Mutual information forecasts for optimal adaptive sampling guidance

We use our mutual information (MI) forecasts between candidate observation locations and future regional field or variables of interest (target or verification field/variables) to predict the most informative future data locations. We use our MIT MSEAS ensemble forecast to predict these MI fields.

For the MI fields that we show in the table below, the goal is to predict the location of candidate future glider observations (temperature and salinity profiles) on different days (00Z Apr 23 to 00Z Apr 28) that are the most informative about the target temperature and salinity, on 00Z Apr 29, 2025, in the region of the forecast LC and its uncertainty.

• The candidate glider observations at each location in the domain occur at the times indicated in the headers of each MI plot. These glider temperature (T) and salinity (S) profiles are assumed taken at the 0m, 50m, 150m, 300m and 500m depths, unless these depths are below the topography, in which case the depths are equally redistributed.
• The target (verification) variables are the temperature and salinity fields in the region of the forecast LC and its uncertainty at 00Z on Apr 29, 2025. These fields are defined by T and S profiles at all the locations shown by green stars taken together, at 00Z on Apr 29, 2025, at 150m, 300m, and 500m depths (in the region of the forecast LC and its uncertainty).

The locations with highest MI (bright yellow) on each day indicate where glider T and S profile observations are predicted to be most informative about the LC and its uncertainty at 00Z on Apr 29, 2025.

00Z Apr 23 2025	00Z Apr 24 2025	00Z Apr 25 2025
00Z Apr 26 2025	00Z Apr 27 2025	00Z Apr 28 2025