CRC LEME OPEN FILE REPORT 163
CONSTRAINED INVERSION OF THE AEM DATA FROM THE LOWER BALONNE AREA, SOUTHERN QUEENSLAND, AUSTRALIA
R Lane , R Brodie and A Fitzpatrick
Borehole conductivity logs, ground electrical data and ground electromagnetic data indicated that conductivity values of several hundred mS/m would be expected at depths below 100m across the St George AEM Survey area. These observations were at odds with the conductivity predictions delivered by the survey contractor. This reflects the inherent ambiguity in estimation of ground conductivity from AEM data, compounded by incomplete knowledge of the relative geometry between the transmitter loop on a fixed wing aircraft, the receiver coils in a towed bird and the ground. In the absence of the independently measured conductivity values mentioned above, the contractor guided the conductivity predictions towards solutions that had low conductivity values at the depth of investigation of the system, between 100 and 200m in this instance.
With the benefit of the additional borehole and ground conductivity data, an alternate set of conductivity predictions were calculated by CRCLEME using EMFlow, guided in this case towards high conductivity values at depth. In addition to this change, small adjustments were made to the geometry parameters assigned to the system. This improved the match between the borehole conductivity logs and predictions of conductivity for observations near to the boreholes. Despite the improvement, this approach lacked control and did not make full use of the known or "a priori" information now available for the AEM system and the survey area.
The observed data are a combination of ground response and primary field. The latter contains important information about the offset of the receiver coils from the transmitter loop. Knowledge of this offset is important for maximising accuracy in near-surface conductivity predictions. Hence, rather than simply remove the primary field fraction from the observed data and focus on the ground response, an iterative inversion procedure was used to progressively refine estimates of both the receiver offset and ground conductivity. However, there is still a degree of ambiguity between the ground conductivity and receiver coil offset. The constrained inversion procedure was designed to utilise a priori information about the conductivity at depth across the survey area and to invert both X and Z component data simultaneously to reduce this ambiguity.
The inversion procedure and the constraints are described in this report. A list of the revised conductivity products is given in Appendix 1. In brief, the inputs to the inversion were;
. the observed data prior to removal of the primary field, and data uncertainty values,
. layered conductivity model reference values and uncertainties,
. layered conductivity model smoothness reference values and uncertainties,
. transmitter loop to receiver coil horizontal and vertical separation reference values and uncertainties, and
. receiver coil pitch angle reference values and uncertainties.
The outputs were;
. the layered conductivity model,
. transmitter loop to receiver coil horizontal and vertical separation values,
. receiver coil pitch angle, and
. the predicted data.
The quality of the inversion output was assessed through comparison of the borehole conductivity values with conductivity predictions from the closest observation to each borehole. The assessment was carried out using conductivity values transformed to logarithm base 10. It was demonstrated that the conductivity predictions from the constrained inversion more closely matched the borehole conductivity information than either of the two previous sets of predictions. Three different measures were used to describe the quality of the inversion output for each 5m interval between the surface and 120m. The capacity to map the variability in borehole conductivity was measured using a correlation coefficient. This parameter varied from 0.64 to 0.91. The bias in the predictions relative to the borehole conductivity values was assessed using the average misfit between these two quantities. This iv parameter was within the range -0.06 to +0.20 decades of conductivity. The magnitude of the differences between borehole conductivity values and the predictions was assessed using the standard deviation of the misfit between these 2 quantities. The standard deviation of the misfit was between 0.09 and 0.28 decades of conductivity.
A "percent data influence" (PDI) parameter was defined based on results for two closely related inversions. PDI profiles as a function of depth were used to arrive at an estimate of the depth of investigation for the survey of 120m.
A comparison of shallow AEM conductivity predictions with EM31 apparent conductivity measurements was carried out in addition to the comparison with borehole conductivity logs. There was a consistent improvement in the statistics used to compare EM31 apparent conductivity observations with each of the 3 generations of 0 to 5m AEM conductivity predictions; from the contractor-supplied EMFlow predictions, the revised EMFlow predictions and finally to the constrained inversion predictions. For the comparison with the latter, the correlation coefficient was ~0.81, misfit mean ~0.18 log10(mS/m) and misfit standard deviation ~0.21 log10(mS/m).
There are fundamental differences in sample volume that limit the degree of correlation that will be observed in any of these comparisons. However, the differences in sample volume can be exploited to obtain the most relevant set of shallow conductivity measurements at different scales of investigation. EM31 or other ground-based devices with a small sampling volume are appropriate for mapping shallow conductivity at paddock to farm scales. AEM measurements with larger sampling volumes provide more comprehensive sampling at sub-catchment to catchment scales.
EM31 apparent conductivity and AEM 0 to 5m conductivity predictions compare more favourably than do borehole 0 to 5m conductivity and AEM 0 to 5m conductivity predictions (e.g., R~0.81 for EM31 and AEM, R~0.64 for boreholes and AEM). This is probably due to a combination of factors such as the smaller sample volume of the borehole measurements, the invasive effects of the drilling process and the vagaries of the smaller number of boreholes (104 boreholes with conductivity logs) compared to the number of EM31 observations.