Can a borehole conductivity log discredit a whole AEM survey?

Abstract

Airborne EM surveys are a well-established method for quickly investigating an area in order to assess what lies beneath the surface. The accuracy of conductivity estimates derived from AEM surveys are negatively impacted by (a) imprecise characterization of the AEM system, (b) miscalibration of data, incorrect estimations on noise levels, (c) assumptions made in the data processing and transformation/inversion to conductivity, and (d) non-uniqueness of the derived models. It is therefore prudent to ?validate? the derived models against independent conductivity measurements. Borehole conductivity induction logs are one common source of independent information. When comparing data sets of different origins, such as AEM and down-hole conductivity logging, caution must be used. Immediately evident is the considerations of scale: the bulk conductivity from an airborne survey is affected by large-scale structures such as discontinuous layers and regional faults; whilst borehole skin depth and the conductivity values are affected by casing materials, voids, temperature and calibration. Conductivities measured from boreholes come from an invasive method, where drilling can disrupt the in situ layering and material conditions. All these aspects impact on the conductivity measurements. Airborne data can also be incorrectly processed and interpreted by using incorrect wave forms, frequencies, and altitudes, or by not taking proper care when correcting for instrument drifts. Borehole conductivity logging can successfully reduce the influence of calibration errors in AEM data but boreholes are usually sparsely concentrated and subject to calibration errors of their own. Therefore, the need of a measure of ?goodness? in the correlation between data sets is desirable. We suggested a generalisation in the regression analysis to assess the ?consistency? and ?misfit? between AEM inversion results and borehole data, based on an extension of the Pearson product-moment coefficient of correlation that accounts for errors on both data sets.