Abstract
In civil and mining engineering, three-dimensional in situ stress states are commonly measured using overcoring (OC) techniques. With these techniques, stresses are usually estimated by classical least-squares regression analysis of measured OC data. However, the estimated stresses are uncertain and may even be unreliable, due to factors such as rock heterogeneity, measurement errors and inadequacy of the regression model. Quantifying such uncertainty is crucial, as doing so both permits quantitative assessment of the reliability of the measured stress state and facilitates application of probabilistic design approaches in rock engineering such as reliability-based design. The classical approach to OC stress estimation suffers various limitations in this respect, particularly the failure in quantifying uncertainty in estimates of principal stresses, and the inability to improve stress estimation by incorporating stress information from other sources (say, stress states measured at nearby locations and orientation of the major principal stress as determined from observations of borehole breakouts). To overcome these limitations, this paper proposes a novel Bayesian approach to OC data analysis that probabilistically quantifies uncertainty in stress estimations and permits formal incorporation of additional stress information in forms of prior distributions. It also discusses the challenges faced in developing the informative prior distributions that are required to allow incorporation of additional stress information.
Similar content being viewed by others
References
Amadei B (2000) CSIRA: a computer program to determine in situ stresses by overcoring in anisotropic rock with CSIR-type triaxial strain cells
Amadei B (1984) In situ stress measurements in anisotropic rock. Int Jo Rock Mech Min Scien Geomech Abstr 21(6):327–338. https://doi.org/10.1016/0148-9062(84)90365-6
Amadei B, Stephansson O (1997) Rock stress and its measurement. Springer, Dordrecht
Ask D (2003a) Analysis of overcoring stress data at the Äspö HRL, Sweden - analysis of overcoring rock stress measurements preformed using the CSIRO HI. Tech. rep., Swedish Nuclear Fuel and Waste Management Co, Stockholm, Sweden
Ask D (2006) Measurement-related uncertainties in overcoring data at the Äspö HRL, Sweden. Part 2: Biaxial tests of CSIRO HI overcore samples. Int Jo Rock Mech Min Sci 43(1):127–138 https://doi.org/10.1016/j.ijrmms.2005.05.012
Ask D (2017) Methodology for determination of the complete stress tensor and its variation versus depth based on overcoring rock stress data. In: Feng XT (ed) Rock Mechanics and Engineering Volume 1, CRC Press, London, UK, chap 8, pp 245–264
Ask D (2003) Evaluation of measurement-related uncertainties in the analysis of overcoring rock stress data from Äspö HRL, Sweden: a case study. Int J Rock Mech Min Sci 40(7–8):1173–1187. https://doi.org/10.1016/S1365-1609(03)00114-X
Ask D, Cornet FH, Stephansson O (2002) Analysis of overcoring stress data at the Äspö HRL, Sweden - analysis of overcoring rock stress measurements preformed using the Borre Probe. Tech. rep., Swedish Nuclear Fuel and Waste Management Co, Stockholm, Sweden
Baecher GB, Christian JT (2005) Reliability and statistics in geotechnical engineering. John Wiley & Sons, Chichester
Barnett V (1976) The ordering of multivariate data. J R Stat Soc Ser A (Gen) 139(3):318–344. https://doi.org/10.2307/2344839
Bozorgzadeh N, Bathurst RJ (2019) Bayesian model checking, comparison and selection with emphasis on outlier detection for geotechnical reliability-based design. Comput Geotech 116:103181. https://doi.org/10.1016/j.compgeo.2019.103181
Bozorgzadeh N, Escobar MD, Harrison JP (2018) Comprehensive statistical analysis of intact rock strength for reliability-based design. Int J Rock Mech Mini Sci 106:374–387. https://doi.org/10.1016/j.ijrmms.2018.03.005
