Deep learning based expert system to automate time-domain electromagnetic data processing

Transient Electromagnetic (TEM) methods are routinely used to obtain detailed understanding of the subsurface, which may be used for a variety of applications such as groundwater mapping and mineral exploration. Modern TEM surveys, employing driving or flying during data collection, result in large datasets that may contain thousands of line kilometers of data. Parts of these data will often be contaminated by interference from man-made conductors, e.g. fences, buried power lines, known as “couplings”. If such disturbed data are inverted, the geological interpretation will be severely degraded in most cases. Therefore, couplings must be identified and removed from the data before inversion. The process of removing couplings is a time-consuming and highly sophisticated manual task. Machine learning based methods have been suggested as obvious automation tools, and the general approach has so far been to use large datasets of manually processed TEM data in a supervised learning approach. The problem with this is that it may be biased to local geological conditions and/or biased toward the individuals who perform the manual assessment (for instance a conservative versus optimistic coupling removal approach).