Data Using a Facies Based Bayesian Seismic Inversion, Forties Field, UKCS Kester Waters* (Ikon Science Ltd), Ana Somoza (Ikon Science Ltd), Grant Byerley (Apache Corp), Phil Rose (Apache UK) Summary The Forties Field, UKCS, is the largest oil field in the UK sector of the North Sea with an estimated STOOIP between 4.2 and 5 billion barrels. Discovered in 1970 by BP, the field was brought online in 1975. In 2003 Apache Corp acquired the Forties field from BP and quickly pursued a program of 4D seismic interpretation to assist in field remediation. After some 40 years of production the field continues to produce at daily production rates exceeding c. 45K bopd. In this paper we present the results of a new inversion technology applied to the Forties field to assist in the detection of direct hydrocarbon indicators and remaining reserves from two vintages of 3D seismic. (a) Introduction We apply a new Bayesian simultaneous inversion technology to two 3D seismic datasets covering the Forties field in the UK North Sea. The inversion scheme jointly solves for seismic facies and their corresponding impedances, without the pre-requisite low frequency background model that other conventional inversion schemes require. The 3D simultaneous AVA inversion was applied separately to two vintages of 3D seismic, acquired in 1988 and 2013 respectively. The objective of the inversions was to characterise the seismic response into facies (overburden shale, Forties shale, Forties brine reservoir and Forties oil reservoir) and impedances and to subsequently use these volumes to identify areas of bypassed pay. The inversion of both the 1988 and 2013 data sets was generally excellent, the sand vs shale facies distribution defining the complex reservoir architecture of the Forties Field which is proved by some 350 well penetrations. The 1988 survey was shot at a point where 1.8 billion barrels of oil had already been produced from the Forties Field since first production in 1975. The 1988 inversion oil sand facies distribution results showed a generally very good calibration to the distribution of remaining hydrocarbon as defined by: water cut evolution in the original development wells; corridors of sweep which can be traced back to high volume injection wells; known water swept responses identified on 4D seismic; mid 1990s infill wells; and calibrated to the near virgin state of the Echo platform area in SE Forties (first production 1987). As expected, the 2013 oil sand facies distribution was much more limited, total production at that time being 2.7 billion barrels. The results calibrated nicely (b) Figure 1: Schematic diagram of conventional model based inversion (a, top) and the Bayesian Joint Impedance and Facies Inversion scheme (b, bottom) to infill drilling results obtained post the 2013 seismic survey, predicting both swept and oil bearing locations. However, it was noticed that the oil sand prediction was less successful in some of the stratigraphically deeper parts of the Forties reservoir and in the Charlie Channel Complex in the west of the field. The Ji-Fi inversion has resulted in valuable new insights to distribution of remaining hydrocarbons in the Forties Field. In this case study it has provided another de-risking tool to be integrated into the Forties bypassed pay target methodology along with the other existing field data such as 4D seismic, previous inversion results, production history and well log data. Page 2871
Theory of the joint impedance and facies inversion Model based seismic inversion is a well-established technique which is widely used in industry. The inversion works by iteratively updating a starting low frequency model by minimising the misfit between synthetic seismic generated from the model and the observed seismic traces, figure 1(a). Figure 2 shows a simple slab model constructed using depth trends and rock property averages from the North Sea Palaeocene Forties formation. A series of 5 synthetic angle stacks were created from 5 to 45 degrees at 10 degree increments. Once the misfit has reached a satisfactory level, the final updated broadband impedance model is output. In the presence of seismic noise, this process alone can be quite unstable, giving large deviations from the starting model and resulting in unphysical impedance estimates. Therefore, a model weight is typically applied, which penalizes large deviations from the starting model and forces more plausible final impedance estimates. The model based inversion scheme is described in equation (1) below. (1) Q seismicmisfit 2 modelmisfit 2 Where Q, is the final broadband impedance output and, is the model weight. The same model based inversion scheme can also be cast as a Bayesian problem (presented in terms of a post stack acoustic inversion for simplicity), as below