Tout positionnement global précis nécessite un repère de référence tel le repère international de référence terrestre (ITRF). La détermination de l’ITRF s’appuie sur des séries temporelles de positions d’instruments géodésiques, en particulier des stations GNSS permanentes. Les séries temporelles de positions de stations GNSS sont corrélées temporellement et spatialement. De nombreuses études ont caractérisé la dépendance temporelle de ces séries et son impact sur la détermination de repères de référence. En revanche, les corrélations spatiales (entre stations proches) des séries GNSS n’ont jusqu’à présent jamais été prises en compte dans le calcul de repères de référence. L’objectif de cette thèse est donc de proposer une méthodologie pour la prise en compte de ces corrélations spatiales et d’évaluer son apport.Les dépendances spatiales entre les séries de 195 stations GNSS sont tout d’abord évaluées à l’aide de variogrammes empiriques confirmant l’existence de corrélations jusqu’à des distances d’environ 5000 km. Des modèles de covariance exponentielle ne dépendant que de la distance inter-stations sont ajustés sur ces variogrammes empiriques.Une méthodologie basée sur un filtre de Kalman est ensuite développée pour prendre en compte les dépendances spatiales des séries GNSS dans le calcul d’un repère de référence. Trois modèles de dépendance spatiale sont proposés : un modèle ne tenant pas compte de la dépendance spatiale (cas actuel du calcul de l’ITRF), un modèle basé sur les covariances empiriques entre séries de différentes stations, et un modèle basé sur les fonctions de covariance exponentielle mentionnées ci-dessus. Ces différents modèles sont appliqués à trois jeux tests d’une dizaine de stations chacun situés en Europe, aux Caraïbes et sur la côte est des États-Unis. Les trois modèles sont évalués à l’aune d’un critère de validation croisée, c’est-à-dire sur leur capacité à prédire les positions des stations en l’absence de données. Les résultats sur les jeux tests d’Europe et des États-Unis montrent une amélioration considérable de cette capacité prédictive lorsque la dépendance spatiale des séries est prise en compte. Cette amélioration est maximale lorsque le modèle de covariance exponentielle est utilisé. L’amélioration est nettement moindre, mais toujours présente sur le jeu test des Caraïbes.Les trois modèles sont également évalués sur leur capacité à déterminer des vitesses de déplacement exactes à partir de séries temporelles de positions courtes. L’impact de la prise en compte de la dépendance spatiale des séries sur l’exactitude des vitesses estimées est significatif. Comme précédemment, l’amélioration est maximale lorsque le modèle de covariance exponentielle est utilisé.Cette thèse démontre ainsi l’intérêt de la prise en compte des dépendances spatiales entre séries GNSS pour la détermination de repères de référence. La méthodologie développée pourra être utilisée pour le calcul de futures versions de l’ITRF. / Any global and precise positioning requires a reference frame such as the International Terrestrial Reference Frame (ITRF). The determination of the ITRF relies on the position time series of various geodetic instruments, including in particular permanent GNSS stations. GNSS station position time series are known to be temporally and spatially correlated. Many authors have studied the temporal dependency of GNSS time series and its impact on the determination of terrestrial reference frames. On the other hand, the spatial correlations (i.e., between nearby stations) of GNSS time series have so far never been taken into account in the computation of terrestrial reference frames. The objective of this thesis is therefore to develop a methodology to account for the spatial correlations of GNSS time series, and evaluate its benefits.The spatial dependencies between the position time series of 195 GNSS stations are first evaluated by means of empirical variograms, which confirm the existence of correlations up to distances of about 5000 km. Exponential covariance models, depending only on the distance between stations, are adjusted to these empirical variograms.A methodology based on a Kalman filter is then developed to take into account the spatial dependencies of GNSS time series in the computation of a terrestrial reference frame. Three models of spatial dependency are proposed: a model which does not account for the spatial dependency between GNSS time series (current case of the ITRF computation), a model based on the empirical covariances between the time series of different stations, and a model based on the exponential covariance functions mentioned above.These different models are applied to three test cases of ten stations each, located in Europe, in the Caribbean, and along the east coast of the US. The three models are evaluated with regard to a cross-validation criterion, i.e., on their capacity to predict station positions in the absence of observations. The results obtained with the Europe and US test cases demonstrate a significant improvement of this predictive capacity when the spatial dependency of the series is taken into account. This improvement is highest when the exponential covariance model is used. The improvement is much lower, but still present with the Caribbean test case.The three models are also evaluated with regard to their capacity to determine accurate station velocities from short position time series. The impact of accounting for the spatial dependency between series on the accuracy of the estimated velocities is again significant. Like previously, the improvement is highest when the exponential covariance model is used.This thesis thus demonstrates the interest of accounting for the spatial dependency of GNSS station position time series in the determination of terrestrial reference frames. The developed methodology could be used in the computation of future ITRF versions.

Physical and chemical heterogeneities cause the porous media transport parameters to vary with scale, and between these two types of heterogeneities geological heterogeneity is considered to be the most important source of scale-dependence of transport parameters. Subsurface processes associated with chemical alterations result in changing reservoir properties with interlinked spatial and temporal scale, and there is uncertainty in the evolution of those properties and the chemical processes. This dissertation provides a framework and procedures to quantify the spatiotemporal scaling characteristics of reservoir attributes and transport processes in heterogeneous media accounting for chemical alterations in the reservoir. Conventional flow scaling groups were used to assess their applicability in scaling of recovery and Mixing Zone Length (MZL) in presence of chemical reactivity and permeability heterogeneity through numerical simulations of CO₂ injection. It was found out that these scaling groups are not adequate enough to capture the scaling of recovery and transport parameters in the combined presence of chemical reactivity and physical heterogeneity. In this illustrative example, MZL was investigated as a function of spatial scale, temporal scale, multi-scale heterogeneity, and chemical reactivity; key conclusions are that 1) the scaling characteristics of MZL distinctly differ for low permeability and high permeability media, 2) heterogeneous media with spatial arrangements of both high and low permeability regions exhibit scaling characteristics of both high and low permeability media, 3) reactions affect scaling characteristics of MZL in heterogeneous media, 4) a simple rescaling can combine various MZL curves by merging them into a single MZL curve irrespective of the correlation length of heterogeneity, and 5) estimates of MZL (and consequently predictions of oil recovery) will fluctuate corresponding to displacements in a permeable medium whose lateral length is smaller than the correlation length of geological formation. We illustrate and extend the procedure of estimating Representative Elementary Volume (REV) to include temporal scale by coupling it with spatial scale. The current practice is to perform spatial averaging of attributes and account for residual variability by calibration and history matching. This results in poor predictions of future reservoir performance. The proposed semi-analytical technique to scale-up in both space and time provides guidance for selection of spatial and temporal discretizations that takes into account the uncertainties due to sub-processes. Finally, a probabilistic particle tracking (PT) approach is proposed to scale-up flow and transport of diffusion-reaction (DR) processes while addressing multi-scale and multi-physics nature of DR mechanisms and also maintaining consistent reservoir heterogeneity at different levels of scales. This multi-scale modeling uses a hierarchical approach which is based on passing the macroscopic subsurface heterogeneity down to the finer scales and then returning more accurate reactive flow response. This PT method can quantify the impact of reservoir heterogeneity and its uncertainties on statistical properties such as reaction surface area and MZL, at various scales. / text