Atlas Deep Geo develops physics-guided seismic reconstruction methods that transform sparse 2D subsurface observations into probabilistic 3D volumes with calibrated P10, P50, and P90 uncertainty envelopes. Its ORCA framework combines a Radial Basis Function smooth baseline with a neural network trained on geologically realistic synthetics, delivering an honest range of outcomes rather than a single best-estimate surface. The company engages through proof-of-concept projects on real datasets before moving toward license-style packaging for repeat workflows.
Funding
Funding not disclosed
Founders
Product
Problem
Subsurface imaging often relies on sparse 2D seismic lines or surveys with gaps, obscuring geological structures and leaving operators without a complete 3D picture. Conventional reconstruction methods produce a single best-estimate surface, concealing the uncertainty inherent in limited spatial sampling and leaving critical decisions uninformed by what the data cannot resolve.
Solution
Atlas Deep Geo provides physics-guided seismic reconstruction through its ORCA framework, turning sparse observations into complete probabilistic 3D volumes. ORCA combines a Radial Basis Function (RBF) smooth baseline that handles the regional trend with a neural network trained on geologically realistic synthetics to predict the residual detail the baseline misses. The framework embeds wave propagation principles, acquisition geometry, and geological prior knowledge as constraints, ensuring outputs are physically plausible rather than purely data-driven. Unlike conventional methods, ORCA generates an ensemble of geologically plausible reconstructions, each consistent with the observed picks, and reports the spread across that ensemble as calibrated P10, P50, and P90 uncertainty envelopes—quantifying exactly what the data supports and where it runs out.
Target Audience
Primary customers are oil and gas operators and seismic data holders with extensive legacy 2D datasets or incomplete 3D surveys who need probabilistic subsurface models for informed development decisions.
Features
- ORCA framework combining an RBF smooth baseline with a neural network trained on geologically realistic synthetics for residual detail prediction
- Physics-informed constraints including wave propagation laws, acquisition geometry, and geological priors that bound every reconstruction
- Ensemble-based output producing calibrated P10, P50, and P90 uncertainty envelopes instead of a single point estimate
- Direct applicability to 2D-to-3D seismic conversion, 3D survey gap infill, imaging beneath obscured zones, and near-offset reconstruction
- Simplicity-first modeling guided by the Occam principle, avoiding the invention of complexity the data does not support
- Published case studies and documented QC for honest validation, including uncertainty comparisons against ground truth when available