Uncertainty quantification for right-sizing computational models of sound propagation in the atmospheric boundary layer

Abstract

Comprehensive modeling of sound propagation through the atmospheric boundary layer is viewed as a judicious combination of accurate computational mechanics models and uncertainty quantification (UQ) methods. The role of numerical models is to represent nominally deterministic phenomena, e.g., geometrical spreading, ground interactions, refraction by mean gradients of wind and temperature. The role of UQ is to characterize the consequences of fundamentally non-deterministic and imprecisely known factors that affect propagation, e.g., turbulence in the atmospheric boundary layer, complex terrain features, and overly sparse spatio-temporal sampling of propagation parameters. High-fidelity wave propagation mechanics cannot compensate for inherent randomness in the environment and insufficient data on the parameters. When uncertainty is significant, the computational cost of high-fidelity models might be better invested in more ensemble simulations with medium-fidelity models and quantifying the payoff from more data about the environment. Work in recent years along three thrusts to enable this form of comprehensive modeling is reviewed: (1) Surrogate modeling based on cluster-weighted models, which are a type of probabilistic generative model, and on statistical learning methods, (2) global sensitivity analysis for assessing the importance of model parameters, and (3) a computational mechanics error budget for rationally analyzing the importance of various sources of uncertainty.Comprehensive modeling of sound propagation through the atmospheric boundary layer is viewed as a judicious combination of accurate computational mechanics models and uncertainty quantification (UQ) methods. The role of numerical models is to represent nominally deterministic phenomena, e.g., geometrical spreading, ground interactions, refraction by mean gradients of wind and temperature. The role of UQ is to characterize the consequences of fundamentally non-deterministic and imprecisely known factors that affect propagation, e.g., turbulence in the atmospheric boundary layer, complex terrain features, and overly sparse spatio-temporal sampling of propagation parameters. High-fidelity wave propagation mechanics cannot compensate for inherent randomness in the environment and insufficient data on the parameters. When uncertainty is significant, the computational cost of high-fidelity models might be better invested in more ensemble simulations with medium-fidelity models and quantifying the payoff from mo...