A BAYESIAN MCMC APPROACH TO MODEL ASSESSMENT, UNCERTAINTY ANALYSIS, AND LITHOLOGICAL PREDICTION FOR AIRBORNE ELECTROMAGNETIC SURVEYS

Abstract

Geophysical data are typically used to infer a single ‘best’ model consistent with observations and prior information. However, because of non-uniqueness, limited resolution, and data errors, many models satisfy both the data and reasonable prior assumptions. Instead of seeking to describe the properties of any single model, we developed a trans-dimensional Bayesian Markov chain Monte Carlo (McMC) algorithm for the analysis of airborne electromagnetic (AEM) surveys that assesses the characteristics of models that are consistent with observations and prior assumptions. This is a powerful tool for model assessment and uncertainty analysis, and provides a wealth of information that can be used to make inferences about plausible subsurface properties. Model assessment and uncertainty analysis is compounded by the large volume of data that are typically acquired for AEM surveys. Here, we discuss the mechanics of a McMC algorithm developed for the analysis of time- or frequency-domain airborne electromagnetic data. Specific aspects of the algorithm that will be discussed include: the trans-dimensional nature of the program, which allows the number of layers to be a free parameter; the capability to assess random and/or systematic data errors as unknown parameters; the use of parallel computing tools to run multiple chains for a single dataset in order to assess convergence, or to analyze many datasets simultaneously; the use of stochastic Newton sampling to optimize sampling efficiency; and the ability to integrate multiple data types to probabilistically assess geological or hydrological properties directly. To illustrate this latter capability, we use AEM data from the San Luis Valley, CO, along with knowledge about the physical characteristics of different lithological units to make probabilistic predictions about their distribution in the subsurface.