, reducing costs and enhancing accessibility. MDIO v1 introduces a structured dataset model that includes Metadata, Variables, Dimensions, and Coordinates, simplifying the organization and use of seismic data. It is interoperable with Python’s Xarray format to ensure compatibility with scientific workflows and natively supports Dask for parallel processing. A native C++ implementation based on Google’s TensorStore library offers options for integrating MDIO with existing processing systems. We present case studies highlighting the benefits of MDIO in seismic workflows, such as rapid SEG-Y ingestion, cloud-based I/O with an example from the AWS Open Data Registry, and the use of structured metadata for pre-stack and post-stack visualization. MDIO enhances efficiency in large-scale geophysical analysis and AI applications by clearly defining the relationships in seismic data.