Global sensitivity analyses of multiple conceptual models with uncertain parameters driving groundwater flow in a regional-scale sedimentary aquifer

Abstract

We rely on various Global Sensitivity Analysis (GSA) approaches to detect the way uncertain parameters linked to diverse conceptual geological models influence spatial distributions of hydraulic heads in a three-dimensional complex groundwater system. We showcase our analyses by considering a highly heterogeneous, large scale aquifer system located in Northern Italy. Groundwater flow is simulated considering alternative conceptual models employed to reconstruct the spatial arrangement of the geomaterials forming the internal makeup of the domain and characterizing the distribution of hydraulic conductivities. For each conceptual model, uncertain factors include the values of hydraulic conductivity associated with the geomaterials composing the aquifer as well as the system boundary conditions. We explore the relative influence of parametric uncertainties to steady-state hydraulic head distributions across the set of conceptual models considered by way of three GSA methodologies, i.e., (a) a derivative-based approach, which rests on the Morris indices; (b) the classical variance-based approach, grounded on the evaluation of the Sobol’ indices; and (c) a moment-based GSA, which takes into account the influence of uncertain parameters on multiple (statistical) moments of a given model output. Due to computational costs, Sobol’ and moment-based indices are obtained numerically through the use of a model-order reduction technique based on the polynomial chaos expansion approach. We find that the sensitivity measures considered convey different yet complementary information. The choice of the conceptual model employed to characterize the lithological reconstruction of the aquifer affects the degree of influence that uncertain parameters can have on modeling results.