Integrated and improved direct hydrocarbon indicators: A step forward in petroleum risk discrimination Open Access

In petroleum exploration, accurate prediction of risk (discriminating between success and failure), along with volume estimation, is essential to constructing an economically successful exploration program. Traditionally, explorationists will define a prospect based on their interpretation of regional and local data, including seismic attributes. Analyzing various lines of evidence for each geological element leads to the geologic chance of success (GCOS) of encountering hydrocarbons (HCs) (e.g., Sykes et al., 2011). Where amenable, based on the ability to image and interpret a fluid change within a reservoir, the incorporation of direct HC indicator (DHI) observations has been shown to roughly double the discrimination rate relative to non-DHI prospects (Rudolph and Goulding, 2017). There is no other individual analysis that discriminates successful predictions as strongly. A key challenge to determining the probability of success is reconciling dissimilar information when GCOS and DHI scores suggest very different predictions for the same prospect. Opposing views can occur, as these independent processes focus on evaluating different aspects of the geologic container. This paper describes a step change in the DHI evaluation process, a drive to integrate with geological risk, and a path to link all observations to volume estimation.

For more than 50 yr, DHIs have been observed in seismic data and supported prospects being drilled (e.g., Forrest, 2003). Through time, DHI analysis has evolved from bright spots, to amplitude versus offset (AVO) class, to today’s multiattribute analysis that leverages improvements in geophysical imaging and interpretation techniques to evaluate the chance of HC presence (Rudolph, 2001; Fahmy and Reilly, 2006; Fawad et al., 2020). Within industry, recent targeting of low-dip, subtle stratigraphic traps necessitated an audit of DHI attribute definitions and performance of the DHI evaluation process. The audit resulted in systematic improvements to the process, including the modernization of DHI attributes and the adaptation of machine learning (ML) facilitated by an industry-leading proprietary database.

To further improve the accuracy of risk prediction, a Bayesian integration process has been implemented that combines the GCOS and DHI score into a final integrated chance of success (iCOS) using GCOS as the starting (prior) value (e.g., Lowry et al., 2005; Simm and Bacon, 2014). The integration is facilitated by a new metric, DHI discernibility (referred to as discernibility), which contextualizes the expectations and confidence of DHI observations and ultimately influences the degree to which a DHI rating can modify the GCOS starting value in a positive or negative vector. Historically, a poorly rated or absent DHI either confirmed a low GCOS prospect or was not used for determining chance of success (COS) and only documented. Incorporating discernibility allows the absence of a DHI, when one is expected, to be treated as a negative line of evidence, which is not consistently applied in industry (e.g., Nixon et al., 2018).

Discernibility and iCOS present an integration framework that ensures geological and geophysical descriptions are internally consistent, in a simple and repeatable process. Guided by discernibility, DHI observations can impact most facets of assessed in-place HCs. Observations can influence ranges for volumetric parameters such as area and thickness (container), porosity (presence/absence of DHI attributes), net-to-gross (NTG) (presence/absence of fluid contact reflection [FCR]), and column height. The updated process ensures a fully integrated description of the subsurface, with accurate descriptions of both COS and prospect materiality to better inform exploration decision making.

The prospect maturation process (Figure 1) begins with the identification of a geological container observed in seismic data and evolves through interpreting independent geological (GCOS) and geophysical risk adequacies (DHI rating). Both values are then integrated through iCOS, leveraging discernibility to establish boundaries for the relative weighting of the lines of evidence. To assess the volume of recoverable HCs (resource uncertainty), reservoir parameters and column height are established, leveraging DHI observations and discernibility for an internally consistent prospect COS and volumes. All of the processes and metrics described herein are underpinned by a robust, industry-leading database.

The primary database used for DHI analysis is a compilation of previously analyzed predictions for wildcat wells and their postdrill results. The proprietary data set comprises DHI and non-DHI wildcats drilled by Exxon Corporation, Mobil Corporation, and ExxonMobil, with additional wells that were operated by others. The data, as of January 2023, represent exploration drilling in >40 countries, with reservoir ages ranging from Triassic to Pliocene, and contain 1008 DHI-rated prospects, of which 466 have been drilled (Figure 2). The DHI database is linked to reservoir parametric data results, permitting rigorous integration and consistency between DHI observations and assessed volumes.

