Assessment of Multiple Third-Party Risk Models Incorporating Uncertainty Quantification and Sensitivity Analysis

Abstract

This study compares three UAS third-party risk (TPR) assessment tools—Quantitative Approach and Departure Risk Assessment (QUADRA), Autonomous Planning and Replanning (AutoPRep), and the Ground Risk Assessment Service Provider (GRASP)—under controlled mission parameters, similar population datasets, and equivalent failure-mode assumptions. Statistical equivalence between QUADRA and AutoPRep is assessed through several means: the two one-sided tests (TOST), null-hypothesis significance testing, effect size (Hedge's g), and post-hoc power analysis results; which show that QUADRA and AutoPRep produce statistically equivalent expected fatality rates when inputs are aligned, while GRASP exhibits non-equivalent behavior due to its difference in population consideration and sampling, discrete lethal crash area structure, and differing treatments of exposure. Uncertainty quantification is integrated into collective risk calculation in QUADRA by calculating gradient-based variance to evaluate sensitivity to stochastic inputs, of which include probability of loss of aircraft, probability of fatality, skid distance, and human height. The analysis demonstrates that although probability of loss drives risk magnitude, overall variance is dominated by uncertainty in probability of fatality and skid distance, clarifying both the equivalence limits and uncertainty drivers of UAS ground-risk assessments.