Comparing Seismic Inversion Methods on a Carbonate Reservoir: A Case-Study from the Mishrif Reservoir, Rumaila Field, Iraq

Abstract

Abstract We present a case-study that compares seismic inversion methods for reservoir characterisation on the Mishrif carbonates in the Rumaila field, Iraq. The two methods interrogated - Deterministic Absolute Acoustic Impedance Inversion (DAAII) and BP's One-Dimensional Stochastic Inversion (ODiSI) - were used to predict porosity. This case-study highlights benefits, limitations and uncertainties associated with these methods. Seismic inversions produce non-unique solutions mainly due to the bandlimited input seismic, as several reservoir property profiles can result in the same seismic trace. DAAII is a widely applied Model- Based Inversion technique, which uses a Low Frequency Model (LFM, derived from well-log impedance) as input to the inversion algorithm. The algorithm uses the LFM as a starting point to seek an impedance profile that best minimises residual error between the resulting modelled synthetic, after convolution with an average of the extracted wavelets, and the input seismic trace. It accepts the inverted impedance profile when the error is minimised. DAAII typically results in a relatively smooth deterministic estimate of absolute Acoustic Impedance (AI), and the output is heavily dependent on robustness of the input LFM. This can be a severe shortcoming in reservoirs with poor well control - simply put, if the input LFM is of questionable quality (e.g. poor well-log data, sampling bias, suboptimal interpolation between wells) then the output impedance is most likely going to be inadequate at accurately predicting reservoir properties. In this case-study, seven wells within the 250km2 test area that had been tied to the seismic were used for the LFM along with three framework surfaces, and of these seven wells, four extracted wavelets were averaged and used for the inversion. Finally, the inverted Acoustic Impedance (AI) volume was used to estimate porosity using a simple linear regression. BPs ODiSI (Connolly & Hughes, 2016) method has been applied to the Mishrif reservoir over the same 250km2 test area within the Rumaila oil field - the first time it has been applied to a carbonate reservoir (Grant et al., 2018). ODiSI is based on matching large numbers of pseudo-wells to each trace from one or more Coloured Inversion (CI) angle stacks - a single CI of full-stack volume was used this case-study. Pseudo-wells are defined as stochastic simulations of 1D stratigraphic profiles with attached physical properties but no lateral information (de Groot et al., 1996). Due to well-log data quality requirements, only one well was used to define the rock physics trends, on which the inversion was based. At each trace location, ODiSI generates a large number of pseudo-wells (20,000 in this case-study) consistent with the input prior information - from this one well, the input CI trace, three framework surfaces and our understanding of bed thicknesses and transitions. ODiSI then generates a synthetic trace for each pseudowell, compares these traces to the CI seismic data and selects the ones that match best (100 in this case- study). These best-match pseudo-wells (BMPWs) were analysed to provide estimates of porosity, uncertainty associated with estimated porosity (e.g. standard deviation), most-likely facies and facies probability at each trace location. Each of these properties, as well as volumes of any intermediate properties such as AI can be generated and examined, allowing for more transparency and opportunities for inversion output QC as a joint inversion scheme. DAAII and ODiSI have been successfully applied to the Mishrif carbonate reservoir in the Rumaila field. This case-study shows how ODiSI results compare against a more established inversion method (i.e. DAAII) in a heterogeneous carbonate reservoir of an onshore field. Since a primary objective of any property estimation procedure like a seismic inversion is to predict the value of a property or attribute at an unmeasured location, several wells were used as blind-tests. More blind-test wells were interrogated with ODiSI compared to DAAII, due to its minimal well input. These tests showed that despite the fewer number of wells required to run ODiSI, the results are generally better than those from the DAAII.