Mud Gas Data Could Reveal a Wealth of Reservoir Information
The utility of mud gas data so far has been limited to fluid typing, formation evaluation, and interwell geological and petrophysical correlation. The ongoing digital transformation has presented the opportunity to increase the utility of, and get more value from, the abundant and rich mud gas data. This article raises the question of whether getting more from mud gas data is possible with machine learning and shares some ideas to expand its utility.
From empirical correlations to machine-learning modeling, the petroleum industry has had some success in indirectly estimating various reservoir properties from wireline logs. Grain size, cementation factor, saturation index, tortuosity, porosity, permeability, water saturation, and lithology are examples of such reservoir properties. This has helped domain experts make decisions ahead of well testing, core description and thin section and special core analyses that would provide equally or more accurate data to validate them. This indirect approach to reservoir characterization, within the range of acceptable accuracy, has been beneficial in making valuable on-time decisions and avoiding the operational costs associated with traditional analyses. The traditional analyses, though more accurate and representing the ground truth, take more time to acquire and require specialized equipment and manpower.
Contrary to wireline logs, the utility of mud gas data has been limited to fluid typing, formation evaluation, and interwell geological and petrophysical correlation. The ongoing digital transformation has presented the opportunity to increase the utility of, and get more value from, the abundant and rich mud gas data. This article raises the question of whether getting more from mud gas data is possible with machine learning and shares some ideas to expand its utility. If successful, these ideas can evolve a new methodology to estimate and predict reservoir properties while drilling and ahead of wireline logging and well testing.
Achieving this objective may evolve a new concept of the reservoir-characterization-while-drilling or real-time-reservoir-characterization work flow. We suggest leveraging the power of machine learning and ubiquitous hardware resources to conduct reproducible and iterative research to investigate the possible linear and nonlinear correlation between mud gas data components and various reservoir properties.
Part of the outcome of the mud-logging process is the mud gas data. The mud-logging process characterizes the liquid and solid components of the mud that are brought to the surface during the drilling of oil and gas wells. The mud gas data is a quantitative measure of the gas liberated from the subsurface formations as the drill bit grinds through rock formations. Gas detectors (chromatographs and spectrometers) are used to separate and quantify the individual natural gas components, ranging from light to heavy as well as organic to inorganic, dissolved in the mud and circulated to the surface. These measurements are recorded in real time in addition to the other drilling parameters such as rate of penetration (ROP), mud weight, flowline temperature, pump pressure, flow rate, and hook load. The lithology of the drill cuttings is also recorded but only in near real time because there is some delay to allow wellsite geologists to do the description. More details about the mud-logging unit and equipment can be found in Whittaker (1991).
The utility of mud gas data is currently limited to fluid typing (Haworth et al. 1985; Pierson 2017), formation evaluation (Walstrom 1972; Hill 2017), and interwell geological and petrophysical correlation (Beeunas et al. 1996). Most of the studies conducted on the use of mud gas data either focused only on the total gas to the exclusion of the individual gases, used only the light gases to the exclusion of the heavier components, or did not leverage the power of machine learning. Similar to the currently wide use of wireline logs to estimate and predict various reservoir properties, mud gas has higher potential to provide real-time information to characterize reservoirs at the early stage of development and improve the safety of drilling personnel. Achieving this is expected to add much more value to the drilling, exploration, and production phases of the upstream business. Using the early acquisition of mud gas data will enhance real-time decisions to possibly prevent drilling problems and improve the safety of drilling personnel.
The focus so far has been on wireline logs (Serra 1984; Kobr 2011) and well/formation test data (Jun and Minglu 2011; Spivey and Lee 2013). While the former provides the earliest available data from the subsurface after drilling, the latter takes measurements as fluids flow from the reservoir. Increasing the utility of the mud gas data to as much as that of wireline log data could add much more value to the drilling, exploration, and production phases of the petroleum business. This short communication focuses more on mud gas data rather than the lithology log component because the latter is not acquired at the same rate in real time as the former. The current level of utility of mud gas data has been appreciated, the limitations have been acknowledged, and the efforts made to overcome the limitations have been properly documented in various publications. Reviewing these publications, however, is beyond the scope of this article.
