Abstract
Real geo-data are three-dimensional (3D) and spatially varied, but measurements are often sparse due to time, resource, and/or technical constraints. In these cases, the quantities of interest at locations where measurements are missing must be interpolated from the available data. Several powerful methods have been developed to address this problem in real-world applications over the past several decades, such as two-point geo-statistical methods (e.g., kriging or Gaussian process regression, GPR) and multiple-point statistics (MPS). However, spatial interpolation remains challenging when the number of measurements is small because a suitable covariance function is difficult to select and the parameters are challenging to estimate from a small number of measurements. Note that a covariance function form and its parameters are key inputs for some methods (e.g., kriging or GPR). MPS is a non-parametric simulation method that combines training images as prior knowledge for sparse measurements. However, the selection of a suitable training image for continuous geo-quantities (e.g., soil or rock properties) faces certain difficulties and may become increasingly complicated when the geo-data to be interpolated are high-dimensional (e.g., 3D) and exhibit non-stationary (e.g., with unknown trends or non-stationary covariance structure) and/or anisotropic characteristics. This paper proposes a non-parametric approach that systematically combines compressive sensing and variational Bayesian inference for statistical interpolation of 3D geo-data. The method uses sparse measurements and their locations as the input and provides interpolated values at unsampled locations with quantified interpolation uncertainty as the output. The proposed method is illustrated using a series of numerical 3D examples, and the results indicate a reasonably good performance.
Similar content being viewed by others
References
Ang AHS, Tang WH (2007) Probability concepts in engineering: emphasis on applications in civil and environmental engineering. Wiley, New York
Beal MJ (2003) Variational algorithms for approximate Bayesian inference. University of College, London
Bishop CM (2006) Pattern recognition and machine learning. Springer, New York, pp 461–517
Bishop CM, Tipping ME (2000) Variational relevance vector machines. In: Proceedings of the sixteenth conference on uncertainty in artificial intelligence, Morgan Kaufmann Publishers Inc., pp 46–53
Buland A, Kolbjørnsen O, Hauge R, Skjæveland Ø, Duffaut K (2008) Bayesian lithology and fluid prediction from seismic prestack data. Geophysics 73:C13–C21
Caiafa CF, Cichocki A (2013a) Computing sparse representations of multidimensional signals using Kronecker bases. Neural Comput 25:186–220
Caiafa CF, Cichocki A (2013b) Multidimensional compressed sensing and their applications. Wiley Interdiscip Rev Data Min Knowl Disco 3:355–380
Candès EJ, Romberg JK (2005) Signal recovery from random projections. In: SPIE international symposium on electronic imaging: computational imaging III, San Jose, California, pp 76–86
Candès EJ, Wakin MB (2008) An introduction to compressive sampling. IEEE Signal Proc Mag 25:21–30
Cao Z-J, Zheng S, Li D, Phoon K-K (2018) Bayesian identification of soil stratigraphy based on soil behaviour type index. Can Geotech J 56(4):570–586
Ching J, Phoon KK, Wu SH (2016) Impact of statistical uncertainty on geotechnical reliability estimation. J Eng Mech. https://doi.org/10.1061/(ASCE)EM.1943-7889.0001075
Cressie N (1993) Statistics for spatial data. Wiley, New York
Dietrich C, Newsam G (1993) A fast and exact method for multidimensional Gaussian stochastic simulations. Water Resour Res 29(8):2861–2869
Dimitrakopoulos R, Mustapha H, Gloaguen E (2010) High-order statistics of spatial random fields: exploring spatial cumulants for modeling complex non-Gaussian and non-linear phenomena. Math Geosci 42(1):65
Dumitru M (2017). Sparsity enforcing priors in inverse problems via normal variance mixtures: model selection, algorithms and applications. arXiv preprint arXiv:1705.10354
