GNGTS 2019 - Atti del 38° Convegno Nazionale

GNGTS 2019 S essione 3.2 709 overestimates the true velocity value. Notwithstanding the velocity inversion, the algorithm accurately predicts a low velocity layer. The uncertainties in the estimated Vs rapidly increase as the depth increases and the velocity of the deepest layer is not well recovered with a MAP solution that significantly underestimates the actual velocity value. The Vp / Vs ratio is again not resolved. Fig 3.c demonstrates that the HMC algorithm correctly identifies the position of the interfaces. Only 30 iterations are enough to converge toward the stationary regime, whereas the observed data is still perfectly matched. Conclusions. We implemented a Hamiltonian Monte Carlo (HMC) algorithm for Rayleigh wave dispersion curve inversion. This approach ensures reliable assessment of the posterior uncertainties in highly non-linear inverse problems and guarantees efficient sampling even in high-dimensional model spaces. This ability rests on the exploitation of derivative information of the misfit function that is not considered by other standard Monte Carlo methods. In this work, the algorithm has been implemented for a Gaussian prior model, but another outstanding benefit of HMC is the possibility to consider either parametric or not-parametric priors. Our tests demonstrated the applicability of the proposed HMC approach for Rayleigh wave dispersion curve inversion and the possibility to derive the optimal model parameterization by adopting standard statistical tools. In particular, the algorithm yielded uncertainty quantifications and model predictions in accordance with the expected model parameter illuminations. The HMC algorithm presented here can be easily extended to include higher modes. We are now testing the algorithm on field data inversions. Furthermore, we are extending the presented HMC approach to full-waveform inversion of surface waves. References Betancourt, M. (2017). A conceptual introduction to Hamiltonian Monte Carlo. arXiv preprint arXiv:1701.02434. Cercato, M. (2011). Global surface wave inversion with model constraints. Geophysical Prospecting, 59(2), 210-226. Duane, S., Kennedy, A. D., Pendleton, B. J., and Roweth, D. (1987). Hybrid Monte Carlo. Physics Letters B, 195, 216–222. Socco L.V. and Strobbia C. (2004). Surface-wave method for nearsurface characterization: a tutorial. Near Surface Geophysics 2, 165–185. ADVANCED GPR DATA ANALYSIS FOR GLACIOLOGY: A MULTI-YEAR STUDY ON THE MARMOLADA GLACIER (DOLOMITES, ITALY) I. Santin 1 , E. Forte 1 , M. Pavan 2 , M. Valt 3 , M. Žebre 4 , R.R. Colucci 5 1 Department of Mathematics and Geosciences, University of Trieste, Italy 2 Department of Earth, Environmental and Life Sciences, University of Genova, Italy 3 ARPAV - Agenzia Regionale per la Prevenzione e Protezione Ambientale del Veneto, Arabba, BL, Italy 4 Department of Geography & Earth Sciences, Aberystwyth University, United Kingdom 5 Department of Earth System Sciences and Environmental Technology, CNR – ISMAR, Basovizza, Italy Introduction. Knowing the evolution of glaciers’ changes in shape and size is extremely useful for both glaciological studies (Carturan et al., 2013), practical application (Diolaiuti et al., 2006) and, even more important, for global climate assessment (Zemp et al., 2013). For this reason, long-term glacier monitoring has been performed generally since the beginning of the last century to understand the physical processes which lead to glaciers’ response to climate change (Haeberli et al., 2013). In order to monitor mass variations of a glacier with time (i.e. its mass balance) and provide more realistic forecasts about its future evolution, detailed information about volume and internal structure is required in addition to classical linear and areal measurements. Ground Penetrating Radar (GPR) has proved to be an efficient instrument