Data conditioning tools are necessary to improve seismic interpretation, which is particularly important when examining legacy data. The tools typically adopted for the purpose are structure-oriented filtering and random noise suppression. While these methods can help improve structural or stratigraphic interpretations, they can also impact amplitude anomalies if they are not done with care, which can present problems when examining seismic data for direct hydrocarbon indicators. The examples presented herein demonstrate the advantages and disadvantages associated with these data conditioning tools.
While new migration algorithms and acquisition techniques are used to get the most accurate and precise representation of the subsurface, they come with an increase in complexity, time and cost. For example, multiwide azimuth acquisitions combined with bottom sea nodes and processed with the most advanced migration algorithms will have a sharper image than legacy data acquired 30 years ago. However, legacy data can be improved to understand geological systems or revisit reservoir areas better, not just for hydrocarbon production but also for CO2 storage or geothermal applications.
Herein, we demonstrate with a few examples two different conditioning tools: structure-oriented filtering and a random noise attenuation filter, or “fxy” from here on. We will show the effects of SOF and fxy separately for structural interpretation and a DHI objective. We showcase our comparison from the point of view of an interpreter, using concise examples from open-access data. These datasets have exciting features that can challenge the filtering tools, interpretation and seismic attributes. We additionally note filtering method differences in the calculation of similarity and DHI-related amplitude changes.
Comparing the Filtering Algorithms
Before describing the datasets, it is helpful to understand the nature of the filtering algorithms we are comparing.
The SOF is a process that smooths along structural planes as defined by a dip field. It is an effective method for removing incoherent noise and improving the continuity of events without smoothing across dipping planes. This process requires information on dip to guide the smoothing, which can be spectrally calculated to improve the results.
On the other hand, regularized non-stationary prediction filtering for fxy is a seismic noise attenuation filter that uses a combination of filters with a predefined range of possible dips to match the original signal in the time and space domains t-x-y to predict the conditioned trace. This filter is computationally inexpensive and sensitive to the number of traces included, yet practical against random noise and gentle with amplitude anomalies.
The first set of data examples are from three 3-D seismic data from offshore California (figure 1), which were acquired using narrow azimuth geometry and processed in 1984 and 1995. The bin size is 25-by-25 meters, dominant frequency 35 hertz with a vertical resolution between 25 and 40 meters. The data are prestack migrated (Kirchhoff) without any post-processing algorithms. The fourth dataset used is from the North Sea, offshore Netherlands, which was acquired in 1998 and is also time migrated.
SOF computation requires inline (IL) and crossline (XL) dip, and an average velocity – typically, 3,000 meters per second – works well for Tertiary sediments. A third parameter is an estimation of similarity (for example: chaos, similarity, coherency) as a measurement for discontinuity. For a faulted dataset, it is recommended to use the energy-ratio similarity.
The fxy filter we implemented consists of a 15-by-15 IL-XL window, and a 250-millisecond rolling window with 50-percent overlapping. The working frequency range we select is between 2 and 72 hertz. Using these values can help us evaluate the differences between SOF and fxy filtering.
The first two datasets have a well-documented gas production history, not to mention a complex structural style. Compressed anticlines and gas saturated sandstones are present throughout the area. On the Arguello seismic dataset (figure 2), a shallow anticline with a horizontal contact is imaged: a flat spot that indicates the gas-water contact. When we use SOF to enhance the structure and the fault north of the section, the edge detection increases, but, the flat spot is transformed into a stair-step reflector. As taking a different section between the input and the filtered versions is a typical practice for evaluating the performance of any filter, in figure 2c we see some leaked signal at the location of the flat spot, and thus our confidence is reduced. SOF has affected the DHI as the search window contains reflectors with contrasting dips, that is, horizontal reflectors surrounded by dipping layers.
A possible solution is to modify the filter search window to include a wider range of dipping values. But as DHIs are influenced by lithology, fluid content and seismic processing, interpreters need to be cautious when conditioning such data. Figure 2 suggests the need to analyze the unfiltered dataset for amplitude related analysis.
