Value creation by a long-term time-lapse seismic processing approach on the Heidrun field

Transcription

1 Value creation by a long-term time-lapse seismic processing approach on the Heidrun field Daniel Fischer, 1 Nils Sørenes, 1 Emanuel Teichmann, 1 Hanna Blekastad, 1 Anita S. Moen, 1 Ivar H. Sollie 1 and Patrick Smith 2* show the value created by time-lapse seismic processing on the Heidrun field and how Statoil has used the technique to cut the timeframe on its 4D seismic. S tatoil has, over the past 15 years, routinely used timelapse (4D) seismic surveys to help maximize recovery and production efficiency of its hydrocarbon reserves. During this time the company has built an extensive portfolio of fields that are covered by multiple vintages of time-lapse seismic data (Figure 1), and a high level of activity is expected to continue for the foreseeable future. Seismic data from the time-lapse projects is integrated into routine reservoir management processes and tends to be used continuously throughout the life of a field. Reservoir engineers therefore need the information from new monitor surveys as soon as possible after data acquisition. However, the conventional approach to time-lapse seismic processing involves co-processing of data from all the available surveys a task that becomes progressively more resource-intensive and time-consuming as the number of vintages increases. Lengthy processing and analysis can result in important reservoir decisions having to be made before the full information from a new monitor survey becomes available, reducing the value of the newly acquired data. While fast-track products using a simplified flow are sometimes useful, they usually provide less reliable information. Statoil is mitigating these challenges by using long-term contracts for time-lapse seismic data processing. This article describes this strategy and illustrates it by reference to the time-lapse seismic campaign on the Heidrun field, offshore Norway. Figure 1 The Statoil portfolio of multi-vintage time-lapse seismic monitoring campaigns. 1 Statoil ASA. 2 WesternGeco. * Corresponding author, psmith11@slb.com 2013 EAGE 93

2 first break volume 31, October 2013 Processing challenges Time-lapse seismic data processing is demanding. The acquired datasets contain inherent variability that can obscure the desired time-lapse signal. Compensating this variability to the necessary degree of accuracy while preserving genuine changes between datasets can be time-consuming, and processing turnarounds of up to six months after completion of data acquisition are commonplace. A particular time-lapse seismic processing flow applied to a new monitor survey may be incompatible with the results produced on previous vintages if hardware and/or software versions have changed in the interim. This is one reason why, when a new survey is acquired, it has been common practice to reprocess all available vintages in order to accurately estimate and compensate variability in a manner that ensures consistency between the different datasets. This adds additional cost requiring the operator to divert resources into activities such as evaluation of the new results and updating interpretations. It also increases the number of processed cubes that must be stored and kept track of. Figure 2 shows the number of datasets generated by a multi-year time-lapse processing campaign on a Norwegian North Sea field. The baseline survey was acquired in 1993 and the first monitor survey in The two surveys were co-processed, generating about eight prime datasets (i.e., those created by the seismic processing contractor), and four 4D datasets (i.e., those created during interpretation). A new monitor survey was acquired in 2007, and all three surveys were reprocessed, generating some 38 datasets. Additional monitor surveys were acquired in 2009 and 2011, with all surveys being reprocessed at each acquisition. The 2011 processing created more than 70 seismic cubes, all of which required quality control (QC) processes and evaluation. As shown in the white bars in Figure 2, it was predicted that continuation of this approach would result in the generation of almost 100 cubes in 2013 and 120 in Instead, a long term contract was awarded in 2011 under which all of the existing datasets were co-processed, and then a new approach was adopted for processing subsequent monitor surveys that would result in the creation of only about 40 datasets at each new acquisition. An alternative processing strategy Statoil is addressing the challenges described above by implementing long-term time-lapse seismic data-processing contracts that incorporate the following features: Firstly, identical processing flows are used from one vintage to the next. The flows are designed such that the new monitor survey can be processed independently of the previous surveys. Hardware and software changes between vintages are addressed by regression testing (described below). Secondly, testing to improve the data processing flow, and any necessary reprocessing, is performed in the periods between acquisition of monitor surveys so that the new flow and reprocessed data are available and interpreted by the time of the next acquisition. Thirdly, the long-term nature of the contracts enables the data processing contractor to maintain, in partnership with Statoil, valuable knowledge and experience of the datasets from one survey to the next. These concepts are probably familiar to those involved in the processing of time-lapse seismic data from permanently emplaced systems (e.g., Van Gestel et al., 2008). We are now essentially applying an equivalent strategy to bring marine streamer time-lapse data processing turnarounds down to a similar order of magnitude to those achieved with permanent systems. Use of the same acquisition contractor and identical configurations for each survey simplifies the processing by eliminating processing steps required to compensate differences in acquisition. While it is not obligatory to use the same contractor for acquisition and processing, this approach can Figure 2 The number of datasets generated by a multi-year time-lapse processing campaign on a field in the Norwegian North Sea. The white bars indicate the predicted number of 3D cubes that would have been generated in 2013 and 2015 if the alternative processing strategy had not been adopted EAGE

