Parcourir la collection par Département "Centre d'hydrogéologie et de géothermie"
Voici les éléments 1 - 20 sur 1022
Résultats par page
Options de tri
- PublicationAccès libre13C- and 15N‑Isotope Analysis of Desphenylchloridazon by Liquid Chromatography−Isotope-Ratio Mass Spectrometry and Derivatization Gas Chromatography−Isotope-Ratio Mass Spectrometry(2019-1)
;Melsbach, Aileen; ;Lihl, Christina; ;Hofstetter, Thomas B.; Elsener, MartinThe widespread application of herbicides impacts surface water and groundwater. Metabolites (e.g., desphenylchloridazon from chloridazon) may be persistent and even more polar than the parent herbicide, which increases the risk of groundwater contamination. When parent herbicides are still applied, metabolites are constantly formed and may also be degraded. Evaluating their degradation on the basis of concentration measurements is, therefore, difficult. This study presents compound-specific stableisotope analysis (CSIA) of nitrogen- and carbon-isotope ratios at natural abundances as an alternative analytical approach to track the origin, formation, and degradation of desphenylchloridazon (DPC), the major degradation product of the herbicide chloridazon. Methods were developed and validated for carbon- and nitrogen-isotope analysis (δ13C and δ15N) of DPC by liquid chromatography−isotope-ratio mass spectrometry (LC-IRMS) and derivatization gas chromatography−IRMS (GC-IRMS), respectively. Injecting standards directly onto an Atlantis LC-column resulted in reproducible δ13C-isotope analysis (standard deviation <0.5‰) by LC-IRMS with a limit of precise analysis of 996 ng of DPC on-column. Accurate and reproducible δ15N analysis with a standard deviation of <0.4‰ was achieved by GC-IRMS after derivatization of >100 ng of DPC with 160-fold excess of (trimethylsilyl)diazomethane. Application of the method to environmental-seepage water indicated that newly formed DPC could be distinguished from “old” DPC by the different isotopic signatures of the two DPC sources. - PublicationMétadonnées seulement222Rn as a partitioning tracer to detect diesel fuel contamination in aquifers: Laboratory study and field observations(1997)
; ;Hoehn, E. ;Höhener, P.Zeyer, J. - PublicationAccès libre3-D groundwater modeling at regional scale(2001)
;Kimmeier, Francesco; Király, LászlóLarge hydrogeological basins are constituted of several superimposed aquifers, separated by geological formations of relatively low permeabilities. The delimitation of the different flow systems is far more difficult to realize for a heterogeneous system. Flux vectors provide valuable indications about groundwater flow paths and hydraulic exchanges between the different geological formations. Hydraulic relationships between two superimposed aquifers can vary locally: an aquifer can «feed» the underlying one at some point and conversely elsewhere. These relationships, which constitute in fact the flow field, will be determined by the structure of the basin as defined by the spatial distribution of the rock permeabilities, and by the boundary conditions, as defined by the locations of the recharge and discharge areas. The aim of this study is to show schematically the deep groundwater flow patterns between the massives of the Aar and the Black Forest. The hydrogeological profiles illustrate three-dimensional flow fields inside a large volume of terrain and represent but one of the numerous solutions of the mathematical modelling realized. Computations have been performed for a steady state flow regime, which means that the boundary conditions do not vary with time.
On the basis of modelling results, it was possible to illustrate schematically the deep flow systems of the most important aquifers between the Aar massives and the Black Forest (Malm, Muschelkalk and upper Cristallin). An approximate but plausible representation of the groundwater circulation in deep aquifers was obtained thanks to the model. We are able to distinguish between the hydraulic relationships of two superimposed aquifers in various regions. The three-dimensional representation shows the outcrop zones of the different geological formations as well as the situation of recharge areas, which are characterised by high potentials, and discharge zones, which are characterised by low potentials in valleys represented by the hydrographic network.
