|
Hyperspectral remote sensing provides a particularly useful means
of determining inherent optical properties of coastal waters where
constituents other than phytoplankton add to the optical
complexity of the water column. The substantial number of
channels, about 200 for most hyperspectral sensors, enables many
of the constituents within the water column to be identified
spectrally. Additionally, in shallow water the water leaving
radiance may include a signal reflected from the sea bottom. A
hyperspectral radiometer was deployed on monthly oceanographic
cruises off the coast near Perth, Western Australia, to make
observations. Field measured reflectance spectra were used as
input into a slightly modified version of a reflectance model
developed by Lee et al, 1999. Products, including the
concentrations of chlorophyll-a (Chl-a), coloured
dissolved organic matter (CDOM), and suspended sediments (SS), as
well as the water column depth (H), were extracted from the
reflectance model by incorporating an optimisation technique. A
Levenberg-Marquardt retrieval scheme was utilised in the
optimisation. This scheme involved minimizing the difference
between the modelled and measured spectral reflectance curves.
Water samples were also collected on the monthly oceanographic
cruises and used to determine the concentrations of
chlorophyll-a, CDOM and SS. Water depth was measured using
the boat's echo sounder. The model-derived products were compared
to in situ measurements. The mean difference between model
retrieved depth and in situ depth was 12.5 % or 1.4 m (R2 = 0.98, N=11). Excluding two field measurements taken in the Marmion Marine Park, the mean RMS difference in depth was 7.6 % or 0.9 m (R2 = 0.99, N=9). The mean RMS difference between retrieved Chl-a concentration and in situ measured Chl-a was 11.1 % or 0.044 mgm-3 (R2 = 0.91, N=8). These preliminary results suggest that the reflectance model works well for depth and Chl-a retrieval for Western Australian coastal waters and their sandy substrate.
|