The global database is categorized by prospect (DHI anomaly) and includes both DHI attributes and geologic metadata. The current DHI attributes are anomaly strength, lateral amplitude contrast, fit to structure, presence of an FCR, and termination distance of the HC wedge. As of January 2023, a subset of the global DHI database (185) has been modernized to be used for ML. Recovering historical ratings to be converted to the current format has been limited by access to heritage seismic data sets.

Examples of context attributes in the database include reservoir depth, dip and trap configuration, year drilled, exploration maturity, country, basin, geologic play, reservoir parameters, fluid type, postdrill HC-water contact elevation (HCWC), etc. Every wildcat is deemed a success or failure on a geological and economic basis in the year of the well’s completion. Geological successes exceed the geological risking minimum established internally. Economic successes have met internal economic criteria and have been entered into the company’s resource base. Geological and economic success thresholds can vary by company, commodity, and geographical and/or geological setting.

The integrated process for determining the COS was implemented in 2021 and blind tested using an integrated database constructed using historical documentation of reservoirs. Approximately 40 reservoir/seal pairs were used to first calibrate the process, with the wells anonymized during evaluation. Since implementing the new approach, another 20 reservoirs have been drilled and more than 40 undrilled reservoirs have been evaluated, with demonstrated improvements in risk prediction described in the Results and Conclusions section.

When evaluating a new prospect, the first step is to establish a GCOS. This process is guided by the fundamental understanding of geologic elements, by leveraging analogues and past drilling results (Rose, 2001). To avoid double counting positive or negative observations, geologic risking must remain independent of DHI attributes. For example, the presence of a DHI does not increase the chance of adequacy (COA) of source presence; the adequacy of source is determined by considering the geologic factors alone. This study uses a nine-element risk model to establish GCOS (Sykes et al., 2011) and a risking process that begins with global base rates established from internal databases (Hood and Steffen, 2018). The approach leverages prior geologic knowledge of petroleum systems and avoids an initial COA of 50%, common in risk matrices. Each of the nine geologic elements of the petroleum system shown in Table 1 requires evaluating independent lines of evidence against the base rate to establish a COA of that particular element. The COA values are then aggregated to determine the final GCOS of a prospect.

The DHI evaluation system has evolved considerably in the past few years to better handle DHI observations across an array of trapping configurations, including low-dip stratigraphic traps. In the 1990s, a DHI rating process was established using multiple seismic and data quality characteristics calibrated to the historical database available at that time (Rudolph, 2001; Fahmy and Reilly, 2006). The historical approach used a risk matrix for expert guided scoring, with prospects being evaluated for both confidence and attribute quality shown in Table 2. Confidence metrics, ranging from very poor to excellent, were evaluated based on data density, data processing, well calibration, and impedance fit to expectations. An effort to include container definition/quality was later added as a metric. The historical DHI system relied on six quality attributes: AVO class, amplitude strength, lateral amplitude contrast, terminations (distance over which the HC leg terminates), fit to structure (FTS), and FCR. These attributes were evaluated on an absolute basis, whereby sharper terminations (shorter distance) and more desirable AVO behavior (class 3) improved the quality of the anomaly, and confidence in the analysis impacted how high or low the DHI rating could ultimately score. The evaluation process before 2020 resulted in a chance of validity score for the DHI observations, risking that the interpreted/evaluated geophysical contact is a fluid contact, independent of accumulated volumes. As DHI analysis has evolved in recent years, internal benchmarking continues to demonstrate that the process remains a well-calibrated prediction system.

Exploration in the Guyana–Suriname Basin, specifically low-dip stratigraphic traps, created a unique challenge for the historical DHI system. Early DHI evaluation resulted in relatively low scores (generally <50%), despite a high success rate, necessitating systematic benchmarking of the global approach in 2018. Thin, shallow-dipping prospects lacked FCRs and have longer termination distances and weaker FTS, as the wedge zone could stretch multiple kilometers. These observations were expected, given the trapping geometry, but were penalized based on absolute metrics. As confidence in the seismic data and well calibration increased, the risk matrix scoring approach also struggled, as the top prospects were now high confidence with only moderate-quality DHI attributes. Benchmarking demonstrated AVO class and amplitude strength (tied to impedance of reservoir) was not correlating to success globally (Figure 3), beyond the Guyana–Suriname Basin observations. Internal benchmarking also demonstrated that the legacy DHI confidence metrics were not predictive of success and were removed from the scoring process.

