Four-dimensional or time-lapse seismic is well established in the industry. The concept involves comparing the differences between two or more datasets shot with a time interval between them. The time interval can be as small as a few months up to several years. The objective of 4-D seismic is to reveal only changes in the reservoir due to production, including phase changes due to gas coming out of solution, compaction of the reservoir interval due to the emptying of the pore spaces and fluid movement over time. Depending on the production method, such fluid movement may be natural as pressure is released or stimulated from injector wells. Nonproduction-related differences are regarded as unwanted noise. Four-dimensional noise is highly detrimental as it can obscure the subtle changes associated with production in the reservoir. The sources of 4-D noise can be many and varied.

Seabed seismic

The typical progression of a 4-D campaign is to initially explore an area with towed streamer seismic acquisition followed by some exploratory wells. If successful, the project will move into a production phase. Prior to any significant production, the first dedicated 4-D seismic shoot will take place. This survey is referred to as the “baseline,” and a large percentage are increasingly likely to be seabed seismic surveys.

The reason for the switch to seabed is twofold. First, production infrastructure does not negatively impact the positioning of seafloor sensors when compared to towed sensors. Second, seabed sensors can be placed with greater accuracy. This is important because successive “monitor” surveys must attempt to mimic the baseline survey as closely as possible, including acquisition equipment, geometry and positioning. If every nuance of the baseline survey is reproduced exactly, there will be minimal 4-D noise and only changes due to production. However, this is rarely the case. In particular, repeating the sensor positions exactly can be challenging. This is one of the motives for using permanent reservoir monitoring programs with fixed receiver positions, but these tend to be prohibitively expensive in most instances.

Methods have evolved over the years to measure and quantify survey repeatability, including measuring the remnant background noise after surveys are differenced. The normal root mean square (NRMS) measurement is perhaps the most established metric. The NRMS equation (Figure 1a) is fundamentally a ratiobased equation expressed as a percentage with a dynamic range of 0% to 200%, with a lower value being better.

Typically, an NRMS measurement is made between survey a and survey b immediately after acquisition and at every step in the processing sequence. Ideally, the value is decreasing or at least not increasing.

Juan Cantillo with Total E&P (2010 SEG expanded abstracts) showed that the NRMS equation can be reformulated to be a function of cross-correlation and amplitude ratios (Figure 1b).

The reworking of the NRMS equation allows a graph (Figure 2) to be made whose axes are amplitude ratio (horizontal axis) and cross-correlation (vertical axis). The contours of the graph are NRMS value. Clearly there appears to be greater sensitivity in the vertical (cross-correlation) axis than the horizontal (amplitude ratios) axis. This is important because it backs up observations on controlled synthetic experiments.

Noise factors

To evaluate influencing 4-D noise factors, ION generated a synthetic geometry where a patch of 4-km by 4-km (2.5-mile by 2.5-mile) shots are fired into a single receiver station. The subsurface consists of 10 flat reflecting interfaces with a simple vertical velocity profile convolved with a 60-Hz Ricker wavelet. The model was laterally homogenous. Although fairly simplistic, this common receiver cube is adequate to show the relative sensitivity of the node to many factors.

A perturbation-free baseline receiver cube was made initially followed by multiple monitor surveys, each with a perturbation representative of what could actually happen in the field. For each perturbation from the baseline, the NRMS noise was measured.

First, the sensitivity to spatial positioning was examined. Spatial positioning can fall into two categories. The first is where the receiver station is some distance from the target but it and the target have known XY locations. The second is where there is some uncertainty in the XY locations.

If the receiver station was 2 m (6.5 ft) from the target and the positions known, the 4-D difference showed negligible NRMS noise. Even if the receiver station was 25 m (82 ft) off target with positions known, the 4-D difference showed negligible NRMS noise on the near offset and only 1.6% noise on the far offset. However, we must keep in mind that this is a laterally homogenous model. Any dip or lateral amplitude changes would see greater values.

Next, 2 m of inaccuracy were examined due to the coordinates being unknowingly in error. The near offset still shows negligible noise, but the far offset now reveals a 7.6% noise level. Increasing the error to 25 m reveals a massive 82.5% noise level. The reason for this is the significant travel time difference on the far offsets between the two surveys, resulting in a poor cross-correlation coefficient (supported by Figure 2).

Intuitively, any timing differences between the surveys (even if their amplitudes match perfectly) should also result in high NRMS noise levels. This is exactly what was observed. For example, a 1-ms node clock drift resulted in a 36.8% NRMS level on both near and far offsets. A 2-ms drift (one seismic sample) resulted in a massive 73.7% NRMS level.

Clearly any amount of time difference between surveys has a huge impact on noise levels. Besides clock drift, the elevation level of the shooting vessel during the cycles of tides can result in travel time differences. An uncorrected 1.5-m (5-ft) tide height difference results in 1-ms time shift or 36.8% NRMS noise. Unfortunately, a simple tidal static correction is not enough. Ghosts and multiples will be embedded in the data with different periods. These must be independently removed in processing.

Similarly, a phase difference between two surveys can result in poor correlation. Geophones tend to have low frequency distortion (roll-off). If these differ between surveys by 10 degrees, it results in 7.6% NRMS noise levels, while a 30-degree difference results in 22.5% NRMS noise levels.

Other noise

Next, the study looked at nonsource-generated noise differences between surveys. In a production environment there are likely many sources of unwanted noise such as those from drilling, shipping and pipeline activity. Adding widespread 10% random noise results in 10.2% NRMS noise.

Finally, the noise associated with simultaneous shooting was assessed. Driven by the quest for faster and cheaper acquisition, there has been a trend to fire sources simultaneously and “de-blend” them in processing. Untreated blended data result in noise levels of 60% to 100%. After de-blending this can reduce the noise levels to 0% to 35%.

In summary, 4-D noise levels are very sensitive to the quality of correlation between data volumes. In that regard, any associated time or phase differences should be minimized as much as possible.

Unknown errors in receiver positioning can result in a correlation error, which increases with offset. To minimize this, acoustic pinging for every receiver station location is recommended. Preferably, the transducer should be built into the sensor unit. Externally strapped-on transducers might unbalance the center of gravity or affect coupling. Moreover, additional positioning quality control via first-arrival analysis is recommended.

Clock drift must be completely eliminated. Even the more accurate chip-scale atomic clocks should include additional residual corrections. GPS satellite synchronization between all receivers would be preferred (as used in cable-based systems). Phase difference between recording systems must be carefully removed. Tidal static corrections should be accurate and followed with quality de-ghosting and de-multiple.

Finally, if simultaneous shooting is planned, it should include a thorough de-blending process.