PyVSP Deep Learning Lookahead
Ahead-of-the-bit VSP velocity inversion with temporal convolutional networks
About
PyVSP Deep Learning Lookahead is the reproducibility pipeline for temporal-convolutional-network inversion of vertical seismic profiling data (Almarzooq et al. 2026). It covers synthetic model generation, hyperparameter optimization, training, ablation studies, blind tests, field inference, robustness analysis, and publication figures and tables.
Why it matters
Estimating velocity ahead of the drill bit can reduce subsurface uncertainty before a formation is reached. The project evaluates data-driven lookahead inversion on deterministic finite-difference simulations and Volve field data, while retaining conventional baselines.
Key result and audience
The repository turns the full manuscript workflow into a configurable pipeline and includes repeatability, bandwidth, and phase-sensitivity experiments. It is aimed at VSP researchers, seismic inversion practitioners, and teams reviewing reproducible machine-learning studies.
Public links
This public SEEM-KFUPM repository is a fork of the manuscript’s reproducibility repository. No software license is declared in the GitHub metadata; consult the repository before reuse.