Bozorgzadeh N, Harrison JP, Escobar MD (2019) Hierarchical Bayesian modelling of geotechnical data: application to rock strength. Géotechnique 69(12):1056–1070. https://doi.org/10.1680/jgeot.17.P.282
Carlin BP, Louis TA (2009) Bayesian methods for data analysis, 3rd edn. CRC Press, Boca Raton
Chaudhuri P (1996) On a geometric notion of quantiles for multivariate data. J Am Stat Assoc 91(434):862–872. https://doi.org/10.1080/01621459.1996.10476954
Christiansson R, Hudson JA (2003) ISRM suggested methods for rock stress estimation-Part 4: quality control of rock stress estimation. Int J Rock Mech Min Sci 40(7):1021–1025. https://doi.org/10.1016/j.ijrmms.2003.07.004
Chung Y, Gelman A, Rabe-Hesketh S, Liu J, Dorie V (2015) Weakly informative prior for point estimation of covariance matrices in hierarchical models. J Educ Behav Stat 40(2):136–157. https://doi.org/10.3102/1076998615570945
Ditlevsen O, Madsen HO (1996) Structural reliability methods. Wiley, New York
Draper NR, Smith H (1998) Applied regression analysis, 3rd edn. John Wiley & Sons, Hoboken
Duncan Fama M, Pender M (1980) Analysis of the hollow inclusion technique for measuring in situ rock stress. Int J Rock Mech Min Sci Geomech Abstr 17(3):137–146. https://doi.org/10.1016/0148-9062(84)90365-60
Duvall WI, Aggson JR (1980) Least squares calculation of horizontal stresses from more than three diametral deformations in vertical boreholes. Tech. Rep. RI 8414, US Bureau of Mines Report of Investigation, Washingtion
Eaton JW, Bateman D, Hauberg S, Wehbring R (2020) GNU Octave version 5.2.0 manual: a high-level interactive language for numerical computations. https://doi.org/10.1016/0148-9062(84)90365-61. Accessed 18 Feb 2020
Fairhurst C (2003) Stress estimation in rock: a brief history and review. Int J Rock Mech Min Sci 40(7–8):957–973. https://doi.org/10.1016/j.ijrmms.2003.07.0022
Feng Y, Bozorgzadeh N, Harrison JP (2020) Bayesian analysis for uncertainty quantification of in situ stress data. Int J Rock Mech Min Sci. https://doi.org/10.1016/j.ijrmms.2020.104381
Gelman A (2006) Prior distributions for variance parameters in hierarchical models. Bayesian Anal 1(3):515–534. https://doi.org/10.1016/0148-9062(84)90365-63
Gelman A, Rubin DB (1992) Inference from iterative simulation using multiple sequences. Stat Sci 7(4):457–472. https://doi.org/10.1016/0148-9062(84)90365-64
Gelman A, Jakulin A, Pittau MG, Su YS (2008) A weakly informative default prior distribution for logistic and other regression models. Ann Appl Stat 2(4):1360–1383. https://doi.org/10.1016/0148-9062(84)90365-65
Gelman A, Carlin JB, Stern HS, Dunson DB, Vehtari A, Rubin DB (2013) Bayesian data analysis, 3rd edn. Chapman and Hall/CRC, Boca Raton
Gelman A, Simpson D, Betancourt M (2017) The prior can often only be understood in the context of the likelihood. Entropy 19(10):555. https://doi.org/10.1016/0148-9062(84)90365-66
Gelman A, Hill J (2006) Data analysis using regression and multilevel/hierarchical models. Cambridge University Press, Cambridge https://doi.org/10.1017/CBO9780511790942
Geweke JF (1991) Evaluating the accuracy of sampling-based approaches to the calculation of posterior moments. Tech. rep, Federal Reserve Bank of Minneapolis, Research Department, Minneapolis, MN
Gray WM, Toews NA (1975) Analysis of variance applied to data obtained by means of a six element borehole deformation gauge for stress determination. In: Proceedings of the 15th US Symposium on Rock Mechanics, ASCE, South Dakota, pp 323–356
Hakala M, Hudson JA, Christiansson R (2003) Quality control of overcoring stress measurement data. Int J Rock Mech Min Sci 40(7–8):1141–1159. https://doi.org/10.1016/0148-9062(84)90365-67
Hakala M, Sjöberg J (2006) A methodology for interpretation of overcoring stress measurements in anisotropic rock. Tech. rep, Posiva Oy, Olkiluoto, Finland
