1 / 43

Interpretation of data for field-scale modeling and predictions

Jean- Raynald de Dreuzy Géosciences Rennes, CNRS, FRANCE. Interpretation of data for field-scale modeling and predictions. Outline. A decision-based framework An elementary example Data interpolation Inverse Problem Conclusion: back to the objectives. Contaminant containment.

fahim
Download Presentation

Interpretation of data for field-scale modeling and predictions

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Jean-Raynald de Dreuzy Géosciences Rennes, CNRS, FRANCE Interpretation of data for field-scale modeling and predictions

  2. Outline A decision-based framework An elementary example Data interpolation Inverse Problem Conclusion: back to the objectives

  3. Contaminant containment Freeze, R. A., et al. (1990), Hydrogeological Decision-Analysis .1. a Framework, Ground Water, 28(5), 738-766.

  4. A decision-basedframework • SM = C – L • SM: safety margin • C: capacity (SC) • L: load (SL) • Probability of failure

  5. A decision-basedframework Freeze, R. A., et al. (1990), Hydrogeological Decision-Analysis .1. a Framework, Ground Water, 28(5), 738-766. • Objective function of alternative j: Fj • Benefits of alternative j: Bj • Costs of alternative j: Cj • Risk of alternative j: Rj • Probability of failure: Pf • Costassociatedwithfailure: Cf • Utility function (risk aversion): g

  6. Accounting for uncertainty Freeze, R. A., et al. (1990), Hydrogeological Decision-Analysis .1. a Framework, Ground Water, 28(5), 738-766.

  7. Decisionframework

  8. Optimal and Acceptable risks

  9. PhD. Etienne Bresciani (2008-2010) Risk assessment for High Level Radioactive Waste storage

  10. Outline A decision-based framework An elementary example Data interpolation Inverse Problem Conclusion: back to the objectives

  11. An elementaryexample Binary distribution of permeabilities Ka=10m/hr, Kb=2.4 m/day L=1km, Porosity=20%, head gradient=0.01 Localization of Ka and Kb?

  12. Some flow and transport values • Extremal values • Kmin=Kb • Kmax=Ka • A random case • K~2.6 m/hr • Advection times

  13. Equivalent permeability distribution for 10.000 realizations Reality is a single realization

  14. Consequences on transport Reality is a single realization

  15. Conditioning by permeability values

  16. Conditioning BY head values

  17. Solution Ka

  18. Outline A decision-based framework An elementary example Data interpolation Inverse Problem Conclusion: back to the objectives

  19. Data interpolation • Accounting for correlation • Inverse of distance interpolation • Geostatistics • Kriging • Simulation • Field examples

  20. Whatiscorrelation

  21. Geostatistical simulation

  22. Outline A decision-based framework An elementary example Data interpolation Inverse Problem Conclusion: back to the objectives

  23. GW Flow & Transport Carrera, J., A. Alcolea, A. Medina, J. Hidalgo, and L. J. Slooten (2005), Inverse problem in hydrogeology, Hydrogeology Journal, 13, 206-222.

  24. inverse problem (Identification of parameters) direct problem T: transmissivity S: storage coefficient Q: source terms bc: boudary conditions h: head inverse problem Trial and error approach: manually change T, S, Q in order to reach a good fit with h Inverse problem: automatic algorithm

  25. inverse problem (Identification of parameters) h(xi) T(xi) i:1…n bc? direct problem T: transmissivity bc: boudary conditions inverse problem Ill-posed problem Under-constrainted (more unknowns than data) km

  26. Specificities of inverse problem in hydrogeology • Model uncertainty: structure of the medium (geology, geophysics) not known accurately (soft data) • Heterogeneity: T varies over orders of magnitude • Low sensitivity: data (h) may contain little information on parameters (T) • Scale dependence: parameters measured in the field are often taken at a scale different from the mesh scale • Time dependence: data (h) depend on time • Different parameters (unrelated): beyond T, porosity, storativity, dispersivity • Different data: simultaneous integration of hydraulic, geophysical, geochemical (hard data)