Modelling results are then compared to available measurements in an attempt to validate the results. It is interesting to notice that it was possible, to a certain extent, to verify the modelling results by deep drillings. Most particularly, measurements of the hydraulic potentials at various depths in these boreholes have revealed upwellings close to the regional discharge areas. - PublicationAccès libre3D Geological Image Synthesis from 2D Examples Using Generative Adversarial Networks(2020-10)
;Coiffier, Guillaume; Lefebvre, SylvainGenerative Adversarial Networks (GAN) are becoming an alternative to Multiple-point Statistics (MPS) techniques to generate stochastic fields from training images. But a difficulty for all the training image based techniques (including GAN and MPS) is to generate 3D fields when only 2D training data sets are available. In this paper, we introduce a novel approach called Dimension Augmenter GAN (DiAGAN) enabling GANs to generate 3D fields from 2D examples. The method is simple to implement and is based on the introduction of a random cut sampling step between the generator and the discriminator of a standard GAN. Numerical experiments show that the proposed approach provides an efficient solution to this long lasting problem. - PublicationAccès libre3D multiple-point statistics simulation using 2D training images(2012-3)
;Comunian, Alessandro; One of the main issues in the application of multiple-point statistics (MPS) to the simulation of three-dimensional (3D) blocks is the lack of a suitable 3D training image. In this work, we compare three methods of overcoming this issue using information coming from bidimensional (20) training images. One approach is based on the aggregation of probabilities. The other approaches are novel. One relies on merging the lists obtained using the impala algorithm from diverse 2D training images, creating a list of compatible data events that is then used for the MPS simulation. The other (s2Dcd) is based on sequential simulations of 2D slices constrained by the conditioning data computed at the previous simulation steps. These three methods are tested on the reproduction of two 3D images that are used as references, and on a real case study where two training images of sedimentary structures are considered. The tests show that it is possible to obtain 3D MPS simulations with at least two 2D training images. The simulations obtained, in particular those obtained with the s2Dcd method, are close to the references, according to a number of comparison criteria. The CPU time required to simulate with the method s2Dcd is from two to four orders of magnitude smaller than the one required by a MPS simulation performed using a 3D training image, while the results obtained are comparable. This computational efficiency and the possibility of using MPS for 3D simulation without the need for a 3D training image facilitates the inclusion of MPS in Monte Carlo, uncertainty evaluation, and stochastic inverse problems frameworks. - PublicationAccès libre3D multiple-point statistics simulations of the Roussillon Continental Pliocene aquifer using DeeSse(2020-10)
; ; ; ;Issautier, BenoîtCabellero, YvanThis study introduces a novel workflow to model the heterogeneity of complex aquifers using the multiplepoint statistics algorithm DeeSse. We illustrate the approach by modeling the Continental Pliocene layer of the Roussillon aquifer in the region of Perpignan (southern France). When few direct observations are available, statistical inference from field data is difficult if not impossible and traditional geostatistical approaches cannot be applied directly. By contrast, multiple-point statistics simulations can rely on one or several alternative conceptual geological models provided using training images (TIs). But since the spatial arrangement of geological structures is often non-stationary and complex, there is a need for methods that allow to describe and account for the non-stationarity in a simple but efficient manner. The main aim of this paper is therefore to propose a workflow, based on the direct sampling algorithm DeeSse, for these situations. The conceptual model is provided by the geologist as a 2D non-stationary training image in map view displaying the possible organization of the geological structures and their spatial evolution. To control the non-stationarity, a 3D trend map is obtained by solving numerically the diffusivity equation as a proxy to describe the spatial evolution of the sedimentary patterns, from the sources of the sediments to the outlet of the system. A 3D continuous rotation map is estimated from inferred paleoorientations of the fluvial system. Both trend and orientation maps are derived from geological insights gathered from outcrops and general knowledge of processes occurring in these types of sedimentary environments. Finally, the 3D model is obtained by stacking 2D simulations following the paleotopography of the aquifer. The vertical facies transition between successive 2D simulations is controlled partly by the borehole data used for conditioning and by a sampling strategy. This strategy accounts for vertical probability of transitions, which are derived from the borehole observations, and works by simulating a set of conditional data points from one layer to the next. This process allows us to bypass the creation of a 3D training image, which may be cumbersome, while honoring the observed vertical continuity. - PublicationAccès libre3D stochastic modeling of karst aquifers using a pseudo-genetic methodology(2013)
; Le but de cette thèse est le développement d'une méthodologie de modélisation des aquifères karstiques. Premièrement, la géométrie des conduits karstiques est simulée. La géométrie de ces conduits est contrôlée à large échelle par la géologie (modèle géologique), et à plus petite échelle par la fracturation (modèle stochastique de fracturation).