In the realm of the petroleum industry’s digital-transformation objectives and in the era of the Fourth Industrial Revolution, there is a need to leverage the abundance of mud gas data to extend its current utility to real-time applications and take the utility closer to the edge (the wellsite). Given the significant cost of acquiring mud gas data, using the power of big data and machine learning technologies would be worthwhile for maximizing its utility for real-time estimation of reservoir properties.
Suggested Research Directions for Possible Exploration
The industry currently is focusing much attention on the use of wireline data for indirectly estimation of most reservoir properties. It is time to extend such attention to mud gas data. We present to the petroleum industry and research community the possibility of predicting those reservoir properties that are currently estimated using wireline logs. Doing this will increase the time value of those reservoir properties and make them available to support real-time decision-making. Geoscientists and petroleum engineers can use such predicted values to make vital decisions in real time during drilling. Such reservoir properties include rock porosity, permeability, pore pressure, lithology, rock strength, formation tops, and hydrocarbon fraction.
Real-Time Porosity Prediction From Mud Gas Data
Geoscientists and petroleum engineers typically obtain indirect estimations of formation porosity from wireline logs (Serra 1984; Al-Anazi and Gates 2012). The petroleum industry has lived with obtaining porosity from direct measurements with core-sample and thin-section analyses and indirectly with wireline log analysis. This information is used for post-drilling formation evaluation, reservoir analyses, and other decision-making processes. Having the capability to obtain porosity measurements while drilling will offer valuable input to critical decisions such as whether to proceed to planned total depth, halt, or sidetrack. With real-time porosity, optimal coring points or formation tests can be identified in real time. For example, the indication of a porous zone could signal a desirable zone to take core samples or conduct tests for possible determination of hydrocarbon shows. A change in porosity might indicate a possible change in lithology, hence providing an indication of a possible formation top or casing point.
In a near-real-time scenario, wellsite geologists typically estimate formation porosity by visually inspecting the rock cuttings brought to the surface by mud circulation during drilling. This requires a detailed rock-typing procedure (Low 1951; Archie 1952; Dunham 1962; Swanson 1981). Because mud and other drilling fluids usually cover the drill cuttings, they need to be washed to ensure the integrity of the estimation. During drilling, geologists take core samples from reservoir rocks and transport them to the laboratory for detailed description. In the laboratory, technicians plug some of the core samples and use part of the plug samples to prepare thin sections. They then use the thin sections to estimate formation porosity using specialized equipment. These measurements are qualitative and highly subjective. The porosity estimation from thin sections requires enormous time and labor.
Predicting porosity from mud gas data while drilling will overcome these challenges, possibly reduce the cost of coring over time, increase the time value of its utility, and assist in real-time decision-making. If necessary, drill cuttings lithology can be integrated to increase the accuracy.
Real-Time Permeability Prediction From Mud Gas Data
The mud logger typically measures the formation permeability by visually inspecting the grain size and sorting (Petrowiki 2021). Similar to the same method for porosity estimation, the measurement is qualitative, highly subjective, and not usually accurate. It may give estimates with little or no confidence at certain conditions. For example, it may be challenging for geologists to adequately quantify the smaller clay-sized particles that may affect permeability estimation if present in substantial quantity. Another limitation occurs because of the possible smearing of the drill cuttings data with various artifacts in the wellbore. This happens because mud loggers collect the drill cuttings at the surface after they have traveled over a relatively long range of depth. This problem typically occurs in reservoir zones where rock properties such as grain size, sorting, and clay content show significant variations (Lewis et al. 2006). Because of this, the method is highly subjective and inaccurate.
Geoscientists also obtain permeability measurements from core samples during a special core analysis process using specialized equipment (Jenkins 1987). Similar to the same method of porosity estimation, this analysis process diminishes the time value of the information and limits it to post-drilling decisions only. They also require enormous time, huge costs, and specialized labor because of the length of time and the specialized equipment used. In the absence of core data, attempts have also been made to estimate permeability indirectly from well logs using correlations such as Timur (1968) and Coates and Denoo (1981) as well as machine learning (Al-Anazi and Gates 2012). This and the methods mentioned earlier are only after the fact.
