Source‐height and frequency dependence in sensitivity analysis of near‐ground sound propagation.

Abstract

Computational forecasts of the near‐ground sound pressure level (SPL) are key features of proposed frameworks for designing acoustic sensor networks. In many cases the local weather and terrain will not be known precisely enough to justify high confidence in forecasts of the probability of detection. The sensitivity of SPL forecasts to parameter variations therefore must be known in order to assess the probability of detecting acoustic disturbances in poorly characterized environments. To facilitate these assessments, we have expanded a recent framework for full‐field sensitivity analysis (FFSA) throughout the parameter space. This new version continues to employ sampling methods, proper orthogonal decomposition, and cluster‐weighted models to develop robust surrogate models for sensitivity analysis. Enhancements shown here include (i) locally linear functions in the cluster‐weighted models, (ii) analytical differentiation in place of local response surfaces for computing sensitivities, and (iii) estimation of the forecast uncertainty in the sensitivities. These capabilities are used for FFSA of the near‐ground SPL due to a harmonic point source operating at several frequencies. The governing parameters and source height are assumed to vary across wide but physically realistic ranges. The dependence of the forecast uncertainty on various factors is examined.