3 Figure 3 Inline from a time-lapse seismic difference cube between the baseline and monitor survey created during acquisition of the monitor survey. Figure 4 Schematic flow chart showing how sailline by sail-line time shifts can be derived without co-processing all available time-lapse vintages. provide additional benefits for turnaround and onboard quality control in the following ways: Firstly, onboard processing (OBP) of the first acquired seismic lines through the early stages of the time-lapse seismic processing flow enables highquality time-lapse comparisons with the existing data, helping to ensure at an early stage that the acquisition and processing are truly identical to the previous datasets. Secondly, a high quality 3D stack cube can be built as the survey progresses so that time-lapse QC can be performed at intervals during acquisition (Figure 3). Acceptance criteria for the acquired lines can then be based on true time-lapse seismic comparisons (Osdal and Alsos, 2010), and strategies put in place at an early stage to address any problems encountered. Thirdly, a significant part of the processing flow and QC can be performed on the seismic vessel during acquisition, with the partially processed data being shipped to the data processing centre for completion. This can substantially reduce data processing turnaround times. Typically, processing steps performed on the vessel are applied to sail-line organised data, although strategies exist for performing 3D demultiple and imaging where necessary. Data perturbations for the newly acquired survey are corrected by referring to the previously processed datasets. Figure 4 shows schematically how this may be done for the derivation of sail-line by sail-line time shifts. Initially, timing measurements from all available surveys are passed to a routine that simultaneously estimates sail-line by sail-line time shifts for each survey. These are applied to the datasets, which are then combined to create a reference cube. Each sail-line of each survey is then compared with the reference cube to re-derive the timing corrections, which should be essentially the same as those originally estimated. The reference cube is then archived and used to derive sail-line by sail-line time shifts for subsequent surveys. Analogous approaches may be used for amplitude corrections and other perturbations. Reference datasets may also be created for time-lapse binning. The results will not be identical to a simultaneous time-lapse binning of all surveys, but, in practice, the differences are usually minor EAGE 95

