I am an applied computational geoscientist, who synthesizes theoretical, observational (e.g., seismology and Bayesian computational methods) and experimental evidence (e.g., mineral physics and geochemistry). I seek to improve our understanding of the chemical composition and physical behavior of the solid earth.
My research into the physical behavior of rock materials (frequency-dependent attenuation) yielded new insights into the nature of the oceanic and continental lithosphere that explained the softness internal to the rigid and permanent continental lithosphere and predicted the abrupt and shallow lithosphere-to-asthenosphere transition in the oceans. I applied novel scattered wave (receiver function, RF) techniques to image the structure of the crust and upper mantle in a variety of settings in the Pacific Ocean, starting with normal ocean sea floor in the Japanese sea as well as anomalous ocean islands like Hawaii, Wake and the Galapagos Islands.
My research has turned to shallower and land-based targets: investigating thetypes of rocks that make up the continents globally, particularly the lower continental crust which is inaccessible to direct sampling. Improved compositional models of the crustal would enlighten us about it’s evolution, history, and reasons for its longevity. Seismological data and techniques are one of the few ways in which this portion of the Earth can be studied directly. I am developing methods to take advantage of the plethora of new data for the continental United States provided by EarthScope, and applying novel data assimilation techniques to jointly invert large and complementary seismological data sets. My goal is to provide a new generation of crustal models with quantitative uncertainties, which can then be combined with geochemical and petrological data to constrain the lithology of the crust.
I will continue to pursue research at the nexus of a variety of geoscientific sub-disciplines. This will include the interpretation of new seismological results (e.g., the presence of shear zones in the crust and its relationship to the origin of ocean islands), testing the predictions of old models (e.g., attenuation in the middle lithosphere and sharpness of the upper mantle discontinuities). I also expect to gain new insights by applying computationally sophisticated modeling methods to the growing dataset of seismic measurements (e.g., Bayesian inversion for azimuthal and radial anisotropy in ambient noise measurements, surface wave ellipticity and other geophysical observables like heat flow and gravity).
Exploring the Pacific Ocean
In the past decade, receiver function studies of the continental and oceanic LAT have focused on the shear wave receiver functions (S-RF) methodology, as primary wave receiver function (P-RF) studies have been difficult to use because waves scattered multiple times within the shallow crust interfere with direct converted phases expected from a deeper LAB conversion. However, P-RFs main advantage compared to S-RFs is substantially greater resolution. This increased resolution is needed in order to test the sub-solidus grain-boundary sliding model (Olugboji et al. 2013), which predicts that gradients in velocity should vary with age. I tested predictions of the anelastic relaxation hypothesis using the P-RF technique for crustal and upper mantle structure beneath ocean bottom stations located on the sea floor of the Shikoku basin and Northwest Pacific, as well as on select ocean islands in the Pacific (Olugboji 2014a, Olugboji and Park 2016a, Olugboji et al. 2016b).
Through collaboration with Japanese scientists, I secured data from oceans underlain by “normal” mantle, far from ocean islands (Shinohara et al., 2006, 2008; Isse et al., 2009, 2010). These data (including buried stations and 10 ocean bottom stations) allowed me to investigate LAT structure unimpeded by complicated crustal structure. Age dependence of LAT depth and sharpness of the velocity gradient was investigated to test the predictions of various models of LAT formation. I also applied a method of harmonic decomposition of P-RFs (e.g., Park and Levin, 2016; Levin and Park, 1998; Bianchi et al., 2010), which can provide constraints on the existence of depth-dependent anisotropic layering underneath a station. The results from this study support the EAGBS hypothesis (see Prague lecture).
Interestingly, in contrast to normal oceans, ocean islands do not fit simple models of ocean formation and evolution, and my study of their crustal structure reveals complexities that suggest a genetic difference from normal oceans (Olugboji and Park 2016a). I find that the crust beneath ocean islands is not only thicker than normal crust, but also comprises multiple layers, most of which exhibit strong evidence for dipping layers sometimes exhibiting anisotropic fabric. The first two characteristics have been documented in the past (e.g., Leahy and Park, 2005; Leahy et al. 2010), but the detection of multiple layers with anisotropy was unexpected. Using the novel technique of harmonic decomposition of receiver functions, I characterized the orientation and depth dependence of the shear zones in the ocean island crust. Future work will involve extending this analysis to a global dataset including the Atlantic and Indian oceans. This will allow us to test new hypothesis for ocean island formation and to compare its crust and lithospheric structure to that of normal oceans (e.g. Park et al., in prep, 2016)
Composition of the Crust with Seismology
Our understanding of crustal composition, especially lower crustal composition, is heavily informed by seismology since most of the deep earth is inaccessible to direct sampling. In-situ observations of seismic wave speeds, from various tectono-geographic locations, are compared to expected relationships between rock composition and seismic wave-speeds derived from laboratory measurements, calculated from thermodynamic relationships or both. Combining the seismological, geochemical and petrological data, it is then possible to deduce (exclude), with some confidence, different models of crustal composition.
