AccuVEL: Migration velocity analysis (MVA) is a procedure to iteratively improve migration velocity models and to construct optimally focused structural images. The analysis starts from Kirchhoff prestack depth migration (MVA product) or common azimuth prestack depth migration (WMVA product) with an initial depth-interval velocity model, which can be derived from sonic logs, CDP velocity analysis, or a constant velocity estimate.

The initial velocity model is usually not accurate, therefore the migrated events are usually not flat on common image gathers. Depending upon the complexity of the investigated geologic structures, an appropriate updating method can be used to correct the residual curvature and to produce an improved velocity model. The associated computational cost is typically proportional to the sophistication and capabilities of the algorithm used. In general, the vertical update based on the Dix method is fast and well suited for flat or slightly dipping layers, while the more sophisticated tomographic update which requires extensive ray tracing for accurate global back projection is best for resolving complex features or for fine tuning velocity models.

Detailed Description
An important step in prestack depth migration is the velocity model building process. The most accurate way of reconstructing velocity models is prestack migration velocity analysis (MVA) based on common image gathers. The CIGs contain redundant structural information that is used to correct the initial velocity model. Furthermore, velocity model building is most accurate if updates are based on prestack gathers generated from the same imaging algorithm that is used for the final imaging step. It is therefore imperative that wave-equation migration imaging is intimately related to wave-equation MVA. For this reason, we perform MVA using angle-domain common image gathers (ACIG) generated by COMAZ. These output image gathers are sorted in either reflection angle or ray parameter.

From the CIGs, we extract residual velocity information by scanning over angle or ray parameter. The scanning formula takes into account the relationships among migration depth, migration velocity, residual velocity, and ray parameter/angle. These scans are typically displayed as semblance panels with peaks that correspond to residual RMS values that flatten the events in the CIGs.

Once the semblance panels are generated, several methods can be used to backproject the residuals to the overburden medium and update the initial velocity model. The semblance peaks can be used directly and automatically to update the velocity, or horizons can be picked and semblances tied to these horizons. The residual semblances can then be used to update the velocity model along vertical rays, along normal rays, or tomographically.

The simplest method is vertical updating. The process can be automated, so that it does not require user intervention or picking of geologic horizons. Given a semblance spectrum of residual velocity, the automatic semblance picking method perturbs the initial velocity profile to search for a new velocity profile that typically follows peak semblance values at every depth. The output from this method is an updated interval velocity profile. Used alone, the method is best suited for preliminary velocity analysis in areas with slightly dipping layers. The advantage of the automated vertical update lies in its efficiency of processing large volume of ACIGs without human intervention. The main drawback is limited noise discrimination and limited geological constraint. A more stable and accurate approach is to tie residuals to geologic horizons and perform horizon-based updates. Geologic horizons usually exhibit strong reflectivity and good lateral continuity and can be picked from stacked image volumes.

Combining vertical update and horizon constraints results in a horizon-based vertical update. In this method, the Dix inversion formula is applied to RMS residual velocity for estimation of interval residual velocity. This algorithm is computationally efficient and involves lateral smoothing of horizon-based residual velocities for each horizon. The update velocity value is the sum of the background and the calculated residual velocity at the grid point of interest. One major limitation of vertical update lies in its simplification of error back-projection methods. In the presence of steep dips, vertical back-projection can wrongfully misplace residual velocity values. With horizon-based normal-ray updating, perturbations occurring along a normal ray to the reflector are taken into account. A specular ray normal to the dipping reflection surface is used to approximate the wave propagation path. The improved accuracy of normal-ray backprojection comes with a slight increase in computation workload for ray tracing.

An even more accurate and computationally intensive method is horizon-based tomographic updating. In this method, a fan of rays with correct wave propagation trajectories is used to backproject residual velocities to the locations where the errors originated. Each tube of rays from an analysis ACIG point illuminates part of the overburden, and several overlapping ray tubes can be used to reconstruct the overburden velocity distribution in a tomographic manner. The method consists of two basic components: forward modeling and tomographic reconstruction. In the forward modeling, ray paths are determined through ray tracing from every analysis CIG point and residual moveout as depth deviation in ACIGs is converted to residual traveltime. The influence of velocity errors on imaged reflector geometry is also taken into account. In the tomographic reconstruction part, ray paths, computed residual traveltimes, and the unknown residual slowness field comprise a linear optimization system, which is solved by the method of conjugate gradients.