An algorithm to retrieve coastal water optical properties, bathymetry, and bottom albedo from hyperspectral imagery

Hyperspectral imagery is a powerful sensing technology for quantitative monitoring of coastal environments. In this paper, we present an algorithm to retrieve water optical properties, bathymetry, and bottom albedo using nonlinear optimization techniques. The proposed method combines the Lee semianalytical model, which relates the quantities of interest to the measured remote sensing reflectance, with a modified version of the Goodman linear mixing model for analysis of the bottom albedo. The estimation problem is posed as a nonlinear least squares problem, where the fractional abundances of the mixing model are linear and the optical properties and bathymetry are nonlinear. A simple two-stage Gauss-Seidel optimization algorithm is employed to compute the estimates and take full advantage of the problem structure. We use both simulated and AVIRIS hyperspectral imagery to compare the combined modeling approach with the Lee approach for retrieval of optical properties and bathymetry and with the unmixing approach of Goodman for determining bottom fractional composition. Results show that the proposed retrieval approach generally produces improved estimates of water optical properties, bathymetry and bottom composition but at a significantly higher computational cost. Results also indicate that although the approach is limited in its capacity to resolve bottom composition as a function of increasing depth and water turbity, it retains a robust capability for estimating water optical properties, even in unfavorable conditions for retrieving bathymetry and bottom albedo.