Hakami E, Hakami H, Cosgrove J (2002) Strategy for a rock mechanics site descriptive model - development and testing of an approach to modelling the state of stress. Tech. rep, Swedish Nuclear Fuel and Waste Management Co, Stockholm, Sweden
Hudson JA, Cornet FH, Christiansson RC (2003) ISRM suggested methods for rock stress estimation—part 1: strategy for rock stress estimation. Int J Rock Mech Min Sci 40(7–8):991–998. https://doi.org/10.1016/0148-9062(84)90365-68
Kruschke JK (2014) Doing Bayesian data analysis: a tutorial with R, JAGS, and Stan, 2nd edn. Academic Press,
Latham JP, Xiang J, Belayneh M, Nick HM, Tsang CF, Blunt MJ (2013) Modelling stress-dependent permeability in fractured rock including effects of propagating and bending fractures. Int J Rock Mech Min Sci 57:100–112. https://doi.org/10.1016/j.ijrmms.2012.08.0029
Leeman ER (1968) The determination of the complete state of stress in rock in a single borehole-Laboratory and underground measurements. Int J Rock Mech Mini Scie Geomech Abstr 5(1):31–38. https://doi.org/10.1016/S1365-1609(03)00114-X0
Lemaire M (2009) Structural reliability. John Wiley & Sons, Hoboken
Lemoine NP (2019) Moving beyond noninformative priors: why and how to choose weakly informative priors in Bayesian analyses. Oikos 128(7):912–928. https://doi.org/10.1016/S1365-1609(03)00114-X1
Liu RY, Singh K (1992) Ordering directional data: concepts of data depth on circles and spheres. Ann Stat 20(3):1468–1484
Liu RY, Parelius JM, Singh K (1999) Multivariate analysis by data depth: descriptive statistics, graphics and inference. Ann Stat 27(3):783–858
Ljunggren C, Chang Y, Janson T, Christiansson R (2003) An overview of rock stress measurement methods. Int J Rock Mech Min Sci 40(7–8):975–989. https://doi.org/10.1016/S1365-1609(03)00114-X2
Lunn D, Jackson C, Best N, Thomas A, Spiegelhalter D (2012) The BUGS book: a practical introduction to Bayesian analysis. CRC Press, Hoboken
Martin CD (1990) Characterizing in situ stress domains at the AECL Underground Research Laboratory. Can Geotech J 27(5):631–646. https://doi.org/10.1016/S1365-1609(03)00114-X3
Martin CD, Christiansson R (1991) Overcoring in highly stressed granite-the influence of microcracking. Int J Rock Mech Min Sci Geomech Abstr 28(1):53–70. https://doi.org/10.1016/S1365-1609(03)00114-X4
Martin CD, Christiansson RC (1991) Overcoring in highly stressed granite: comparison of USBM and modified CSIR devices. Rock Mech Rock Eng 24(4):207–235. https://doi.org/10.1007/BF010450325
McElreath R (2019) Statistical rethinking: a Bayesian course with example in R and Stan, 2nd edn. Chapman and Hall/CRC, New York, NY, https://doi.org/10.1201/9781315372495
Moore DS, McCabe GP, Craig BA (2016) Introduction to the practice of statistics, 9th edn. W. H Freeman, New York
Moos D, Zoback MD (1990) Utilization of observations of well bore failure to constrain the orientation and magnitude of crustal stresses: application to continental, Deep Sea Drilling Project, and Ocean Drilling Program boreholes. J Geophys Res 95(B6):9305–9325. https://doi.org/10.1016/S1365-1609(03)00114-X6
R Core Team (2018) R: A language and environment for statistical computing. https://doi.org/10.1016/S1365-1609(03)00114-X7. Accessed 20 Oct 2019
Serfling R (2004) Nonparametric multivariate descriptive measures based on spatial quantiles. J Stat Plan Inference 123(2):259–278. https://doi.org/10.1016/S1365-1609(03)00114-X8
Sjöberg J, Klasson H (2003) Stress measurements in deep boreholes using the Borre (SSPB) probe. Int J Rock Mech Min Sci 40(7–8):1205–1223. https://doi.org/10.1016/S1365-1609(03)00114-X9
Sjöberg J, Christiansson R, Hudson JA (2003) ISRM Suggested Methods for rock stress estimation—Part 2: overcoring methods. Int J Rock Mech Min Sci 40(7–8):999–1010. https://doi.org/10.2307/23448390
