Co-Location Of Surveys In Space And Time For Improved Accuracy And Sensitivity In 4D Marine Surveys

Time lapse, or 4D, seismic surveys are designed to monitor changes in the seismic response of hydrocarbon<br>reservoirs caused by production. In many reservoirs, the acoustic impedance changes induced by production<br>are small and therefore we are looking for weak 4D signals. This means that 4D surveying requires a high<br>degree of accuracy in order to detect a reliable 4D signal. In marine 4D surveys, variations in geometry caused<br>by currents, variations in wave height and variations within the water column all combine to reduce the<br>repeatability of the recording. Successful 4D processing must take account of these factors<br>Interpretation of 4D results has often been done on difference sections obtained by simply taking the difference<br>between two vintages of data previously processed for 3D purposes. Variations in the individual survey results<br>will be caused by variations in recording equipment and geometry, variations in feathering and infill, variations<br>within the water column and variations in processing. In order to compensate for these differences, matching<br>filters are applied to the datasets. In this method, it is assumed that matching the overburden will compensate<br>for all of the above factors in a single filter (perhaps a single filter per trace) and thus reveal the true change at<br>the reservoir level. Although this approach implies that the biggest difference between the two surveys will be<br>outside the matching design window, it does not mean that these changes are only caused by reservoir<br>production. Indeed, the difference between the datasets is often found to increase everywhere outside the<br>design window. For example, data shot over the Oseberg field in 1989 and 1991 were normalised by applying<br>scalars derived in a 150ms window around the water bottom for each trace. Equalising the amplitudes around<br>the water bottom INCREASED the difference energy over a larger window from 0s to 3s, illustrating the dangers<br>inherent in simply matching different datasets to optimise the 4D signal.<br>Since the differences between two datasets are caused by a number of factors, it is likely to be better to address<br>each problem separately rather than attempting to do so with a single matching filter. The first, obvious step is<br>to process both datasets simultaneously for 4D purposes. This eliminates differences in processing methods<br>and algorithms and also testing of whether or not particular processing steps improve the 4D difference. Within<br>this processing, variations in positioning (in space and time) should be addressed separately from, say, wavelet<br>changes. Finally, matching filters may still be appropriate, but now they are addressing only residual differences<br>in the seismic system.<br>To examine how sensitive taking the difference between two datasets can be, consider two 40 Hz Ricker<br>wavelets separated by a 1ms static shift. Figure 1 shows the two wavelets and the difference between them. The<br>peak difference is about 25% of the peak value of each wavelet. Increasing the static shift to 4ms causes the<br>peak value and the RMS of the difference to approximately equal the equivalent values for the individual<br>wavelets. Static shifts greater than this cause the difference to have peak and RMS values up to twice that of the<br>individual wavelets. This demonstrates that we will have to account for small timing and positioning shifts<br>between vintages if we are to reduce the background difference ‘noise’ and detect true changes caused by<br>hydrocarbon production. Moreover, many reservoirs require sensitivities of less than 25% in order for the<br>production induced seismic changes to be detectable i.e accuracy to better than 1ms for the above synthetic<br>wavelets.