Shallow water substrate mapping using hyperspectral remote sensing

During April 2004 the airborne hyperspectral sensor, HyMap, collected data over a shallow coastal region of Western Australia. These data were processed by inversion of a semi-analytical shallow water optical model to classify the substrate. Inputs to the optical model include water column constituent specific inherent optical properties (SIOPs), view and illumination geometry, surface condition (based on wind speed) and normalised reflectance spectra of substrate types. A sub-scene of the HyMap data covering approximately 4 km2 was processed such that each 3×3 m2 pixel was classed as sand, seagrass, brown algae or various mixtures of these three components. Coincident video data were collected and used to estimate substrate types. We present comparisons of the habitat classifications determined by these two methods and show that the percentage validation of the remotely sensed habitat map may be optimised by selection of appropriate optical model parameters. The optical model was able to retrieve classes for approximately 80% of all pixels in the scene, with validation percentages of approximately 50% for sand and seagrass classification, and 90% for brown algae classification. The semi-analytical model inversion approach to classification can be expected to be applied to any shallow water region where substrate reflectance spectra and SIOPs are known or can be inferred.

► We present habitat maps derived using a shallow water semi-analytical model.
► Video data are compared to the habitat maps derived from the airborne hyperspectral data.
► Approximately 80% of all pixels in the scene were successfully classified.
► Validation percentages for sand, seagrass and algae ranged from approximately 50% to 90%.