Abstract
Sedimentary deposits constitute the primary record of changing environmental conditions that have acted on Earth’s surface over geologic time. Clastic material is eroded from source locations (parents) in sediment routing systems and deposited at sink locations (children). Both parents and children have characteristics that vary across many different dimensions, including grain size, chemical composition, and the geochronologic age of constituent detrital minerals. During transport, sediment from different parents is mixed together to form a child, which in turn may serve as the parent for other sediment farther down-system or later in time when buried sediment is exhumed. The distribution of detrital mineral ages observed in parent and child sediments allows for investigation of the proportion of each parent in the child sediment, which reflects the properties of the sediment routing system. To model the proportion of dates in a child sample that comes from each of the parent distributions, we use a Bayesian mixture of Dirichlet processes. This model enables us to estimate the mixing proportions with associated uncertainty while making minimal assumptions. We also present an extension to the model whereby we reconstruct unobserved parent distributions from multiple observed child distributions using mixtures of Dirichlet processes. The model accounts for uncertainty in both the number of mineral formation events that constitute each parent distribution and the mixing proportions of each parent distribution that constitutes a child distribution. To demonstrate the model, we perform analyses using simulated data where the true age distribution is known as well as using a real-world case study from the coast of central California, USA.
Similar content being viewed by others
References
Amidon WH, Burbank DW, Gehrels GE (2005a) Construction of detrital mineral populations: Insights from mixing of U-Pb zircon ages in Himalayan rivers. Basin Res 17(4):463–485
Amidon WH, Burbank DW, Gehrels GE (2005b) U-Pb zircon ages as a sediment mixing tracer in the Nepal Himalaya. Earth Planet Sci Lett 235(1–2):244–260
Berliner LM (2003) Physical-statistical modeling in geophysics. J Geophys Res Atmos 108(D24)
Blake WH, Boeckx P, Stock BC, Smith HG, Bodé S, Upadhayay HR, Gaspar L, Goddard R, Lennard AT, Lizaga I et al (2018) A deconvolutional Bayesian mixing model approach for river basin sediment source apportionment. Sci Rep 8(1):1–12
Chang W, Haran M, Applegate P, Pollard D (2016) Calibrating an ice sheet model using high-dimensional binary spatial data. J Am Stat Assoc 111(513):57–72
Chen W (2012) Guillaume M (2012) HALS-based NMF with flexible constraints for hyperspectral unmixing. EURASIP J Adv Signal Process 1:54
Chen JH, Moore JG (1982) Uranium-lead isotopic ages from the Sierra Nevada batholith, California. J Geophys Res Solid Earth 87(B6):4761–4784
Cherniak D, Watson E (2001) Pb diffusion in zircon. Chem Geol 172(1–2):5–24
Cooper RJ, Krueger T (2017) An extended Bayesian sediment fingerprinting mixing model for the full Bayes treatment of geochemical uncertainties. Hydrol Process 31(10):1900–1912
de Valpine P, Turek D, Paciorek C, Anderson-Bergman C, Temple Lang D, Bodik R (2017) Programming with models: writing statistical algorithms for general model structures with NIMBLE. J Comput Graph Stat 26:403–413
Diebolt J, Robert CP (1994) Estimation of finite mixture distributions through Bayesian sampling. J R Stat Soc Ser B (Methodological) :363–375
Donoho D, Stodden V (2004) When does non-negative matrix factorization give a correct decomposition into parts? In: Advances in neural information processing systems, pp 1141–1148
Ferguson T S (1973) A Bayesian analysis of some nonparametric problems. Ann Stat 209–230