Sjöberg J, Lindfors U, Perman F, Ask D (2005) Evaluation of the state of stress at the Forsmark site. Tech. rep, Swedish Nuclear Fuel and Waste Management Co, Stockholm, Sweden
Stan Development Team (2018) RStan: the R interface to Stan. https://doi.org/10.2307/23448391. Accessed 21 Oct 2019
Stephansson O, Zang A (2012) ISRM Suggested Methods for rock stress estimation—Part 5: establishing a model for the in situ stress at a given site. Rock Mech Rock Eng 45(6):955–969. https://doi.org/10.1007/s00603-012-0270-x2
Tukey JW (1975) Mathematics and the picturing of data. International Congress of Mathematicians. Vancouver, Canada, pp 523–531
Wasserman L (2004) All of statistics: a concise course in statistical inference. Springer Texts in Statistics, Springer, New York, New York, NY, https://doi.org/10.1007/978-0-387-21736-9
Wiprut D, Zoback M, Hanssen TH, Peska P (1997) Constraining the full stress tensor from observations of drilling-induced tensile fractures and leak-off tests: application to borehole stability and sand production on the Norwegian margin. Int J Rock Mech Min Sci 34(3–4):1–365. https://doi.org/10.2307/23448393
Worotnicki G (1993) CSIRO triaxial stress measurement cell. In: Hudson JA (ed) Comprehensive Rock Engineering, Pergamon Press, Oxford, vol 3, pp 329–394
Zang A, Stephansson O (2010) Stress field of the earth’s crust. Springer, Dordrecht
Zoback MD (2010) Reservoir geomechanics. Cambridge University Press, Cambridge
Zoback MD, Moos D, Mastin L, Anderson RN (1985) Well bore breakouts and in situ stress. J Geophys Res Solid Earth 90(B7):5523–5530. https://doi.org/10.2307/23448394
Zoback MD, Barton CA, Brudy M, Castillo DA, Finkbeiner T, Grollimund BR, Moos DB, Peska P, Ward CD, Wiprut DJ (2003) Determination of stress orientation and magnitude in deep wells. Int J Rock Mech Min Sci 40(7–8):1049–1076. https://doi.org/10.1016/j.ijrmms.2003.07.0015
Zou DH (1995) Statistical regression applied to borehole strain measurements data analysis. Geotech Geol Eng 13(1):17–27. https://doi.org/10.2307/23448396
Zuo Y, Serfling R (2000) General notions of statistical depth function. Ann Stat 28(2):461–482
Acknowledgements
We acknowledge the financial support from the Chinese Scholarship Council and NSERC Canada (Discovery Grant RGPIN-2016-06722). We also thank Prof. Daniel Ask of the Luleå University of Technology for his comprehensive compilation of the raw overcoring stress data that made the analyses presented here possible, and Prof. Michael Escobar of the University of Toronto for his valuable and enlightening comments on numerous aspects of Bayesian inference.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors declare that they have no conflict of Interest.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
Appendix A: Overcoring Stress Transformation
The transformation of stress tensors from one (e.g. global) Cartesian coordinate system to another (e.g. local) coordinate system is written as:
where \(\varvec{\mathbf {R}}\) denotes a \(3 \times 3\) transformation matrix of stress tensors between the two Cartesian coordinate systems, and can be conveniently calculated using the standard basis of the two coordinate systems (i.e. unit vectors of the axes of a Cartesian coordinate system) by:
In the context of overcoring analysis, given the global geographical coordinate system (x East, y North and z vertically upwards) and the local coordinate system attached to the overcoring borehole with trend \(\alpha\) and plunge \(\beta\), the transformation matrix \(\varvec{\mathbf {R}}\) is obtained as:
Substituting Eq. (13) into Eq. (11) and rearranging in a stress vector form give:
where \(\varvec{\mathbf {T}}\) is the transformation matrix of stress vectors from the global geographical coordinate system to the local coordinate system attached to the overcoring borehole.
