Voici les éléments 1 - 8 sur 8
  • Publication
    Accès libre
    Stochastic multiple data integration for the characterization of quaternary aquifers
    La gestion des ressources en eaux souterraines nécessite souvent le développement de modèles géologiques et hydrogéologiques. Cependant, la construction de modèles précis peut s’avérer une tâche difficile et longue, en particulier dans les vastes zones présentant des dépôts quaternaires complexes. Or, ces zones sont souvent celles qui sont le plus fréquemment soumises à l’exploitation des ressources et à la pollution. Pour résoudre ce problème, plusieurs études ont proposé des méthodologies innovantes pour intégrer différents types de données, notamment des données sur les puits, des données géophysiques et des données hydrogéologiques. L’objectif est de faciliter la construction de ces modèles dans des cadres cohérents et reproductibles avec une estimation robuste des erreurs. Nous présentons ici quatre études qui proposent de nouvelles méthodologies pour relever ce défi. La première étude présente un vaste et dense ensemble de données électromagnétiques dans le domaine temporel (TDEM) acquises dans la haute vallée de l’Aar, en Suisse, afin d’améliorer la connaissance des variations spatiales des dépôts quaternaires. Les modèles de résistivité inversée dérivés de cette acquisition ont été publiés et pourraient être utilisés pour diverses études futures. Cette étude met également en évidence le potentiel de l’ensemble de données pour le développement d’algorithmes d’intégration de données en raison de l’abondance de diverses données librement disponibles sur la même zone. La deuxième étude propose une nouvelle méthodologie pour combiner les forages et les données géophysiques avec une propagation de l’incertitude pour prédire la probabilité d’argile à l’échelle d’une vallée. Une fonction de translation variant dans l’espace a été utilisée pour estimer la fraction d’argile à partir de la résistivité. Les paramètres de cette fonction sont inversés en utilisant la description des forages comme points de contrôle. Ils combinent cette estimation de la fraction d’argile avec un cadre d’interpolation stochastique 3D non déterministe basé sur un algorithme de statistiques à points multiples et une fonction aléatoire gaussienne afin d’obtenir un modèle 3D réaliste à haute résolution spatiale de la fraction d’argile pour la haute vallée de l’Aar. L’étude démontre la qualité des valeurs prédites et leurs incertitudes correspondantes en utilisant la validation croisée. La troisième étude porte sur la possibilité d’intégrer des données de forage, géophysiques et hydrogéologiques, tout en conservant la cohérence du concept géologique des modèles. Nous avons utilisé un générateur stochastique de modèles géologiques pour construire un ensemble de modèles préalables basés sur les forages. Nous proposons ensuite une approche d’inversion multi-échelle qui combine des modèles peu fidèles et moins précis avec des modèles plus fidèles et plus précis afin de réduire le temps nécessaire à la convergence de l’inversion. Les données géophysiques et hydrogéologiques sont intégrées à l’aide d’un algorithme ES-MDA (Ensemble Smoother with Multiple Data Assimilation Algorithm). Le flux de travail garantit que les modèles sont géologiquement cohérents et estime de manière robuste l’incertitude associée au modèle final. L’étude démontre l’efficacité de cette approche pour un cas synthétique contrôlé. Elle montre que ArchPY et ES-MDA sont capables de générer des réalisations plausibles de la subsurface pour les modèles sédimentologiques du Quaternaire. Enfin, la quatrième étude présente une méthodologie innovante qui combine l’algorithme ES-MDA avec un code de modélisation géologique hiérarchique open-source pour intégrer des sources de données multiples et construire des modèles géologiquement cohérents avec une estimation d’erreur robuste. La méthodologie est appliquée à un cas de terrain dans la haute vallée de l’Aar, en Suisse. Un cadre de validation croisée est mis en oeuvre afin d’évaluer la méthodologie. L’approche aboutit à des modèles finaux qui équilibrent efficacement la précision et l’incertitude et qui