Because permeability is controlled by such factors as pore size and pore-throat geometry, as well as porosity, the amount of gas liberated from the formation could have a direct relationship with pore geometry. Predicting permeability from mud gas data while drilling also will overcome these challenges, possibly reducing the cost of coring, increase the time value of its utility while drilling, and assisting in real-time decision-making. If also necessary, drill cuttings lithology can be integrated to increase accuracy.
Real-Time Pore Pressure Prediction From Mud Gas Data
Apart from porosity and permeability, there is also a potential opportunity to estimate, while drilling, the pore pressure of a reservoir from the real-time integration of drilling and mud gas data. Geoscientists and engineers typically are able to correlate variations in pore pressure, particularly in transition zones (from normal to geopressures), with various real-time measurable attributes. Some of the methods used to do this include “D exponent,” connection gas, and cuttings analysis (Abass 2020). Real-time formation pressure estimation will allow, at a minimum, more frequent calibration of pressure models. This is in addition to providing valuable input for work flows that help identify and address geohazards such as kicks, overpressures, and blowouts in real time.
Real-Time Grain Size Prediction From Mud Gas Data
For grain size estimation, the requirement of reservoir rock fragments with preserved pore systems creates a major disadvantage for drill cuttings analysis while drilling. This problem becomes worse when drilling poorly consolidated or unconsolidated sands. In addition, the use of certain bit designs such as polycrystalline diamond compact bits reduces most of the collected and observed cuttings to fine-sized fragments, thereby reducing the chance for a realistic grain size estimation (Petrowiki 2021). Among the latest efforts to estimate reservoir rock grain size indirectly is by advanced image processing (Stewart et al. 2009) and using wireline logs (Anifowose et al. 2020). Exploring the relationship between grain size and sorting on one hand and pore geometry, grain size, and the volume of liberated gas on the other, grain size can be related to mud gas data.
In the light of this, having the capability to predict grain size from mud gas data, in real time, will help to overcome these challenges and add a lot more value to this important reservoir property.
Real-Time Lithology Prediction From Mud Gas Data
In a similar fashion, geologists traditionally obtain lithology from the analysis of drill cuttings (in near real time) and core samples (after the fact). Both methods are subjective and prone to error. There are also issues with time lag and depth match with the mud gas data. The latter especially takes time on the scale of months, requires specialized labor, and is subjective to human expertise and condition. Direct estimation of lithology from certain logs such as gamma ray, spontaneous potential, caliper, density, and neutron porosity, taken individually or combined, is not always accurate (Hancock 1992). The more recent approaches to estimating lithology from wireline logs using statistical methods (Busch et al. 1987; Ward and Waltho 1988), empirical approach using the Walsh transform (Maiti and Tiwari 2005), and machine learning (Delfiner et al. 1987; Hsieh et al. 2005; Lopes et al. 2019; Bressan et al. 2020) are equally after the fact. Therefore, they will not meet the real-time needs. Having the capability to predict lithology from mud gas data in real time would not only help overcome these challenges but also would add a lot more time value to its utility.
If that can be achieved for lithology, then it easily can be extended to reservoir rock facies classification and bed boundaries identification. The latest attempts of rock facies classification using machine-learning techniques have been driven by wireline logs (Qi and Carr 2006; Duboisa et al. 2007) while bed boundaries identification has been studied using an extended wavelet analysis (Davisa and Christensen 2013).
Real-Time Hydrocarbon Fraction Prediction From Mud Gas Data
Hydrocarbon fraction is the proportion of hydrocarbon component in a sample of reservoir fluid (Vladislav et al. 2019). Petrophysicists typically compute it from wireline logs. Similar to porosity and permeability, this reservoir property is not available until long after completing and analyzing the drilling process and wireline logging data, respectively. Because this reservoir property is computed from wireline logs, its availability and accuracy depend largely on the quality of wireline logs. We know through Serra (1984) and Kobr (2011) that wireline logs may not be available at certain intervals because of tool failure or bad borehole conditions. All these limitations in wireline logs will affect the availability and quality of hydrocarbon fraction, if the former is used to estimate the latter.