in equation (2) where the posterior probability, q, of the acoustic impedance given the seismic, is proportional to the likelihood function, L, of the seismic (or in this case forward modelled synthetic) given an acoustic impedance, multiplied by the probability distribution function, p, of the acoustic impedance prior model. (2) q AI seismic L seismicai p AI The Bayesian joint impedance and facies inversion scheme is presented as a schematic diagram in figure 1(b). Unlike the conventional model based inversion, there is no requirement for a single multi-facies low frequency model. Rather, depth dependent trends are established for each of the expected facies - typically reservoir and non-reservoir. Given the prior facies based trends, the seismic data and wavelets and a set of prior facies proportions, the inversion then solves for both the most likely facies model and its corresponding set of impedances. Equation (2) is thus extended to contain an additional facies term as described in equation (3) below (adapted from Kemper and Gunning, 2014). (3) AI seismic L seismicai p AI facies p facies Figure 2: Slab model constructed using depth trends and average properties showing acoustic impedance (top), 5 degree near stack synthetic (middle) and 45 degree far stack (bottom) (blue = hard) The synthetic data were then inverted using a conventional model based inversion scheme and the Ji-Fi technique, Figure 3. In order to compare final facies estimates, a second step of Bayesian classification was also applied to the conventional model based inversion results, using the same depth trends as were input to Ji-Fi. For both inversion schemes, the recovered Facies image is exact and correct when compared to the input model. However, in order to achieve this result using the conventional model based inversion scheme we required Page 2872
Figure 3: Left column from top to bottom shows one of the four low frequency models input into the Ji-Fi process, the inverted acoustic impedance and corresponding facies model. Right column from top to bottom shows the single low frequency model required to correctly invert, via simultaneous inversion to acoustic impedance in order to derive, via Bayesian classification the facies image. precise knowledge of the reservoir geometry and the presence and location of the saturating fluid types, along with the correct merge frequency for the model. This example therefore illustrates the primary challenge with conventional inversion. The Bayesian joint inversion we present in this paper requires no detailed model building and provides absolute impedances and facies estimates directly. This ensures total consistency of the rock physics, impedance cubes and facies volumes without the prerequisite and often heavily biased background model. The facies inversion results were extracted from the 1988 inversion results, figure 4, and compared to active production (green) and injection (blue) wells at the time of and preceding seismic acquisition. In general an excellent correlation is observed between the mature production areas to the North and far South West and the initial production around the clustered production wells in the Echo area in the South East. The facies inversion does a good job of identifying the sweep pattern and flood front, with brine bearing predictions forming halos around a few of the more mature injector/producer locations, however there are areas where the inversion notably over-predicts the oil distribution, in particular around the Charlie channel system, in the west of the field, and in the stratigraphically deeper Upper Main Sand. The oil facies predictions in the east of the field are broadly commensurate with post 1988 sweep responses identified from 4D seismic analysis using the 1988 base line survey together with the 5 later monitor surveys. Figure 4: Stratal slice through 1988 Facies cube, 50ms below top Sele Fm Application of the joint inversion algorithm to the Forties field, UKCS Ten key exploration and appraisal wells were selected for the study and used to define the compaction trends and rock physics models. The wells were drilled over a period of time spanning the production period and seismic vintages that were used. The remaining several hundred production and injection wells were used in the quality control of inversion results over the full field area and with knowledge of their age and production profiles relative to the seismic vintages. Two seismic datasets were used during the analysis. The 1988 and 2013 seismic data sets consisted of five angle stacks, with central angles of 9, 18, 27, 35 and 42 degrees. Structural / stratigraphic horizons were available for the Sele Fm, Forties Fm and Maureen Fm. Wavelets were estimated for each of the angle stacks and for each vintage, giving a total of ten wavelets which were used in the study. Figure 5: Facies extraction in top 20m of major sand bodies based on 2013 inversion results Figure 5 shows an extraction from the 2013 facies cube for the upper 20m of the Forties reservoir zone. The definition of the OWC is much improved over the 1988 seismic data set with less oil being predicted beyond the field limits. The increased maturity of the field is clearly reflected by Page 2873
