Transcription
Geological Modeling:IntroductionDr. Irina OvereemCommunity Surface Dynamics Modeling SystemUniversity of Colorado at BoulderSeptember 2008
Course ObjectiveGeoscientists find resources by assessing the characteristics andconstraints of the earth subsurface. The subsurface has been formedover millions of years, and by the interaction of a host ofsedimentary processes and time-varying boundary conditions likeclimate, sea level and tectonics. This course aims at exploringGeological Modeling techniques as: Learning tools to disentangle complex interactions of sedimentarysystems and time-varying boundary conditions. Quantitative tools to create 3D geological models of the subsurface,including properties like grain size, porosity and permeability. A means to quantify uncertainties in the subsurface models.2
Course outline 1 Lectures by Irina Overeem: Introduction and overviewDeterministic and geometric modelsSedimentary process models ISedimentary process models IIUncertainty in modelingLecture by Overeem & Teyukhina: Synthetic migrated data3
Geological ModelingPrimary objective of geological characterizationis concerned with predicting the spatialvariation of geological variables.VariableAny property of the geological subsurface that exhibitsspatial variability and can be measured in terms of realnumerical values.Spatial VariationTypically the subsurface is anisotropic, spatially complexand sedimentary bodies are internally heterogeneous.4
Geological Modeling ReservoirArchitecture Modeling Construction (e.g. Westerscheldt tunnel)Groundwater flow models for drinkwater and irrigationMapping of ore deposits, or gravel & sand miningMapping for mine burial, naval warfare 5
Contaminant transport at Gardermoen Airport, NOpebbly sandcoarse sandsilty sandHydraulic conductivities vary within topset,foreset, and bottomset sedimentary layers.KTFS 6.3 * 10 -4 , KFFS 3.2 * 10Assess risk for contaminant transport need a subsurface flow model-6m/sGroundwater flow in the coarse sandy units canbe extremely rapid ( 500 m/day).6
Seafloor variability, New Jersey Margin, USANew Jersey shallow shelf.Assess variability inseafloor properties forsonar signal propagation(US Navy).Geostatistics of seabedheterogeneity plottedusing semivariograms.(Data courtesy: ChrisJenkins, CSDMS)
Well data correlation in the shallow subsurface ofthe Tambaredjo Field, SurinamTambaredjo Reservoir in fluvial deposits, Staatsolie Suriname NV Assess connectivity of sandbodies to optimize recovery Data Courtesy: Applied Earth Sciences, Delft University of Technology8
Introduction Modern reservoir characterisation started around 1980: Reason: deficiency of oil recovery techniques (inadequate reservoirdescription) Aim: predict inter-well distributions of relevant properties (φ, K) Subsurface (inter-well) heterogeneity cannot be measured: Seismic data (large support, low resolution) Well data (small support, high resolution) Complementary sources of information: Geological models Statistical models Combine data and models ‘static’ reservoir model9
Some thoughts on Support andresolution Seismic data (large support, low resolution) What are typical sizes of a 3D seismic dataset? What is typical resolution of 3D seismic data?Well data (small support, high resolution) What is the typical size of a well? Spacing? What fraction of the subsurface is sampled? What is typical resolution of well data?10
Static reservoir models Reservoir geology is the science (art?) of building predictivereservoir models on the basis of geological knowledge ( data,interpretations, models) A reservoir model depicts spatial variation of lithology (porosityand permeability): “static” model Simulations of multi-phase flow (“dynamic” models) require highquality “static” reservoir models Static reservoir models are improved through analysis of dynamicdata: iterative process11
Geological Modeling: different tracksReservoir DataSeismic, borehole and wirelogsData-driven modelingDeterministicModelProcess modelingStochastic ModelStaticReservoir ModelUpscalingFlow ModelSedimentaryProcess Model
Geological model Elements of the geologicalmodel: Bounding surfaces Distributions of physicalproperties betweensurfaces Faults OWC, GWC, GOC Conditioned to well data ?13
Concepts: Deterministic Models Deterministic models involve data collection andinformation processing to infer correlations and developunderstanding of stratal geometry. The deterministic model inferred fully acknowledges the data;the model contains no random components; consequently,each component and input is determined exactly.Computer visualization of known faultsExample from RML-Geosim
Concepts: Stochastical Models Statistics: science of exploring, analyzing and summarizing dataStatistical model: deterministic summary of the data with quantifieduncertainty. Stochastic Deterministic Random Noise is random by definition, most data are stochastic Apparent randomness implies sensitivity to initial conditions Stochastic simulation: generation of hypothetical data (realizations) froma statistical model by feeding it (pseudo)random input values. MOST COMMONLY USED IN PETROLEUM INDUSTRYExamples: PETREL (Shell), RML-Geosim (IFP), these techniques will beused in Production Geology Course!August 5, 200915
Concepts: Sedimentary ProcessModels Sedimentary Process Models consist of causative factors(input) that undergo dynamical physical processes and resultin an prediction of stratigraphy (output).prograding topsetssandy turbiditesriver plume mudsSimulation of 12,000 yrs of glacio-fluvial sedimentation in Arctic setting- sea level variation –40m, 5m, 15m- seasonal time-steps, Holocene climate
Why is geological modeling difficult?The output of many natural systems exhibits apparentrandomness, which is usually caused by extreme sensitivity toinitial conditions. Initial conditions and physical laws of suchsystems cannot be inferred from the output. Measurements are a finite sample of the output (all possiblerealisations of the system). Statistical models may be used to describe such measurements inthe absence of a physical model. Geological modeling software (a worst-case scenario): Designed by statisticians who know little about geology Applied by geologists / engineers who know little aboutstatisticsMany things can and will go wrong ! 17
Upscaling issues In addition to the natural scales of heterogeneity in the systemand the scale of the measurements, there is also the scale of thediscrete elements (grid blocks) in a reservoir model.Upscaling measurements to grid-block scale is a critical issue ingeological modeling and the object of active researchCommon errors in numerical reservoir models: Discretisation errors Upscaling errors Input errorsGeological modeling aims at minimizing the input errors toimprove reservoir-model performance18
Useful references on statisticalanalysis of geological data Jensen, J.L., Lake, L.W., Corbett, P.W.M., Goggin, D.J., 2000. Statistics forpetroleum engineers and geoscientists – 2nd Edition. Elsevier, Amsterdam, 338 p.(devoted to geostatistical modelling, fairly advanced level, poor graphics, quiteexpensive) Davis, J.C., 2002. Statistics and data analysis in geology - 3rd Edition. Wiley, NewYork, 638 p. (comprehensive text on statistical analysis of geological data, nomodelling, very well written – recommended) Swan, A.R.H., Sandilands, M., 1995. Introduction to geological data analysis.Blackwell, Oxford, 446 p. (simplified and abbreviated version of Davis) Houlding, S., 1994. 3D geoscience modeling; computer techniques for geologicalcharacterization. Springer-Verlag, Berlin. (specifically for 3D geological models)19
Final remark Different approaches to modeling, my personal philosophy is that theyneed to be mixed. Statistics is a very powerful geological modeling tool, but only when it isfirmly supported by geological knowledge“No matter what prediction technique we apply to a variable weare unlikely to achieve an acceptable result unless we takegeological effects into account.”(Houlding, 1994)August 5, 200920
Concepts: Sedimentary Process Models Sedimentary Process Models consist of causative factors (input) that undergo dynamical physical processes and result in an prediction of stratigraphy (output). Simulation of 12,000 yrs of glacio-fluvial sedimentation in Arctic setting - sea level variation -40m, 5m, 15m
