A comparative study of reservoir modeling techniques and their impact on predicted performance of fluvial- dominated deltaic reservoirs: Reply Peter E. K. Deveugle 1 , Matthew D. Jackson 2 , and Gary J. Hampson 3 INTRODUCTION We thank Li et al. (2018, this issue) for their discussion of our paper (Deveugle et al., 2014), which assessed the impact of using different stochastic-reservoir-modeling techniques to capture geologic heterogeneity and fluid-flow behavior, via comparison with a reference model constructed from a fluvial-dominated deltaic reservoir outcrop analog (Deveugle et al., 2011). The stochastic models in Deveugle et al. (2014) were constructed using a sparse data set of pseudowells, synthetic three-dimensional (3-D) seismic data, and geologic interpretations to mimic a reservoir-modeling project to support early field development. The first part of the discussion of Li et al. (2018, this issue) raises the interesting issue of how to deal with bias in well data and provides us with the opportunity to clarify some misconceptions regarding the recognition and handling of unrepresentative well data in reservoir-modeling studies. The second part of their discussion seeks fuller explanation of our calculation of facies probabilities and their implemen- tation in generating facies proportions in the models of Deveugle et al. (2014). In this reply, we address each part of the discussion of Li et al. (2018, this issue) in turn, and then we conclude with an assessment of how their discussion points impact the main find- ings of our original paper (Deveugle et al., 2014). RECOGNITION AND HANDLING OF BIAS IN WELL DATA Wells sample a tiny proportion of any reservoir (e.g., eight pseudowells sample 0.00001% of the volume of the reservoir models presented in Deveugle et al., 2014). The problem of small sample size is com- pounded by the nonperiodic and nonstationary character of facies distributions in many reservoirs (e.g., “jigsaw puzzle” and “labyrinth” reservoir types of Weber and Van Guens, 1990), with the result that wells are highly unlikely to sample representatively the facies proportions and distributions within the reservoir. Thus, there is commonly some bias in the facies proportions and distributions sampled by wells (e.g., Pyrcz et al., 2006). As outlined by Li et al. (2018, this issue), there are several established techniques to decluster or debias well data. Declustering methods rely on weighting the data sampled by the wells to account for spatial representativeness but assume that the entire range of the true data distribution (i.e., all facies types) has been sampled (e.g., Journel, 1983; Isaaks and Srivastava, 1989). In the absence of a clear and persistent spatial trend in facies between wells, de- biasing of well data is achieved by using secondary data, such as a conceptual geologic model, to adjust the primary data distribution of the well samples (e.g., Frykman and Deutsch, 1998; Pyrcz et al., 2006; Ma, 2009). However, the reservoir geoscientist does not know a priori the “correct” conceptual geologic model or concept-derived scenario that pertains to any given reservoir. She or he should instead in- vestigate “several diverse scenarios . . . perhaps based on a range of appropriate ancient, modern, and ex- perimental analogs. In fluviodeltaic reservoirs, the scenarios should capture uncertainty in facies pro- portions . . . [which is one of several] critical aspects of facies architecture that control sweep efficiency” Copyright ©2018. The American Association of Petroleum Geologists. All rights reserved. 1 12 Archdeacon Street, Nedlands, Western Australia 6009, Australia; deveugle@ gmail.com 2 Department of Earth Science and Engineering, Imperial College London, South Ken- sington Campus, SW7 2AZ London, United Kingdom; m.d.jackson@imperial.ac.uk 3 Department of Earth Science and Engineering, Imperial College London, South Ken- sington Campus, SW7 2AZ London, United Kingdom; g.j.hampson@imperial.ac.uk Manuscript received December 4, 2017; final acceptance January 8, 2018. DOI:10.1306/01081817409 AAPG Bulletin, v. 102, no. 8 (August 2018), pp. 1664–1667 1664