Beginning in 2020, the first of two major adaptations to the global DHI rating process was to modify attributes and their definitions. The AVO class and amplitude strength were removed from the evaluation process and replaced with anomaly strength and anomaly consistency. Adjustments were made to all attributes measured at the contact (terminations, FTS, FCR) to define them relative to expectations for a prospect-specific container geometry, ensuring that low-dip (e.g., Guyana) and high-dip (e.g., Angola) trapping configurations could be evaluated on equivalent standards. Scoring of the updated attributes and definitions demonstrates correlation with success rate (Figure 4).

The second major adjustment to the DHI process was to transition from expert-guided scoring via a risk matrix to a supervised ML (SML) algorithm, which was fully implemented in 2021. The matrix approach required experts to choose a score within a chance of validly bucket based on the quality/confidence combination—for example, 0.3–0.7 for moderate quality, low confidence. A risk-averse culture resulted in any score >40% generally succeeding, reducing the discriminatory ability of the system. At this junction, anomaly consistency was also removed as a DHI attribute, as it was the least predictive and not required to balance out a risk matrix, which required an even number of attributes. Leveraging an SML algorithm for scoring removes human bias in assigning the risk value, though it does result in unique obstacles. As more prospects are added to the database, the scores can drift. As a result, the database requires upkeep to ensure accuracy.

Of the database of 400+ drilled DHI prospects, 185 globally diverse prospects were re-evaluated with updated definitions and attributes for inclusion in the DHI SML database (Table 3). Examples of the five attributes are shown in Figure 5. The database is additionally trained on geological success, with the resulting score now a COS rather than chance of validity, which did not have a volumetric component for success. Having the database and an SML scoring algorithm allows for rapid evaluation and implementation of future improvements; the algorithm can simply be swapped out for a more accurate or optimized artificial intelligence/ML routine and rapidly scored to demonstrate improvements in prediction. Thus, the DHI system has evolved from an expert-directed, absolute attribute quality-based system, to an expectation-based, expert-guided, SML rating system that is better optimized for fluid discrimination and calibrated to historic success rate through an industry-leading database.

When integrating two individual systems (GCOS and DHI), the relative contribution of each is critical to accurate predictions. The contextualization of determining a GCOS is standard procedure in exploration prospect evaluation (Rose, 2001); however, the same context for geophysical COS has not been previously established. The iCOS process establishes this context through a new metric called “discernibility.” This metric is divided into two evaluation criteria: expectations and confidence.

The goal of the discernibility expectations metric is to determine whether DHI observations are expected based on container expression and anticipated rock properties, independent of the prospect’s observed DHI attributes. Reservoir properties can be estimated leveraging depth, age, and analogues. Given the anticipated reservoir properties for a prospect, should one expect to see a difference in seismic response between HC and water? Expectations also consider trap geometry and depositional fill, aiming to characterize whether structural or stratigraphic complexity may lead to obscuring or creating false DHI attributes. The compilation of expectations will lead to an overall rating of likely, more likely than not, less likely than not, or unlikely to observe DHI attributes for the prospect. Expectations are evaluated regardless of the quality of the seismic data to adequately observe their expression.

The discernibility confidence metric describes the ability for an interpreter to define a container and make DHI observations in seismic data consistent with the response of the reservoir and fluids. Confidence is divided into two categories: seismic and geological. Seismic confidence assesses whether data coverage (spatial and offset) and data maturity (processing) are adequate for DHI analysis. The critical question to address is will the seismic limitations materially impact DHI observations for the specific prospect? Geological confidence assesses the understanding of container definition, rock property models, and if the reservoir observations are fitting expectations. The confidence criteria result in a rating of high, moderate, low, or no.

The final discernibility score is determined by a cross plot of expectations and confidence (Figure 6). The compilation of metrics into one value uses a least common denominator methodology to emphasize that one suboptimal attribute will limit discernibility. For a prospect with no discernibility—implying the inability to observe a fluid change within reservoir (e.g., challenged imaging, most carbonate reservoirs)—it is not advised to conduct DHI analysis. It is important to highlight that direct calibration is not required for high discernibility, nor is “perfect” seismic data. The philosophy is to describe the geological or geophysical circumstances that are expected to impact the interpretation of the DHI score within the integration process. In practice, discernibility is determined prior to DHI risk evaluation, as shown in Figure 1.