Ferguson TS (1983) Bayesian density estimation by mixtures of normal distributions. In: Recent advances in statistics. Elsevier, pp 287–302
Flowers RM, Ketcham RA, Shuster DL, Farley KA (2009) Apatite (U-Th)/He thermochronometry using a radiation damage accumulation and annealing model. Geochimica et Cosmochimica Acta 73(8):2347–2365
Fosdick JC, Grove M, Graham SA, Hourigan JK, Lovera O, Romans BW (2015) Detrital thermochronologic record of burial heating and sediment recycling in the Magallanes foreland basin. Patagonian Andes. Basin Research 27(4):546–572
Frühwirth-Schnatter S (2006) Finite mixture and Markov switching models. Springer Science & Business Media
Garver JI, Brandon MT, Roden-Tice M, Kamp PJ (1999) Exhumation history of orogenic highlands determined by detrital fission-track thermochronology. Geol Soc Lond Spec Publ 154(1):283–304
Gehrels G (2012) Detrital zircon U-Pb geochronology: current methods and new opportunities. Tectonics of Sedimentary Basins: Recent Advances, pp 45–62
Gehrels G (2014) Detrital zircon U-Pb geochronology applied to tectonics. Annu Rev Earth Planet Sci 42:127–149
Ghosal S (2010) The Dirichlet process, related priors and posterior asymptotics. Bayesian Nonparametrics 28:35
Green PJ (1995) Reversible jump Markov chain Monte Carlo computation and Bayesian model determination. Biometrika 82(4):711–732
Guan Y, Haran M, Pollard D (2018) Inferring ice thickness from a glacier dynamics model and multiple surface data sets. Environmetrics 29(5–6):e2460
Haario H, Saksman E, Tamminen J (2001) An adaptive Metropolis algorithm. Bernoulli 7(2):223–242
Hamilton N (2018) ggtern: An extension to ggplot2, for the creation of ternary diagrams. R package version 2(2):2
Hefley TJ, Brost BM, Hooten MB (2017) Bias correction of bounded location errors in presence-only data. Methods Ecol Evol 8(11):1566–1573
Irwin WP, Wooden JL (1999) Plutons and accretionary episodes of the Klamath Mountains, California and Oregon. US Geological Survey Open-File Report 99-374
Ishwaran H, James LF (2001) Gibbs sampling methods for stick-breaking priors. J Am Stat Assoc 96(453):161–173
Ishwaran H, Zarepour M (2002) Exact and approximate sum representations for the Dirichlet process. Can J Stat 30(2):269–283
Jasra A, Stephens DA, Gallagher K, Holmes CC (2006) Bayesian mixture modelling in geochronology via Markov chain Monte Carlo. Math Geol 38(3):269–300
Kimbrough DL, Grove M, Gehrels GE, Dorsey RJ, Howard KA, Lovera O, Aslan A, House PK, Pearthree PA (2015) Detrital zircon U-Pb provenance of the Colorado River: A 5 my record of incision into cover strata overlying the Colorado Plateau and adjacent regions. Geosphere 11(6):1719–1748
Lock EF, Dunson DB (2015) Shared kernel Bayesian screening. Biometrika 102(4):829–842
Mason CC, Fildani A, Gerber T, Blum MD, Clark JD, Dykstra M (2017) Climatic and anthropogenic influences on sediment mixing in the Mississippi source-to-sink system using detrital zircons: Late Pleistocene to recent. Earth Planet Sci Lett 466:70–79
Miao L, Qi H (2007) Endmember extraction from highly mixed data using minimum volume constrained nonnegative matrix factorization. IEEE Trans Geosci Remote Sens 45(3):765–777
Miller JW, Harrison MT (2018) Mixture models with a prior on the number of components. J Am Stat Assoc 113(521):340–356
Moecher DP, Samson SD (2006) Differential zircon fertility of source terranes and natural bias in the detrital zircon record: Implications for sedimentary provenance analysis. Earth Planet Sci Lett 247(3–4):252–266
Papaspiliopoulos O, Roberts GO (2008) Retrospective Markov chain Monte Carlo methods for Dirichlet process hierarchical models. Biometrika 95(1):169–186
Paterson GA, Heslop D (2015) New methods for unmixing sediment grain size data. Geochem Geophys Geosyst 16(12):4494–4506
Puetz SJ, Ganade CE, Zimmermann U, Borchardt G (2018) Statistical analyses of global U-Pb database 2017. Geosci Front 9(1):121–145
