Tsunami-Surrogate Benchmark: Processed Multi-Reference Dataset for Neural-Operator Emulation of Tsunami-Like Wave Propagation

Processed benchmark data and result artifacts for “Tsunami-Surrogate: A Controlled Multi-Fidelity Benchmark for Neural-Operator Emulation of Tsunami-Like Wave Propagation.” This archive contains model-ready processed datasets for hydrostatic reconstruction, MUSCL-HR, and elevation-only Boussinesq reference simulations; metadata-filtered distribution-shift suites; strict held-out bathymetry experiments; cross-resolution processed data; real-bathymetry diagnostic suites; scenario manifests; normalization statistics; checksums; and JSON/CSV result files underlying the paper tables and figures. The benchmark is designed for research on fast forward emulation of solver-generated tsunami-like wavefields. It is not intended for operational tsunami warning, navigation, safety-at-sea use, or site-specific hazard assessment. Code and scripts are available at: https://github.com/1zuki/tsunami-surrogate The full raw rollout archive is not included in this DOI archive because of its size. It can be regenerated from the released scripts and configuration files. A large auxiliary mirror is available separately.