Imagerie sismique quantitative de la marge convergente d'Equateur-Colombie : Application des mèthodes tomographiques aux données de sismique réflexion multitrace et réfraction-réflexion grand-angle des campagnes SISTEUR et SALIERI

摘要

This work's main aim is the estimation of the physical properties and the analysis of the geodynamic process of the Ecuador-Colombian subduction zone. I adapted and developped seismic imaging tools based on diffraction tomography. This tools are applied to multichannel seismic reflection (MCS) data and reflection/refraction wide-angle (WA) data acquired during SISTEUR and SALIERI cruises. Quantitative seismic imaging is presented for three zones : shallow zone (~ 0-3 km), intermediate zone (~ 3-10 km) and deep zone (~ 10-30 km). In the shallow zone the profile SIS-40 on the Colombian margin was processed and physical properties of the Bottom Simulating Reflector (BSR) were estimated. Along the BSR regions with relative increase of the velocity (1470-1650 m/s), were obtained and associated with the presence of hydrates above the BSR and regions with relative decrease of the velocity (~1200 m/s), associated with gaz below the BSR. In the intermediate zone, I analysed the subduction channel (profil SIS-72). This structure is between the decollement (top of the subduction channel) ant the top of the oceanic crust (bottom). In some area the decollement exhibits relative velocity decrease possibly due to the presence of fluids inside and below the decollement. I design an integrated approach to obtain the small scale velocities around the decollement area. The integrated approach is based on 2 steps: (1) asymptotic waveform inversion with iterative correction of the velocity macro-model and to obtain a 2-D quantitative depth model for velocity; (2) an automated post-processing procedure to eliminate the source signature from the tomographic images and to estimate the absolute values of the velocity along the decollement. The post-processing is formulated as an automatic non-linear inverse problem where the data space is composed of several one-dimensional logs extracted for different offset from the depth migrated image. The model space is composed of a family of realistic impulse layered models in depth, parameterized by a limited number of parameters (random velocity amplitude and a random thickness for each layer). These models mimick the logs of the physical model searched. To build the predicted dataset, the tested logs are converted from space to time using the velocity of the background medium and are convolved with the source wavelet. To estimate the source wavelet we use an average of the direct wave. The predicted dataset are computed by convolution of the depth-to-time converted impulse models with the source wavelet and compared with the tomographic models. The inverse problem is solved by a random exploration of the model space for each offset, using the very fast simulating anealing algorithm (VFSA). A small scale velocity model is obtained. Positve velocity perturbations appear at the top of the oceanic crust. The decollement is characterized by regions with positve velocity perturbations probably associated with fluids diffusion up to the decollement and regions with negative velocity perturbations probably related to the channeling of the fluids. Deep structures are imaged by a combined approach of MCS and WA data to improve the spatial resolution of the Moho and of interplate contact and to establish a possible relation with the seismogenic zone. A well constrained velocity model between 0 and 25 km is obtained and deep reflectors such as the Moho and the interplate contact are imaged as well the existence of the splay fault is validated (the splay fault was not clearly identified on the time stack section