4 first break volume 31, October 2013 If the hardware and software systems are not frozen between acquisitions, regression testing is used to demonstrate the compatibility of the current configuration with those used previously. This involves the archiving of flows and example input and output datasets for every step in the seismic processing sequence. A few months before the next acquisition, the processing flows are applied to the archived input data using the current hardware and software versions. The new outputs are compared with the previous ones by visual inspection of differences and by computation of NRMS difference attributes (Kragh and Christie 2001). If the NRMS difference exceeds a pre-defined threshold, remedial action is taken, usually by reverting to a previous version of the algorithm responsible for the differences. On rare occasions this is not feasible, and reprocessing of the older surveys from a certain point in the flow is required. Early identification of the issue enables mitigation without delay to the project. Application to the Heidrun time-lapse seismic campaign Heidrun is an oil and gas field in the Norwegian Sea on the southernmost part of the Nordland Ridge (Figure 5). There are three main reservoirs: the Fangst Group and the Tilje and Åre Formations, with the Fangst Group being subdivided into the Garn and Ile Formations. The reservoirs are spread out at depths ranging from around 2100 to 3400 m. Expected stock tank oil initially in place (STOIIP) is about 432 x 10 6 Sm 3 of oil, and original gas in place (OGIP) is about 88 x 10 9 Sm 3. Production started in 1995 and, by early 2012, about x 10 6 Sm 3 of oil and 14 x 10 9 Sm 3 of gas had been produced. The field is produced from a single platform located approximately in the centre of the field, connected to five subsea templates. The baseline time-lapse seismic dataset for the Heidrun field comprises a merge of two 3D surveys, one acquired in 1986 covering the area south-west of the platform, and one from 1991 covering the area to the north-east. Monitor surveys have been acquired by WesternGeco in 2001 and 2004 (covering the south-west area) and in 2006, 2008 and 2011 (covering the entire field). Undershooting of the platform took place in 2008 and All monitor surveys were identically parameterized. Table 1 shows the time-lapse processing campaigns in the Heidrun field. In 2001 the first monitor survey was processed in parallel with the 1986 baseline, with the moni- Figure 5 Location and stratigraphy of the Heidrun field EAGE

5 Table 1 Time-lapse processing campaigns in the Heidrun field. tor being downgraded to match the poorer quality of the baseline survey. In 2004 all three surveys were reprocessed in order to take advantage of the high-quality comparisons available between the 2001 and 2004 surveys. The baseline survey was successfully upgraded to match the bandwidth of the newer datasets, although, as expected, the general data quality of the baseline was worse than that of the newer surveys. In 2006 a subset of the 2006 survey, covering the same area as the earlier monitor surveys, was processed through the 2004 processing flow and delivered six weeks after completion of acquisition. The full survey area of all surveys was then processed through a new flow, taking advantage of newer technology and spending considerable time trying to address the limitations of the baseline surveys. In 2008 the entire 2008 monitor survey was processed through the existing flow and delivered in eight weeks. In 2010 the processing flow was revised to include 3D SRME (Dragoset et al., 2008) and further improve the baseline survey. This delivered substantial improvements in data quality, but took some time to perform. In 2011 the monitor survey was processed through the 2010 flow and delivered in 9 1/2 weeks. In 2011 the formal long-term contract on Heidrun began and the obvious benefits of the previous continuity of processing were one of the driving forces behind this strategy. As the same contractor was acquiring and process- ing the data, OBP was used to further improve turnaround. Figure 6 shows the timeline of the 2011 monitor survey. Prior to acquisition, the seismic data processing centre created the necessary OBP and onshore processing flows and performed regression testing. The flows were transferred to the vessel, together with all required datasets, such as velocity files and seismic reference cubes. A 3D stack was created from the first line acquired and differenced with an equivalent 3D stack for the 2008 survey. No obvious problems were seen. Each acquired line was processed, during acquisition, through navigation-seismic merge, QC, noise attenuation and signal processing, with a 3D QC stack volume being created progressively throughout acquisition. 4D QC of the onboard 3D stack volume against that of the 2008 survey highlighted a small consistent phase difference of about minus 6 degrees between the 2008 and 2011 surveys, as shown in Figure 7. This turned out to be due to the 2011 processing flow using a version of the acquisition filter that included the hydrophone impulse response, whereas the 2008 flow had used a filter that did not include the hydrophone response. Early identification of this subtle error enabled correction without delay to the project. The Statoil line acceptance criteria included evaluation of source and receiver repetition accuracy versus the 2008 pre-plot, and an evaluation of the likely impact of acquisition issues on the seismic data processing. For example, strict noise specs were imposed to avoid further testing of the processing flow, and out-of-spec lines were reshot in their entirety to avoid complicating the processing. Due to other activities in the area, the seismic acquisition could not be completed as planned and the vessel moved to another project before returning later in the year to complete the survey. The remainder of the processed data and 3D stack cube were transferred onshore, with a full 4D QC being available within two days of data receipt. Subsequent processing took 9 1/2 weeks to complete. The delay in acquisition put at risk the original project plan, but the rapid data processing turnaround ensured that the final 4D data was still available in time for the planned reservoir interventions. Figure 6 Timeline of the 2011 Heidrun time-lapse seismic monitor survey EAGE 97