Deuxièmement, ces modèles géométriques (formations 3D et conduits ensemble) sont utilisés comme base pour la simulation d'écoulement et de transport.
En dernière partie, le simulateur de conduits SKS ("Stochastic Karst Simulator") est couplé avec la simulation d'écoulement et transport pour investiguer une approche inverse qui permette d'utiliser cette méthodologie dans une étude d'incertitude.
La méthodologie de simulation des conduits karstiques développée ici se base sur une approche dite "pseudo-génétique", càd qui mime les résultats des processus de spéléogénèses, sans pour autant simuler toute la dynamique complexe de ces processus, comme la dissolution et le transport réactif de calcite, etc. Dans cette approche pseudo-génétique les conduits karstiques sont simulés par une physique approchée, qui se base sur le principe de minimisation de l'énérgie. L'eau se déplace dans un milieu en cherchant le chemin de moindre résistance. Ce principe est utilisé ici, par l'utilisation d'un algorithme de Fast Marching, qui permet de calculer le chemin de moindre effort., The focus of this thesis is the development of a methodology to model karst aquifers. First the geometry of the karst conduits is simulated. Their geometry is controlled by the geology at large scale (geological model) and by the fracturation at smaller scale (stochastic model of fractures).
Secondly, these geometrical models (3D geological formation together with conduits) are used as base for flow and transport simulation.
Finally, the karst conduit generator called SKS ("Stochastic Karst Simulator") is coupled with the physical simulation (flow and transport) to investigate an inverse approach which allows to use this methodology in an uncertainty analysis.
The karst conduits simulation methodology is called pseudo genetic because it mimic the results of the speleogenetic processes, without simulating all the complex kinetic of karst systems genesis, like reactive transport, calcite dissolution and precipitation. In this approach, the karst conduits are simulated by approaching the physical principle of minimization of energy using a Fast Marching Algorithm. This algorithm allow to compute the minimum effort path, which is assumed to be the one used by water, and consequently the preferential dissolution., Das Ziel dieser Dissertation besteht in der Entwicklung einer Methode zur Modellierung von Karstaquiferen. In einem ersten Schritt wird die Geometrie des Karstsystems simuliert. Im grossen Massstab wird die Geometrie des Karstsystems durch die Geologie bestimmt (geologisches Modell), in kleinerem Massstab durch Klüfte (stochastisches Modell zur Kluftgenese). In einem zweiten Schritt wird dieses Modell als Grundlage für die Modellierung von Grundwasserströmung und Stoffransport im Karstsystem verwendet.
Schliesslich wird die Simulation des Karstsystems mittels des Karst-Simulator (SKS, "Stochastic Karst Simulator") mit Simulationen von Grundwasserströmung und Stoffransport gekoppelt, um einen inversen Ansatz zu untersuchen, der es erlauben würde diese Methode im Rahmen einer Unsicherheitsanalyse zu verwenden.
