Event Resolution Imaging™

Event Resolution Imaging™ (ERI) confronts the challenging task of mathematically resolving the features in 3D seismic data which statistically distinguish the characteristics of high producing wells from those of dry holes. The basic question which Event Resolution Imaging asks is, “What pattern of information contained in the 3D seismic data best separates good high producing wells from bad low producing wells for a particular zone of interest?”.

Seismic Insight uses Event Resolution Imaging to combine production information from producing wells with 3D seismic information to generate a very useful new type of information which we call a 3D Seismic Contrast Volume™.

We use Event Resolution Imaging to find, delineate and high grade the sweet spots, and to distinguish them from areas that are unlikely to produce economic wells. Event Resolution Imaging allows the sheep (example trace segments around good wells) and goats (trace segments surrounding dry holes) to separate and self distribute however they will across a given separation landscape without regard to how close the sheep and the goats come to the separation fence (separation threshold) which is the statistical threshold.

We let the sheep and goats self distribute to plus and minus infinity and get as far away from the fence as they wish--even off to infinity. In this way, our separation learning algorithms apply no artificial statistical mixing pressures on the "animals" (trace segments) as they distribute.

The separation landscapes are spanned by mathematical characteristics or attributes such as frequencies, latencies, amplitudes and phase relationships that are mathematically found in segments of 3D seismic data which are centered on, or in the vicinity of, potential pay zones of interest.

Our Event Resolution Imaging learning algorithms do NOT use neural networks. They do not get stuck in local minima. They do not use gradient descent, and no type of least squared minimization algorithm is used. We do not use linear regression, multiple regression or any type of discriminate analysis to minimize error cost functions, but rather we use a statistical sorting and counting method which simply steps through multiple attribute separation landscapes to find where the sheep (good wells) separate best from the goats (dry holes).

Simply stated, Event Resolution Imaging™ involves the processes outlined in the figure to your right.  As outlined in the yellow Learning box, a few sample traces are analyzed by the software to learn the seismic profile of a desirable result -- along with the contrast of an undesirable result.  Example traces are linked with their appropriate labels (such as "good/high production" and "bad/dry hole"), then mathematically analyzed, compared, and contrasted to develop an "Interpretation Key." This Interpretation Key can be used later for analysis of "raw data." In the Classification process (green box), raw data traces are classified and assigned a rating value -- either a desirable prospect or a risk rejection.  Once the classifications are complete, the Validation procedures (white box, partly in green) are implemented to quantify the confidence level of the results.  True labels are used to compare with predictive results for accuracy.  If classification results can successfully validate additional known well sites (good and bad), confidence rises -- especially if this can be accomplished for wells in outlying areas.  And if the preponderance of evidence is positive, risk is reduced and drilling confidence increases.

First we have the software examine, trace by trace, the seismic properties of some known state A condition (perhaps a known good well) and an opposing state B condition (perhaps a known bad well).  After the software has "learned" from those states, it develops a profile (mathematical encoding of seismic wave feature patterns) we call an Interpretation Key, represented in the image below. This could be thought of as asking each new trace:  "Are you more similar to State A (the good trace examples) or to State B (the bad trace examples).  If the trace is VERY similar, it is given an "high + score".  If slightly similar, it gets a moderately positive score.  If the trace is dissimilar, it gets a prorated negative score.  A collection of traces within a given 3D, such as traces in the vicinity of various well locations, are rated and plotted together in an identified group.

We have used a simple +1 to -1 system to put everything on a "relative degree of similarity" basis.  As each trace is classified, it's assigned a + or - similarity index value as shown below:

• A full value of +1 represents a high degree of similarity to State A traces, while a value of +.3, for example, represents "somewhat" similar.
• A full value of -1 represents great dissimilarity from State A traces with high simmilarity to State B traces.

Once these values associated with individual traces are grouped, they begin to suggest a definite separation.  Some locations give a very strong indication which we like to think of as a "pool of similarity".

