BEGIN:VCALENDAR VERSION:2.0 PRODID:-//Talks.cam//talks.cam.ac.uk// X-WR-CALNAME:Talks.cam BEGIN:VEVENT SUMMARY:Uncovering hidden scales – AI-driven\, Bayesian imaging and moni toring of subwavelength microstructures with finite-frequency waves - Ivan Vasconcelos (Universiteit Utrecht) DTSTART:20230131T141500Z DTEND:20230131T150000Z UID:TALK194524@talks.cam.ac.uk DESCRIPTION:More often than not\, important geologic processes occur at mi cro-scales\, e.g.\, fluid flow\, mineral-phase changes\, chemically-induce d alteration\, rock-frame compaction\, or even mechanical ruptures/instabi lities leading to large earthquakes. However\, reliably imaging material p roperties at such scales from remote long-wavelength information contained in either seismic or EM fields has long been a challenge to the geophysic al\, engineering and material science communities. In this talk\, we prese nt a general framework for the estimation of sub-wavelength material prope rties from long-scale waves\, building on recent advances on statistical m icrostructure descriptors (SMDs) within the field of material science. &nb sp\;In geoscience\, traditional approaches to describing material microhet erogeneity rely on either analytical inclusion-based models\, or in sample -based digital rocks: each of these having their pros and cons. Here\, we by discussing the role of SMDs and more importantly  \;the so-called & lsquo\;Reconstruction&rsquo\; problem\, to statistically describe microhet erogeneous geo-materials in a manner that is capable of generalizing compl ex geometrical information hidden in microstructures\, while also retainin g realism and sample fidelity. To these advances in material descriptions with imaging\, we rely on wave-equation-based Strong Contrast Expansions ( SCEs) to predict frequency/scale-dependent effective wave properties for a coustic\, elastic and EM waves. \; We briefly discuss how SMD-describe d microstructures affect long-wave properties &ndash\; and in particular h ow they not only predict frequency-dependent attenuation due to sub-wavele ngth scattering\, but that attenuation is particularly sensitive to micros tructure when compared to effective wavespeeds. When it comes to the estim ation of microstructure properties from wave observations\, the problem be comes substantially more difficult because realistic microscale parameters could in principle have far too many degrees of freedom than what is obse rvable from finite-frequency wave data. As such\, it is key that any metho d that aims at realistically retrieving microstructure information from lo ng-scale wave data accounts for uncertainty\, while also handling the high ly nonlinear nature of microstructure-dependent effective wave properties. To that end\, we combine our SMD and SCE approaches for effective wave pr operties with the supervised machine-learning method of Random Forests to construct a Bayesian approach to infer microstructure properties from effe ctive wave parameters as observables. This method yields full posterior di stributions for microstructure parameters (e.g.\, property contrast\, volu me fraction\, and geometry information) from frequency-dependent observati ons of wave velocities and attenuation. We present several examples of inf erence scenarios\, showing\, for example\, that i) attenuation is key to m icrostructure imaging\, and ii) microgeometry information can only be reli ably retrieved if either contrast or volume fraction are well known a prio ri. We illustrate of inference approach with several examples of analytica l and real microstructures\, including data from a \; laboratory compa ction experiment controlled by microscale CT imaging. LOCATION:Seminar Room 1\, Newton Institute END:VEVENT END:VCALENDAR