The open file 2,700-sq-km (1,042-sq-mile) Willem 3-D seismic survey adjacent to the giant Io/Jansz, Wheatstone and Pluto gas fields in the Carnarvon Basin on the northwest shelf of Australia has been reprocessed and inverted to provide a significantly improved dataset to reduce the risk of the presence of hydrocarbon charge and top seal leakage in an underexplored yet prolific hydrocarbon area. The seismic survey was originally acquired and processed by Woodside Petroleum in 2006. The Urania-1 gas discovery and northern extent of the Pluto Field were detected on the original data. Despite a high chance of success driven by 3-D seismic data and an amplitude vs. offset- (AVO-) rich area, some wells within the survey failed to encounter hydrocarbons when testing seismic amplitudes that were subsequently found to be caused by the tuning effect on the edge of the local depositional center (e.g., Ixion-1). Improvements in seismic processing, including broadband technology and prestack depth migration as well as improvements in quantitative interpretation through facies-based inversion, suggested that significant value could be added to the dataset for a fresh look at areas previously passed over.

Reprocessing
The original processing was regarded as good for the time; however, a number of limitations were clear: poor imaging, amplitude attenuation, multiple energy and signal-to-noise (S/N) drop below the Cretaceous. The causes of these problems also were clear: the strong bathymetric relief, the complex overburden and shallow anomalies associated with potential leakage features. The reprocessing was designed to address all of these issues. A key element of the workflow was the use of broadband de-ghosting that has now become a common tool for conventionally shot streamer data. The de-ghosting improved the de-multiple sequence by having a zero phase wavelet with minimal sidelobes. Surface-related multiple elimination worked quite well, and having a better understanding of the phase allowed for a tighter premigration radon. Integrating regional geological knowledge early improved the initial velocity analysis. Furthermore, going into the velocity modeling, there was less ambiguity of the events in the overburden, which allowed for significant additional detail in the velocity model. Having reduced tuning effects/constructive interference on the flanks of channels and thin beds, the tomography could accurately define the velocity contrasts and helped account for the ray path distortion seen on the far offsets of the original dataset.

Depth migration consisted of five iterations of constrained global tomography to generate a final velocity model and two anisotropic updates. Emphasis was placed on properly resolving the channel features. Resolving the deeper complex features aided in identifying shallow anomalies. The anisotropic model was layer-constrained using regional lithological knowledge and quality-controlled, partly through a focus on the deep AVO response on the far offsets. The detailed interpretation resulted in high-resolution velocity and anisotropy models. Post-migration, the detailed velocity model allowed for a more effective de-multiple as tighter constraints could be applied and the anisotropic imaging ensured a stable offset response for stacking. The data were stacked into partial angle stacks for simultaneous inversion.

Inversion
Inversion was carried out to improve the interpretability of the seismic data. The type of inversion chosen was a deterministic facies-based simultaneous inversion. That is, the partial angle stack seismic data are simultaneously inverted to elastic rock properties and a most likely facies model. There are a number of reasons for selecting such an inversion. First, simultaneous inversion combines the well-known benefits of inversion with the AVO in the seismic to produce, at least, acoustic and shear (S) impedance models of the subsurface, which when combined allow for a better definition of lithologies and fluids. Second, combining simultaneous inversion with a facies inversion removes the requirement for an initial low-frequency model and introduces constraints that potentially improve the definition of thin and thick beds. The lack of a requirement for a low-frequency model is important in this case as there is limited well control within a very large geographical area. The simple interpolation/extrapolation of a few wells across such a large area would be meaningless.

Including facies within the inversion limits the required input to time-dependent rock physics relationships for each of the chosen facies complete with an assessment of uncertainty. Of course, the assumption is that these rock physics models are applicable over the entire area and can be extrapolated to depths greater than the well control. The low-frequency components of the resulting absolute elastic property models are determined by the predicted vertical distribution of facies and the trends associated with those facies.

The facies chosen for inversion were based on five open file wells: Ixion-1, Bellatrix-1, Urania-1, Pluto-4 and Guilford-1. Five facies were selected based on the geological sequence and upscaled elastic properties as the facies must have some degree of separation in the elastic domain to allow the seismic reflectivity to drive the final distribution. These facies were brine sand, gas sand, shale, limestone and marl. The rock physics models were defined by only two wells with a complete suite of elastic logs: Bellatrix-1 and Urania-1. The rock physics models consist of time-dependent trends of compressional (P)-velocity and relationships between P-velocity and S-velocity and between P-velocity and density. The additional inputs to the inversion are wavelets for each angle stack, S/N estimates for each stack and prior probabilities for each facies.

(Source: Ikon Science)

Results
The results can be assessed by the conformity of the facies distribution to geological expectation and the match at all of the wells. Given that the rock physics models were based on only two wells where S-sonic was measured, the prediction of facies at all the wells is very encouraging (see figure). It is interesting to note that although no Cretaceous gas sands were available for input into the rock physics model, the prediction of the gas sand in the Guilford-1 well is accurate. It is also interesting to note that at the two wells where no gas sands were encountered, although they had originally been drilled based in part on seismic amplitudes, the inversion predicts that there are no gas sands.

The Pyxis-1 discovery by Woodside Petroleum was announced a very short time before the inversion part of the project was initiated. Only the well location and the fact that gas had been discovered were known; neither the depth nor the stratigraphic level of the find were known. A gas sand was predicted at the Pyxis-1 well location within the Upper Jurassic, possibly Tithonian interval. The discovery is of significant size, covering an area of about 20 sq km (7.7 sq miles), and the inversion has provided a clear definition of the dimensions, including the prediction of a very thin gas leg separated from the main accumulation by a small graben. The inversion predicts a gas-bearing sand with a thickness of about 18.3 m (60 ft) at the well location, which compares favorably with the 19.5 m (64 ft) announced by Woodside Petroleum.

Acknowledgements
The authors would like to thank Searcher Seismic for allowing access to the data and permission to publish the paper; Finder Exploration, which provided the interpretation; and Ikon Science, which provided the facies-based inversion. This article is an abridged version of an article that ran in the January 2016 issue of The Leading Edge. The Society of Exploration Geophysicists owns the copyright and has granted permission for its reprint. Mark Sams, Shane Westlake, Josh Thorp and Ebrahim Zadeh (2016). “Willem 3D: Reprocessed, inverted, revitalized.” The Leading Edge, 35 (1), 22-26. doi: 10.1190/tle35010022.1.