The 2026.3 release spans performance, functionality, and platform improvements across seismic QC, rock physics, SDC, and visualization workflows. Several updates carry direct relevance to Rock Physics work in particular - a new wavelet conditioning method that operates in the complex domain, faster classification exports, a new spectral attribute type, and a quantitative well-tie QC measure built into the seismic track. The thread running through them is the same: better-conditioned inputs and tighter, more directed QC, with less time lost to steps that used to sit on the critical path. Here is a closer look at each update and how to put it to work.
The Multi-well Wavelet tool now offers Complex Wavelet Averaging as a new option for averaging and conditioning a set of extracted wavelets. It treats each frequency in the wavelet as what it actually is — a single complex value with coupled amplitude and phase — and solves the average directly in the complex domain, with a frequency-dependent uncertainty model behind every bin.
The mechanism is worth understanding because it explains the benefits. At each frequency, amplitude and phase are two coordinates of one complex number, and their uncertainties are coupled: amplitude error acts radially along the complex vector, phase error acts tangentially, and because both directions are rotated by the wavelet's own phase, the true uncertainty at each bin is a tilted ellipse in the real–imaginary plane. Complex Wavelet Averaging propagates each input wavelet's amplitude and phase error bars into exactly that ellipse, then computes a generalized least-squares (precision-weighted) average of the complex values across wells: wavelets with tighter, more reliable uncertainty at a given frequency carry more weight, less certain ones carry less. Phase circularity is handled naturally, because the averaging happens in complex space rather than on a wrapped phase angle. The output is a representative wavelet together with a defensible amplitude and phase uncertainty at every frequency.\
The standout benefit is bin-to-bin stability across the spectrum. The method scores each frequency bin independently — on phase coherence across wavelets, amplitude consistency, how many wavelets actually contribute, whether one wavelet dominates, and how locally smooth the spectrum is — and conditions the result accordingly. In reliable regions it stays close to the raw data-driven estimate; in noisy or low-energy regions it leans toward a smoothed trend; and where the high-frequency tail collapses it tapers gracefully rather than letting noise propagate into the final wavelet. The conditioned result is therefore smooth where the data don't support structure and faithful where they do. And where the formal error bars and the actual inter-well disagreement are inconsistent — the data scattering more than the supplied errors claim — the covariance is inflated accordingly, so the reported uncertainty stays honest rather than overconfident.
This matters most in two situations. The first is when wavelets extracted across wells show meaningful phase variability — a common outcome in fields with variable overburden, differing acquisition geometry across vintages, or wells drilled at different periods. The complex approach captures that variability faithfully and produces a representative wavelet without artificial smoothing of the phase or the wrap artefacts that arise when phase is averaged directly. The second is sparse datasets, where the statistical basis for averaging is limited. Here the per-bin reliability scoring and conditioning do the heavy lifting, stabilising the result from a small number of inputs in a principled way rather than relying on noise cancelling out across a thin average.
To try it, open the Multi-well Wavelet tool and select Complex Wavelet Averaging from the averaging method options. The conditioning behaviour is controlled by a profile — Smoothest, Balanced, or Data-Honouring — with Balanced as the default; the other two are worth reaching for when you want either a more aggressively stabilised wavelet or one that stays as close as possible to the raw estimates. A useful first pass is to review the phase spectra and the match between the resulting synthetic and the seismic at each well, particularly on projects spanning different areas or vintages where phase consistency is not guaranteed.
Bayesian Classification exports in RD2D and RD3D are now significantly faster in 2026.3 - approximately 2x faster in PDF mode and 2.5x faster in DTA mode - with no change to the underlying outputs. This is a shared improvement across Rock Physics and Reservoir Characterisation licenses, developed in direct response to client requests, and its practical effect depends on how frequently classification is run as part of your project workflow.
For iterative Rock Physics work - testing different prior distributions, evaluating the sensitivity of classification results to fluid substitution outputs, or running classification across multiple wells as part of a QC pass - the shorter export time means more scenarios fit within a working session. Rather than treating a classification export as a step that requires dedicated time to run, faster exports make it more practical to use classification results as a routine check throughout the project, not just at the end. For projects involving large classification volumes or iterative scenario testing, this translates directly into reduced turnaround time. If you have been deferring classification to the final stages of a Rock Physics project because of the time cost, 2026.3 is a good point to reconsider where it fits in your workflow.