Reiners PW, Brandon MT (2006) Using thermochronology to understand orogenic erosion. Annu Rev Earth Planet Sci 34:419–466
Reiners PW, Campbell I, Nicolescu S, Allen CM, Hourigan J, Garver J, Mattinson J, Cowan D (2005) (U-Th)/(He-Pb) double dating of detrital zircons. Am J Sci 305(4):259–311
Richardson S, Green PJ (1997) On Bayesian analysis of mixtures with an unknown number of components (with discussion). J R Stat Soc Ser B (Stat Methodol) 59(4):731–792
Romans BW, Castelltort S, Covault JA, Fildani A, Walsh J (2016) Environmental signal propagation in sedimentary systems across timescales. Earth-Sci Rev 153:7–29
Saylor JE, Sundell K, Sharman G (2019) Characterizing sediment sources by non-negative matrix factorization of detrital geochronological data. Earth Planet Sci Lett 512:46–58
Sethuraman J (1994) A constructive definition of dirichlet priors. Stat Sin 639–650
Sharman GR, Johnstone SA (2017) Sediment unmixing using detrital geochronology. Earth Planet Sci Lett 477:183–194
Sharman GR, Sylvester Z, Covault JA (2019) Conversion of tectonic and climatic forcings into records of sediment supply and provenance. Sci Rep 9(1):4115
Sickmann ZT, Paull CK, Graham SA (2016) Detrital-zircon mixing and partitioning in fluvial to deep marine systems, central California, USA. J Sediment Res 86(11):1298–1307
Stephens M (2000) Dealing with label switching in mixture models. J R Stat Soc Ser B (Stat Methodol) 62(4):795–809
Stock GM, Ehlers TA, Farley KA (2006) Where does sediment come from? Quantifying catchment erosion with detrital apatite (U-Th)/He thermochronometry. Geology 34(9):725–728
Sundell KE, Saylor JE (2017) Unmixing detrital geochronology age distributions. Geochem Geophys Geosyst 18(8):2872–2886
Tipton JR, Hooten MB, Pederson N, Tingley MP, Bishop D (2016) Reconstruction of late Holocene climate based on tree growth and mechanistic hierarchical models. Environmetrics 27(1):42–54
Tipton J, Hooten M, Goring S (2017) Reconstruction of spatio-temporal temperature from sparse historical records using robust probabilistic principal component regression. Adv Stat Climatol Meteorol Oceanogr 3(1):1–16
Tipton JR, Hooten MB, Nolan C, Booth RK, McLachlan J (2019) Predicting paleoclimate from compositional data using multivariate Gaussian process inverse prediction. Ann Appl Stat 13(4):2363–2388
Tye A, Wolf A, Niemi N (2019) Bayesian population correlation: a probabilistic approach to inferring and comparing population distributions for detrital zircon ages. Chem Geol 518:67–78
Vehtari A, Gelman A, Gabry J (2017) Practical Bayesian model evaluation using leave-one-out cross-validation and WAIC. Stat Comput 27:1413–1432
Vermeesch P (2012) On the visualisation of detrital age distributions. Chem Geol 312:190–194
Ward EJ, Semmens BX, Schindler DE (2010) Including source uncertainty and prior information in the analysis of stable isotope mixing models. Environ Sci Technol 44(12):4645–4650
Weltje GJ (1997) End-member modeling of compositional data: numerical-statistical algorithms for solving the explicit mixing problem. Math Geol 29(4):503–549
Weltje GJ, Prins MA (2007) Genetically meaningful decomposition of grain-size distributions. Sediment Geol 202(3):409–424
Wotzlaw JF, Schaltegger U, Frick DA, Dungan MA, Gerdes A, Günther D (2013) Tracking the evolution of large-volume silicic magma reservoirs from assembly to supereruption. Geology 41(8):867–870
Author information
Authors and Affiliations
Corresponding author
Appendix
Appendix
In the body of the text, we explain in detail how the scientific understanding of the geologic processes that generated the data led to the development of the statistical model. Here in the appendix, we define the statistical models directly without the scientific motivation for clarity of model definition.