Die hier entwickelte Methodik zur Simulierung von Karstaquiferen basiert auf einem sogenannt "pseudogenetischen" Ansatz, weil sie speleogenetische Prozesse nachbildet, ohne die komplexe Dynamik dieser Prozesse, wie Auflösung und reaktiver Transport von Kalzit etc., im Detail zu simulieren. Die Karstgenese wird in dieser Methodik vielmehr durch einen Ansatz angenähert, der auf dem Prinzip der Energieminimierung beruht. Dabei wird ein Fast Marching Algorithmus verwendet, um zwischen den Stellen mit Wasserzuflüssen ins Karstsystem und den Karstquellen den Weg des geringsten Widerstandes zu berechnen., Lo scopo di questa tesi, è lo svipuppo di una metodologia di modellizzazione realistica degli acquiferi carsici. Nella prima parte viene modellizzata la geometria dei condotti carsici. La geometria di questi ultimi, è controllata a larga scala dalla geologia (modello geologico) e a piu piccola scala dalla fratturazione (modello di fratturazione stocastico).
In seguito questi modelli geometrici vengono usati come base per la simulazione di fluidi e trasporto di contaminanti. E infine, il simulatore di condotti carsici SKS ("Stochastic Karst Simulator") è usato insieme alla simulazione di fluidi e trasporto per valutare l'applicabilità di un approcio inverso che permetta di utilizzare questa metodologia di simulazione.
La metodologia di simulazione dei condotti carsici è detta "pseudo genetica", perchè tenta di approssimare la fisica complessa della speleogenesi (come il trasporto reattive, la dissoluzione/precipitazione della calcite) senza dover risolverla numericamente. In effetti si basa sul principio fisico della minimizzazione dell'energia, usando un algoritmo di Fast Marching per calcolare il cammino di minor resistenza (cioè quello che dovrebbe seguire l'acqua). In pratica questa metodologia simula dei sistemi carsici maturi, direttamente nel loro stato finale. - PublicationAccès libreA 2D hyperspectral library of mineral reflectance, from 900 to 2500 nm(2019-11)
; ; ; Mineral identification using machine learning requires a significant amount of training data. We built a library of 2D hyperspectral images of minerals. The library contains reflectance images of 130 samples, of 76 distinct minerals, with more than 3.9 million data points. In order to produce this dataset, various well-characterized mineral samples were scanned, using a SPECIM Short Wave Infrared (SWIR) camera, which captures wavelengths from 900 to 2500 nm. Minerals were selected to represent all the mineral classes and the most common mineral occurrences. For each sample, the following data are provided: (a) At least one hyperspectral image of the sample, consisting of 256 wavelengths between 900 and 2500 nm. The raw data, the high dynamic range (HDR) image, and the masked HDR image are provided for each scan (each of them in HDF5 format). (b) A text file describing the sample, providing supplementary information for the subsequent analysis (c) RGB images (JPEG files) and automated 3D reconstructions (Stanford Triangle PLY files). These data help the user to visualize and understand specific sample characteristics. 2D hyperspectral images were produced for each mineral, which consist of many different spectra with high diversity. The scans feature similar spectra than the ones in other available spectral libraries. An artificial neural network was trained to demonstrate the high quality of the dataset. This spectral library is mainly aimed at training machine learning algorithms, such as neural networks, but can be also used as validation data for other types of classification algorithms. - PublicationAccès libreA 3D geological model of a structurally complex Alpine region as a basis for interdisciplinary research(2018-10)
; ;Mariéthoz, GrégoireCertain applications, such as understanding the influence of bedrock geology on hydrology in complex mountainous settings, demand 3D geological models that are detailed, high-resolution, accurate, and spatially-extensive. However, developing models with these characteristics remains challenging. Here, we present a dataset corresponding to a renowned tectonic entity in the Swiss Alps - the Nappe de Morcles - that does achieve these criteria. Locations of lithological interfaces and formation orientations were first extracted from existing sources. Then, using state-of-the-art algorithms, the interfaces were interpolated. Finally, an iterative process of evaluation