Development of a New Parallel Code for 3-D Joint Refraction and Reflection Travel-Time Tomography of Wide-Angle Seismic Data. Synthetic and Real Data Applications to the Study of Subduction Zones

Abstract

This dissertation is devoted to seismic tomography. I have implemented a new modelling tool for 3-D joint refraction and reflection travel-time tomography of wide-angle seismic data (TOMO3D). The reason behind this central objective is the evidence that the information based on 2-D seismic data does not allow to capture the structural complexity of many 3-D targets, and in particular that of the seismogenic zone in subduction margins. The scientific rationale for this statement, which justifies the central part of my thesis work, is based on the analysis of 2-D models obtained in the convergent margin of Nicaragua, a seismically active area where a textbook example of tsunami earthquake took place in 1992. In this application I modelled two perpendicular wide-angle seismic profiles for the characterisation of the overriding plate and the interplate fault. To do this, I applied TOMO2D, a state-of-the-art joint refraction and reflection 2-D travel-time tomography code. The inversion outcomes are two 2-D velocity models along both profiles, together with the 1-D geometry of the interplate boundary. In combination with other geophysical data measurements, namely coincident multichannel seismic profiles and gravity data, these models provide new constraints on the nature and structure of the margin, and in particular add new insights on the nucleation and propagation of the said earthquake and its tsunamigenic behaviour. Ultimately, this case study evidences the aforementioned limitations of 2-D modelling in the investigation of 3-D geological structures and phenomena. Following from this first application and with the idea of increasing the amount of data used in travel-time tomography, I focused on an a priori paradoxical phenomenon related to water-layer multiple phases, that under certain circumstances, is observed on wide-angle record sections. The interest of this study lies in the fact that this phenomenon can provide additional constraints on travel-time tomography models. First, I propose and corroborate the hypothesis explaining the apparent paradox, and then derive the most favourable geological conditions for the phenomenon to occur. Subsequently, the possibility to model this multiple-like phases is introduced in TOMO3D.

The development of TOMO3D, which constitutes the core of my work, is founded on TOMO2D, from which it inherits the numerical methods for solving the forward and inverse problems. Source files have been rewritten, redefining and introducing the necessary variables and functions to handle 3-D data inversion. The tests made with the sequential version of the code emphasise the need of parallelisation for practicality reasons. Indeed, the increasing size of data sets along with the modelling of the additional spatial dimension results in computationally demanding inversions. Hence, I parallelised the forward modelling part of the code, which takes up to 90% of the computing time, with a combination of multiprocessing and message-passing interface extensions. Subsequently, the parallel version of TOMO3D is applied to a complex synthetic case simulating a subduction zone. This first 3-D application serves to evaluate the correctness of the code's programming, and as step-by-step description of the modelling procedure, with particular attention on the layer-stripping strategy used to successively model several reflectors. The outcomes demonstrate the ability of the code and the chosen inversion strategy to accurately recover the velocity distribution and the geometry of the two reflectors. Finally, TOMO3D is applied to a real 3-D wide-angle seismic data set acquired at the Pacific margin of Ecuador and Colombia to extract a 3-D velocity model of the overriding and incoming plates, which is then compared to previous results obtained with an extensively tested and used 3-D refraction travel-time tomography code (FAST). The comparison indicates that TOMO3D is more accurate than FAST but at the same time it is computationally more demanding. However, the parallelisation of TOMO3D allows using high-performance computing facilities, which is not the case of FAST or most of the existing codes.

Aquesta tesi està dedicada a la tomografia sísmica. Concretament, he implementat una eina de modelització 3D per a la tomografia conjunta de temps de trajecte de refraccions i reflexions (TOMO3D). La raó darrere d'aquest objectiu és l'evidència de que la informació basada en dades sísmiques 2D no permet copsar la complexitat de gran part dels cossos geològics, i en particular de la zona sismogènica en marges de subducció. El desenvolupament del TOMO3D es basa en el TOMO2D, un codi d'avantguarda per a la tomografia conjunta de refraccions i reflexions en 2D. Els arxius de codi han estat reescrits, redefinint i introduint les funcions necessàries per dur a terme la inversió 3D. Els testos fets amb la versió seqüencial del codi posen de manifest la necessitat de paral·lelització ja que l'increment de la mida dels conjunts de dades així com la modelització de la dimensió espacial afegida fan que les inversions siguin computacionalment exigents. La versió paral·lelitzada del TOMO3D ha sigut aplicada a un cas sintètic complex que simula una zona de subducció. Aquesta primera aplicació 3D serveix per avaluar la correcció de la programació del codi, i com a descripció pas a pas del procediment de modelització. Els resultats demostren la capacitat del codi per recuperar acuradament la distribució de velocitat i la geometria dels dos reflectors. Finalment, el TOMO3D és aplicat a un conjunt 3D de dades de sísmica de gran angle adquirit al marge pacífic d'Equador i Colòmbia per extreure'n un model 3D de la velocitat de les plaques cavalcant i subduïda, que és comparat amb el resultat obtingut amb un codi 3D de tomografia de temps de trajecte de refraccions (FAST). La comparació indica que el TOMO3D és més acurat que el FAST però al mateix temps és computacionalment més exigent. Tot i així, la paral·lelització del TOMO3D permet utilitzar plataformes de supercomputació, a diferència del que passa amb el FAST i la majoria de codis existents.