Multi-Dimensional Ambient Noise Model - Phase II --- 18-117

At its core, the process of predicting the spectral and directional structure of the ambient noise field, and how it manifests as beam-level noise, relies on properly accounting for the uncertainty that arises from the confluence of several types of noise sources and physical processes (i.e. wind, rain, shipping, ice noise, sound speed variability, scattering, etc.). Existing TDAs rely heavily on static databases when making ambient noise predictions, and this can lead to common situations in which TDA outputs differ substantially from current sensor observations. This has the ultimate effect of fostering mistrust in the TDAs. The basic approach of this effort is to make better use of all available information sources, including live sensor data for near-term (small tau) forecasts, and to combine these with modern machine learning and Monte Carlo algorithms designed to properly account for parameter-level uncertainty. This will result in better estimates of the horizontal and vertical noise field directionality, which can then be rendered into beam-level noise for different sensor types, sensor depths, sensor orientations, and operating frequencies.