Toward Decision-Conditioned Credibility for an Unvalidatable UAS Acoustic-Disturbance Digital Twin

Small uncrewed aircraft systems (UASs) increasingly operate near noise-sensitive animals, and physics-based acoustic digital twins are now used to plan "wildlife-safe" flight altitudes and buffer distances. Such decision-support models face a structural obstacle: in situ validation against the sound pressure levels an animal actually perceives is rarely feasible, so operators are asked to trust a model that cannot be validated in the conventional sense. This work explores a question other than "is the model correct?": whether the operational decision the model informs actually depends on what remains unknown.

To address this, we are developing a reproducible, standards-grounded credibility-evaluation workflow for an unvalidatable UAS acoustic-disturbance twin (ISO 9613-2 propagation). The workflow combines solver verification, input-pedigree classification of aleatory and epistemic uncertainty, uncertainty quantification following GUM/JCGM 101 (Monte Carlo with a polynomial-chaos surrogate), and global sensitivity analysis (Morris screening to Sobol indices). Its central idea is decision-conditioned credibility: sensitivity and credibility are assessed against the decision indicator, whether the predicted level exceeds a species audiogram threshold, rather than the raw output distribution. This reframes an empty validation factor from a fatal gap into a testable question, and points toward a ranked, value-of-information measurement priority.

The approach is illustrated on a case study of multirotor operation near a sensitive terrestrial receptor (domestic-horse audiogram), with the aim of deriving a minimum trustworthy operating altitude and horizontal buffer under a bounded exceedance-probability constraint. The intended outcome is a transparent basis for trusting model-based operational limits when validation data cannot be obtained.