1.1 Top-Down Mixing Model
In the top-down mixing model, observations are made on both the child sediments and parent sediments. The \(n_y\) observed age measurements for the child sediment are given by the vector \(\mathbf {y} = (y_1, \ldots , y_{n_y})'\) and are reported with an observed analytic measurement standard deviation \(\varvec{\sigma }_y = (\sigma _{y1}, \ldots , \sigma _{yn_y})'\) associated with each observation. Assuming a normal distribution for the measurement process, the latent true ages are defined as \(\tilde{\mathbf {y}} = (\tilde{y}_1, \ldots , \tilde{y}_{n_y})'\) and are modeled by
where \(N(\mathbf {x} | \varvec{\mu }, \varvec{\varSigma })\) is a multivariate normal distribution with data vector \(\mathbf {x}\), mean vector \(\varvec{\mu }\), and covariance matrix \(\varvec{\varSigma }\). The notation \({\text {diag}} (\varvec{\sigma ^2})\) represents a diagonal covariance matrix with i, ith element \(\sigma ^2_i\) and off-diagonal elements all equal to 0.
Likewise, the \(b = 1, \ldots , B\) parent observations each comprise \(n_b\) observations and are given by the vector \(\mathbf {z}_b = (z_{b1}, \ldots z_{bn_b})'\) and are reported with an observed analytic measurement standard deviation \(\varvec{\sigma }_b = (\sigma _{b1}, \ldots , \sigma _{bn_b})'\) associated with each observation. Assuming a normal distribution for the measurement process, the latent true ages are defined as \(\tilde{\mathbf {z}}_b = (\tilde{z}_{b1}, \ldots , \tilde{z}_{bn_b})'\) and are modeled by
The latent parent age distributions for the \(b = 1, \ldots , B\) parents are modeled using a finite mixture of K Gaussian distributions
where \(\varvec{\mu } = (\mu _1, \ldots , \mu _K)'\) and \(\varvec{\sigma }^2 = (\sigma ^2_1, \ldots , \sigma ^2_K)'\) are the mean and variance of the mixture distributions (which are shared across each of the parents) and \(\mathbf {p}_{b} = (p_{b1}, \ldots , p_{bK})'\) are mixture weights where for \(k = 1, \ldots , K,\) \(p_{bk} > 0\) and \(\sum _{k=1}^K p_{bk} = 1\). For \(k = 1, \ldots , K\), the mixture distribution means are assigned independent, vague priors \(N(\mu _\mu = 150 \text{ Myr }, \sigma ^2_\mu = 150^2 \text{ Myr}^2)\), where Myr represents a million years. To ensure the mixing distributions are relatively concentrated with respect to geologic time, the mixing kernel standard deviations are assigned independent truncated half-Cauchy priors \(\sigma _k \sim {\text {Cauchy}}^+(0, 25 \text{ Myr})I\{0< \sigma _k < 50 \text{ Myr }\}\), which enforces the mixing distribution scales (which represent geologic mineral formation events) to be small relative to the range of dates from about 0 to 300 Myr.
For each parent \(b = 1, \ldots , B\), the mixing probabilities \(\mathbf {p}_b\) are modeled by introducing \(k = 1, \ldots , K-1\) independent and identically distributed random variables \(\tilde{p}_{bk} \sim Beta(1, \alpha _b)\) and transforming the \(\tilde{p}_{bk}\)s by
which induces a finite approximation to the stick-breaking representation of a Dirichlet process so long as K is chose large enough (Section 3 Ishwaran and James 2001 and Ishwaran and Zarepour (2002)). For \(b = 1, \ldots , B\), \(\alpha _b\) is assigned a gamma(1, 1) prior with \( \varvec{\alpha } = (\alpha _1, \ldots , \alpha _B)'\) defined as the vector of concentration parameters.
Combining the parent distributions, the unobserved, latent ages are modeled using the finite mixture of mixtures
where the notation \(\{\mathbf {p}_b\}_{b=1}^{B}\) denotes the set of parameters \(\{ \mathbf {p}_1, \ldots , \mathbf {p}_B\}\). The parameter \(\varvec{\phi } = (\phi _1, \ldots , \phi _B)'\) models the proportion of the child sediment \(\phi _b\) that comes from parent b where \(\phi _b > 0\) and \(\sum _{b=1}^B \phi _b = 1\). The mixing proportion \(\varvec{\phi }\) is assigned a \(Dirichlet(\alpha _\phi \mathbf {1})\) prior where \(\mathbf {1}\) is a vector of ones of length B and \(\alpha _\phi \) is assigned a gamma(1, 1) prior.
