Resolution analyses for selecting an appropriate airborne electromagnetic (AEM) system

Abstract

The choice of an appropriate airborne electromagnetic system for a given task should be based on a comparative analysis of candidate systems, consisting of both theoretical considerations and field studies including test lines.It has become common practice to quantify the system resolution for a series of models relevant to the survey area by comparing the sum over the data of squares of noise-normalised derivatives. We compare this analysis method with a resolution analysis based on the posterior covariance matrix of an inversion formulation. Both of the above analyses depend critically on the noise models of the systems being compared. A reasonable estimate of data noise and other sources of error is therefore of primary importance. However, data processing and noise reduction procedures, as well as other system parameters important for the modelling, are commonly proprietary, and generally it is not possible to verify whether noise figures have been arrived at by reasonable means. Consequently, it is difficultsometimes impossibleto know if a comparative analysis has a sound basis. Nevertheless, in the real world choices have to be made, a comparative system analysis is necessary and has to be approached in a pragmatic way involving a range of different aspects.In this paper, we concentrate on the resolution analysis perspective and demonstrate that the inversion analysis must be preferred over the derivative analysis because it takes parameter coupling into account, and, furthermore, that the derivative analysis generally overestimates the resolution capability. Finally we show that impulse response data are to be preferred over step response data for near-surface resolution.

Enhancing SNMR model resolution by selecting an optimum combination of pulse moments, stacking, and gating

“…The surface nuclear magnetic resonance (SNMR) inverse problem is non-linear (equation ( 2)), and although the methodology of using the covariance matrix is strictly valid only for linear inverse problems, it has been successfully applied for non-linear inverse problems (Behroozmand et al 2013;Christensen and Dodds 2007;Christensen and Lawrie 2012). We decided to investigate the model space using a genetic algorithm (GA) in order to ensure that the linearized covariance calculation gives acceptable results.…”

“…The parameter uncertainty estimates are then obtained by the square root of the diagonal elements of C est . A similar approach for accessing the parameter uncertainty has been reported in (Christensen and Dodds 2007;Christensen and Lawrie 2012).…”

Characterising cover and exploring under cover with AEM

“…A traditional approach used to define the noise of an electronic system requires noise measures for each component of the system and an understanding of how each component contributes to overall system noise. Due to the complexity of AEM systems this approach is difficult to undertake and therefore noise is typically approximated by analysing observed data from multiple flights (Brodie & Fisher, 2008;Christensen & Lawrie, 2012;Green & Lane, 2003). Using this method the additive noise term is identified as the clearest indicator of system noise and resolvability.…”

Handling noise in AEM inversion – implications for subsurface characterisation

“…The method by Green & Lane (2003), requires the estimation of a system noise level by analysis of the standard deviation of data measured at high altitude (additive noise) and the repeatability of data measured on multiple occasions over the same ground (multiplicative noise). Given the relative simplicity and robustness of this method, it is a regularly used approach for estimating the noise present in AEM data sets (see, for example, Annetts & Hauser, 2015;Brodie & Fisher, 2008;Brodie & Sambridge, 2009;Christensen & Lawrie, 2013;Ley-Cooper et al, 2015). The method developed by Auken et al (2009), requires the allocation of noise levels or removal of noisy data points in a dataset which is typically done via a manual process.…”