peuvent prendre en compte diverses sources de données, y compris des données géophysiques, des connaissances conceptuelles régionales, des forages et des mesures hydrogéologiques à l’échelle d’une vallée. En résumé, cette thèse présente plusieurs méthodes innovantes qui pourraient être appliquées à la réalisation de modèles hydrogéologiques à petite ou grande échelle. ABSTRACT Groundwater resource management often requires the development of geological and hydrogeological models. However, constructing accurate models can be a challenging and time-consuming task, especially in large areas with complex Quaternary deposits. However, these areas are often the most frequently subject to resource exploitation and pollution. To address this issue, several studies have proposed innovative methodologies to integrate various types of data, including wells, geophysical, and hydrogeological data. The objective is to facilitate the construction of these models within coherent and reproducible frameworks with robust error estimation. In these, we present four studies that present novel methodologies to address this challenge. The first study presents a large and dense Time Domain ElectroMagnetic (TDEM) dataset acquired in the upper Aare Valley, Switzerland, to improve knowledge of the spatial variations of Quaternary deposits. The inverted resistivity models derived from this acquisition were published and could be used for various future studies. It also highlights the data set’s potential for data integration algorithm development because of the abundance of various freely available data on the same zone. The second study proposes a new methodology to combine boreholes and geophysical data with a propagation of the uncertainty to predict the probability of clay at the scale of a valley. A spatially varying translator function was used to estimate the clay fraction from resistivity. The parameters of this function are inverted using the description of the boreholes as control points. They combine this clay fraction estimation with a nondeterministic 3D stochastic interpolation framework based on a Multiple Points Statistics algorithm and Gaussian Random Function to obtain a 3D realistic high spatial resolution model of clay fraction for the upper Aare valley. The study demonstrates the quality of the predicted values and their corresponding uncertainties using cross-validation. The third study addresses the possibility of integrating boreholes, geophysical, and hydrogeological data, while keeping the geological concept of the models coherent. We used a stochastic geological model generator to construct a set of prior models based on the boreholes. We then propose a multiscale inversion approach that combines low-fidelity and less accurate models with high-fidelity and more accurate models to reduce the time needed for the inversion to converge. Both geophysical and hydrogeological data are integrated, using an Ensemble Smoother with Multiple Data Assimilation Algorithm (ES-MDA) algorithm. The workflow ensures that the models are geologically consistent and robustly estimate the associated uncertainty with the final model. The study demonstrates the effectiveness of this approach for a controlled synthetic case. It shows that ArchPY and ES-MDA are capable of generating plausible subsurface realizations for Quaternary Sedimentological Models. Finally, the fourth study presents an innovative methodology that combines the ES-MDA algorithm with an open-source hierarchical geological modeling code to integrate multiple data sources and construct geologically consistent models with robust error estimation. The methodology is applied to a field case in the upper Aare Valley, Switzerland. In order to benchmark the methodology, a cross-validation framework is implemented. The approach results in final models that effectively balance accuracy and uncertainty and can take into account various data sources, including geophysical data, regional conceptual knowledge, boreholes, and hydrogeological measurements at a valley scale. In summary, this thesis presents several innovative methods that could be applied on small to large scale hydrogeological model realization.
