GNGTS 2024 - Atti del 42° Convegno Nazionale

Session 3.3 GNGTS 2024 Introducton In areas rich with ancillary data, their integraton in the inversion is a must, for validaton as well as for enhancing sensitvity. However, data integraton can be a tricky process for many reasons: biased data, diference in supportng volume along with their locaton, or they may have been acquired in diferent periods, with variatons occurring in between due, for instance, to the depleton of groundwater resources or seawater intrusion. Confictng data in an inversion process can easily prevent the proper convergence of the inversion, but culling too much data out might throw out important informaton. The removal of confictng informaton is even more difcult when there is a signifcant amount of ancillary informaton, acquired over a long period of tme. To solve this challenge, we propose to use a generalizaton of the minimum support norm (Last and Kubik, 1983; Portniaquine and Zhdanov 1999), namely the asymmetric generalized minimum support AGMS norm (Fiandaca et al., 2015), for identfying outliers in a joint inversion of AEM data, vertcal electrical soundings (VES) and borehole resistvity logs. We test the method on a synthetc example, mimicking a joint inversion of AEM data and borehole logs, with both correct and incorrect logging, as well as real data. The feld case consists of a SkyTEM survey carried out in 2022, complemented with a vast and open-source database of ashore resistvity logs, as well as VES, acquired over many decades. Method and results The inversion of AEM, VES and borehole logs is carried out in EEMverter (Fiandaca et al., 2024), a new inversion algorithm in which diferent norms are applicable in the objectve functon for both data misft and regularizaton through the iteratvely reweighted least squared (IRLS) inversion scheme (Farquharson and Oldenburg, 1998). In partcular, the penalty of the data misft x=d-f between data and forward response is expressed through the AGMS norm (Fiandaca et al., 2015) as: (1) where . (2) ϕ ( x ) = α −1 ( 1 − β ) ( x 2 σ 2 ) p 1 1 + ( x 2 σ 2 ) p 1 + β ( x 2 σ 2 ) p 2 1 + ( x 2 σ 2 ) p 2 β = ( x 2 σ 2 ) ( p 1 , p 2 ) 1 + ( x 2 σ 2 ) ( p 1 , p 2 )