Fenton GA (1999a) Estimation for stochastic soil models. J Geotech Geoenviron Eng 125(6):470–485
Fenton GA (1999b) Random field modeling of CPT data. J Geotech Geoenviron Eng 125(6):486–498
Gilanifar M, Wang H, Sriram LMK, Ozguven EE, Arghandeh R (2019) Multi-task Bayesian spatiotemporal Gaussian processes for short-term load forecasting. IEEE Trans Ind Electron. https://doi.org/10.1109/TIE.2019.2928275
Griffiths DV, Marquez RM (2007) Three-dimensional slope stability analysis by elasto-plastic finite elements. Geotechnique 57(6):537–546
Hack R, Orlic B, Ozmutlu S, Zhu S, Rengers N (2006) Three and more dimensional modelling in geo-engineering. B Eng Geol Environ 65(2):143–153
Hillier MJ, Schetselaar EM, de Kemp EA, Perron G (2014) Three-dimensional modelling of geological surfaces using generalized interpolation with radial basis functions. Math Geosci 46(8):931–953
Hong Y, Wang L, Zhang J, Gao Z (2020) 3D elastoplastic model for fine-grained gassy soil considering the gas-dependent yield surface shape and stress-dilatancy. J Eng Mech 146(5):04020037
Horta A, Correia P, Pinheiro LM, Soares A (2013) Geostatistical data integration model for contamination assessment. Math Geosci 45:575–590
Hu Y, Wang Y, Zhao T, Phoon KK (2020) Bayesian supervised learning of site-specific geotechnical spatial variability from sparse measurements. ASCE-ASME J Risk Uncertain Eng Syst Part A Civ Eng 6(2):04020019
Huang Y, Beck JL, Wu S, Li H (2016) Bayesian compressive sensing for approximately sparse signals and application to structural health monitoring signals for data loss recovery. Probab Eng Mech 46:62–79
Huysmans M, Dassargues A (2009) Application of multiple-point geostatistics on modelling groundwater flow and transport in a cross-bedded aquifer (Belgium). Hydrogeol J 17(8):1901
Ji S, Xue Y, Carin L (2008) Bayesian compressive sensing. IEEE Trans Signal Process 56:2346–2356
Kroese DP, Botev ZI (2015) Spatial process simulation. In: Stochastic geometry, spatial statistics and random fields, Springer, pp 369–404
Kroonenberg PM (2008) Applied multiway data analysis. Wiley, Hoboken
Lamorey G, Jacobson E (1995) Estimation of semivariogram parameters and evaluation of the effects of data sparsity. Math Geol 27(3):327–358
Largueche FZB (2006) Estimating soil contamination with Kriging interpolation method. Am J Appl Sci 3(6):1894–1898
Li J, Heap AD (2011) A review of comparative studies of spatial interpolation methods in environmental sciences: performance and impact factors. Ecol Inform 6(3–4):228–241
Liu LL, Deng ZP, Zhang SH, Cheng YM (2018) Simplified framework for system reliability analysis of slopes in spatially variable soils. Eng Geol 239:330–343
Luo Z, Atamturktur S, Juang CH (2013) Bootstrapping for characterizing the effect of uncertainty in sample statistics for braced excavations. J Geotech Geoenviron Eng 139:13–23
Mariethoz G, Caers G (2015) Multiple-point geostatistics: stochastic modeling with training images. Wiley, London
Matheron G (1963) Principles of geostatistics. Econ Geol 58:1246–1266
Matheron G (1973) The intrinsic random functions and their applications. Adv Appl Probab 5(3):439–468
MathWorks I (2020) MATLAB: the language of technical computing. http://www.mathworks.com/products/matlab/. Accessed 6 May 2020
Moja SS, Asfaw ZG, Omre H (2018) Bayesian inversion in hidden Markov models with varying marginal proportions. Math Geosci 51(4):463–484
Murphy KP (2012) Machine learning: a probabilistic perspective. The MIT Press, London, pp 731–766
Petersen KB (2004) The matrix cookbook. Technical University of Denmark. http://www.cim.mcgill.ca/~dudek/417/Papers/matrixOperations.pdf. Accessed 13 Apr 2019
Phoon KK, Kulhawy FH (1999) Characterization of geotechnical variability. Can Geotech J 36(4):612–624
Rasmussen CE, Williams CKI (2006) Gaussian processes for machine learning. MIT Press, London
Remy N, Boucher A, Wu J (2009) Applied geostatistics with SGeMS: a user’s guide. Cambridge University Press, Cambridge
Salomon D (2007) Data compression: the complete reference. Springer, New York
Schnabel U, Tietje O, Scholz RW (2004) Uncertainty assessment for management of soil contaminants with sparse data. Environ Manag 33(6):911–925
