Practical Inversions For Helicopter Electromagnetic Data

Discrete-layer inversions are the most accurate method of converting geophysical<br>data to a geoelectrical model of the earth. Smooth-model inversions and transforms<br>(Occam’s inversions and CDI sections) are more robust to execute, but only discretelayer<br>inversions can provide direct measurement of depth to a target layer, which is<br>necessary for engineering purposes. However, there can be a problem with non-unique<br>solutions – there may be two or more solutions that fit the data within the specified<br>accuracy. Also, discrete-layer inversions tend to be model specific. They generally use<br>a constant number of layers, similar conductivity contrast for the starting model, etc.<br>This can present a problem when the geological model changes within a data set,<br>changing the conditions or dropping data from high signal down to zero, for example.<br>An inversion process with built-in geological and geophysical “intelligence” can<br>overcome many of these limitations. Input data are weighted based on the signal level.<br>Questionable data, or data clearly influenced by non-target effects (e.g. power lines) are<br>rejected. Existing data, such as drill holes, are used to generate the best possible starting<br>model and the inversion process constrained to honour these data. Known geophysical<br>parameters, such as the conductivity of bedrock, or conductivity of a water layer, can be<br>fixed. Any of these factors can change across a data set or region, and the inversion will<br>adjust itself to match the changes.<br>Layer extraction algorithms have been developed to measure the depth to specific<br>layers from smooth sections, or to provide starting models from smoothed sections for<br>discrete-layer inversions.<br>This “intelligent” inversion process was used to generate depth-to-bedrock maps<br>over a sink-hole where the overburden changed from conductive clay to resistive sand.<br>Drill and seismic information was used to help generate the starting model for each data<br>point, and the inversion was constrained when close to these data points. Magnetically<br>permeable geology and cultural interference were identified, and the input data were<br>adjusted to minimize the effects of these non-conductivity effects.