Spatial uncertainty in higher fidelity acoustic models, and impact on model-data comparisons

Acoustic propagation models, such as the parabolic equation (PE), are typically parameterized by a single source depth and one or more receiver depths and ranges. At the higher frequencies, propagation angles, and ranges enabled by advances in computer hardware and refinement of algorithms, the acoustic field predicted by the model can contain features at considerably finer scale than the resolution of the requested output points. The typical practice of simply sampling the field at the output points can introduce artificial large scale structure in the output, due to aliasing. However, averaging out the fine-scale spatial variability can also result in a misleadingly precise value for transmission loss (TL) at a given point. We show that reconciliation of acoustic model predictions with experimental data can be greatly improved by presenting the TL variability due to the expected uncertainty of source and receiver position within an otherwise deterministic model. Even with good estimates of source and receiver positions, this variability can be greater than that due to uncertainty in the model's environmental inputs. Model-data comparisons are demonstrated with examples from a recent ONR effort in the eastern Gulf of Mexico which included controlled TL measurements at 0.4–2 kHz in shallow water.