|
<< Click to Display Table of Contents >> Navigation: Modeling > FD-RTM 2D modeling |
FD-RTM 2D modeling performs Finite-Difference Reverse Time Migration (FD-RTM) forward modeling in two dimensions. The module synthesizes seismic shot records and produces depth-domain images and angle-domain common image gathers (ADCIGs) from a supplied interval velocity model. It is used for model-building QC, migration operator testing, and velocity analysis workflows where a full two-way wave-equation solution is required.
The algorithm uses a staggered-grid finite-difference acoustic wave equation solver with a Perfectly Matched Layer (PML) absorbing boundary. For each active source position the module runs two forward-propagation passes: one through the full velocity model to record observed synthetic data, and one through a smooth background velocity model to record modeled data. The residual between the two shot records is then back-propagated, and the cross-correlation of the forward and backward wavefields accumulates both a stacked depth image and angle-domain gathers. The Ricker wavelet source signature is synthesized internally from the specified peak frequency.
The full interval velocity model in the depth domain, supplied as a velocity gather (VistaGatherSettingsVelocity format). This model drives the synthetic wavefield propagation that produces the observed shot records. Trace spacing in this gather defines the horizontal spatial sampling of the model grid; the vertical sample interval is read directly from the gather and used as the depth grid step. Make sure the model covers the entire area to be imaged and that the velocity values are geologically realistic, as the PML damping boundary is calibrated from the maximum velocity found in this model.
A spatially smoothed version of the depth velocity model, also in VistaGatherSettingsVelocity format and with the same dimensions as the full model. This smooth background model is used for the second forward-propagation pass that produces the modeled (reference) shot record. The residual between the full-model and smooth-model shot records drives the back-propagation and cross-correlation imaging step. Typically this is prepared by applying a moderate spatial smoothing filter to the Depth velocity model so that it preserves the long-wavelength velocity trends while removing short-wavelength reflectivity structure.
A boolean flag (default: on) that enables a reduced test-mode run for parameter verification. When enabled, the module processes a limited subset of the data so that geometry, grid spacing, and wavefield parameters can be quickly inspected before committing to a full production run. Disable this option when you are satisfied with the setup and are ready to compute the complete image.
A parameter group that defines the finite-difference grid spacing used in the wave equation solver. Both Delta X and Delta Z must be chosen to satisfy the numerical dispersion criterion: the grid spacing should be no larger than approximately one-tenth of the shortest wavelength present in the data (wavelength = velocity / maximum frequency). Using finer spacing improves accuracy but increases computation time and memory requirements significantly.
The horizontal grid spacing of the finite-difference computation grid, in metres (default: 5 m, minimum: 1 m). This spacing controls how finely the model is sampled in the lateral direction during wavefield propagation. Use a value that is small enough to avoid numerical dispersion at the highest frequency of interest. As a rule of thumb, Delta X should satisfy: Delta X <= V_min / (10 * Frequency), where V_min is the minimum velocity in the model and Frequency is the peak source frequency.
The vertical (depth) grid spacing of the finite-difference computation grid, in metres (default: 5 m, minimum: 1 m). This value is set independently of Delta X and controls depth-direction numerical accuracy. In practice, Delta Z is often set equal to Delta X for isotropic accuracy, but it can differ when the model has strongly anisotropic structure. The same anti-dispersion rule applies as for Delta X. Note that the depth sample interval of the input velocity gather is used separately to read velocity values from the model; Delta Z here is the computation grid step, not the input model step.
Specifies how the module identifies the free surface (Earth's surface) within the velocity model grid. Three options are available:
No topo — the surface is assumed to be flat and coincides with the top row of the velocity grid. Use this option for marine data or flat-terrain land data.
By air velocity (default) — the module detects grid cells whose velocity is equal to or below the specified Air velocity and treats those cells as the air/near-surface layer above the true Earth surface. This is the recommended choice for land data with variable topography when the velocity model contains explicit air-layer values.
By headers — the surface elevation is taken directly from the trace headers of the input velocity gather. Use this option when the model was built with explicit elevation information in the headers.
