Extracted Data from a Scoping Review of Machine Learning Approaches for Wave Propagation Modeling in Seismology

1) General Information (General Reading) - Paper type / category: - Measurement paper? - Analysis of an existing system/method? - New method proposal? - Research prototype / proof of concept? - Review / survey? - Context and related work: - Which prior papers is it related to? - What theoretical foundations are used (PDE theory, inverse theory, optimization, numerical analysis, etc.)? - Correctness and assumptions: - Are the assumptions stated clearly? - Do they appear valid for seismic wave modeling? - Are limitations acknowledged? - Main contributions: - What are the key contributions? - What is new compared to prior work? - Clarity and quality of writing: - Is the paper well written and well structured? - Are the methods reproducible? 2) Neural Network Architecture (ML Design) - Which neural network architecture is used? - Advantages and disadvantages of the chosen architecture: - Reported strengths: - Reported limitations (scalability, stability, generalization, etc.): - Frameworks and implementation tools: - ML framework (PyTorch, TensorFlow, JAX, etc.): - Physics/PDE framework used for forward modeling (if any): 3) Forward Modeling (Direct Problem) - Did the paper use a standard numerical method as reference? - If yes, which software/framework was used? - SPECFEM - FEniCS - DUNE - Devito - Madagascar - In-house solver - Other (specify) 4) Type of data used for inversion: - Real data - Synthetic data - Both - If synthetic: how was it generated? 5) Dimensionality and Scalability - Dimensionality of the modeled problem: - 1D / 2D / 3D - Impact of dimensionality on ML complexity (if discussed): - Did model size increase significantly? - Did the number of parameters increase? - Did training time increase? - Was memory/GPU a limitation? 6) Governing Equations and Physical Model - Which equation(s) are modeled? - Acoustic wave equation - Elastic wave equation - Viscoelastic / attenuation models - Anisotropic models - Other PDEs (specify) - How does the equation type influence the method? 7) Hybrid and Combined Strategies (Traditional + ML) - How are traditional methods combined with ML? - ML as a surrogate forward solver - ML to accelerate simulation - ML to improve inversion (regularization, priors) - Hybrid loss functions (data + physics) - Coupling numerical solvers with neural networks - Other (specify) - What was the reported benefit of the hybrid strategy? - Speedup - Better accuracy - Better generalization - Improved stability - Lower memory/computation cost 8) Inverse Problems (Applications) - Does the paper solve an inverse problem? - Yes / No - If yes, which one? (FWI, tomography, source inversion, parameter estimation, etc.) 9) Limitations and Future Work - What limitations does the paper acknowledge? - Scalability to 3D - Realistic Earth models - Noise robustness - Computational cost - Generalization across velocity models - Other (specify) - What future work is proposed? - What is explicitly left as future work? 10) Evaluation and Comparative Results - How is error measured? - L2 error - Relative error - Misfit function - Travel-time error - Signal-to-noise metrics - Structural similarity (SSIM) - Other (specify) - If comparison is qualitative, how is it done? - Visual comparison of wavefields - Comparison of seismograms - Comparison of reconstructed models - Comparison of arrival times - Other (specify)