Discernibility is utilized to weight the impact of the geophysical observations (DHI) in the final prediction. In the iCOS framework, discernibility becomes the key metric for informing how much modification of the GCOS prior by DHI observations is warranted (e.g., Nixon et al., 2018). Discernibility boundaries are constructed to have maximum weight where discernibility is high to moderate and minimal weight where low or absent (Figure 7). Dividing discernibility into three categories allows for the integration process to be simple and flexible, while accounting for the uniqueness of each prospect. It is important to note that the discernibility boundaries established are guidelines and can be altered if the observations warrant modification. For example, a moderately discernible prospect with a moderate DHI rating could be viewed as only a slightly positive line of evidence and fall within the low-discernibility bounds. The integration process is prospect specific, and there is no set outcome for a given GCOS/DHI pair, regardless of discernibility. Each prospect integration is therefore dependent on the subregional understanding of the geological and geophysical circumstances.

Through the iCOS process, the intent is to integrate, rather than reconcile, the GCOS and DHI values. The Bayesian integration approach is well-documented in industry (Houck, 1999; Stabell, 2012), and within this framework, the user must identify success (e.g., oil) and failure cases (e.g., water or nonreservoir) and assign associated likelihoods. These success and failure cases and their likelihoods are convolved to establish the final COS. The iCOS process uses GCOS as a prior value, which is then modified as a function of DHI score, using discernibility to modulate the impact of the geophysical evidence on the final iCOS (Figure 7). For example, in basins or plays where prospects are expected to have DHI support, the iCOS process takes all lines of evidence into account through the lens of discernibility. The lack of DHI attributes is integrated as a negative line of evidence and would reduce the prior GCOS. The new process has been demonstrated to improve discrimination and accuracy over the GCOS and DHI processes independently, as shown in the Results and Conclusions section.

Geophysical observations and discernibility impact not only the integrated risk value, but also the volumetric parameterization. Using discernibility for a volumetric consistency check has been formally integrated into the workflow. For moderate and highly discernible prospects, the process requires the range of volumetric parameters to be fully consistent with the geophysical observations and expectations. For low discernibility, the most likely volumetric parameters are required to be permissible within the boundaries of geophysical observations. For prospects that rate as no discernibility, geophysical tie is not required; however, volumetric parameters cannot violate the root cause of the no-discernibility valuation.

The lateral edges and up-dip extent of a prospect container is relatively clear when defined by DHI attributes. A typical pitfall is when DHI attributes cover only a portion of the container, and the positive observations are extrapolated well beyond the DHI limits. Down-dip, the container extent is defined by the HCWC. The interpretation of HCWC is typically the most significant uncertainty impacting total volumes. When constructing a probabilistic assessment, an empirical relationship has been established between DHI score and column height weighting (Figure 8). High DHI scores (>0.50 rating) weight the HCWC at the rated DHI elevation to 95% of the total trials. Lower DHI scores (<0.50 rating) weight the HCWC at the DHI elevation relative to the rating outcome (double the DHI score for weighting value). The lower end of the DHI database is poorly constrained, as very few prospects (<10) with weak DHI support have been drilled and encountered geological success.

When above tuning, top and base reservoir are easily interpreted and not discussed further. Below tuning, geoscientists are advised to map and assess band limited thickness and apply an NTG ratio based on attribute strength and geologic constraints.

The NTG can be a large uncertainty for predrill resource estimation and often requires significant geological context to assign viable parameter ranges, which are frequently tied to reservoir parameter databases. Seismic integration is key to establishing reasonable bounds. For single-cycle DHIs at or below tuning thickness, amplitude increases linearly with increasing HC pore thickness (HCPT). This can be further simplified to NTG increasing with amplitude response when isolating for constant porosity and saturation (Connolly, 1999). Figure 9 shows the forward seismic model of an outcrop and demonstrates the relationship between NTG and amplitude. At the exploration scale, this can be done qualitatively (e.g., map-based amplitude transforms), and at the development or production scale, this can be done quantitatively through inversions (Monigle et al., 2022).

The presence of an FCR can also be used to help constrain NTG values used in the volumetric assessment. Figure 10 summarizes the results of simple two-dimensional modeling to demonstrate the impact of NTG within a container on the ability to observe an actual FCR versus common amplitude terminations. The modeling shows that reservoirs with an FCR have higher NTG (e.g., 40%–85% NTG) than those without an FCR (e.g., 18%–55% NTG). Other factors affecting FCR quality/strength include reservoir properties (porosity, fluid), stacking patterns (massive, interbedded), bed dip and thickness, and data quality (tuning).

