“Using potential field data and stochastic optimisation to refine 3D geological models” by Richard Lane (Geoscience Australia, firstname.lastname@example.org), Phil McInerney (Intrepid Geophysics, email@example.com), Ray Seikel (Intrepid Geophysics, firstname.lastname@example.org), and Antonio Guillen (BRGM and Intrepid Geophysics, email@example.com). Paper presented at the Geophysics Session, PDAC, Tuesday, March 4 2008, Toronto, Canada. Abstract : As a geoscience agency, Geoscience Australia has sought a platform that allows us to integrate complimentary but diverse sources of information into consistent products. Several groups have made progress by blending 3D geological mapping and potential field modelling. We describe the approach implemented in GeoModeller software and illustrate typical workflows using a synthetic example and a case study involving the San Nicolas volcanogenic massive sulphide deposit. Starting with an initial 3D geological map, typically based on sparse surface observations, we utilise potential field data to investigate the viability of the proposed configuration of geological units at depth. Forward modelling of the property distribution derived from the 3D geological map and supplied rock property estimates allows us to simulate any of the gravity and magnetic fields or their associated vector or gradient tensor components. A visual comparison of the calculated and observed potential field data provides immediate feedback on the consistency between the 3D geological map and the observed potential field data. We may also use a bounded property optimisation procedure to derive an alternate combination of properties for the geological units (i.e., the combination that would best reproduce the supplied potential field observations). A review of the results obtained with these two simple procedures is used to identify any significant changes that are required for the 3D geological map or our estimates of the properties. Several iterations of geological mapping, forward modelling and property optimisation are generally required to derive a “reasonable” candidate 3D geological map for further consideration. At this point, a powerful geometry optimisation procedure can be used to fully invert the potential field data. The ambiguity that is inherent in this process is reduced by simultaneously inverting any number of gravity and magnetic data types and by doing so with strong geological constraints. The procedure utilises random elements and statistical decision rules to produce a large number of viable models, in contrast to the more common deterministic approach that results in a single “best” model. Statistical techniques are then used to analyse the acceptable models and identify important features of the 3D geological maps that are consistent with both geological and geophysical observations.