Data assimilation improvements for acoustic propagation prediction and underwater noise forecast. Storto, Andrea; Falchetti, Silvia; Oddo, Paolo; Jiang, Yong-Min. CMRE-MR-2018-016. May 2019.
This report investigates the benefits of assimilating oceanic observations on acoustic propagation prediction by means of a simulation study. Observing System Simulation Experiments (OSSEs) were carried out by using a regional coupled oceanic-acoustic prediction system developed within CMRE EKOE2 POW. The setup of the experiments for this research is as follows: 1) Oceanic sound speed field generation through a coupled nature run (the ground truth), a control run (no data assimilation) and three data assimilation schemes with different algorithmic complexities and computational requirements. Specifically, the assimilation schemes considered here are 3DVAR (with standard formulation of the three-dimensional variational analysis scheme), HYBRID (with hybrid ensemble-variational backgrounderror covariances), and 4DVAR (with simplified four-dimensional variational data assimilation). The data to be assimilated were drawn from in-situ profiles and altimetry data from the LOGMEC17 campaign; 2) Acoustic TL prediction at 75 and 2500 Hz using the range dependent sound propagation modeling tool RAM. The oceanic sound speed fields generated in the previous step were used as inputs; 3) Diagnoses on the sources of uncertainties, such as lateral boundary conditions, ocean stochastic physics and geo-acoustic parameterizations, for both oceanic and the acoustic fields; and 4) Benefits of data assimilation schemes on acoustic TL prediction evaluation. The ensemble experiments show that the lateral boundary conditions, ocean stochastic physics and geoacoustic parameterizations are the major sources of uncertainties in the oceanic sound speed fields, and thus in the acoustic TL predictions. Further, the assimilation of oceanic observations significantly improves the acoustic TL predictions at both 75 and 2500 Hz. In particular, both 4DVAR and HYBRID ensemble-variational data assimilation outperformed 3DVAR.