The absence of an FCR does not preclude a moderate (∼50%) NTG, but presence does preclude low NTG. Globally, FCR strength has not been tied to reservoir quality, but this relationship has been observed subregionally. In the North Sea, zones with a moderate to strong FCR all had greater than 60% NTG, whereas zones with weak or no FCR all had less than 50% NTG. In West Africa, where reservoirs are generally steeper dip and more heterogeneous, fluid contact observations are more variable, and a simple relationship to NTG is not observed.

Seismic amplitude is a function of HCPT, and, therefore, DHIs provide boundaries on porosity. The DHIs are not observed at low porosities, and, empirically, DHIs are not observed below 14%. Impedance and/or AVO class can help inform relative porosity. For a given fluid type and encasing shale impedance, class 3 reservoirs will have higher porosity than class 2 or 1 (increasing impedance with decreasing porosity). The properties of the overlying and underlying shale are critical to the relationship between porosity and seismic response. For example, high-porosity sands (∼25%) have been observed as class 1 reservoirs, where the encasing shales are anomalously slow (∼2100–2500 m/s). This effect contributes to AVO class not being a predictor of success in DHI analysis.

To demonstrate the application of the new integrated risk process, a series of examples are highlighted below. The examples range from high-dip, thick reservoir (“classic DHI”) to low-dip, thin reservoir (“subtle DHI”) and span a range of outcomes enabled by the flexibility of the iCOS framework. The examples illustrate the critical role of DHI observations, including discernibility, in the integration process. Examples also highlight the link between reservoir parameters and geophysical observations, which impacts the volumetric predictions. The prospects have been anonymized, and scales have been generalized to allow for disclosure.

The Liza prospect (Price et al., 2021), located offshore Guyana, was drilled in 2015 and is a good example to illustrate the need for improvements in the historic COS process (Figure 11). At the time, the DHI system (Table 2) determined Liza to be of low confidence due to lack of nearby well calibration and resulting ambiguity in rock property expectations. All attributes contributing to heritage DHI score are shown in the “radar plot” in Figure 11B. Although the amplitude strength was considered high (consistently above background), it was penalized for AVO, as the prospect dims on the Liza three-dimensional data set. It was also penalized for lacking an FCR and for poor FTS, while scoring moderate on terminations. All seismic observations were internally consistent with the container geometry (Figure 11A). In the historical system, with low attribute quality and low confidence, the DHI outcome range was 20%–50%, and the result was an expert-judged 40% DHI rating. With the updated DHI process, Liza would now rate as 70%, which was not possible for a play test (low confidence) and the absolute attribute metrics of the heritage system. When Liza was drilled, the total GCOS was 22% (seal was key risk), with no reconciliation required between GCOS and DHI, as they were considered sufficiently close. During the iCOS benchmarking, Liza was determined to be low discernibility and estimated to have a 38% iCOS based solely on predrill knowledge (GCOS 22%, DHI 40%), a substantial decrease in risk. With the updated DHI system (Table 3) and contact weighting guidance, the predrill volumes estimation for Liza would also have increased by ∼20% over the original contact weighting, which was the discovered HCWC.

Prospect A (Figure 12) was a thick, steeply dipping container where all the traditional DHI attributes were observed. The prospect was a play extension in a proven HC system with primary risk in seal adequacy that resulted in a GCOS of 58%. The geophysical lines of evidence were positive to very positive, resulting in a DHI rating of 80% in the historic system, which was in place at the time of maturation. In the new system, predicted lower net reservoir within the container (no observed FCR) contributed to a moderate discernibility rating with expectations and confidence both being moderate. The container geometry results in sharp amplitude terminations and tight FTS, conforming to expectations. Although the GCOS was already higher than most exploration prospects, the final prediction was further improved by the positive geophysical observations and resulted in an iCOS of 85%. Prospect A was a success, and the DHI-predicted contact was confirmed. The lack of an observed FCR was confirmed to be the result of low-moderate NTG. Well results include a 220-m-thick section at ∼20° dip, 29% NTG, and 29% porosity, with a substantial column height.

