Issues and potential improvement of multiband models for remotely estimating chlorophyll-A in complex inland waters

Remote estimation of chlorophyll-A (chl-a) in complex freshwaters remains a challenging problem due to the rapid spatial variability and wide range as influenced by terrestrial constituents. A controversial issue is whether or not 2-B models possess sufficient wavelength information for accurately estimating Chl-a concentrations from remote sensing data for freshwater environments. This study introduced a systemic approach and proved that adding additional wavelength information to 2-B model could not significantly improve the estimation of freshwater chl-a, but acted to increase model uncertainty. This convincing solution was based on a large synthetic data set (38 937 samples) combined with a set of in situ data (51 samples) collected in three cruises in Lake Huron. The synthetic data set has two distinct features: 1) large data items and 2) covers a broad range of chl-a (0-1000;mg/m³), colored dissolved organic matter (CDOM) (0-50;m^-1), and NAP (nonalgal particles) (0-500 mg/l). Additionally, this study reveals how hyperspectral wavelength selection, number of bands, bandwidth, and parameter calibration are associated with the uncertainty in remote sensing of chl-a. The systematic analysis approach was used to evaluate 34 chl-a algorithms by using optimal location and number of wavelengths as well as calibrated parameters. The study introduced a set of new 2-B, 3-B, and 4-B models derived also from using optimized parameters, suggested wavelengths, and bands available in MERIS and MODIS satellite images. Validation results demonstrated that these models are suitable to general freshwater environments because of broad ranges of biochemical and physical properties in both synthetic and in situ data.