Therefore, predicting hydrocarbon fraction from real-time mud gas data using machine learning will not only overcome the limitations imposed on it by wireline logs but also will increase its time value. This has the potential to create an opportunity for hydrocarbon fraction estimation while drilling.
In addition to the real-time porosity, permeability, pore pressure, grain size, lithology, and hydrocarbon fraction estimation opportunities discussed, extensive investigations also can be conducted on other possibilities such as real-time drill bit selection, dipole shear sonic imager (DSI) logs (compressional, shear, and Stoneley wave velocities), formation temperature, and formation resistivity predictions. Drill bit selection has been modeled with wireline logs using artificial neural networks (ANN) (Yilmaz et al. 2002). DSI logs have been estimated indirectly from conventional wireline logs using machine learning (Rajabi et al. 2010). One of the earliest works to simulate formation temperature used well geometry, fluid and flow characteristics, and initial formation temperature as input parameters (Garcia et al. 1998, Santoyo et al. 2000) while more recent efforts used ANN (Bassam 2010), an empirical derivation (Ricard and Chanu 2013), and a rational polynomial method (Wong-Loya et al. 2015). Estimations of formation resistivity have been attempted using a quasiexponential function with a 1D spline interpolation (Dutta 1997).
Exploring the possibility of estimating these drill bit and reservoir properties would help add more time value for them by making them available while drilling. The capability to make intelligent decisions on bits selection while drilling will reduce costs and save money. Obtaining DSI logs, formation temperature, and petrophysical logs such as resistivity will assist and enhance critical decisions in real time. Such decisions will improve safety, increase exploration success rates, help identify on-time efficient recovery methods, and reduce cost.
Summary and Conclusion
This article appraised the prevailing use of wireline data as a significant resource for indirect estimation of most reservoir properties. We advocated that the same or a greater level of utility be extended to mud gas data. We observed that, despite mud gas data being rich, abundant, and available in almost all wells, it has been grossly underused. Its current utility is limited to fluid typing, formation evaluation, and interwell geological and petrophysical correlation. Leveraging the power of machine learning, big data, processing, and storage technologies, and considering the prevailing wave of digital transformation across the petroleum industry, we attempted to provide answers to the following questions and suggest the exploration of such possibilities:
- Can we get more values from mud gas data?
- Can we estimate reservoir properties such as porosity, permeability, grain size, water saturation, and hydrocarbon fraction in real time from mud gas data?
- Can we evolve a new concept of reservoir characterization while drilling or real-time reservoir characterization leveraging the rich mud gas data?
Leveraging the power of machine learning and its capability to quickly explore various relations and modeling possibilities, the answer to these questions is definitely yes.
We hope that this article succeeds in highlighting the importance of increasing the utility of mud gas data beyond the current limited window to almost unlimited opportunities. We also hope that it will open a promising ground for a rich discussion on this subject. It ultimately has the potential to contribute to accelerating the digital transformation of the industry through more effective use of abundantly available and rich data such as the mud gas.
We would like to acknowledge the help of our colleague, Christopher Ayadiuno, in reviewing the initial draft of this manuscript. His objective and critical comments added much value and increased the quality.
Abass, A.E., Teama, M.A., Kassab, M.A., and Elnaggar, A.A. 2020. Integration of Mud Logging and Wire‑Line Logging To Detect Overpressure Zones: A Case Study of Middle Miocene Kareem Formation in Ashrafi Oil Field, Gulf of Suez, Egypt. Journal of Petroleum Exploration and Production Technology 10: (515–535).
Al-Anazi, A.F. and Gates, I.D. 2012. Support Vector Regression To Predict Porosity and Permeability: Effect of Sample Size. Computers & Geosciences 39: 64–76.
Anifowose, F., Shahrani, S., and Mezghani, M. 2020. Clastic Reservoir Rock Grain Size Estimation from Wireline Logs Using a Random Forest Model: Initial Results. Proc., EAGE Digitalization Conference and Exhibition, Vienna, Austria, 30 November–3 December.