and re-interpolation was undertaken. The geology was satisfactorily reproduced; modelled interfaces correspond well with the input data, and the estimated volumes seem plausible. Overall, 18 formations, including their associated secondary folds and selected faults, are represented at 10m resolution. Numerous environmental investigations in the study area could benefit from the dataset; indeed, it is already informing integrated hydrological (snow/surface-water/ groundwater) simulations. Our work demonstrates the potential that now exists to develop complex, highquality geological models in support of contemporary Alpine research, augmenting traditional geological information in the process. - PublicationAccès libreA comparison of FBG- and Brillouin-strain sensing in the framework of a decameter-scale hydraulic stimulation experiment(: American Rock Mechanics Association, 2018-6-18)
;Krietsch, H. ;Gischig, V. ;Jalali, R. ;Doetsch, J.; Amann, F.In the framework of the In-situ Stimulation and Circulation (ISC) experiment Fiber-Bragg-Grating (FBG) and Brillouin strain sensing systems were installed to monitor deformation during six hydraulic shearing and six hydraulic fracturing experiments. Three boreholes were dedicated to strain monitoring. Both systems are installed in the same boreholes, offering a unique opportunity to compare these systems with respect to their applicability in hydraulic stimulation tests. A total of 60 FBG sensors with 1 m base length were installed across fractures, shear zones and intact rock. Along the entire borehole length, pre-stressed optical cables for Brillouin distributed strain (DBS) sensing were embedded in grout with two installation methods: a bare cable and a cable packed and fixed with glue every 0.65 m. The strain signals were compared as time series for a given borehole depth and as profiles along the borehole axis. The study reveals that the FBG system gives a high accuracy (0.04 µ-strain) and temporal resolution (>1s) with pointwise measurements. The bare DBS leg yield good quantitative strain data with poorer strain accuracy (>500 times poorer than FBG) and poorer temporal resolution (factor of >100). The packed DBS leg provide no meaningful information about the strain field. - PublicationMétadonnées seulementA deterministic approach to the coupled analysis of karst springs' hydrographs and chemographs(2003)
;Grasso, D Alessandro; During the chemically based recession flow phase of karstic springs the carbonate (dissolved limestone) concentration can be expressed as negative power of the flow rate. The empirically determined Conc/Q relationship allows two parameters (alpha and A) to be defined, of which one (alpha) depends on the geometric dimensions of the saturated (submerged) karstic network. In this paper we present a deterministic model which simulates the concentration of carbonate at the outlet of a network of circular rectilinear conduits as a function of flow rate. This model, based on hydraulic principles and the calcite dissolution kinetics, allows the sensitivity of the alpha and A parameters to be studied under different chemical, physical and geometric scenarios. Simulation results show that A is a function of the calcite saturation concentration, whereas alpha depends on the spatial dimensions of the karstic network (void length and aperture). The deterministic model results were applied to real karstic systems to evaluate the geometric dimensions of submerged karstic networks. (C) 2002 Elsevier Science B.V. All rights reserved. - PublicationAccès libreA dynamic model of the Aral Sea water and salt balance(2004)
;Benduhn, FrançoisThe Aral Sea is shrinking rapidly since the 1960s mainly because of the diversion of the Amu Darya and Syr Darya rivers for irrigation purposes. Since then, the evaporation became the most important component of the water balance of the Sea and led to a concentration of the remaining salts. In this article, we investigate through a coupled mathematical model of water and salt balance of the Aral Sea, the dynamic evolution of the sea. The water balance considers river inflow, groundwater inflow, atmospheric precipitation and evaporation. The salt balance considers the dominant ions and the chemical precipitation of gypsum, epsomite and mirabilite. The evaporation rates are calculated with a modified Penman equation accounting for the salinity of the lake and using statistical climatic data.