I am working on addressing some challenges attending the use of seismological models to inferring crustal composition. The most obvious of these is the limited geographic distribution of the seismological data from which the nature of the crust is derived. For example, the most popular global crustal model, CRUST1.0 (Laske et al., 2013) and regional models like NACR14 (Tesauro, 2014) do not use all the available seismic data (e.g. ambient noise surface wave dispersion) to constrain seismic parameters. I remedy this by taking advantage of available seismic data (especially in the continental USA from EarthScope USArray). New seismic data is a great advantage (especially in the US). Equally important, however, is asking the question “How good is CRUST1.0 in light of new data (e.g. grumpygeophysicist.com)?” or “To what degree and in what regions can existing models be trusted?” I address these questions using a the Bayesian inverse approach (more below). With model uncertainties, I can approach this question in the framework of statistical inference (see poster, Olugboji et al., 2016 and manuscript in prep).
The over-arching goal is to improve on these crustal models using available seismic data, while reducing the biases and uncertainties that can be introduced by extrapolating constraints from one region to another through tectonic analogy. Also, by combining the rich techniques now routine in global seismology – ambient noise phase dispersion, receiver functions, and surface wave ellipticity – we can constrain more than just crustal Vs, but also its Poisson’s ratio (derived from Vp/Vs), radial and azimuthal anisotropy, the presence of internal discontinuities, as well as how gradational or sharp these discontinuities may be. With a multiple elastic parameter crustal model, derived completely from seismological data, the trade-offs in the second stage inference of composition using petrology and geochemistry can be reduced.
‘Big’ and Multifaceted Geophysical Data meets High Performance Computing
One of the challenging and exciting components of combining multiple datasets using the THBI
approach is the need to interface with computing technology. The amount of seismic data that needs to be processed is already massive, and will continue to grow rapidly. From the spectral methods for automating phase velocity measurements of ambient noise waveforms, to the rapid and structured data processing workflows applied to receiver function measurements, and the strategies of grouping waveforms using cluster-analysis during surface wave ellipticity measurements, the need to rethink traditional approaches to data analysis has become more important than ever before. In addition, when performing large inversions that involve ~1,600 seismic stations with multiple stage inversions and with model parameters in the tens of hundreds, it can take days to weeks to achieve convergence on a several-core desktop computing infrastructure. At the University of Maryland (UMD), I make use of the Deepthought2 high performance-computing (HPC) and MARC/BlueCrab cluster to increase efficiency in order to partially overcome the curse of dimensionality for large model space search techniques and tackle problems of unprecedented complexity using datasets of unprecedented size.
- Vtunes optimizing intel HPC code
Joint ‘Transdimensional’ Inversion of Multiple Seismic Datasets
In order to assimilate these different seismological datasets, I use an inverse technique k
nown as transdimensional, hierarchical Bayesian inversion (inference) (THBI). This technique enables one to assimilate datasets with different sensitivities, by quantifying the optimal weighting parameters. Importantly, the dimension of the model is also a parameter to be constrained by the data themselves (hence “transdimensional”). This means that we can determine the number of layers, say, that are required in the crust without making any limiting assumptions before commencing the inference (e.g., 3 layer vs. 2 layer crust). By conducting an exhaustive, non-restrictive, Markov chain Monte Carlo search of the dimension (e.g., number of layers), weighting (noise) parameters, isotropic and anisotropic elastic parameters, Bayesian inference allows us to quantify the full uncertainties associated with the crustal model. These uncertainties can then be propagated through to the second stage inference-using laboratory and thermodynamic measurements in order to quantify uncertainty associated with inferences of Earth’s composition.
I have applied this approach to quantifying the full uncertainties in ambient noise phase dispersion maps by constructing phase dispersion maps using THBI (Olugboji et. al., in prep). Using phase dispersion curves and their associated uncertainties extracted from these maps, I am inverting for a model of crustal Vs and associated uncertainties. The ultimate goal is to extend this approach to derive a full model that will include compressional wave speeds, density, and anisotropic perturbations in wave-speeds using receiver function data, and surface wave ellipticity (e.g., Chao & Lekic, 2014).
Back to top