A new Spectral Attribute type is available across RD2D, RD3D, and RD4D, displaying the per-trace frequency spectra of a given seismic quantity between two user-defined frequency bounds. It is created and managed in the same way as other attribute types, bringing spectral QC into the standard attribute workflow with no separate analysis step required.
For Rock Physics work, spectral analysis of the seismic input is a natural part of evaluating what frequency content is available for wavelet extraction and well-tie. A useful approach is to create a Spectral Attribute over the frequency band of your wavelet and review it before committing to a single wavelet for a project. Significant spatial variation in the attribute, particularly if it correlates with acquisition footprint or with the locations of your wells, is a signal that the seismic spectral content is not consistent across the area, and that a single-wavelet assumption may be a poor fit. In that case, spatially varying wavelets or zone-based extraction strategies are worth considering before the well-tie is finalised.
The attribute is also useful as a quick QC of frequency content after any processing step that modifies the spectrum, such as spectral balancing or structure-oriented filtering. Comparing the attribute before and after gives an immediate read on where the processing has changed the frequency character and whether that change is spatially consistent.
The seismic track in RokDoc 2026.3 now supports a sliding correlation window that shows the running correlation between the selected seismic and synthetic directly in the track. This provides a quantitative well-tie QC measure without requiring a separate workflow step or switching between displays — addressing a gap that has been raised consistently by clients and sharpening the well-tie QC experience for everyone working with synthetics.
During an iterative well-tie, the correlation window is most useful when kept active while adjusting the wavelet or the time-depth relationship. A localised drop in the running correlation at a specific depth is often a more reliable indicator of where the synthetic and seismic diverge than visual inspection alone — particularly in sections where amplitude or frequency differences are subtle. When a drop is identified, it narrows the diagnostic focus: the issue is likely the time-depth relationship in that interval, the wavelet phase or amplitude in that frequency range, or a genuine mismatch between the elastic log and the seismic at that depth. Using the correlation window as an ongoing guide during the well-tie turns what is often an iterative trial-and-error process into a more directed one.
Several Platform license updates to the User-Defined Plot are relevant to Rock Physics interpretation workflows, and reflect a continued effort to keep the plot in step with how petrophysicists and rock physicists actually use it.
Interactive polygon selection in the multi-Crossplotter allows users to draw polygons on the crossplot to highlight and filter data. For Rock Physics crossplot work - fluid and lithology discrimination, AVO classification, rock physics template fitting - a practical approach is to define your expected cluster regions on a reference crossplot using polygons, then apply those polygons to filtered subsets of the data to test whether the separation holds across different depth intervals, wells, or saturation conditions. The polygon acts as a reproducible, documented region that can be revisited and adjusted as the rock physics model evolves, rather than requiring a fresh visual assessment each time.
Core data can now be co-rendered alongside continuous log data in the User-Defined Plot. For Rock Physics projects where core measurements inform the elastic model — plug measurements of velocity and density, saturations, mineralogy from XRD — displaying these against the log response at the same depth helps identify where the log is a reliable proxy for core and where there is scatter that needs to be understood before the model is built. To set this up, add core measurements as a track in the User-Defined Plot on a shared depth axis with the log curves of interest.
Beyond these, UDP axes now support configurable tick mark density, from Minimal through Very Dense, with a minor tick toggle and persistence in layout templates, and users can designate a directory of custom images as a source for lithology patterns, available in both global and project settings. All of these features are part of the Platform license and available to all users.
The Project Viewer now sorts Volumes under Surveys by Log Type, with Pre-stack Volumes moved into a dedicated P Stack folder - completing the folder structure alignment across Horizons and Volumes that was started in the previous release. Multiple subdirectories from within the same parent directory can now be added to the RokDoc Directories dialog in a single Ctrl+click operation, eliminating the need to add each entry individually.
Several of the features in this release would support a dedicated tutorial or technical post. Complex Wavelet Averaging applied to a specific dataset type - phase-variable wells, sparse acquisition, or vintage differences - would make a practical follow-up. A step-by-step guide to building spectral QC into a well-tie workflow using the Spectral Attribute is another candidate. If there is a topic from the 2026.3 updates you would like covered in more detail, or a how-to request from your own project work, let us know in the comments or get in touch directly. Your suggestions shape what we write next.
We look forward to continuing to hear your feedback for future RokDoc releases!