Multivariate approach for chlorophyll-a and suspended matter retrievals in Case II type waters using hyperspectral data

Remote sensing has become promising in providing temporal and spatial information on biogeodynamics in large and open freshwater bodies. In optically complex environments, such as in the Western Basin of Lake Erie (WBLE), the water contains multiple biogeochemical constituents or colour producing agents (CPAs), such as phytoplankton, suspended matter and dissolved organic carbon; identifying and analysing such in-water constituents is crucial for understanding and assessing many biogeochemical processes. For example, concentrations of chlorophyll-a and total suspended matter can be used as proxies to assess phytoplankton dynamics and particulate loading. However, quantitative estimation of their concentrations from satellite observations is complicated when working with mixed spectral signatures. Hyperspectral remote sensing is fast emerging as a key technology for advanced and improved understanding of optically complex waters. This study estimates concentrations of chlorophyll-a and total suspended matter (TSM) in the WBLE by applying the partial least squares (PLS) method to a full range (400-900nm) of continuous narrow spectral bands. The PLS method models the covariance between hyperspectral bands and CPAs, and identifies the optimal bands that characterize most of the variance in the CPAs. This method avoids the curse of dimensionality and the effects of multi-collinearity, a challenge that is associated with new-generation hyperspectral satellite sensors. Validation parameters for the PLS-based models produced R-2 of 0.84 for chlorophyll-a (RMSE=1.18g/L), and R-2 of 0.90 for TSM (RMSE=1.26mg/L), illustrating the potential of the PLS method for isolating and extracting absorption features characterizing the various CPAs in optically complex Case II type waters.