Shekaramiz M, Moon TK, Gunther JH (2017) Sparse Bayesian learning using variational Bayes inference based on a greedy criterion. In 2017 51st Asilomar conference on signals, systems, and computers, pp 858–862
Shekaramiz M, Moon TK, Gunther JH (2019) Exploration vs data refinement via multiple mobile sensors. Entropy 21(6):568
Shi C, Wang Y (2020) Non-parametric and data-driven interpolation of subsurface soil stratigraphy from limited data using multiple point statistics. Can Geotech J. https://doi.org/10.1139/cgj-2019-0843
Sivia D, Skilling J (2006) Data analysis: a Bayesian tutorial. OUP Oxford, New York
Strebelle S (2002) Conditional simulation of complex geological structures using multiple-point statistics. Math Geol 34(1):1–21
Tipping ME (2001) Sparse bayesian learning and the relevance vector machine. J Mach Learn Res 1:211–244
Wang Y, Zhao T (2017a) Bayesian assessment of site-specific performance of geotechnical design charts with unknown model uncertainty. Int J Numer Anal Methods Geomech 41:781–800
Wang Y, Zhao T (2017b) Statistical interpretation of soil property profiles from sparse data using Bayesian compressive sampling. Géotechnique 67:523–536
Wang Y, Huang K, Cao Z (2013) Probabilistic identification of underground soil stratification using cone penetration tests. Can Geotech J 50(7):766–776
Wang Y, Cao Z, Li D (2016) Bayesian perspective on geotechnical variability and site characterization. Eng Geol 203:117–125
Wang H, Wellmann JF, Li Z, Wang X, Liang RY (2017) A segmentation approach for stochastic geological modeling using hidden Markov random fields. Math Geosci 49(2):145–177
Wang X, Wang H, Liang RY, Zhu H, Di H (2018) A hidden Markov random field model-based approach for probabilistic site characterization using multiple cone penetration test data. Struct Saf 70:128–138
Wang Y, Zhao T, Hu Y, Phoon K-K (2019) Simulation of random fields with trend from sparse measurements without detrending. J Eng Mech 145:04018130
Williams CK (1998) Prediction with Gaussian processes: from linear regression to linear prediction and beyond. In: Learning in graphical models, Springer, Dordrecht, pp 599–621
Xiao T, Li DQ, Cao ZJ, Au SK, Phoon KK (2016) Three-dimensional slope reliability and risk assessment using auxiliary random finite element method. Comput Geotech 79:146–158
Yamamoto JK (2008) Estimation or simulation? That is the question. Comput Geosci 12(4):573–591
Yu L, Wei C, Jia J, Sun H (2016) Compressive sensing for cluster structured sparse signals: variational Bayes approach. IET Signal Process 10(7):770–779
Zhang T (2008) Incorporating geological conceptual models and interpretations into reservoir modeling using multiple-point geostatistics. Earth Sci Front 15(1):26–35
Zhang J, Zhang LM, Tang WH (2009) Bayesian framework for characterizing geotechnical model uncertainty. J Geotech Geoenviron Eng 135:932–940
Zhao T, Wang Y (2018a) Interpretation of pile lateral response from deflection measurement data: a compressive sampling-based method. Soils Found 58:957–971
Zhao T, Wang Y (2018b) Simulation of cross-correlated random field samples from sparse measurements using Bayesian compressive sensing. Mech Syst Signal Process 112:384–400
Zhao Q, Zhang L, Cichocki A (2015) Bayesian sparse Tucker models for dimension reduction and tensor completion. arXiv preprint arXiv:1505.02343
Zhao T, Hu Y, Wang Y (2018a) Statistical interpretation of spatially varying 2D geo-data from sparse measurements using Bayesian compressive sampling. Eng Geol 246:162–175
Zhao T, Montoya-Noguera S, Phoon KK, Wang Y (2018b) Interpolating spatially varying soil property values from sparse data for facilitating characteristic value selection. Can Geotech J 55(2):171–181
Acknowledgements
The work described in this paper was supported by grants from the Research Grants Council of the Hong Kong Special Administrative Region, China (Project Nos. CityU 11213117 and CityU 11213119) and the Fundamental Research Funds for the Central Universities. The financial supports are gratefully acknowledged.