  • Publication
    Accès libre
    An Attempt to Boost Posterior Population Expansion Using Fast Machine Learning Algorithms
    In hydrogeology, inverse techniques have become indispensable to characterize subsurface parameters and their uncertainty. When modeling heterogeneous, geologically realistic discrete model spaces, such as categorical fields, Monte Carlo methods are needed to properly sample the solution space. Inversion algorithms use a forward operator, such as a numerical groundwater solver. The forward operator often represents the bottleneck for the high computational cost of the Monte Carlo sampling schemes. Even if efficient sampling methods (for example Posterior Population Expansion, PoPEx) have been developed, they need significant computing resources. It is therefore desirable to speed up such methods. As only a few models generated by the sampler have a significant likelihood, we propose to predict the significance of generated models by means of machine learning. Only models labeled as significant are passed to the forward solver, otherwise, they are rejected. This work compares the performance of AdaBoost, Random Forest, and convolutional neural network as classifiers integrated with the PoPEx framework. During initial iterations of the algorithm, the forward solver is always executed and subsurface models along with the likelihoods are stored. Then, the machine learning schemes are trained on the available data. We demonstrate the technique using a simulation of a tracer test in a fluvial aquifer. The geology is modeled by the multiple-point statistical approach, the field contains four geological facies, with associated permeability, porosity, and specific storage values. MODFLOW is used for groundwater flow and transport simulation. The solution of the inverse problem is used to estimate the 10 days protection zone around the pumping well. The estimated speed-ups with Random Forest and AdaBoost were higher than with the convolutional neural network. To validate the approach, computing times of inversion without and with machine learning schemes were computed and the error against the reference solution was calculated. For the same mean error, accelerated PoPEx achieved a speed-up rate of up to 2 with respect to the standard PoPEx.
  • Publication
    Accès libre
    An optical laser device for mapping 3D geometry of underwater karst structures: first tests in the Ox Bel’Ha system, Yucatan, Mexico
    (2016)
    Schiller, A
    ;
    In the course of extended hydrological studies in the coastal Karst plain of Yucatan, near the town of Tulum amongst others, a novel laser scanning device was developed and applied for the acquisition of the 3d-geometry of ground water conduits. The method is derived from similar industrial systems and for the first time adapted to the specific measurement conditions in underwater cave systems. The device projects a laser line over the whole perimeter at a certain position. This line represents the intersection of a plane with the cave walls. The line is imaged with a wide angle camera system. Through proper design and calibration of the device it is possible to derive the true scale geometry of the perimeter via special image processing techniques. By acquiring regularly spaced images it is possible to reconstruct the true scale and 3 d-shape of a tunnel through the incorporation of location and attitude data. In a first test in the Ox Bel Ha under-water cave system, about 800 metres of tunnels have been scanned down to water depths of 20 metres. The raw data is further interpolated using the ODSIM-algorithm in order to delineate the 3D geometry of the cave system. The method provides easy, operable acquisition of the 3-D geometry of caves in clear water with superior resolution and speed and significantly facilitates the measurement in underwater tunnels as well as in dry tunnels. The data gathered represents crucial input to the study of the state, dynamics and genesis of the complex karst water regime., Durante el transcurso de intensivos estudios hidrológicos realizados en la llanura costera kárstica de Yucatán, cerca de la ciudad de Tulum entre otras, se desarrolló un novedoso dispositivo de escaneo láser, que se aplicó a la adquisición de la geometría 3D de conductos de agua subterránea. El método se deriva de sistemas industriales similares y que ha sido adaptado por primera vez a las condiciones de medición específicas de los sistemas de cuevas submarinas. El dispositivo proyecta una línea láser sobre todo el perímetro en una localización dada. Esta línea representa la intersección de un plano con las paredes de las cuevas. La línea es fotografiada con un sistema de cámara de gran angular. A través de un apropiado diseño y calibración del dispositivo es posible obtener la geometría verdadera del perímetro a través de técnicas especiales de procesamiento de imágenes. De este modo, adquiriendo regularmente imágenes a intervalos espaciados es posible reconstruir la escala verdadera y la forma 3D de un túnel con la incorporación de los datos de posición e inclinación. En una primera prueba en el sistema de la cueva submarina Ox Bel Ha, se escanearon alrededor de 800 metros de túneles hasta profundidades, bajo el agua, de 20 metros. Los datos en bruto son interpolados utilizando el algoritmo de ODSIM para delinear la geometría 3D del sistema de cuevas. El método proporciona una adquisición sencilla y operativa de la geometría tridimensional de cuevas submarinas con aguas claras, con muy buenas resolución y velocidad lo que facilita la medición en conductos submarinos así como en túneles subaéreos. Los datos recogidos representan una información fundamental para el estudio del estado, dinámica y génesis del complejo régimen del agua kárstica.
