A Process for Identifying Requirements for Physical Referent Data to Support Computational Model Validation

Abstract

This paper is one of a series being presented at this conference summarizing the technical outcomes of NATO AVT-297. It presents the framework for a process that may be used to identify the requirements for physical referent data in support of computational model validation. This relies upon applying the principles of systems engineering to develop systematic abstractions and decompositions of the problem space. Rather than contriving to prescribe a specific, structured series of activities a priori, the framework relies on empirical observation and learning to facilitate the identification of candidate Points of Entry into model validation experiments. In order to avoid the inappropriate conflation of one form of abstraction with another, the framework builds on three architectures: functional, physical and modeling. Guidelines are provided as to how these may be distinguished and traversed consistently as they, themselves, mature. Particular attention is drawn to (i) the challenges that may be faced when handling mismatches in abstraction typical of those encountered in multidisciplinary modeling scenarios, and (ii) the potential utility of multi-fidelity analyses as mechanisms for estimating model form uncertainty and assessing the suitability of candidate Points of Entry into model validation experiments. The importance of technique verification and validation dialog is reinforced throughout, highlighting the mutual accountability of those engaged in the computational and physical sciences to provide the learning and, subsequently, the capabilities that will be required to realize our digital transformation ambitions.