Comparing the Performance of Empirical, Semi-empirical, and Curve Fitting Models in Predicting Cyanobacterial Pigments

摘要

This study presents a comparative analysis of several algorithms for the estimation of cyanobacterial pigments chlorophyll a (CHL) and phycocyanin (PC) from hyperspectral reflectance, while also providing a consistent basis for determining the best performing models in predicting both CHL and PC concentrations. Simple band ratio algorithms, band tuning methods, semi-empirical algorithms, and the modified Gaussian model (MGM) parameters were used to estimate CHL and PC from multiple source spectral datasets of two eutrophic central Indiana reservoirs: Eagle Creek and Morse. The spectral datasets were collected over a three-year period (2005 to 2007) using two field-based (Asp Field-Spec; Ocean Optics USB4000) spectroradiometers Spectral parameters used in these mapping algorithms were examined for their correlation to the CHL and PC concentrations. The results demonstrate for estimating CHL, simple and modified band ratios performed well; for PC estimation, the highest performing models include the algorithms using MGM strength and band tuning methodology.