Inversion of multichannel seismic data by combination of travel-time and full-waveform tomography

Abstract

This thesis presents the implementation and application of a procedure combining different geophysical techniques to extract high-resolution information that helps characterizing the structure and properties (p-wave velocity, Vp) of the subsurface by using marine multichannel seismic (MCS) data alone. The challenge is overcoming the inherent non-linearity and non-uniqueness of inverse methods, in general, and of full-waveform inversion (FWI), in particular, which are especially critical for short-offset, band-limited seismic data. I have designed and applied a modelling sequence consisting of: (1) data re-datuming or downward continuation (DC) by back-propagation of the recorded seismograms to the seafloor; (2) travel-time tomography (TTT) using especially the first arrivals of the re-datumed shot gathers, and (3) FWI of the original shot gathers using the model obtained by TTT as initial reference. This workflow is first tested with synthetic data, and then applied to field data acquired in the Alboran Sea (SE Iberia).

Due to the short source-receiver offset and the depth of the water column, refractions are hardly identified as first arrivals. To solve this problem, I changed the reference datum of the data set from the sea surface to the seafloor, by implementing a DC code that uses a solver of the acoustic wave equation developed at Barcelona-CSI [Dagnino et al., 2016]. By modifying the MCS records to simulate a seafloor-type acquisition it is possible to recover refracted phases, crucial in Vp modelling, as first arrivals.

Then, I performed TTT using the travel times of the DC first arrivals to obtain a coarse, but kinematically correct, Vp model. This TTT Vp model has the correct low-wavenumber information because the waveforms simulated with the inverted model and the recorded ones are not cycle-skipped. To better constrain the result, particularly in the deep parts of the model, I have incorporated the seismic phases corresponding to a major reflecting interface (top of the basement, TOB) and performed a joint refraction and reflection TTT combining the original and the DC field data.

Finally, finer structural details are progressively introduced in the initial model by applying iterative, multi-scale FWI to the original MCS data. The results confirm that the combination of data re-datuming and TTT provides reference models that are accurate enough to apply FWI to relatively short offset streamer data in deep-water settings as the ones used, even if records lack low frequencies (< 4 Hz). I also show that, when the initial model is not kinematically correct, the FWI falls into a local minimum.

The application of the workflow to the Alboran field data reveals a number of geological structures in the FWI Vp model that cannot be appreciated in the TTT Vp model, nor easily interpreted based on MCS images alone. A sharp strong velocity contrast defines the geometry of an irregular TOB that includes high velocity volcano-like structures. The model clearly images steeply dipping Vp changes at the flanks of the basin that may correspond to faults. Moreover, it displays a 200–300 m thick high-velocity layer that could probably correspond to evaporites deposited during the Messinian crisis in the Mediterranean. The result is validated by comparing the two-way time-transformed Vp model and the time-migrated MCS image, showing that velocity changes coincide with major reflectivity contrasts.

Overall, this study shows that by using an appropriate workflow, in our case including DC of MCS data to the seafloor, joint refraction and reflection TTT, and FWI, accurate, geologically meaningful Vp models can be obtained even for non-optimal data sets. In particular, our results provide information that improves the geological characterization and interpretation of the subsurface of the Alboran Basin. The main results have recently been published in Solid Earth [https://doi.org/10.5194/se-2019-46].

Aquesta tesi presenta el desenvolupament, implementació i aplicació d'un procediment que combina diferents tècniques d'inversió tomogràfica per extreure informació d'alta resolució que permeti caracteritzar l'estructura i propietats (velocitat d'ona p, Vp) del subsòl marí, utilitzant exclusivament dades de sísmica de reflexió multicanal (MCS). El repte principal és el de superar els problemes inherents de no-linealitat i no-unicitat dels mètodes d'inversió, en general, i de la inversió de forma d’ona completa (FWI), en particular, especialment crítics en registres sísmics que manquen de baixes freqüències (<4 Hz) i han estat adquirits amb un abast experimental relativament curt (~6 km). Per afrontar el problema, l’estudi proposa i segueix un flux de treball que primer es posa a prova amb dades sintètiques, i després s'aplica a dades de camp adquirides al mar d’Alboran.

Primerament es va desenvolupar un codi que modifica les dades sísmiques retro-propagant-les a la superfície del fons marí, simulant així una adquisició virtual en aquesta superfície. Els registres resultants permeten identificar les refraccions com a primeres arribades, que aporten informació robusta i essencial per modelar la Vp. Posteriorment, amb els temps d'aquestes arribades es va dur a terme la tomografia de temps de trajecte (TTT). En l'aplicació amb dades de camp, es va afegir la reflexió del sostre del basament (TOB) per acotar millor el resultat. Els models obtinguts mostren la variació de Vp correcta del subsòl, ja que els sismogrames simulats amb aquests models i els que es pretenen reproduir no mostren desfasaments importants. Aquest fet possibilita la correcta aplicació de tècniques més complexes com la FWI utilitzant el model de TTT com a inicial. La FWI proporciona un model de Vp d’alta resolució del medi utilitzant tot el camp d'ones de les dades originals. Aquesta tesi és la primera aplicació pràctica de FWI amb dades de camp a nivell nacional, i s’han utilitzat essencialment codis desenvolupats al BCSI. Els resultats revelen diverses formes geològiques d'interès, la geometria irregular del TOB, causada per possibles estructures volcàniques, i falles; a més d'una capa d'alta Vp que pot correspondre a evaporites dipositades durant el Messinià. Els resultats principals s’han publicat recentment a Solid Earth [https://doi.org/10.5194/se-2019-46].