Bassam, A., Santoyoa, E., Andaverde, J., Hernandez, J.A., and Espinoza-Ojeda, O.M. 2010. Estimation of Static Formation Temperatures in Geothermal Wells by Using an Artificial Neural Network Approach. Computers & Geosciences 36: 1,191–1,199.
Beeunas, M.A., Baskin, D.K., Schoell, M. 1999. Application of Gas Geochemistry for Reservoir Continuity Assessment and Identification of Fault Seal Breakdown, South Marsh Island 61, Gulf of Mexico. AAPG Hedberg Research Conference on Natural Gas Formation and Occurrence, Colorado.
Bressan, T.S., de Souza, M.K., Girelli, T.J., and Junior, F.C. 2020. Evaluation of Machine Learning Methods for Lithology Classification using Geophysical Data. Computers & Geosciences 139: 1–13.
Busch, J.M., Fortney, W.G., and L.N. Berry. Determination of Lithology From Well Logs by Statistical Analysis. SPE Form Eval 2 (1987): 412–418. http://dx.doi.org/10.2118/14301-PA.
Coates, G., and Denoo, S. 1981. The Productibility Answer Product. Schlumberger Tech Rev 29 (2): 54–63.
Davisa, A.C., and Christensen, N.B. 2013. Derivative Analysis for Layer Selection of Geophysical Borehole Logs. Computers & Geosciences 60: 34–40.
Delfiner, P., Peyret, O., and Serra, O. 1987. Automatic Determination of Lithology From Well Logs. SPE Form Eval 2: 303–310. http://dx.doi.org/10.2118/13290-PA.
Duboisa, M.K., Bohlinga, G.C., Chakrabarti, S. 2007. Comparison of Four Approaches to a Rock Facies Classification Problem. Computers & Geosciences 33: 599–617.
Dunham, R.J. 1962. Classification of Carbonate Rocks According to Depositional Texture. In Classification of Carbonate Rocks. W.E. Ham ed., Memoir 1, 108–121, AAPG, 1962.
Dutta, D.J. 1997. EXTRA: A Fortran Program To Calculate Resistivity Departure Curves With a Quasi-Exponential Transition Zone. Computers & Geosciences 23 (3): 251–257.
Garcia, A., Hernandez, I., Espinosa, G., and Santoyo, E. 1998. TEMLOPI: A Thermal Simulator for Estimation of Drilling Mud and Formation Temperatures During Drilling of Geothermal Wells. Computers & Geosciences 24 (5): 465–477.
Hancock, N.J. 1992. Quick-Look Lithology From Logs: Part 4. Wireline Methods, Development Geology Reference Manual, Oil—Methodology and Concepts. Methods in Exploration, pp. 174–179.
Haworth, J.H., Sellens, M., and Whittaker, A. 1985. Interpretation of Hydrocarbon Shows Using Light (C1–C5) Hydrocarbon Gases From Mud-Log Data. The American Association of Petroleum Geologists Bulletin 69 (8): 1,305–1,310.
Hill D.G. 2017. Formation Evaluation. Hsu C.S. and Robinson P.R., eds. Springer Handbook of Petroleum Technology. Springer Handbooks.
Hsieh, B., Lewis, C., and Lin, Z. 2005. Lithology Identification of Aquifers From Geophysical Well Logs and Fuzzy Logic Analysis: Shui-Lin Area, Taiwan. Computers & Geosciences 31: 263–275.
Jenkins, R. E. 1987. Typical Core Analysis of Different Formations. Bradley, H.B., ed., Petroleum Engineering Handbook, Society of Petroleum Engineers.
Jun, Y. and Minglu, W. 2011. Software Programming of Streamline Numerical Well Testing Interpretation. Streamline Numerical Well Test Interpretation, Gulf Professional Publishing.
Kobr, M., Geophysical Well Logging. Gupta H.K. eds., Encyclopedia of Solid Earth Geophysics. Encyclopedia of Earth Sciences Series. Springer.
Lewis, M.A., Cheney, C.S., and Ódochartaigh, B.É. 2006. Guide to Permeability Indices. British Geological Survey Open Report, CR/06/160N.
Lopes, D.M.R. and Andrade, A.J.N. 2019. Lithology Identification on Well Logs by Fuzzy Inference. Journal of Petroleum Science and Engineering 180: 357–368.