The velocity threshold used to identify air-layer grid cells when Detect topography is set to By air velocity, in m/s (default: 310 m/s). Any model grid cell with a velocity at or below this value is classified as air and placed above the modeled free surface. The default of 310 m/s is slightly below the speed of sound in air at sea level (~343 m/s), which is suitable when the velocity model uses a representative air-layer value. This parameter is hidden when Detect topography is set to any other option.
The spacing between consecutive source positions, expressed as a number of velocity-model grid cells (default: 5, minimum: 1). Sources are placed at every Nth lateral grid node of the velocity model, where N is this value. The actual distance between shots in metres equals Source step multiplied by Delta X. A larger step reduces the total number of shots and speeds up computation, but produces a coarser illumination of the subsurface. Set this to 1 to place a source at every grid node for maximum illumination density.
The maximum half-aperture of the receiver spread, in grid cells, measured on either side of each source position (default: 100, minimum: 1). For a source at grid position iSRC, receivers are placed from (iSRC - Receivers count) to (iSRC + Receivers count), clipped to the model boundary. The maximum offset in metres is therefore Receivers count multiplied by Delta X. Increase this value to capture longer-offset reflections and improve deep illumination; note that wider apertures also increase computation time proportionally.
The spacing between consecutive receiver positions within the spread, expressed as a number of grid cells (default: 1, minimum: 1). A value of 1 places a receiver at every grid node within the spread aperture. Increasing this value decimates the receiver density and reduces the number of traces per shot, which speeds up both forward propagation and back-propagation at the cost of reduced offset sampling. For most modeling purposes a value of 1 gives the best image quality.
The width of the PML (Perfectly Matched Layer) absorbing boundary zone, in grid cells, applied to all four sides of the computation grid (default: 20, minimum: 1). The PML boundary absorbs outgoing waves at the edges of the model and prevents artificial reflections from the grid boundaries that would contaminate the image. A larger padding zone provides stronger absorption but increases grid size and memory use. Values between 10 and 30 cells are typical; very low values (below 10) may produce visible edge artefacts in the output image.
The number of discrete angle bins used to accumulate the angle-domain common image gather (ADCIG) output (default: 20, minimum: 1). The total opening-angle range is divided uniformly into this many bins. The ADCIG is produced by decomposing the cross-correlation of the source and receiver wavefields into angular contributions at each depth point. More angle bins give a finer angular resolution in the gather, which is useful for amplitude-versus-angle (AVA) analysis and velocity QC, but increases memory consumption proportionally.
The peak (dominant) frequency of the Ricker wavelet source signature, in Hz (default: 20 Hz, minimum: 1 Hz). The Ricker wavelet is synthesized internally; no external wavelet file is required. A higher frequency produces finer temporal and spatial resolution in the synthetic shot records but demands a smaller grid spacing (Delta X, Delta Z) and a smaller Time step to avoid numerical dispersion and instability. Set this to match the dominant frequency of the real seismic data you are trying to model or the bandwidth of the target reflection event.
The temporal sampling interval of the finite-difference time-stepping scheme, in seconds (default: 0.004 s, minimum: 10 ns). This value must satisfy the Courant-Friedrichs-Lewy (CFL) stability condition: Time step <= min(Delta X, Delta Z) / (V_max * sqrt(2)), where V_max is the maximum velocity in the model. Violating the CFL condition causes the simulation to become numerically unstable and produce meaningless output. Smaller time steps increase accuracy and stability but multiply computation time by the ratio of the reduction. The total modeled record length equals Time step multiplied by Samples count.
The total number of time samples in each synthetic shot record (default: 100, minimum: 1). The modeled record length in seconds equals Samples count multiplied by Time step. Set this value so that the total record length is long enough to capture all significant reflections from the target depth. For example, to model a target at 3 km depth with an average velocity of 2000 m/s, the two-way travel time is approximately 3 s, requiring at least 3 / Time step samples. The boundary-saving buffer used for reverse-time back-propagation is allocated based on this value, so very large counts increase memory requirements.