The ML-SCOPE Team will soon be publishing a special issue “Machine Learning for Ocean Modelling” in the journal Ocean Modelling. Prof. Pierre Lermusiaux and the other ML-SCOPE co-PIs will serve as guest editors. Submissions are open until September 30, 2022. Additional information can be found on the journal website.

Developing accurate and computationally efficient models for underwater sound propagation in the uncertain, dynamic ocean environment is inherently challenging. In this work, we evaluate the potential of dynamic reduced-order modeling for stochastic ray tracing. We obtain and implement the stochastic dynamically-orthogonal (DO) differential equations for Ray Tracing (DO-Ray). With stochastic DO-Ray, we can start from non-Gaussian environmental uncertainties and compute the stochastic acoustic ray fields in a dynamic reduced order fashion, all while preserving the dominant complex statistics of the ocean environment and the nonlinear relations with ray dynamics. We develop varied algorithms and discuss implementation challenges and solutions, using direct Monte Carlo for comparison. We showcase results in an uncertain deep-sound channel example and observe the ability to represent the stochastic ray trace fields in a dynamic reduced-order fashion.

Onboard forecasting is challenging but essential for unmanned autonomous ocean platforms. Due to the numerous operational constraints of these platforms, efficient adaptive Reduced-Order Models (ROMs) are needed. In this work, we employ the incremental Low-Rank Dynamic Mode Decomposition (iLRDMD), which is an adaptive, data-driven, DMD-based ROM that enables efficient forecast compression, transmission, and onboard forecasting. We demonstrate the algorithm on 3D multivariate Hybrid Coordinate Ocean Model (HYCOM) ocean fields in the Middle Atlantic Ridge (MAR) region. We further demonstrate that these iLRDMD ocean forecasts can be used for interdisciplinary applications such as underwater acoustics predictions. Here, acoustics fields computed from the ocean iLRDMD forecasts are compared to those computed from HYCOM fields. We also illustrate the application of a joint ocean-acoustics iLRDMD model for predetermined acoustics configurations. In the MAR region, we find that iLRDMD models are sufficiently accurate and efficient for onboard ocean and acoustic forecasting of temperature, salinity, velocity, and transmission loss fields.

Real-time sea experiments often involve large computational costs and software development associated with running numerical ocean simulations. Effective visualization tools that interpret the results of these simulations are therefore a necessity, and must overcome the challenges of plotting large, high-resolution, three-dimensional, time-dependent, and probabilistic ocean fields and associated quantities in real-time. Although disparate visualization tools aimed at ocean forecasting exist, a complete, integrated visualization suite that is efficient, interactive, and has 3D capabilities is still needed. In this work, we present the MSEAS high-performance visualization suite for real-time sea experiments. It processes multidisciplinary oceanographic fields in a computationally efficient manner and creates easy-to-use, portable, and interactive visualizations. The suite includes static visualization tools based on NCAR Graphics and MATLAB; the interactive web-based tool 2DSeaVizKit built using leaflet and D3.js for interactive 2D visualization on the world map; and 3DSeaVizKit, a browser-based, interactive 3D visualization tool built using Plotly and WebGL for exploratory 3D analysis of ocean forecasts. It can provide standard 2D cross-sections for scalar-valued data; quiver plots, pathlines, and streamtubes for vector-valued data; Lagrangian products (such as trajectories, Lagrangian Coherent Structures, etc.); isosurfaces for 3D data; and an interactive graphical user interface for selecting the quantities, times, and sub-domains of interest. We showcase applications of the visualization suite during three recent exercises that took place in the Gulf of Mexico, the Clarion-Clipperton Fracture Zone in the Pacific Ocean, and the Balearic sea.

Congratulations to Abhinav Gupta on his graduation! Abhinav successfully defended and received his PhD from Mechanical Engineering for his research on “Scientific Machine Learning for Dynamical Systems: Theory and Applications to Fluid Flow and Ocean Ecosystem Modeling” with our MSEAS group at MIT. We wish all the best to Abhinav for his next steps!