With this model, we obtain an estimate of the evaporation flux (between 1100 and more than 1200 mm/year depending on the salinity) larger than earlier estimates. The estimated groundwater discharge into the sea is also larger than earlier estimates and is highly variable from year to year. The last point is that the model is able to simulate rather well the evolution of the salinity until the 1980s, but it does not reproduce accurately the chemical evolution of the lake during the most recent period and needs further improvements. - PublicationAccès libreA dynamic model of the Aral Sea water and salt balanceThe Aral Sea is shrinking rapidly since the 1960s mainly because of the diversion of the Amu Darya and Syr Darya rivers for irrigation purposes. Since then, the evaporation became the most important component of the water balance of the Sea and led to a concentration of the remaining salts. In this article, we investigate through a coupled mathematical model of water and salt balance of the Aral Sea, the dynamic evolution of the sea. The water balance considers river inflow, groundwater inflow, atmospheric precipitation and evaporation. The salt balance considers the dominant ions and the chemical precipitation of gypsum, epsomite and mirabilite. The evaporation rates are calculated with a modified Penman equation accounting for the salinity of the lake and using statistical climatic data. With this model, we obtain an estimate of the evaporation flux (between 1100 and more than 1200 mm/year depending on the salinity) larger than earlier estimates. The estimated groundwater discharge into the sea is also larger than earlier estimates and is highly variable from year to year. The last point is that the model is able to simulate rather well the evolution of the salinity until the 1980s, but it does not reproduce accurately the chemical evolution of the lake during the most recent period and needs further improvements. (C) 2004 Elsevier B.V. All fights reserved.
- PublicationMétadonnées seulementA fast algorithm for the estimation of the equivalent hydraulic conductivity of heterogeneous media(2000-1-10)
; ;Le Loc'h, Gaëlle ;Ledoux, Emmanuel ;De Marsily, GhislainMackay, Rae - PublicationAccès libreA Fast Algorithm for the Estimation of the Equivalent Hydraulic Conductivity of Heterogeneous Media(2000)
; ;Le Loc’h, Gaëlle ;Ledoux, Emmanuel ;de Marsily, GhislainMackay, RaeFast upscaling of hydraulic conductivity is a recurrent problem in modeling flow through heterogeneous porous media. We propose a new renormalization technique. It is based on the iterative application of the Cardwell and Parsons [1945] bounds on elementary groups of cells. The combination of the bounds with a heuristic formula allows anisotropy to be taken into account. The new technique is tested and compared with other fast techniques. Among the tested techniques the two most reliable ones are the tensorial renormalization and the new simplified renormalization. The numerical efficiency of the simplified renormalization leads us to recommend it when a diagonal tensor of equivalent conductivity is sufficient. - PublicationAccès libreA finite element formulation of the outlet gradient boundary condition for convective-diffusive transport problems(2004-11-05)
; ; Diersch, Hans-JörgA simple finite element formulation of the outlet gradient boundary condition is presented in the general context of convective-diffusive transport processes. Basically, the method is based on an upstream evaluation of the dependent variable gradient along open boundaries. Boundary normal unit vectors and gradient operators are evaluated using covariant bases and metric tensors, which allow handling finite elements of mixed dimensions. Even though the presented method has implications for many fields where diffusion processes are involved, discussion and illustrative examples address more particularly the framework of contaminant transport in porous media, in which the outlet gradient concentration is classically, but wrongly assumed to be zero. - PublicationMétadonnées seulementA fluid-pressure feedback model of dehydration reactions: experiments, modelling, and application to subduction zones(2003)
