b'AEM 2023Short abstractsCalifornias Central Valley and will include the collection of infillIn this study we present a way to integrate AEM data and AEM data, as well as other ground-based geophysical surveys. other types of resistivity data (boreholes electrical logging and vertical electrical soundings, in this case), through an inversion Closing the gap between galvanic and inductivescheme that identify automatically conflicting data without methods: EEMverter, a new 1D/2D/3D inversion toolpreventing the general convergence of the process. To do so, we make use of a generalisation of the minimum support norm, the for electric and electromagnetic data with focus onasymmetric generalised minimum support (AGMS) norm, for induced polarisation defining the data misfit in the objective function of an iterative reweighted least squared (IRLS) gauss-newton inversion. The Gianluca Fiandaca1, Bo Zhang2, Jian Chen1, Alessandro Signora1,AGMS norm in the data misfit puts a cap on the weight of Francesco Dauti1, Stefano Galli1, Nicole Anna Lidia Sullivan1,non-fitting data points, allowing for the inversion to focus on Arcangela Bollino1 and Andrea Viezzoli3 the data points that can be fitted. Outliers are identified after 1. University of Milano, Milano, Italy the AGMS inversion and excluded, in order to complete the 2. Jilin University, Jilin, China inversion process with a classic L2 misfit.3. EMergo srl, Cascina (PI)The interest on Induced Polarisation (IP) in AEM data (AIP)The development of the TEMPEST AEM systemhas significantly increased in recent years, both within theAndrew Greenresearch community and in the industry. However, the inversion of AIP data is particularly ill-posed, especially when spectralOTBC Pty Ltd, Pymble, NSW, Australiamodelling, such as Cole-Cole modelling, is used. Furthermore, the comparison of AIP and galvanic ground IP inversion modelsTEMPESTs origins lie in the difficulty half-sine AEM systems had is hindered by the fact that the IP effect is usually modelledin mapping the Australias dryland salinity. This resulted in the differently in the inductive and galvanic computations. development of the SALTMAP system, a collaboration between World Geoscience Corp and CSIRO. This was a 500 Hz square In this study we present a new inversion software, EEMverter,wave system with excellent high frequency response, full-which has been developed to model IP in electric andwaveform digital acquisition, processing, calibration and bird electromagnetic (EM) data within the same inversionpositioning. With the advent of the CRC for Australian Mineral framework. In particular, three specific goals have beenExploration Technologies (CRCAMET) and an industry push for identified within EEMverters development: i) to allow multiplean Australian system with deeper penetration, the SALTMAP inversion cycles that mix, sequentially or simultaneously, 1D,System was taken to a lower base frequency (25 Hz) and higher 2D and 3D forward modelling, for diminishing the inversionpower while retaining as much higher frequency response as burden; ii) to allow the joint inversion of AIP, ground EM-IP andpossible. The previously implemented signal processing and ground galvanic IP data; iii) to allow time-lapse inversions of AIP,calibration was retained enabling a reliable conversion to Step EM and galvanic IP data. Response for ease of interpretation.EEMverter has been tested on several AEM and AIP surveys, alsoThe development history of TEMPEST is a result of collaboration in conjunction with ground EM and ground galvanic IP databetween company, university and government research. Funding in joint inversion. In this study, the inversion of the VTEM AIPcame from a diverse range of sources, government grants, survey over the Valen Cu-Ni deposit is presented, highlightingcollaborative industry funds and WGC. However, like most other the improvements in model resolution when compared tofixed-wing systems, it was caught up in the consolidation and standard inversion approaches. subsequent decadal changes in ownership that started after TEMPEST first became operational. But the consolidation was Automated integration of AEM data, VES andgood for TEMPEST. At the end of 2000 it was operating on a borehole logs platform that had limited power and an airframe that constrained the bird to a shape that made coil motion noise difficult to Stefano Galli1, Frans Schaars2, Frank Smits3,4, Lucas Borst5,reduce. The merger with Geoterrex brought new aircraft, better Arianna Rapiti6 and Gianluca Fiandaca1 coil suspension and extensive operational experience that took TEMPEST to another level of operational efficiency.1. Universit degli Studi di Milano, Milano, ITALY, Italy2. Artesia Water, Schoonhoven, The Netherlands3. Waternet, Amsterdam, The Netherlands Deep learning for the inversion of Airborne EM data4. Technical University of Delft, Delft, The Netherlands Eldad Haber1 and Cristoph Schwartzbach25. PWN, Velsenbroek, The Netherlands6. Emergo srl., Cascina (PI), Italy 1. University of British Columbia, Vancouver, BC, Canada2. CGI, VancouverAirborne electromagnetic (AEM) surveys are widely used for hydrogeological applications. The areas targeted for AEMIn the recent decade Deep Learning have revolutionised fields campaigns may present a great deal of ancillary informationsuch as computer vision and image understanding. However, its (e.g. resistivity logs, lithology, etc.) and integrating it with AEMuse for the solution of inverse problems have been limited. In data is fundamental. Yet, using this information either as a-priorithis work we examine the use of deep learning for the processing or a-posteriori may bring out conflict between different datasets,and inversion of airborne EM data. Preliminary results show preventing reconciliation everywhere. For instance, somethat by incorporating deep learning it is possible to eliminate borehole drillings may have been logged inaccurately, AEM datamany of the artefacts that are commonly observed in airborne may present bias, or data may have been acquired at differentinversion allowing us to obtain much more reliable inversions times, with variations occurring in between. that fit not only the data, but also our a-priori information.AUGUST 2023 PREVIEW 56'