  27. First approach: Cauchy problem • Interpolation of heads • Determination of flow tubes • Each tube contains a known permeability value • Determination of head everywhere by: • Drawbacks • Instable (small h0 errors induce large T0 errors) • Strong unrealistic transmissivity gaps between flow tubes • Independence between transmissivity obtained between flow tubes

  28. Use of geostatistics and cokrigeage • The Co-kriging equation uses the measured values of Φ =h-H, of Y, and the strcutures (covaraince, variogram, cross-variogram of Y, Φ and Y- Φ) which are known (Y) or calculated analytically from the stochastic PDE. • The inverse problem is thus solved without having to run the direct problem and to define an objective function. • Sometimes the covariance of Y is assumed known with an unknown coefficient which is optimized by cross-validation at points of known Y • Principle: express permeability as a linear function of known permeability and head values

  29. GW Flow & Transport Example of CokrigING • Advantages • No direct problem • Almost analytical • Additional knowledge on uncertainties • Drawbacks • Limited to low heterogeneities • Requires lots of data [Kitanidis,1997]

  30. Otherwise: Optimization of an objective • Objective function • Minimize head mismatch between model and data [Carrera, 2005]

  31. Inverse problem issues • Unstable parameters from data • Restricts instability of the objective funtion • Solution: regularization • More parameters than data (under-constrained) • Reduce parameter number drastically • Reduce parameter space • Acceptable number of parameters • gradient algorithms requiring convex functions: <5-7 parameters • Monté-Carlo algorithms: <15-20 parameters • Solution: parameterization

  32. GW Flow & Transport Simulated annealing interlude on traveling salesman problem

  33. Addition of a permeability term Regularization plausibility • Which proportion between • goodness of fit • plausibility • l?

  34. Illustration of regularization “True” medium [Carrera, Cargèse, 2005]

  35. p2 p2 p1 p1 Interpretation of regularization p2 Long narrow valleys Hard convergence and instability p1 Reduces uncertainty Smooths long narrow valleys Facilitates convergence Reduces instability and non-uniqueness [Carrera, Cargese, 2005]

  36. Parameterization • Relevant parameterization depends • on data quantity • on geology • on optimization algorithm [de Marsily, Cargèse, 2005]

  37. Comparison of 7 inverse methods • Zimmerman, Marsily, Carrera et al, 1998 for stochastic simulations, 4 test problems • 3 based on co-kriging • Carrera-Neuman, Bayesian, zoning • Lavenue-Marsily, pilot points • Gomez-Hernandez, Sequantial non Gaussian • Fractal ad-hoc method

  38. Results • If test problem is a geostatistical field, and variance of Y not too large, (variance of Log10T less than 1.5 to 2) all methods perform well • Importance of good selection of variogram • Co-kriging methods that fit the variogram by cross-validation on both Y and h’ data perform better • For non-stationary “complex” fields • The linearized techniques start to break down • Improvement is possible, e.g. through zoning • Non-linear methods, and with a careful fitting of the variogram, perform better • The experience and skill of the modeller makes a big difference…

  39. GW Flow & Transport Test cases K: 6 teintes de gris couvrant chacune un ordre de grandeur entre 10-7 et 10-2. cas 1: champ gaussien de log transmissivité (log10(T)) moyenne et variance de -5.5 et 1.5 respectivement et de longueur de corrélation 2800 m. cas 2 moyenne et variance plus importantes de -1.26 et 2.39. cas 3 comprend un milieu hétérogéne de log transmissivité et variance -5.5 et 0.8 et des chenaux de log transmissivité -2.5. cas 4 comprend de larges chenaux de forte transmissivité avec une distribution de log transmissivité de moyenne et variance -5.3 et 1.9.

  40. GW Flow & Transport Results for test cases 1 and 3

  41. Outline A decision-based framework An elementary example Data interpolation Inverse Problem Conclusion: back to the objectives

  42. Gary Larson, The far side gallery

  43. Consequence of data scarcity and geological complexity: UNCERTAINTY Example of protection zone delineation Pochon, A., et al. (2008), Groundwater protection in fractured media: a vulnerability-based approach for delineating protection zones in Switzerland, Hydrogeology Journal, 16(7), 1267-1281.

More Related