2.3. Seagrass Class Data

A previous, supervised classification of the study area’s benthic habitat [15] was employed in the present study. A brief summary of the classification methodology is presented here. A maximum likelihood classifier was applied to remote sensing data, of the study area, that were recorded during the time-of-year when the benthic signal was assumed to dominate the variation in the top-of-water reflectances with negligible contribution from the water column. Five Landsat images, from each year in the period 2007–2011, along with Florida Bay Fisheries Habitat Assessment Program in situ surveys of seagrass cover, conducted in the spring of each year, were employed to train and validate the classifier. Data from 2009–2011 were used to train the classifier while data from 2007 and 2008 were used for validation. Pixels were classified as (1) medium-dense seagrass; (2) low seagrass; (3) sparse seagrass; or (4) turbid, as determined from three depth-invariant bands derived from the visible wavelength bands [15].

Sparse and low classes were combined in the present study so that a 2-class scheme was used to distinguish between all study area benthic habitats. The benthos were classified each year (1998–2008) from spring and early summer images when phytoplankton concentrations are generally low. The classifications produced from the spring/early summer data were assumed constant throughout the calendar year (from the January before the classified image to the December following the classified image). New maps, describing the mode of a 1 km radius around each 30 m Landsat grid cell, were created to account for SeaWiFS spatial resolution of 1.1 km at nadir.