Prospect B (Figure 13) was a play extension that lacked DHI attributes in a geological and geophysical setting in which they were expected. Prospect B carried primary risk in seal adequacy and reservoir quality that resulted in a GCOS of 46%. The prospect rated as moderate discernibility, and the geophysical lines of evidence were not consistent with an HC response, resulting in a low DHI score of 5%. Although the geologic understanding of the prospect suggested higher COS based on discoveries nearby, the final prediction was significantly impacted by the absence of geophysical observations, resulting in an iCOS of 8%. With the opportunity to test and calibrate a play extension, the prospect was drilled but was unsuccessful. Although the well results were not a success for exploration, they were anticipated and the most likely outcome based on the lack of geophysical support. The well encountered 41-m reservoir in a confined deepwater channel complex with ∼1° dip, 94% NTG, and 27% porosity, high-quality wet reservoir.

The results of prospects A and B are tied closely to geophysical predictions and demonstrate the predictive power of DHI analysis. The next two examples highlight challenges with subtle reservoir responses and/or container complexity, the latter of which often presents the most substantial challenge in DHI evaluation. Subtle DHI scenarios continue to pose challenges for interpretation and integration and illustrate why processes that determine COS need to continuously improve.

Prospect C (Figure 14) was an example of a play extension proximal to a recent discovery that carried minor risk in seal adequacy and resulted in a GCOS of 72%. Like prospect B, prospect C lacked DHI attributes. Expectations were reduced (“less likely”) as the reservoir fairway is discontinuous across the interpreted contact and interpreted to be low net, resulting in low discernibility. The geophysical observations were neutral to negative, and the DHI score was 30%. Since the geophysical context suggested that the DHI analysis may not be predictive, the DHI score was given no weight in the integration process. The iCOS value remained the same as the prior GCOS at 72%. As enforced by the integration process, predrill reservoir parameters were low NTG, matching the geophysical response. Prospect C was a success. Well results include 64-m-thick section with ∼1° dip. The well targeted a high-response area and encountered 69% NTG, 24% porosity, and a column height of >300 m.

Prospect D (Figure 15) was a play test in a proven basin. Prospect D carried much of the risk in seal adequacy (fault, top, and lateral) that resulted in a GCOS of 36%. The prospect had positive DHI attributes, but expectations were lowered due to structural complexity. Coupled with low confidence, this prospect was assessed low discernibility. The geophysical observations ranged from very positive to neutral, resulting in a moderate DHI score of 45%. Since the geophysical context was unclear, the DHI score was given minor positive weight and resulted in an iCOS of 46%. Prospect D was unsuccessful and encountered 231-m gross deepwater channel reservoir dipping at ∼6°, 94% NTG, and 27% porosity–high-quality wet reservoir.

As the prior examples show, the integrated system allows for flexibility, given the DHI score and discernibility. In cases where a geophysical response is expected, the integration applies maximum weighting to the DHI analysis (positive or negative). Where the geophysical response is less clear, the process allows interpreters to use geologic knowledge and the understanding of the basin to have the flexibility to weigh all lines of evidence appropriately for the final prediction.

The DHI audit and benchmarking in 2018 led to two substantial changes in the geophysical risking process, both of which improved the accuracy of the prediction based on the Brier score. The Brier score is the mean square difference between prediction and actual outcome, with lower values representing better predictions. The first adaptation was moving from absolute attribute metrics to relative/expectations-based metrics and modernizing the attributes. The initial adaptation improved the Brier score from 0.22 to 0.17 (Figure 16A, B). The second adaptation, introducing the SML algorithm, further reduced the Brier score to 0.14 (Figure 16C), representing an overall increase in accuracy of the prediction by ∼35% to the historical system. The SML rating and later adjustments to remove sentiment-based scoring descriptors are designed to reduce cognitive bias and continue to improve discrimination (difference between predicted success and failure versus actual success and failure).

The DHI integration with geologic observations has further improved prediction by using the iCOS process guided by the new discernibility metric. To mitigate hindsight bias, when the new process was implemented, benchmarking was conducted using historical data and only predrill knowledge. As of January 2023, the results of the benchmarking demonstrate that iCOS shows an ∼50% improvement at discrimination (30% compared to ∼15%) and has ∼30% improvement in accuracy (0.15 Brier compared to ∼0.23) when compared to the independent systems (Figure 17).