; ;van der Zee, Wouter ;Olgaard, David L.Connolly, James A. D.Dehydration and melting reactions generate large volumes of fluid in the crust and upper mantle, and play an important role in subduction zone seismicity. The fluid pathway must evolve from isolated pockets in low porosity, low permeability rock, coalescing to interconnected permeable pathways to the surface. When fluid pressures generated from a dehydration or melting reaction are sufficient to induce hydrofracture, then hydrofracture significantly influences the porosity-permeability structure within the dehydrating/melting horizon. If a low fluid-pressure boundary is introduced to the dehydrating rock, then fluid will be driven from the rock along the evolved permeable network toward that boundary. The resulting pressure reduction can then accelerate the dehydration reaction and further drive the flow. The sudden introduction of a low fluid-pressure boundary may occur by the co-seismic (dilatant) rupturing of a pressure seal that connects different fluid pressure states. This mechanism is invoked to explain the observed post-seismic evolution of wave velocities (V-p/V-s,) following the 1995 Antofagasta, Chile earthquake. We show experimental results and introduce a conceptual and numerical model that reflects this scenario. The model couples the mechanical and thermodynamic effects of fluid pressure with devolitization kinetics, and is quantitatively consistent with experimental studies of the dehydration of gypsum and serpentine. The experimental results show that dehydration is controlled by access to a free (drained) boundary. The model provides a mechanistic explanation for the experimental observations and has applications in understanding the role of transient transport networks on the large-scale behavior of dehydrating and melting systems. (C) 2003 Elsevier B.V. All rights reserved. - PublicationMétadonnées seulementA forward model for earthquake generation on interacting faults including tectonics, fluids, and stress transfer(2001)
;Fitzenz, Delphine D.We present a forward model of interacting faults for systems of any geometry. The model generalizes that of Miller et al. [1996,1999] to a fully three-dimensional model where faults of any strike and geometry interact through an elastic matrix using the general solutions of Okada [1992]. The model includes large-scale plate motion loading and increasing pore pressures from a source term, undrained poroelastic effects, large coseismic hydraulic property changes, and porosity creation through dilatant slip. To illustrate the basic behavior and utility of the model, results are presented of the long-term evolution (approximate to 9300 years) for a generic case of a blind, dipping fault and a subvertical strike-slip fault in a transpressional environment. We show the stress state evolution along both faults, seismicity time lines, quasi-static rupture propagation including rake angle changes, local and regional stress buildup and rotations, static and dynamic fault interactions, and Delta CFS (changes in Coulomb Failure Stress) within the fault system. Large compartments of varying overpressure result on both faults from coseismic pore pressure changes and contribute to the complexity of the stress state. For the considered case, we find that the poroelastic effects on the receiver fault are about twice the change in the shear stress, providing a significant contribution to the Delta CFS. Regional stress rotations in response to the model seismicity indicate that further model developments must include dynamic generation of new faults in response to the evolving tectonic regime. - PublicationAccès libreA Framework for the Cross‐Validation of Categorical Geostatistical Simulations(2020-6)
; ; The mapping of subsurface parameters and the quantification of spatial uncertainty requires selecting adequate models and their parameters. Cross‐validation techniques have been widely used for geostatistical model selection for continuous variables, but the situation is different for categorical variables. In these cases, cross‐validation is seldom applied, and there is no clear consensus on which method to employ. Therefore, this paper proposes a systematic framework for the cross‐validation of geostatistical simulations of categorical variables such as geological facies. The method is based on K‐fold cross‐validation combined with a proper scoring rule. It can be applied whenever an observation data set is available. At each cross‐validation iteration, the training set becomes conditioning data for the tested geostatistical model, and the ensemble of simulations is compared to true values. The proposed framework is generic. Its application is illustrated with two examples using multiple‐point statistics simulations. In the first test case, the aim is to identify a training image from a given data set. In the second test case, the aim is to identify the parameters in a situation including nonstationarity for a coastal alluvial aquifer in the south of France. Cross‐validation scores are used as metrics of model performance and quadratic scoring rule, zero‐one score, and balanced linear score are compared. The study shows that the proposed fivefold stratified cross‐validation with the quadratic scoring rule allows ranking the geostatistical models and helps to identify the proper parameters.