PHYSCON 2009, Catania, Italy, September, 1–September, 4 2009 INTEGRATED INVERSION OF NUMERICAL GEOPHYSICAL MODELS USING ARTIFICIAL NEURAL NETWORKS Agnese Di Stefano Dip. Ing. Elettrica, Elettronica e dei Sistemi University of Catania Ist. Naz. Geofisica e Vulcanologia Sezione di Catania, Italy distefano-a@ct.ingv.it Gilda Currenti Ist. Naz. Geofisica e Vulcanologia Sezione di Catania Italy currenti@ct.ingv.it Ciro Del Negro Ist. Naz. Geofisica e Vulcanologia Sezione di Catania Italy delnegro@ct.ingv.it Luigi Fortuna Dipartimento Ing. Elettrica, Elettronica e dei Sistemi University of Catania Italy lfortuna@diees.unict.it Giuseppe Nunnari Dipartimento Ing. Elettrica, Elettronica e dei Sistemi University of Catania Italy gnunnari@diees.unict.it Abstract A unified modelling procedure is proposed to jointly interpret the variations observed in geophysical data and to properly take into account the relationship be- tween the intrusive processes and the geophysical vari- ations expected at the ground surface. We focus on the joint inversion of geophysical data by a procedure based on Artificial Neural Network (ANN) for the esti- mation of the volcanic source parameters. As forward model, we developed a 3D numerical model based on Finite Element Method (FEM) for computing ground deformation, magnetic and gravity changes caused by magmatic overpressure sources, with the aim to con- sider a more realistic description of Etna volcano, in- cluding the effects of topography and medium hetero- geneities. Key words Identification, Modeling, Numerical methods. 1 Introduction Geodetic and potential fields investigations have been playing an increasingly important role in Mt. Etna eruptive processes ([Bonaccorso et al, 1999]; [Bon- forte et al, 2008]; [Del Negro et al, 2004]; [Napoli et al, 2008]; [Carbone et al, 2007]; [Carbone et al, 2008]). The amount of available data collected repre- sents a valuable database, but limited efforts have been made for an effective integration of different data. Even if complicated models have been proposed, ground de- formation, magnetic and gravity data are usually inter- preted separately from each other and the joint mod- elling has remained elusive, despite the obvious ben- efit in constraining the solution. When the cause of their variations can be ascribed to the same volcanic source, a joint inversion would be advisable in order to identify the source parameters with a greater degree of confidence [Nunnari et al, 2001]. For an integrated inversion modelling, complex inverse methods are re- quired to combine forward models with appropriate op- timization algorithms and automatically find the best set of parameters that well matches the available obser- vations. Hence, the rationale of the inversion modelling approach requires: (i) solution of forward models, (ii) numerical inversion procedure. The forward problem consists in deriving a relationship between sources and observations. Over the last decades, straightforward analytical solutions for simplified geometric sources have been devised under the assumption of homoge- neous elastic half-space medium ([Mogi, 1958]; [Sa- sai, 1991]; [Hagiwara, 1977]). To overcome this in- trinsic limitation and provide more realistic models, which consider various geometries as well as compli- cated distribution of medium properties, numerical so- lutions can be investigated. With the aim to consider a more realistic description of Mt Etna, we developed a numerical procedure based on Finite Element Method (FEM) to evaluate geophysical changes caused by over- pressure source in a 3D formulation. The FEM-based numerical model is able to include not only compli- cated distribution of both rock magnetization and elas-