CONFOUNDING CONSTITUENTS IN REMOTE SENSING OF PHYCOCYANIN

Abstract

This project examines the impact of confounding variables that have limited the accuracy of remotely predicting phycocyanin in three Indiana drinking and recreational water reservoirs. In-situ field reflectance spectra were collected from June to November 2006 over a wide range of algal bloom conditions using an ASD Fieldspec (UV/VNIR) spectroradiometer. Groundtruth samples were analyzed for chlorophyll a, phycocyanin, total suspended matter, and other water quality constituents. Previously published spectral algorithms for the detection of phycocyanin were evaluated against lab measured pigment concentrations using linear least squares regression. Algorithm performance varied across study sites (best performing models by reservoir resulted in r2 values of 0.32 to 0.84). Residuals of predicted versus measured pigment concentrations were analyzed against concentration of potential confounding water constituents. Residual analysis revealed optically active constituents contributed between 25% and 95% of original phycocyanin model errors. Inclusion of spectral variables into models to account for significant confounders resulted in improved spectral estimates of phycocyanin (r2 = 0.56 to 0.93).