|
<< Click to Display Table of Contents >> Navigation: Migration > RTM3D depth distributed |
This module performs 3D Reverse-Time Migration (RTM) in the depth domain using a distributed (cluster) architecture. RTM is a two-way wave-equation migration method that produces high-quality depth images by forward-propagating the source wavefield and back-propagating the recorded receiver wavefield, then applying an imaging condition (cross-correlation) at each depth point. Because the full wave equation is solved, RTM correctly images complex overburden structures, steeply dipping reflectors, and salt flanks that are difficult to handle with ray-based methods.
The distributed implementation splits the input source gathers across compute nodes, with each node processing one chunk of shots independently. Partial images from all nodes are then summed to produce the final migrated volume. Both CPU and GPU execution are supported. Use this module for large 3D surveys where single-node computation would be impractically slow.
Path and filename for the output RTM depth image (GSD format). The file will contain the stacked or sub-stacked migrated volume depending on the Save type setting.
Determines how individual shot contributions are accumulated into the output. Create stack sums all shot images into a single migrated volume. Create sub-stacks (default) saves each source-gather image as a separate sub-stack, enabling angle-domain common-image gather construction or shot-domain noise attenuation in post-processing.
Directory path where intermediate wavefield snapshots are stored during the RTM correlation step. This directory must be accessible by all compute nodes in a distributed run and should reside on fast local or shared storage. The temporary files are deleted after the migration completes.
3D depth velocity model (P-wave interval velocity cube) used for wavefield propagation. The velocity model must span the full lateral and depth extent of the survey. Resolution should be consistent with the DX and DY grid spacing set in the Parameters section.
Handle to the input pre-stack seismic data in SEG-Y format. Traces should be organised by source gather and contain valid source and receiver coordinate headers.
Trace header vector providing geometry for each input trace, including source and receiver positions used to partition the data into shot gathers for RTM processing.
Controls how surface topography is handled at the top of the velocity model. No topo assumes a flat surface at zero depth. By air velocity (default) detects the earth surface by identifying where the velocity drops below the Air velocity threshold, filling the region above with air velocity. By headers reads surface elevation from trace headers. Use By air velocity when the velocity model contains an air layer; use By headers for land data with significant elevation variation.
Velocity in m/s used to fill the air (above-surface) layer in the model (default: 310 m/s). Only used when Detect topography is set to By air velocity. This value acts as the threshold: model cells with velocity below this value are treated as air.
Integer factor controlling source-gather decimation (default: 1, all sources). Setting to 2 uses every other source point, reducing computation time by approximately half. Use values greater than 1 for quick QC images or when the acquisition is significantly oversampled.
Finite-difference grid spacing in the X (crossline) direction in metres (default: 20 m). This value must match the lateral sampling of the input velocity model. Setting DX too coarse relative to the dominant wavelength will cause numerical dispersion artefacts; as a rule of thumb, DX should be no larger than the minimum wavelength divided by four.
Finite-difference grid spacing in the Y (inline) direction in metres (default: 20 m). Apply the same selection rule as for DX above.
Additional lateral margin in metres (default: 500 m) added around each shot gather's receiver footprint when extracting the local velocity model sub-volume for that shot. Increasing this value improves illumination of steeply dipping events near the aperture edge at the cost of processing more velocity model points per shot.
Number of absorbing boundary layer cells added around the finite-difference model to suppress artificial reflections from the grid edges (default: 20). Larger values produce cleaner boundaries at the cost of a larger computational domain. A minimum of 20 cells is recommended for most datasets.
Peak frequency of the modelled source wavelet in Hz (default: 25 Hz). This controls the stability check for the finite-difference time step. Set this to the dominant frequency of the recorded data. Higher frequencies require a smaller time step (DT) and finer spatial grid spacing to satisfy the dispersion condition.
Finite-difference time step in seconds (default: 0.0001 s, i.e. 0.1 ms) used for the wave equation propagation. This must satisfy the Courant stability condition: DT < DX / (sqrt(3) * Vmax). Smaller time steps improve numerical accuracy but increase computation time proportionally.
Time interval in seconds at which wavefield snapshots are saved to disk for the cross-correlation imaging condition (default: 0.001 s, i.e. 1 ms). This must be greater than or equal to the propagation Time step. A larger correlation step saves fewer snapshots and reduces disk I/O but may slightly reduce image quality for very high frequencies. This value must be a multiple of the propagation Time step.
Total number of time samples for the forward and backward wave propagation (default: 100). The maximum imaging time is NT * DT seconds. Set this to cover the full two-way travel time of the deepest target: NT = round(max_TWT / DT). For example, to image to 3 s two-way time with DT = 0.0001 s, set NT = 30000.