A sequential hydrogeophysical approach to quantify model structural uncertainty

Abstract:

Geophysical data- and in particular airborne electromagnetic (AEM) surveys- are increasingly being used to help develop large-scale hydrogeologic models. Model structural error is a key source of uncertainty in hydrologic predictions, but is not often taken into account during the model development process. We propose a sequential approach, whereby many plausible structural models are generated according to geophysical model uncertainty and available borehole data, and are subsequently calibrated to hydrologic observations. In the first step, geophysical model uncertainty is quantified using a Bayesian Markov chain Monte Carlo analysis that estimates the posterior probability distribution of electrical resistivity at multiple depths for each of the thousands of densely spaced AEM sounding locations. Next, geophysical probabilities at each location are translated into probabilistic estimates of structure by calculating the area under each geophysical probability distribution associated with the resistivity distribution that is assumed or empirically estimated for different geological units (e.g. sand, silt, or clay). Many plausible structural models are then generated using a sequential indicator simulation algorithm, with sparse borehole data as hard constraints and the dense geophysically-derived structural probabilities as soft data. Finally, hydrologic observations are being used to calibrate each of the simulated structural models to determine the best-fitting hydraulic parameters. We will discuss the implementation of this sequential modeling approach, and also the ability to estimate uncertainty of both model parameters and hydrologic response.