the more limited distribution of remaining pay. As 37 wells have been drilled by Apache since the acquisition of the 2013 survey, it is possible to score the fluid and lithology predictive capability of seismic characterization methods against the well results. Both the Ji-Fi facies volume and an oil probability volume derived from conventional simultaneous inversion of the 2013 data were scored in this way. When scoring the results, the reservoir unit was recorded as follows: Charlie shallow stratigraphy in the West; Upper Main Sand deeper stratigraphy important in the South and West; Delta shallow sands in the center of the field; and Alpha-BravoEcho shallow sands in the East. When scoring the results, a 1 was assigned for an accurate prediction of reservoir and pay geometry, declining to a 0.5 for a correct prediction with only moderate geometric fidelity. 0 was assigned for an incorrect prediction. The results were also split depending on whether the target was successful or a dry hole. As an illustration of this scoring method, Figure 6 shows a Ji-Fi extraction at the successful T280 location, showing the successful pilot, A62, wet horizontal section A62Z and successful horizontal completion A62Y. The Ji-Fi volume successfully predicted the presence of oil at A62 and A62Y and a dry hole at A62Z. However the thickness of the oil column was over predicted and the sand shale ratio under predicted at the A62 pilot-hole location. This target was assigned a score of 0.8 as part of the successful target Alpha-Bravo-Echo dataset. Figure 6: Cross section along appraisal and development well T280 The results of the score card assessment of the Ji-Fi and simultaneous inversion predictions of post 2013 wells are summarized in table 1. It was felt that to be a useful prediction the score for both dry holes and successful targets would need to be greater than 0.5. The results confirmed that Ji-Fi over-predicted pay in the UMS, scoring 0 for the dry holes, and under-predicted pay in the Delta sands, scoring 0.33 for successful targets. The prediction for the Charlie sands was considered moderate (a tendency for over prediction of pay resulted in a 0.5 score Table 1: Score chart for simultaneous inversion and Ji-Fi results for dry holes) and good for the Alpha-Bravo-Echo sands in the East, where both successful wells and dry holes scored well above 0.5. The scoring pattern for the simultaneous inversion oil prediction was similar to Ji-Fi results, but it generated a poor prediction for the Alpha-Bravo-Echo trend with significant over-prediction of pay sands, scoring only 0.3 for dry holes. Conclusions We have shown proof of concept and initial results of a new Bayesian simultaneous inversion which jointly solves for impedances and facies. It does not require the construction of a conventional low frequency model and therefore reduces bias in the output impedance cubes. It also allows more comprehensive scenario testing to assess the sensitivity of the inversion to e.g. Net:Gross and hydrocarbon prior probabilities. The results show generally good calibration to the available well data, particularly in the younger reservoirs, however the predictive power of the inversion results diminish in the deeper section. The results of the inversion whilst encouraging, do not on their own provide the basis for a drilling campaign. These new results which indicate potential direct hydrocarbon indication are being integrated into the appraisal and development process presented in this mature field. Low risk drillable targets identified by this new technique must also be consistent with offset well logs, production, lithology prediction and 4D seismic response analysis. Acknowledgements We would like to acknowledge Apache UK for permission to show these results and CSIRO and the various staff at Ikon Science Ltd in their generation. Page 2874
EDITED REFERENCES Note: This reference list is a copyedited version of the reference list submitted by the author. Reference lists for the 2016 SEG Technical Program Expanded Abstracts have been copyedited so that references provided with the online metadata for each paper will achieve a high degree of linking to cited sources that appear on the Web. REFERENCE Kemper, M. A. C., and J. Gunning, 2014, Joint impedance and facies inversion seismic inversion redefined: First Break, 32, 89 95. Page 2875