All combined, the top-down mixing model posterior is
where each line on the right-hand side of the proportional symbol is the data, process, and prior model, respectively.
1.2 Bottom-Up Unmixing Model
In the bottom-up unmixing model, observations are made on \(d = 1, \ldots , D\) child sediments whereas the parent sediments are unobserved. For each of the \(d = 1, \ldots , D\) children, the \(n_d\) observed age measurements are given by the vector \(\mathbf {y}_d = (y_{d1}, \ldots , y_{dn_d})'\) and are reported with an observed analytic measurement standard deviation \(\varvec{\sigma }_d = (\sigma _{d1}, \ldots , \sigma _{dn_d})'\) associated with each observation. Assuming a normal distribution for the measurement process, the latent true ages are defined as \(\tilde{\mathbf {y}}_d = (\tilde{y}_{d1}, \ldots , \tilde{y}_{dn_d})'\) and are modeled by
As none of the parent ages are observed, the latent parent age distributions for the \(b = 1, \ldots , B\) parents are represented as a finite mixture of K Gaussian distributions
where \(\varvec{\mu } = (\mu _1, \ldots , \mu _K)'\) and \(\varvec{\sigma }^2 = (\sigma ^2_1, \ldots , \sigma ^2_K)'\) are the mean and variance of the mixture distributions (which are shared across each of the parents) and \(\mathbf {p}_{b} = (p_{b1}, \ldots , p_{bK})'\) are mixture weights where for \(k = 1, \ldots , K,\) \(p_{bk} > 0\) and \(\sum _{k=1}^K p_{bk} = 1\). For \(k = 1, \ldots , K\), the mixture distribution means are assigned independent, vague priors \(N(\mu _\mu = 150 \text{ Myr }, \sigma ^2_\mu = 150^2 \text{ Myr}^2)\) where Myr represents a million years. To ensure the mixing distributions are relatively concentrated with respect to geologic time, the mixing kernel standard deviations are assigned independent truncated half-Cauchy priors \(\sigma _k \sim {\text {Cauchy}}^+(0, 25 \text{ Myr})I\{0< \sigma _k < 50 \text{ Myr }\}\), which enforces the mixing distribution scales (which represent geologic mineral formation events) to be small relative to the range of dates from about 0 to 300 Myr.
Because none of the parent ages are observed, the parent distributions are estimated entirely using child sediment observations. Assuming a fixed and known number of parents B, the bottom-up process model for the dth child is
where the B-dimensional vector of mixture proportions \(\varvec{\phi }_d = \left( \phi _{d1}, \ldots , \phi _{dB} \right) '\) models the proportion of the dth child sediment that can be attributed to each of the B parents where for \(b = 1, \ldots , B\), \(\phi _{db}>0\) and \(\sum _{b=1}^B \phi _{db} = 1\). For each of the \(d = 1, \ldots , D\) children, the mixing proportions \(\varvec{\phi }_d\) are assigned an independent \(Dirichlet(\alpha _{d} \mathbf {1})\) prior, where \(\mathbf {1}\) is a vector of ones of length B and each \(\alpha _{d}\) is assigned a gamma(1, 1) prior. The priors for \(\varvec{\mu }\), \(\varvec{\sigma ^2}\), and \(\{\mathbf {p}_{b}\}_{b=1}^B\) (and their respective hyperparameters \(\varvec{\alpha } = (\alpha _1, \ldots , \alpha _B)'\)) are the same as in the top-down mixing model.
Thus, the bottom-up unmixing model posterior distribution is
where each line on the right-hand side of the proportional symbol is the data, process, and prior model, respectively.
Rights and permissions
About this article
Cite this article
Tipton, J.R., Sharman, G.R. & Johnstone, S.A. A Bayesian Nonparametric Approach to Unmixing Detrital Geochronologic Data. Math Geosci 54, 151–176 (2022). https://doi.org/10.1007/s11004-021-09961-x
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11004-021-09961-x