This dataset exhibits a prominent submarine canyon system running almost north-south, which is evident on the bathymetric profiles taken on the Pacific Ocean floor. We tested the effect of SOF and fxy on the sweetness attribute. Sweetness combines the amplitude envelope and instantaneous frequency attributes and is commonly used to distinguish sandstones from shales. When there is sufficient contrast between the sands and shales, high values of sweetness are usually interpreted as sweet sands.
In figure 3 we show horizon slices from the sweetness attribute computed on the input seismic data, after running SOF and fxy processes. We notice the edges of the channel system exhibiting braided channels and point bars on all three displays, but the sand distribution associated with higher values of sweetness are seen in figure 3b and c, that is, after running sweetness on the input seismic data with SOF and fxy respectively. However, SOF data shows a sharper distribution of the braided channels and their limits.
High values of sweetness are commonly interpreted as sweet sands. In figure 3 the channel system exhibits braided channels and point bars. In both conditioned cases – SOF and fxy – sweetness values are higher than the conditioned version. On the other hand, SOF shows a sharper distribution of the braided channels and its limits.
Santa Clara 3-D
The Santa Clara 3-D seismic volume (with a size of 180 square kilometers) exhibits multiple DHI and is therefore suitable for studying the effect of conditioning tools. A brightening of amplitude seen on top of the anticline, as well as a change of polarity, is suggestive of gas presence, which is not surprising, considering the paleovolcanic activity in the area. Besides, a large number of legacy well clusters in this area classified as gas producers again suggest that the amplitude anomalies are hydrocarbon related. Though the anomaly has structural conformance, the reflectors beneath the anomaly appear disordered and exhibit a sag, the latter being another indicator that suggests the presence of gas. The breaking of amplitude at the top of the structure could indicate that the trap integrity is compromised. The amplitude anomaly looks preserved and exhibits different dips to the surrounding reflectors.
A comparison of the seismic sections is shown in figure 4, from input data (4a), after SOF (4b), and their difference (4c). We notice that the flatness of the anomaly looks disordered on the difference section. Also, the SOF application appears to have impacted the strength of the amplitude anomaly. The flatness of the amplitude anomaly looks disordered and the amplitude is reduced.
The fxy filter is commonly used for suppression of random noise during processing of seismic data. Depending on the parametrization used, fxy can be made to have a gentle effect on amplitude anomalies. Figure 5 shows how the fxy filter removes smearing in and around a bright spot, particularly at the top. The difference section in figure 5c shows very little leaked signal. Interestingly, the amplitude levels before and after fxy filter application remain the same, which was not the case with SOF, where higher amplitude values were seen eliminated, perhaps indicating more effective conditioning.
F3 Seismic Volume
As seen in Figure 6, the application of SOF appears to enhance the main faults at the location of the yellow arrow as well as to the left of the red line. The continuity of the faults is seen to be improved in both the horizontal and vertical directions. The red line is the location of the time slice exhibited below from Sobel filtered and multispectral Sobel filter attribute volumes. SOF has been successfully employed on F3 data volume in other studies for machine learning applications, which suggests a significant benefit for using SOF for facies classification, and image segmentation problems like salt detection of fault prediction.
In conclusion, SOF conditioning is more effective for fault plane discontinuities as well as channel edges, and can be used iteratively, depending on the signal-to-noise ratio of the input seismic data. It can be a useful tool for removing migration and other artifacts. The amplitude response associated with bright spots appears to be changed somewhat, and deformations on the flat spots are also seen and should be used with care in such cases. Depending on the parametrization, fxy filter could be used to have milder effect on the amplitudes, especially those associated with bright spots. It should not be used iteratively, as it could influence the polarity of the anomalies.
Thus, the data conditioning tools discussed in this article can influence the amplitude response and interpreters should use them with this awareness. Of course, the correct filtering technique employed would depend on the objective, but should be applied with care, especially in the presence of DHIs.