Author information
Authors and Affiliations
Corresponding author
Appendices
Appendix A: Construction of B −3Dt
In this appendix, the 3D basis function \( \underline{{\mathbf{B}}}_{t}^{3D} \) is reconstructed from columns of three independent 1D basis function matrices, i.e., B1, B2, and B3, which have dimensions of N1 × N1, N2 × N2, and N3 × N3, respectively. The columns of B1, B2, and B3 represent orthonormal basis functions, e.g., discrete cosine functions. B1, B2, and B3 can be obtained using the formula for discrete cosine functions (see Salomon 2007) or constructed readily using a built-in function in commercial software, such as “dctmtx” in MATLAB (Mathworks 2020), which requires only N1, N2, and N3 as the input. A 3D basis function is then constructed as \( \underline{{\mathbf{B}}}_{i,j,k}^{3D} = {\varvec{b}}_{i}^{1} \otimes {\varvec{b}}_{j}^{2} \otimes {\varvec{b}}_{k}^{3} \) (i = 1, 2, …, N1; j = 1, 2, …, N2; k = 1, 2, …, N3). The subscript of \( \underline{{\mathbf{B}}}_{i,j,k}^{3D} \), i.e., “i, j, k” changes to “t” later in this paragraph. \( {\varvec{b}}_{i}^{1} \), \( {\varvec{b}}_{j}^{2} \), and \( {\varvec{b}}_{k}^{3} \) represent the ith, jth, and kth columns of B1, B2, and B3, respectively; “\( \otimes \)” represents an outer product and an element of \( \underline{{\mathbf{B}}}_{i,j,k}^{3D} \), such as the element indexed by (m, n, l), i.e., \( \underline{{\text{B}}}_{i,j,k}^{3D} (m,n,l) \), is expressed as \( \underline{{\text{B}}}_{i,j,k}^{3D} (m,n,l) = b_{i}^{1} (m)b_{j}^{2} (n)b_{k}^{3} (l) \) (Kroonenberg 2008). \( b_{i}^{1} (m) \), \( b_{j}^{2} (n) \), and \( b_{k}^{3} (l) \) are the mth, nth, and lth elements of \( {\varvec{b}}_{i}^{1} \), \( {\varvec{b}}_{j}^{2} \), and \( {\varvec{b}}_{k}^{3} \), respectively. After the construction of \( \underline{{\mathbf{B}}}_{i,j,k}^{3D} \), the subscript “(i, j, k)” of \( \underline{{\mathbf{B}}}_{i,j,k}^{3D} \) changes to t for derivation convenience. “t” is numbered in increasing order of i, followed by j and k, respectively. It is worth noting that although the discrete cosine function is adopted in this study to construct the 3D basis function, other basis functions (e.g., wavelets functions) can also be used in the proposed method. The discrete cosine function is adopted here because it has analytical function forms and the basis function can be readily obtained.
Appendix B: Derivation of Eq. (15)
The expression of KL divergence defined in Eq. (14) is expanded to
In accordance with the rules of conditional probability, \( p(\hat{\varvec{\omega }}^{3D} ,\varvec{\alpha},\gamma ,\tau |\varvec{y}^{3D} ) = p(\hat{\varvec{\omega }}^{3D} ,\varvec{\alpha},\gamma ,\tau ,\varvec{y}^{3D} )/p(\varvec{y}^{3D} ) \). Substituting this expression into Eq. (27) and rearranging the terms lead to
Note that ln[p(y3D)] is independent of the distribution \( q(\hat{\varvec{\omega }}^{3D} ,\varvec{\alpha},\gamma ,\tau ) \). Therefore, \( \int {q(\hat{\varvec{\omega }}^{3D} ,\varvec{\alpha},\gamma ,\tau )} \ln p(\varvec{y}^{3D} ){\text{d}}\hat{\varvec{\omega }}^{3D} {\text{d}}\varvec{\alpha}{\text{d}}\gamma {\text{d}}\tau = \ln p(\varvec{y}^{3D} ) \). As a result, Eq. (28) is simplified as
Let \( L(q) = \int {q(\hat{\varvec{\omega }}^{3D} ,\varvec{\alpha},\gamma ,\tau )} \ln \left( {\frac{{q(\hat{\varvec{\omega }}^{3D} ,\varvec{\alpha},\gamma ,\tau ,\varvec{y}^{3D} )}}{{q(\hat{\varvec{\omega }}^{3D} ,\varvec{\alpha},\gamma ,\tau )}}} \right){\text{d}}\hat{\varvec{\omega }}^{3D} {\text{d}}\varvec{\alpha}{\text{d}}\gamma {\text{d}}\tau \). Equation (29) can then be rewritten as KL(q||p) = − L(q) + ln[p(y3D)]. Subsequently, ln[p(y3D)] = L(q) + KL(q||p), i.e., Equation (15) can be obtained.