  • Publication
    Accès libre
    Stochastic simulation of rainfall and climate variables using the direct sampling technique
    An accurate statistical representation of hydrological processes is of paramount importance to evaluate the uncertainty of the present scenario and make reliable predictions in a changing climate. A wealth of historic data has been made available in the last decades, including a consistent amount of remote sensing imagery describing the spatio-temporal nature of climatic and hydrological processes. The statistics based on such data are quite robust and reliable. However, to explore their variability, most stochastic simulation methods are based on low-order statistics that can only represent the heterogeneity up to a certain degree of complexity.
    In the recent years, the stochastic hydrogeology group of the University of Neuchâtel has developed a multiple-point simulation method called Direct Sampling (DS). DS is a resampling technique that allows the preservation of the complex data structure by simply generating data patterns similar to the ones found in the historical data set. Contrarily to the other multiple-point methods, DS can simulate either categorical or continuous variables, or a combination of both in a multivariate framework.
    In this thesis, the DS algorithm is adapted to the simulation of rainfall and climate variables in both time and space. The developed stochastic weather or climate generators include the simulation of the target variable with a series of auxiliary variables describing some aspects of the complex statistical structure characterizing the simulated process. These methods are tested on real application cases including the simulation of rainfall time-series from different climates, the variability exploration of future climate change scenarios, the missing data simulation within flow rate time-series and the simulation of spatial rainfall fields at different scales. If a representative training data set is used, the proposed methodologies can generate realistic simulations, preserving fairly well the statistical properties of the heterogeneity. Moreover, these techniques result to be practical simulation tools, since they are adaptive to different data sets with minimal effort from the user perspective. Although leaving large room for improvement, the proposed simulation approaches show a good potential to explore the variability of complex hydrological processes without the need of a complex statistical model.
  • Publication
    Accès libre
    Analog-based meandering channel simulation
    (2014-1-10) ;
    Comunian, Alessandro
    ;
    Irarrazaval, Inigo
    ;
  • Publication
    Accès libre
    Grid-enabled Monte Carlo analysis of the impacts of uncertain discharge rates on seawater intrusion in the Korba aquifer (Tunisia)
    (2010) ; ;
    Lecca, Giuditta
    ;
    Tarhouni, Jamila
    L'aquifère de Korba, situé au nord de la Tunisie, est gravement touché par une salinisation du à l'intrusion marine. En 2000, l'aquifère a été exploité par plus de 9000 puits. Le problème, c'est qu'il n'y a pas d'information précise concernant les débits de pompage, leur répartition dans l'espace ainsi que leur évolution dans le temps. Dans cette étude, un modèle géostatistique des débits d'exploitation a été construit en se basant sur une régression multilinéaire combinant des données directes incomplètes ainsi que des données secondaires exhaustives. Les impacts de l'incertitude associée à la distribution spatiale des débits de pompage sur l'intrusion marine ont été évalués en utilisant un modèle tridimensionnel d'écoulement et de transport à densité variable. Pour contourner les difficultés liées à de longs temps de calcul, nécessaires pour résoudre des problèmes en régime transitoire, les simulations ont été réalisées en parallèle sur une grille informatique de calcul mise à disposition par le projet “Enabling Grid for E-Science in Europe”. Les résultats des simulations de Monte Carlo ont montré que 8.3% de la surface de l'aquifère est affectée par l'incertitude liée aux données d'entrée., The Korba aquifer, located in the north of Tunisia, suffers heavily from salinization due to seawater intrusion. In 2000, the aquifer was exploited from more than 9000 wells. The problem is that no precise information was recorded concerning the current extraction rates, their spatial distribution, or their evolution in time. In this study, a geostatistical model of the exploitation rates was constructed based on a multi-linear regression model combining incomplete direct data and exhaustive secondary information. The impacts of the uncertainty on the spatial distribution of the pumping rates on seawater intrusion were evaluated using a 3-D density-dependent groundwater model. To circumvent the large amount of computing time required to run transient models, the simulations were run in a parallel fashion on the Grid infrastructure provided by the Enabling Grid for E-Science in Europe project. Monte Carlo simulations results showed that 8.3% of the aquifer area is affected by input uncertainty.