As the groups take shape, there will be a "separation of the sheep from the goats".  If we are comparing well locations, the plot will usually indicate a pronounced agreement of whether the well was productive or not.  The confidence level is highest when all or nearly all wells in a producing zone -- with a proved productive or unproductive status (but not used in Learning) -- begin to manifest agreement with the Interpretation Key.

Thus the application of the Interpretation Key to specific map areas in seismic horizons of interest, produces collective groups which can be plotted together.  We call the resulting visual a "Key Activation Plot," shown below.

The green colored groups of dots in the plot represent traces in the vicinity of drilled wells that were proven to be productive oil producers.  The blue dots represent traces around wells that were proven to be dry holes.  In this hypothetical example, the first group of traces (green dots around wells W1, W2, and W3) are used as "Learn State A" traces.  The second group of traces (blue dots around wells W4, and W5) are used as "Learn State B" traces.  Taken together, these two groups of traces are the only traces used for learning and the generation of the Interpretation Key.  The remaining traces (shown in purple in this example) are used for Validation and assessment of Confidence.  Suppose wells W6 through W10 are also drilled wells that have been proven to be either productive or non-productive.  Traces around these wells can be Classified with the Interpretation Key just generated and assigned their relative values between plus 1 and minus 1.  If most of the traces corresponding to oil producing wells rank positive (close to +1), while most of the traces corresponding to dry wells rank negative (close to -1), then we have validated an Interpretation Key and found that it has high confidence or predictive value.  We can know with a reasonable degree of certainty that as we classify additional traces within the 3D, we will have a map that discriminates between areas of high and low probability of hydrocarbon presence.

The plot below is an example Key Activation Plot generated by applying an Interpretation Key to traces surrounding 12 wells in Southwest Texas.  At the time the Interpretation Key was created, production information was available for only the first 6 wells (groups 1-6 in the columns below):

• Wells 1 and 2 (green) were known good wells.
• Wells 3 and 4 (yellow) were known bad wells.
• Wells 5 and 6 (brown) were "wet."
• Wells 7 - 12 (rose) were unknown "testing" wells, set aside to validate the Interpretation Key at the conclusion of the analysis.

From the analysis, it was concluded that of the unknown wells, numbers 8 and 11 would have the highest probability of being productive wells.   Assuming the Interpretation Key had already been Validated, these would be the first places we would drill (drilling well 11 first, as it has the most high positive trace rankings).  We'd like to drill Well #8 next, but only after additional analysis after the results came in from drilling Well #11.   The remaining unknown wells would be expected to be poor areas or dry holes.

Since wells 7-12 had been set aside to validate the Interpretation Key, we compared the results from the plotted classifications with the known, proved status of each well.   Each well absolutely validated the projected outcome:

• 7, 9, 10, and 12 were not productive wells
• 8 and 11 were good wells

This Key was thus shown to be valid and consistent with proved well control with high statistical confidence.   It could now be expected that as we applied the Interpretation Key to other areas within the 3D (to create a Contrast Volume™), we would be unveiling a map that would have high predictive value for areas with and without hydrocarbon shows.

Continuing from the example above, we applied the Interpretation Key over the entire 3D seismic volume to create the Contrast Volume™ (CV).  We have included two different images below to compare the Contrast Volume with the original Amplitude map. The first image is the Amplitude map, which would have been the conventional view used to identify potential drilling "bright spots." The second image (CV) is a map resulting from assigning the color palette to the -1 < 0 (dark to light blues) through 0 < +1 (green, orange yellow, and white) value scale we introduced earlier.

These maps follow a single pay horizon corresponding to the zone where the analyzed wells had been perforated.   Locations of existing well sites have been plotted on both maps. The white "open plus signs" are existing dry holes (there are 8) and the 4 black dots are successful oil wells.  Note that the CV shows bright spots in different locations from the Amplitude map, thus telling a different story.

Just below is a type of seismic display called an "Amplitude Map."  Using this map to make the choice of possible drilling sites, you can see why the well locations might have been chosen where they were. (Click maps for enlargements)

Each pixel (trace location) on the CV (below) answers the question "How similar am I to the features of the productive trace samples versus the unproductive trace samples?" (Click maps for enlargements)