Appendix C: Derivation of q(\( \hat{\varvec{\omega }}^{3D} \)), q(α), q(γ), and q(τ)
In this appendix, the framework for using VBI to derive the tractable distribution is introduced. Let Θ = [θ1, θ2, …, θn]T represent a set of random variables and “p(Θ|Data)” represent the true posterior PDF of Θ updated by “Data.” Suppose that “p(Θ|Data)” has no analytical solution and VBI is adopted to seek an approximate distribution \( q(\varvec{\varTheta}) \) that can properly represent p(Θ|Data). As mentioned in the main text, \( q(\varvec{\varTheta}) \) is usually factorized as \( q(\varvec{\varTheta}) = \prod\nolimits_{i = 1}^{n} {q(\theta_{i} )} \). In accordance with the derivation by Bishop (2006) (pp. 461–517), the q(θi) that minimizes the KL divergence between q(Θ) and p(Θ|Data) is then expressed as
where Θ−i represents Θ with θi removed and “const” represents a term that ensures that the integration of q(θi) = 1.
In this paper, the random variables of interest are \( \hat{\varvec{\omega }}^{3D} \), α, γ, and τ, and their approximate distribution can be individually derived using Eq. (30). Consider, for example, q(\( \hat{\varvec{\omega }}^{3D} \)). In accordance with Eq. (30), \( q(\hat{\varvec{\omega }}^{3D} ) \) is expressed as Eq. (17). Substituting Eqs. (6), (8–11), and (18) into Eq. (17) leads to
where “const2” represents the term that does not involve \( \hat{\varvec{\omega }}^{3D} \). Note that q(α) and q(γ) are independent of \( \ln p(\varvec{y}^{3D} |\hat{\varvec{\omega }}^{3D} ,\tau ) \). Therefore, \( \int {q(\varvec{\alpha})q(\gamma )q(\tau )\ln p(\varvec{y}^{3D} |\hat{\varvec{\omega }}^{3D} ,\tau )} {\text{d}}\varvec{\alpha}{\text{d}}\gamma {\text{d}}\tau = \int {q(\tau )\ln p(\varvec{y}^{3D} |\hat{\varvec{\omega }}^{3D} ,\tau )} {\text{d}}\tau \). Similarly, \( \int {q(\varvec{\alpha})q(\gamma )q(\tau )\ln p(\hat{\varvec{\omega }}^{3D} |\alpha )} {\text{d}}\varvec{\alpha}{\text{d}}\gamma {\text{d}}\tau = \int {q(\varvec{\alpha})\ln p(\hat{\varvec{\omega }}^{3D} |\varvec{\alpha})} {\text{d}}\varvec{\alpha} \). Subsequently, substituting Eqs. (6) and (8) into Eq. (31) and rearranging the terms lead to
where “const3” represents the terms that do not involve \( \hat{\varvec{\omega }}^{3D} \). Completing the square for \( \hat{\varvec{\omega }}^{3D} \) in Eq. (32) and rearranging the terms lead to
Let \( {\varvec{\Sigma}}_{{\hat{\varvec{\omega }}^{3D} }} = [{\mathbf{A}}^{T} {\mathbf{A}}E(\tau ) + E({\mathbf{D}}^{\alpha } )]^{ - 1} \). Equation (33) is rewritten as
where “const4” is a term that incorporates “const3” and new terms that do not involve \( \hat{\varvec{\omega }}^{3D} \). Let \( \varvec{\mu}_{{\hat{\varvec{\omega }}^{3D} }} = {\varvec{\Sigma}}_{{\hat{\varvec{\omega }}^{3D} }} {\mathbf{A}}\varvec{y}^{3D} E(\tau ) \). Equation (34) is rewritten as