  • Publication
    Accès libre
    Geological stochastic imaging for aquifer characterization
    (2009)
    Mariéthoz, Grégoire
    ;
    Accurately modeling connectivity of geological structures is critical for flow and transport problems. Using multiple-points simulations is one of the most advanced tools to produce realistic reservoir structures. It proceeds by considering data events (spatial arrangements of values) derived from a training image (TI). The usual method consists in storing all the data events of the TI in a database, which is used to compute conditional probabilities for the simulation. Instead, the Direct Sampling method (DS) proposed in this thesis consists in sampling directly the TI for a given data event. As soon as the data event in the TI matches the data event at the node to simulate, the value at its central node is directly pasted in the simulation. Because it accommodates data events of varying geometry, multi-grids are not needed. The method can deal with categorical and continuous variables and can be extended to multivariate cases. Therefore, it can handle new classes of problems. Different adaptations of the DS are proposed. The first one is aimed at reconstructing partially informed images or datasets. Instead of inferring data events from a TI, a training dataset is used. If the density of measurements is high enough, significant non-parametric spatial statistics can be derived from the data, and the patterns found in those data are mimicked without model inference. Therefore, minimum assumptions are made on the spatial structure of the reconstructed fields. Moreover, very limited parameterization is needed. The method gives good results for the reconstruction of complex 3D geometries from relatively small datasets. Another adaptation of the DS algorithm is aimed at performing super-resolution of coarse images. DS is used to stochastically simulate the structures at scales smaller than the measurement resolution. These structures are inferred using a hypothesis of scale-invariance on the spatial patterns found at the coarse scale. The approach is illustrated with examples of satellite imaging and digital photography. Parallelization is another important topic treated in this thesis. The size of simulation grids used for numerical models has increased by many orders of magnitude in the past years. Efficient pixel-based geostatistical simulation algorithms exist, but for very large grids and complex spatial models, computational burden remains heavy. As cluster computers become widely available, using parallel strategies is a natural step for increasing the usable grid size and the complexity of the models. These strategies must take profit of the possibilities offered by machines with a large number of processors. On such machines, the bottleneck is often the communication time between processors. This thesis presents a strategy distributing grid nodes among all available processors while minimizing communication and latency times. It consists in centralizing the simulation on a master processor that calls other slave processors as if they were functions simulating one node every time. The key is to decouple the sending and the receiving operations to avoid synchronization. Centralization allows having a conflict management system ensuring that nodes being simulated simultaneously do not interfere in terms of neighborhood. The strategy is computationally efficient and is versatile enough to be applicable to all random path based simulation methods. In addition to the preceding topics, a new cosimulation algorithm is proposed for simulating a primary attribute using one or several secondary attributes known exhaustively on the domain. This problem is frequently encountered in surface and groundwater hydrology when a variable of interest is measured only at a discrete number of locations and when a secondary variable is mapped by indirect techniques such as geophysics or remote sensing. In the proposed approach, the correlation between the two variables is modeled by a joint probability distribution function. A technique to construct such relations using latent variables and physical laws is proposed when field data are insufficient. The simulation algorithm proceeds sequentially. At each node of the grid, two conditional probability distribution functions (cpdf) are inferred. The first is inferred in a classical way from the neighboring data of the main attribute and a model of its spatial variability. The second is inferred directly from the joint probability distribution function of the two attributes and the value of the secondary attribute at the location to be simulated. The two distribution functions are combined by probability aggregation to obtain the local cpdf from which a value is randomly drawn. Various examples using synthetic and remote sensing data demonstrate that the method is more accurate than the classical collocated cosimulation technique when a complex relation links the two attributes.