Dependence of predictive skill for outdoor narrowband and broadband sound levels on the atmospheric representation

Outdoor sound propagation predictions are usually made from averaged, instantaneous, or modeled vertical wind and temperature profiles. Because vertical profiles do not capture the full spatial and temporal variability of the atmosphere, the predictive skill is inherently limited. This study compares bias and root-mean-square (rms) errors for predictions made from the actual mean vertical profiles, approximate vertical profiles based on Monin-Obukhov similarity theory (MOST), and profiles for a windless, neutral atmosphere. The errors are calculated by comparing predictions to “ground-truth” data generated by numerically propagating sound through high-resolution large-eddy simulations of the atmosphere. Predictive errors are found to be similar when actual mean and MOST profiles are used. Assuming a windless, neutral atmosphere leads to a large, positive bias errors in upward refraction, but does not affect skill as much in downward refraction. Analyses are presented for single-frequency signals as well as broadband signals (such as impulse events). It is shown that predictive skill improves as bandwidth is increased, particularly for downwind propagation.