Search

"Mining operations generate large amounts of wastes which are usually stored into large-scale storage facilities which pose major environmental concerns and must be properly monitored to manage the risk of catastrophic failures and also to control the generation of contaminated mine drainage. In this context, non-invasive monitoring techniques such as time-lapse electrical resistivity tomography (TL-ERT) are promising since they provide large-scale subsurface information that complements surface observations (walkover, aerial photogrammetry or remote sensing) and traditional monitoring tools, which often sample a tiny proportion of the mining waste storage facilities. The purposes of this review are as follows: (i) to understand the current state of research on TL-ERT for various applications; (ii) to create a reference library for future research on TL-ERT and geoelectrical monitoring mining waste; and (iii) to identify promising areas of development and future research needs on this issue according to our experience. This review describes the theoretical basis of geoelectrical monitoring and provides an overview of TL-ERT applications and developments over the last 30 years from a database of over 650 case studies, not limited to mining operations (e.g., landslide, permafrost). In particular, the review focuses on the applications of ERT for mining waste characterization and monitoring and a database of 150 case studies is used to identify promising applications for long-term autonomous geoelectrical monitoring of the geotechnical and geochemical stability of mining wastes. Potential challenges that could emerge from a broader adoption of TL-ERT monitoring for mining wastes are discussed. The review also considers recent advances in instrumentation, data acquisition, processing and interpretation for long-term monitoring and draws future research perspectives and promising avenues which could help improve the design and accuracy of future geoelectric monitoring pr

New processors are increasingly supporting half-precision floating-point numbers, often with a significant throughput gain over single-precision operations. Seismic modeling, imaging, and inversion could benefit from such an acceleration, but it is not obvious how the accuracy of the solution can be preserved with a very narrow 16-bit representation. By scaling the finite-difference expression of the isotropic elastic wave equation, we have found that a stable solution can be obtained despite the very narrow dynamic range of the half-precision format.We develop an implementation with the CUDA platform, which, on most recent graphics processing units (GPU), is nearly twice as fast and uses half the memory of the equivalent single-precision version. The error on seismograms caused by the reduced precision is shown to correspond to a negligible fraction of the total seismic energy and is mostly incoherent with seismic phases. Finally, we find that this noise does not adversely impact full-waveform inversion nor reverse time migration, which both benefit from the higher throughput of half-precision computation.

Applying deep learning to 3D velocity model building remains a challenge due to the sheer volume of data required to train large-scale artificial neural networks. Moreover, little is known about what types of network architectures are appropriate for such a complex task. To ease the development of a deep-learning approach for seismic velocity estimation, we have evaluated a simplified surrogate problem — the estimation of the root-mean-square (rms) and interval velocity in time from common-midpoint gathers — for 1D layered velocity models. We have developed a deep neural network, whose design was inspired by the information flow found in semblance analysis. The network replaces semblance estimation by a representation built with a deep convolutional neural network, and then it performs velocity estimation automatically with recurrent neural networks. The network is trained with synthetic data to identify primary reflection events, rms velocity, and interval velocity. For a synthetic test set containing 1D layered models, we find that rms and interval velocity are accurately estimated, with an error of less than 44 m/s for the rms velocity. We apply the neural network to a real 2D marine survey and obtain accurate rms velocity predictions leading to a coherent stacked section, in addition to an estimation of the interval velocity that reproduces the main structures in the stacked section. Our results provide strong evidence that neural networks can estimate velocity from seismic data and that good performance can be achieved on real data even if the training is based on synthetics. The findings for the 1D problem suggest that deep convolutional encoders and recurrent neural networks are promising components of more complex networks that can perform 2D and 3D velocity model building.

la Terre, et ce, à une variété d’échelles. Le sismologue voulant étudier la structure du manteau terrestre, le géophysicien d’exploration cherchant un nouveau gisement, l’hydrogéologue souhaitant modéliser l’écoulement de contaminants dans un aquifère; tous ont besoin d’estimer la structure et les propriétés physiques du sol. Une des méthodes les plus avancées pour étudier l’intérieur de la sous-surface a été développée au courant des dernières décennies: l’inversion de formes d’ondes complètes. L’engouement envers cette technique provient du fait qu’elle permettrait, à terme, de reconstruire de multiples propriétés mécaniques simultanément à une résolution inégalée. Pour l’instant, l’inversion en formes d’onde complètes est utilisée surtout dans l’industrie pétrolière pour obtenir des modèles de vitesse des ondes de compression plus précis. L’inversion de multiples paramètres (propriétés des ondes de cisaillement, les facteurs d’atténuation, les coefficients d’anisotropie, etc.) reste marginale en raison de son coût élevé en calcul et des difficultés de convergence lors de l’inversion. Cette thèse porte sur le développement d’une approche d’inversion de formes d’ondes complètes prenant en compte les phénomènes élastiques et les pertes visqueuses, c’est-à-dire l’inversion viscoélastique. La base de cette approche est le calcul du gradient par la méthode adjointe. Les équations adjointes sont ainsi développées pour la formulation vitesse-contrainte de l’équation d’onde, largement utilisée pour la modélisation sismique viscoélastique. Le deuxième élément de cette approche est la résolution numérique des équations directes et adjointes à l’aide de plateformes de calcul parallèles. En utilisant les divers degrés de parallélisme accessibles sur les plateformes de calcul modernes, il est possible d’accélérer significativement les calculs de l’inversion en formes d’onde complètes. Notamment, le calcul sur des cartes graphiques avec OpenCL mène à une accélération jusqu’à