NGH-Seis v1.0: A physics-based synthetic marine seismic dataset for gas hydrate and free-gas characterization

It contains 1,000 quality-controlled geological realizations balanced across five geological scenario families. Each realization links geological and petrophysical fields, P-wave velocity, density, acoustic impedance, hydrate saturation, free-gas saturation, 25-shot towed-streamer observations, an independently generated reverse-time-migration image, spatial masks, coordinates, and generation metadata.

The dataset provides a stratified 800/100/100 IID training, validation, and test split, together with five leave-one-family-out structural-OOD folds. It supports two benchmark tasks: reconstruction of P-wave velocity and acoustic impedance, and estimation of hydrate and free-gas saturation. All realizations are synthetic outputs of two-dimensional variable-density acoustic simulations and should not be interpreted as a digital twin or site-specific inversion of a particular marine area.