Drilling success rates in the Guyana–Suriname Basin continue to exceed global benchmarking. Looking exclusively at the Guyana–Suriname Basin, predrill predictions are closer to actual success rates when binned into thirds (Figure 18) and discrimination is upward of 40% (Figure 17). Discrimination is especially apparent for high-risk prospects before 2019 (Figure 18A), with success rate substantially higher than evaluated risk. After adopting iCOS (Figure 18B), discrimination increases and predicted success tracks with actual success more closely.

The integration plot shown in Figure 19A demonstrates the improvement in discrimination as the successful prospects generally cluster above the midline and the failures below. The number of prospects with GCOS (prior) values in the 40%–60% range is quite high, whereas the iCOS (posterior) values are more separated. As shown in Figure 19B, more discernible prospects are modified more substantially by the DHI process. Volumetric parameterization is also enforced to be consistent with geophysical lines of evidence (as a function of discernibility), with guidelines derived from historical database(s).

In frontier and emerging basins, quantifying the COS, coupled with volumetric guidelines, can have a significant impact on the economic outcome of an exploration portfolio. Discriminating between success and failure impacts drilling order, drilling optimization, and timing of development projects. The goal of this new methodology was to improve prediction through a simple, repeatable, and fully integrated process to support business decisions across a global portfolio of opportunities. These results demonstrate that when fully integrated, DHIs have the power to be highly discriminative of economic prospects when used appropriately with the correct context.

The authors thank ExxonMobil, our partners, and our government partners in Angola, Guyana, and Australia for permission to publish. We also acknowledge the welcomed guidance of Patty Walker, who was instrumental in the implementation of the updated workflows. Thank you to Wayne Camp, Daniel Minisini, and three anonymous reviewers; and to our editors that provided invaluable feedback: Reece Murrell, Joe Reilly, Ken Hood, Jonathan Stewart, and Alex Martinez.

The concepts presented herein were developed and refined over 30+ years of direct hydrocarbon indicator (DHI) evaluation at ExxonMobil and have built upon the contributions of many. We would like to specifically thank Bill Fahmy, Cody MacDonald, Guy Medema, Kurt Rudolph, Kurt Steffen, Jie Zhang, and the DHI research team for their foundational efforts to improve our systems.

This work is only possible because of the entirety of the integrated teams that worked cooperatively to develop integrated chance of success (iCOS) and discernibility. The global iCOS team is Andrew Budnick, Sarah Gelman, Juliet Irvin, Chris Jenkins, Reece Murrell, Vanessa Nenna, Audrey Parnell, Ed Penrose, Will Waters, and Luke Wilson. The global discernibility team is Julia Howell, David Abt, Kun Guo, Gary Lewis, Kathleen McManus, Reece Murrell, Mark Nell, Willie Meyer, Ed Penrose, and Susanna Webb.

We also thank various individuals for their contribution to this publication through figure support. The DHI Atlas map is courtesy of Joshua Johnson; fluid contact reflection/net-to-gross forward models were courtesy of Becky Simon; and seismic and other examples were supported by Shlomo Honig, Gerrick Jensen, Rachel Paez, Janelle Sherman, Steven Spencer, Will Waters, and Mark Widmer.

Patrick Monigle is a geophysical advisor at ExxonMobil with 10 years of experience in petroleum geoscience, encompassing exploration, development, and research with notable expertise in the Guyana–Suriname Basin. His academic journey took him from NC State to Oregon State University, where he earned his Ph.D. in geophysics in 2014. His interests include geophysical reservoir characterization, hydrocarbon response, and the integration/intersection of geophysical methods, geomodeling, and stratigraphic analysis.

Tiffany Hedayati is a geoscience advisor at ExxonMobil with 16 years of experience in petroleum geoscience, spanning roles from exploration to production and specializing in risk analysis. She earned her geoscience degree from The University of Texas at Austin in 2008. She is dedicated to accurately characterizing risks and quantifying subsurface prospects. Her mission is to revolutionize and standardize risk analysis across exploration, driving more informed and confident decision-making.

Frank J. Goulding is the chief geoscientist at ExxonMobil, bringing 37 years of experience in petroleum geoscience across exploration, development, production, and research assignments. He graduated from the University of Calgary with a degree in geophysics in 1987. His interests include seismic interpretation, hydrocarbon response, stratigraphy, and production and development geoscience.