Appendix B: Basics of Bayesian Statistics
1.1 B.1 Bayes’ Rule
Similar to frequentist methods, Bayesian methods are used to make statistical inference based on observed data. A fundamental theoretical difference between the two approaches is that frequentist methods treat observations (i.e., data) as random variables, and the unobservable quantities we wish to learn about (i.e., statistical parameters such as mean, standard deviation and covariance) as fixed unknowns, while the Bayesian methods treat both the data and statistical parameters as random variables (Gelman et al. 2013).
To make Bayesian inference for a set of statistical parameters \(\theta\) (e.g., the mean and variance of a population distribution), we begin by defining a full joint probability distribution for all observable quantities y and unobservable quantities \(\theta\) as
where \(f(\theta )\) is known as the prior distribution and reflects the state of knowledge about the parameters \(\theta\) in the form of a probability density function prior to observing the data y. The term \(f(y|\theta )\) is known as the likelihood function, and is the probability distribution of the data y assuming they arise from a statistical model with the parameters \(\theta\).
Normalizing by f(y), a constant that depends only on the known values of the data y to ensure unity of posterior distributions, leads to Bayes’ rule
where \(f(\theta |y)\) is known as the posterior distribution and reflects the updated state of knowledge about the parameters \(\theta\) after observing the data y in terms of probability. The posterior distribution represents the probability of parameters \(\theta\) over all plausible values, from which the usual point estimate (i.e. posterior mean) and probabilistic measure of uncertainty (e.g. 95% credible intervals) regarding the parameters \(\theta\) can be readily made.
An important feature of the Bayesian approach is that it allows commonsense probabilistic interpretations of statistical conclusions. For instance, a Bayesian credible interval can be directly regarded as an interval that a specified probability of containing the true value of an unknown parameter of interest, in contrast to a frequentist confidence interval that must be strictly interpreted only in relation to a hypothetical sequence of similar inferences that might be made in repeated practice (Gelman et al. 2013).
1.2 B.2 Prior Distribution
A distinctive advantage of the Bayesian approach over the classical frequentist approach is the use of the prior distribution \(f(\theta )\) for statistical inference. This allows additional information from other sources to be logically incorporated into statistical analyses. Prior distributions can be summarized from relevant historical data, engineering judgment or a combination of both. For example, the posterior distribution from one relevant Bayesian analysis can be used directly, or with some modification, as a prior distribution for a subsequent analysis.
Prior distributions that are based on information (e.g. historical data or engineering judgment) other than the immediate data at hand are known as informative priors, and can strongly influence the posterior distribution through Eq. (16). On the other hand, priors that have no preference for any particular parameter values over a wide range are called vague (or diffuse, flat) priors, such as a uniform (0, \(10^6\)) or a half-normal (0, \(5\times 10^5\)) for a standard deviation parameter. Vague priors are often used as the default prior choice in the absence of specific prior information. However, in cases of limited data and/or complex models, such default priors can sometimes be undesirably informative in that they assign approximately equal probability density to an unrealistically wide range of parameter values, thereby pulling the posterior distributions towards extreme values and hence biasing the inferences (Gelman et al. 2013, 2017).
For the purpose of more robust Bayesian inference, a related prior type—weakly informative priors which are based on the general understanding of the data at hand (i.e. contextual/background knowledge)—has recently been strongly advocated over vague priors as the default prior choice (Gelman et al. 2013, 2008; Chung et al. 2015; Gelman 2006; Lemoine 2019; Gelman et al. 2017). The principled weakly informative priors that carry some amount of realistic information can regularize the posterior distributions and prevent unrealistic posterior parameter values. As weakly informative priors generally allow the data to dominate the posterior distributions while providing regularization, adopting them generally results in statistical inferences similar to those obtained with frequentist methods relying solely on the data. Arguably, for Bayesian analysis of geotechnical data weakly informative priors should be used as the default prior, considering the usual lack of sufficient data in geotechnical engineering.
Appendix C: Stan Model
Stan Model
Rights and permissions
About this article
Cite this article
Feng, Y., Harrison, J.P. & Bozorgzadeh, N. A Bayesian Approach for Uncertainty Quantification in Overcoring Stress Estimation. Rock Mech Rock Eng 54, 627–645 (2021). https://doi.org/10.1007/s00603-020-02295-w
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00603-020-02295-w