where \( \frac{{(\varvec{\mu}_{{\hat{\varvec{\omega }}^{3D} }} )^{T} ({\varvec{\Sigma}}_{{\hat{\varvec{\omega }}^{3D} }} )^{ - 1}\varvec{\mu}_{{\hat{\varvec{\omega }}^{3D} }} }}{2} \) is incorporated into the term “const5”. Therefore, \( q(\hat{\varvec{\omega }}^{3D} ) \) can be derived as
A close examination of Eq. (36) shows that \( \frac{1}{{\sqrt {(2\pi )^{N} \det ({\varvec{\Sigma}}_{{\hat{\varvec{\omega }}^{3D} }} )} }}\) \(\exp \Big[ { - \frac{{(\hat{\varvec{\omega }}^{3D} -\varvec{\mu}_{{\hat{\varvec{\omega }}^{3D} }} )^{T} ({\varvec{\Sigma}}_{{\hat{\varvec{\omega }}^{3D} }} )^{ - 1} (\hat{\varvec{\omega }}^{3D} -\varvec{\mu}_{{\hat{\varvec{\omega }}^{3D} }} )}}{2}} \Big] \) is the multivariate normal distribution, and its integration with respect to \( \hat{\varvec{\omega }}^{3D} \) is therefore 1. In addition, note that \( q(\hat{\varvec{\omega }}^{3D} ) \) is a PDF, which leads to \( \int {q(\hat{\varvec{\omega }}^{3D} )} {\text{d}}\hat{\varvec{\omega }}^{3D} = 1 \). As a result, the term “\( \sqrt {(2\pi )^{N} \det ({\varvec{\Sigma}}_{{\hat{\varvec{\omega }}^{3D} }} )} \exp (const_{5} ) \)” in Eq. (36) is equal to 1. In such a case, Eq. (36) is simplified as Eq. (19) in the main text.
Similarly, q(α) is derived as
where \( a_{t} = E[(\hat{\omega }_{t}^{3D} )^{2} ] \), \( b_{t} = E(\gamma ) \), p = − 1/2, and \( q(\alpha_{t} ) \) = \( \exp \left[ { - \frac{{a_{t} \alpha_{t} + b_{t} \alpha_{t}^{ - 1} }}{2}} \right](\alpha_{t} )^{p - 1} \times \frac{{(a_{t} /b_{t} )^{p/2} }}{{2K_{p} (\sqrt {a_{t} b_{t} } )}} \), which is a generalized inverse Gaussian (GIG) PDF (Zhao et al. 2015; Dumitru 2017). The mean or expectation of αt is shown in Eq. (21a). In addition, note that E(α −1t ) is needed in the proposed method [see Eq. (24)], which cannot be directly evaluated even if q(αt) is available. This is because 1/αt is a non-linear function of αt. To address this problem, the PDF of 1/αt, i.e., q(α −1t ) is derived as (Ang and Tang 2007)
Equation (38) shows that (1/αt) follows a GIG with parameters bt, at, and −p. The mean of (1/αt), i.e., E(α −1t ), is obtained as Eq. (24).
In a manner similar to the derivation of q(\( \hat{\varvec{\omega }}^{3D} \)) and q(α), both q(τ) and q(γ) are derived to follow a Gamma distribution, which is shown in Eqs. (39) and (40), respectively
Rights and permissions
About this article
Cite this article
Zhao, T., Wang, Y. Statistical Interpolation of Spatially Varying but Sparsely Measured 3D Geo-Data Using Compressive Sensing and Variational Bayesian Inference. Math Geosci 53, 1171–1199 (2021). https://doi.org/10.1007/s11004-020-09913-x
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11004-020-09913-x