Rapid detection and classification of airborne time-domain electromagnetic anomalies using weighted multi-linear regression

Abstract

Listen

Translate article

We propose a rapid and efficient methodology for the detection and interpretation of airborne time-domain electromagnetic anomalies generated by thin sheet-like volcanogenic massive sulphides (VMS) deposits in a resistive environment, which are representative of VMS deposits in the Canadian Shield.In the first step of the approach, we use high-order statistics for the detection and the recognition of a MEGATEM anomaly as indicating a thin sheet-like VMS deposit with respect to three criteria of detection: the minimum level of detection, the length of detection, and the coherence of detection over time. We adapt these criteria in order to optimise the detection of thin sheet-like VMS deposits against geological noise models. Once the anomaly is detected and recognised as the response to a thin sheet conductor, we interpret the model geometry and physical property using attributes calculated from the MEGATEM anomaly. We develop a system of weighted multi-linear regression to find the most significant attributes to estimate the dip, depth, conductance, and dimensions of a thin sheet-like VMS deposit. Stepwise regression suggests that shape attributes are most significant to estimate dip while depth is most strongly estimated by size attributes. The most significant attribute to estimate the conductance is the time constant. The size is best estimated by attributes related to the size of the anomaly. We test the regression system on thin sheet models with excellent performance. Most of the parameters of the thin sheet models were estimated within an interval of confidence about the initial property. We further test the system by estimating properties of three VMS deposits in the Abitibi Greenstone Belt, Québec, Canada, for which the geometries and geological properties are known. Most parameters are estimated within the interval of confidence for ISO, a thin sheet body, while the estimates for New-Insco and Gallen show more variability caused by departure from the reference thin sheet model.