Bridging to a Big Data World
As databases grow, they can easily become overwhelming. Aspects of this problem include the sheer volume of computed attributes, problems with data access, and problems with data visualization. If the data and access to the data are not managed reasonably, the interpreter is forced to focus on functional issues rather than on the interpretation questions. Painless data access, presentation, and visualization are just as important to the act of interpretation, as the content itself.
Condition the Data: Make sure that the data has fidelity and that it is decomposed to a set of attributes that in combination, provide insight that is greater than the sum of the parts.
Beware the Hype: Remember that in an AI assisted world, the interpretation “building blocks” required to get the job done will remain the same. It’s just that AI will make it easier to consider the breadth of information and dimensions.
Stage Investigations: Approach the interpretation questions via reconnaissance scoping before embarking on targeted and detailed analysis and interrogation. During initial stages, detection and awareness of anomalies is more important than resolution of details. Be open to surprises.
Facilitate Interaction with Data and with the Extended Team: Within any team, there is often a wide range of interpretation experience, knowledge and skills. Not everyone will always be equally proficient with seismic visualization and the available tools. Don’t overwhelm those who are not experts in seismic interpretation or seismic processing. Present the information in a way that facilitates interaction. The easier it is to navigate and investigate, the more likely important features will be identified and localized. Remember that integrating and calibrating early and often, reduces dead-ends and cycle-time.
At OpenGeoSolutions, one of the approaches that we are taking to make the transition more manageable, is to build (for each project area that we work on) a comprehensive library of pre-rendered images that covers the dimensionality of the data, and that can be accessed via a browser. This means creating literally millions of images that expose the full dimensionality of the information that is available. By generating both plan-view and cross-section images at the sample rate of the incoming data, we incorporate the element of motion into the visualization. By using RGB blending, semi-transparent overlays, etc…, we combine attribute/dimensional characteristics in a way that makes the whole greater than the sum of the parts. The goal is not to subjectively guide the interpreter along a pre-identified, rigid, hard-wired path through the data and dimensions. The interpreter will never want to examine all of the millions of images that are created. Instead, the goal is to facilitate the examination of the imaging that is available within the data, and to provide the freedom and flexibility to identify, track, and home-in on any/all interesting features. If an interesting anomaly is spotted along the way, it can be followed along the many available dimensions. With the conditioning/preparation that is required to generate such a library, the by-product is a comprehensive set of decomposed characteristics that can be leveraged by AI/ML.