Dataset for Sensitivity-Driven Scaling Enables Real-Time Inference for High-Dimensional PDEs: Tsunami Case Study

This dataset supports the experiments presented in “Sensitivity-Driven Scaling Enables Real-Time Inference for High-Dimensional PDEs.” It contains the data used for the tsunami case study, including source representations, forcing conditions, and corresponding simulation outputs generated by shallow-water equation (SWE) models. The dataset is designed for training and evaluating neural operator models, particularly the Sensitivity-Constrained Neural Operator (SC-NO) and its Fourier-based implementation (SC-FNO). It enables both forward prediction of tsunami propagation and inverse reconstruction of seafloor deformation from sparse observations. Contents include: Processed input fields representing initial seafloor deformation and forcing conditions Time-resolved solution fields (water surface elevation and related variables) Preprocessed data used for training, validation, and testing Auxiliary files required to reproduce the tsunami forecasting and inversion experiments The data are generated using physics-based numerical solvers for the shallow water equations and are structured to align with the model inputs and outputs described in the manuscript. Usage This dataset can be used to: Train neural operators for spatiotemporal PDE systems Benchmark forward prediction accuracy and rollout stability Evaluate inverse problems such as source reconstruction from sparse sensor data The corresponding source code for model training, inference, and evaluation is available at:https://doi.org/10.5281/zenodo.19923479 Notes The dataset is provided in a processed format suitable for machine learning workflows Users should refer to the associated repository for data loading and preprocessing scripts This dataset focuses on the tsunami case study; additional datasets for other PDE systems may be released separately If you want, I can also: make a short version (Zenodo summary field) or prepare the exact BibTeX dataset citation (AGU-ready)