Low, J.W. 1951. Examination of Well Cuttings. Quarterly of the Colorado School of Mines 46 (4).
Maiti, S., and Tiwari, R.K. 2005. Automatic Detection of Lithologic Boundaries Using the Walsh Transform: A Case Study From the KTB Borehole. Computers & Geosciences 31: 949–955.
Pierson, R.M. 2017. The ELLA GRA Process—Concepts and Methods for the Prediction of Reservoir Hydrocarbon Type Using Ratios of Gas Chromatography C1–C5 Gases. Article #42122, Search and Discovery, presented at the 2017 AAPG Rocky Mountain Section Meeting, Billings, Montana, June 25–28.
Qi, L., and Carr, T.R. 2006. Neural Network Prediction of Carbonate Lithofacies From Well Logs, Big Bow, and Sand Arroyo Creek Fields, Southwest Kansas. Computers & Geosciences 32: 947–964.
Rajabi, M., Bohloli, B., and Ahangar, E.G. 2010. Intelligent Approaches for Prediction of Compressional, Shear, and Stoneley Wave Velocities From Conventional Well Log Data: A Case Study from the Sarvak Carbonate Reservoir in the Abadan Plain (Southwestern Iran). Computers & Geosciences 36: 647–664.
Ricard, L.P., and Chanu, J. 2013. GeoTemp 1.0: A MATLAB-Based Program for the Processing, Interpretation, and Modeling of Geological Formation Temperature Measurements. Computers & Geosciences 57: 197–207.
Santoyo, E., Garcia, A., Espinosa, G., Hernandez, I., and Santoyo, S. 2000. STATIC_TEMP: A Useful Computer Code for Calculating Static Formation Temperatures in Geothermal Wells. Computers & Geosciences 26: 201–217.
Serra, O. 1984. Fundamentals of Well-Log Interpretation. Developments in Petroleum Science, Elsevier.
Spivey, J.P., and Lee, J. 2013. Applied Well Test Interpretation. SPE Textbook Series, Volume 13. Society of Petroleum Engineers.
Stewart, L.K., Kostylevc, V.E., and Orpin, A.R. 2009. Windows-Based Software for Optimizing Entropy-Based Groupings of Textural Data. Computers & Geosciences 35: 1,552–1,556.
Swanson, R.G. 1981. Sample Examination Manual. Tulsa, Oklahoma: American Association of Petroleum Geologists.
Timur, A. 1968. An Investigation of Permeability, Porosity, and Residual Water Saturation Relationships for Sandstone Reservoirs. The Log Analyst 9 (4).
Torlov, V.S., Bonavides, C.S., Al-Qubaisi, K.A., and Fernandes, A.K. 2019. Utilization of Mud Logs for Oil Fraction Quantification in Transition Zones and Low Resistivity Reservoirs. Paper presented at the Abu Dhabi International Petroleum Exhibition and Conference, Abu Dhabi, UAE, November 2019. http://dx.doi.org/10.2118/197403-MS.
Walstrom, J.E. 1972. A Review of Formation Evaluation. Paper presented at the SPE California Regional Meeting, Bakersfield, California, November. http://dx.doi.org/10.2118/4187-MS.
Ward, C.R., and Waltho, A.E. 1988. A BASIC Program for In-Field Entry of Lithologic Descriptions in Borehole Logs to a Hand-Held Portable Computer System. Computers & Geosciences 14 (1): 83–97.
Whittaker, A. 1991. Mud Logging Handbook: Englewood Cliffs, NJ, Prentice-Hall.
Wong-Loya, J.A., Santoyo, E., Andaverde, J.A., Quiroz-Ruiz, A. 2015. RPM-WEBBSYS: A Web-Based Computer System To Apply the Rational Polynomial Method for Estimating Static Formation Temperatures of Petroleum and Geothermal Wells. Computers & Geosciences 85: 45–59.
Yilmaz, S., Demircioglu, C., and Akin, S. 2002. Application of Artificial Neural Networks to Optimum Bit Selection. Computers & Geosciences 28: 261–269.