MASTR Real-Time Sea Experiment 2024
Gulf of Mexico – February 2024
MSEAS Mutual Information (MI) Optimal ROCIS Flight Paths: Background Information
The objective of our optimal path planning is to maximize information about the western wall of the Loop Current (LC) and the associated (cyclonic) eddies.
The ROCIS flights sample surface velocities along their tracks. They define the candidate sample observations. Commonly, a flight is planned for a future time or date, and sampling region, within all operational constraints. For our information-theoretic objective function, an optimal flight path maximizes the predicted information contained in its sampled surface velocities (the candidate observations) about the western wall of the LC and its associated eddies (the target fields or verification fields). These candidate observations, target fields, and the objective function are defined as follows:
- The candidate observations are surface U and V data along a feasible flight track, in accord with all operational constraints and planned feasible observation times or data times.
- The target or verification fields are defined by their time(s), region(s), and field variable(s).
- The target times can be a single time, an integral over time, or any vector of discrete times. In our case, we select a set of future times (e.g. same day, and every day or couple days thereafter) and evaluate the mutual information content for each of these times.
- The target region is a grid covering the western wall of the LC and the associated eddies, including hotspot regions of interest.
- The target field variables are either surface sigma-t (density of seawater at standard pressure) or surface sigma-t and surface U, V over the target region. Subsurface fields relevant to the western wall of the LC can also be considered.
- The objective function optimized is the mutual information (MI) between the candidate observations and the target fields.
- Note that if candidate observations are to be made at a single physical point or if the target variables are defined at a single spatiotemporal location, the MI is a scalar field defined over the verification region or candidate observation region, respectively.
- In some sense, MI generalizes covariances to vector variables and non-Gaussian variables.
Back to the MASTR Sea Exercise Home Page