6 first break volume 31, October 2013 The upper part of Figure 8 lists average NRMS difference values for three different areas of the survey, as shown colour-coded on the map. The NRMS difference values for the comparison are essentially identical to those of the comparison, even though, as described above, the 2006 and 2008 surveys were co-processed, whereas the 2011 dataset was processed in isolation. The level of 4D noise on the time-lapse difference comparisons in the lower part of Figure 8 are very similar between the and comparisons, and the waterfront movement is clearly visible in both cases. We conclude from this and other examples that the long term strategy is delivering data of consistent quality. Impact of the 2011 Heidrun time-lapse seismic monitor survey The 2011 Heidrun monitor survey data had an immedi- ate impact on the planning of intervention operations for oil producer well A-44_A. The Ile Formation in this well had been plugged-off in December 2007 due to increased water production. The first panel of Figure 9 shows the change in acoustic impedance on inverted data from 2004 to Water has moved northwards from a down-flank water injector up to A-44_A, explaining the increased water production. The second panel shows additional minor water movement from the south between 2006 and Also, a clear waterfront is seen coming from the west due to oil production farther north. However, the area around Well A-44_A, and to the north east, is largely unchanged. This suggested that reopening the well in a different zone of the Ile Formation might enable production of the remaining oil. The first attempt at well intervention was unsuccessful and planning for a second attempt was underway in the autumn of 2011, incorporating information from the new 4D moni- Figure 7 Phase difference between the 2011 onboard 3D QC stack and the earlier stack from the 2008 survey. Figure 8 Examples of time-lapse data quality for the and comparisons EAGE

7 Figure 9 Time-lapse seismic data examples at well A-44_A. tor survey. The amplitude difference (panel C of Figure 9) shows that the waterfront had now passed the well, making the planned intervention unnecessary. This saved around NOK million ( million) and also freed up the well slot for sidetracking to other targets. The 2011 data was inverted to acoustic impedance immediately after receipt, and both the amplitude and acoustic impedance datasets were used in a number of ongoing well planning projects. The rapid delivery of the 2011 monitor survey ensured that the data represented an accurate snapshot of current reservoir conditions. Conclusions The use of long-term seismic processing contracts can reliably and consistently deliver high-quality time-lapse seismic results in short timeframes. Further time savings can be attained by using the same contractor for acquisition and processing where possible. Reduced turnaround can have significant economic impact in terms of both efficiency and optimal use of the data. These concepts are being used by Statoil on a number of North Sea fields, and their use is expected to increase in the future. Acknowledgements The authors would like to thank the Statoil Heidrun PTC asset for permission to publish this article and the WesternGeco processing team, particularly Anna Smith and Aleksandra Handzlik, for their efforts to ensure the success of the 2011 monitor survey. We would also like to thank the partners ConocoPhillips Skandinavia, Eni Norge and Petoro for allowing publication of this work. References Dragoset, B., Moore, I., Yu M. and Zhao, W. [2008] 3D general surface multiple prediction: An algorithm for all surveys. 78 th SEG Annual International Meeting, Expanded Abstracts, 27, 1, Kragh, E. and Christie, P. [2001] Seismic Repeatability, Normalized RMS and Predictability. 71 st SEG Annual Meeting, Expanded Abstracts, 20, 1, Osdal, B. and Alsos, T. [2010] Norne 4D and Reservoir Management The Keys to Success. 72 nd EAGE Conference & Exhibition, Expanded Abstracts, L012. Van Gestel, J.P., Kommedal, J.H., Barkved, O.I., Mundal, I., Bakke, R. and Best, K.D. [2008] Continuous seismic surveillance of Valhall Field. The Leading Edge, 27, EAGE 99