Project

Back to overview

GEOFACES: GEOphysics-based FAlsification and Corroboration in the Earth Sciences

Applicant Linde Niklas
Number 184574
Funding scheme Project funding (Div. I-III)
Research institution Institut des sciences de la Terre Université de Lausanne
Institution of higher education University of Lausanne - LA
Main discipline Geophysics
Start/End 01.11.2019 - 31.10.2023
Approved amount 808'937.00
Show all

All Disciplines (2)

Discipline
Geophysics
Hydrology, Limnology, Glaciology

Keywords (6)

Inversion; Model selection; Uncertainty quantification; Markov chain Monte Carlo; Geostatistics; Geophysics

Lay Summary (French)

Lead
Ce projet aborde comment on peut transformer les signaux géophysiques en images réalistes de propriétés (hydro)géologiques ou de représentations de processus avec une quantification appropriée de l'incertitude. L'accent est mis sur l’utilisation des données géophysiques pour classer des modèles conceptuels décrivant les processus et les propriétés de la subsurface.
Lay summary
Le projet GEOFACES vise à créer des représentations probabilistes qui intègrent des informations géologiques et hydrogéologiques à priori. Les progrès de l’apprentissage automatique, de la statistique mathématique et des disciplines connexes seront utilisés pour construire des images réalistes de la subsurface. Plus spécifiquement, le projet vise à relever trois défis principaux. (i) Comment construire efficacement des images géologiquement réalistes et en accord avec des données géophysiques qui capturent les incertitudes découlant de données imprécises et d'un échantillonnage incomplet? Pour ce faire, nous nous appuierons sur les avancées récentes en matière d’apprentissage approfondi, de géostatistique et d’optimisation. (ii) Comment pouvons-nous transformer des données géophysiques intrinsèquement sensibles aux propriétés physiques en images de propriétés (hydro)géologiques par le biais de relations incertaines? Des nouvelles méthodes statistiques exactes et des approximations itératives seront utilisées. (iii) Comment pouvons-nous tester des hypothèses scientifiques ou des modèles conceptuels concurrents en utilisant des données géophysiques? Plus spécifiquement, nous aborderons les hypothèses diffèrent concernant la dynamique des sols gelés et leur impact sur l'écoulement de l'eau en haute montagne dans le Colorado, aux États-Unis. Ce travail a des implications importantes pour les utilisations scientifiques et commerciales en géophysique puisqu'il cherche à supprimer le stade interprétatif ambigu et subjectif avec une approche intégrée et objective dans laquelle les connaissances spécialisées sont intégrées dès le départ.
Direct link to Lay Summary Last update: 25.06.2019

Responsible applicant and co-applicants

Employees

Project partner

Associated projects

Number Title Start Funding scheme
155924 Summary statistics and model selection in Hydrogeophysics 01.02.2015 Project funding (Div. I-III)
132249 Integrated methods for stochastic ensemble aquifer modelling (ENSEMBLE) 01.03.2011 Sinergia

Abstract

Geophysical data play a determinant role in many Earth science disciplines. Today, it is becoming possible to realistically simulate geophysical responses for virtually any distribution of physical properties and experimental designs. Geophysical data are routinely inverted by regularized deterministic inversions that provide upscaled estimates of subsurface properties. Corresponding localized uncertainty quantification and resolution analysis originating from linear inverse theory provide a partial assessment of model uncertainty and informs about the scales at which subsurface properties are resolved. In addition, upscaling methodology, detailed laboratory analyses, and borehole logging have enhanced our understanding of the links between the physical properties that are sensed by geophysical data and the primary properties (e.g., porosity, permeability) and state variables (e.g., salinity, temperature, pressure, water content) that are of interest to the wider Earth science community. Nevertheless, prevailing approaches to geophysical inversion build on strong inherent assumptions about underlying conceptual models, they are largely unsuitable to formally test scientific or practical hypotheses about subsurface structures of governing processes. This is also true for theoretically more solid global Bayesian inversion methods that aim at inferring the full posterior distribution of model parameters. There has been several eloquent criticisms within the Earth science community highlighting the limits of the prevailing-essentially positivist-approach employed in geophysics and other quantitative Earth sciences. Yet there are few practical alternatives available in the literature and even fewer field demonstrations of a more Popperian perspective for non-trivial Earth science problems. There is, thus, a need to develop a practical and sound scientifically-based hypothesis-driven approach, which is designed such that geophysical data are directly used to answer research questions and solve practical applications in the Earth sciences. That is, to move towards formal testing of scientific hypotheses and competing conceptual models against geophysical data. The project GEOFACES aims ultimately at developing a general approach to falsify and corroborate conceptual Earth science models using multi-method geophysical data that is robust and feasible for large-scale problems. The term conceptual models refer here both to alternative geological descriptions of Earth structure (spatial heterogeneity) and process understanding (governing equations). The framework and its associated methodological components will be used to answer complex science-driven research questions with specific emphasis on an alpine catchment; an environment that is particularly sensitive to climate change. To accommodate this framework while further advancing geophysical inverse theory, new methodological developments are proposed, notably to encode geological realism in deterministic and probabilistic inversions; to condition geostatistical realizations to geophysical tomograms; to account for petrophysical and upscaling errors in inversions; and to enable (exact and approximate) estimates of the evidence (marginal likelihood) needed for Bayesian model selection.
-