Computing the parameter sensitivities of outdoor sound propagation in a random environment

Abstract

Computational forecasts of near-ground sound levels are compromised by uncertainty (e.g., randomness) and error (e.g., grid resolution) in the atmosphere and terrain representations, and by simplified or incorrect physics. For an incompletely known environment, a model's predictive power cannot be assessed without first quantifying the sensitivity of its forecasts to the full range of every parameter. Knowledge of these sensitivities throughout the spatial domain also is essential for effectively investing data-gathering resources to support sound propagation forecasts. Sensitivity analysis therefore is central to raising the relevance of computational acoustics in practical applications. These considerations should motivate practitioners of computational acoustics to adopt a consistent framework for sensitivity and uncertainty analyses. Topics to be discussed include: (1) standard uncertainty taxonomies in computational mechanics, (2) why uncertainty about a parameter should be distinguished from sensitivity of a model to that parameter, (3) sources of uncertainty in the near-ground acoustics, (4) a sampling-based sensitivity analysis framework that facilitates estimating typical and extreme values of sensitivities at each point in the spatial domain (i.e., full-field sensitivities), (5) factors to be aware of when applying sensitivity analysis to forecasts of near-ground sound propagation, and (6) ways of representing sensitivity estimates to facilitate insight.