Improved hyperspectral inversion of aquatic reflectance under non-uniform vertical mixing.

Estimating the concentration of water constituents by optical remote sensing assumes absorption and scattering processes to be uniform over the observation depth. Using hyperspectral reflectance, we present a method to direct the retrieval of the backscattering coefficient (b(λ)) from reflectance (> 600 nm) towards wavebands where absorption by water dominates the reflectance curve. Two experiments demonstrate the impact of hyperspectral inversion in the selected band set. First, optical simulations show that the resulting distribution of b(λ) is sensitive to particle mixing conditions, although a robust indicator of non-uniformity was not found for all scenarios of stratification. Second, in the absence of spectral backscattering profiles from in situ data sets, it is shown how substituting the median of b(λ) into a near infra-red / red band ratio algorithm improved chlorophyll-a estimates (root mean square error 75.45 mg m became 44.13 mg m). This approach also allows propagation of the uncertainty in b estimates to water constituent concentrations.