A Tale of Two Algorithms: Analysis of PHYDOTax vs. MESMA in the San Francisco Bay Area

Abstract:

Accurate interpretation of phytoplankton functional types (PFTs) is important for monitoring harmful algal blooms and assuring public safety in aquatic environments. Previous work has suggested that it is possible to determine PFTs using remote sensing techniques on hyperspectral imagery such as Airborne Visible/InfraRed Imaging Spectrometer (AVIRIS) data. PHYtoplankton Detection with Optics Taxonomy (PHYDOTax), for example, is an algorithm written specifically for determining PFTs using AVIRIS data. However, its simplicity comes at the cost of high uncertainty depending on the complexity of the environment in the image. In contrast, Multiple Endmember Spectral Mixture Analysis (MESMA) is a spectral unmixing technique primarily used to classify terrestrial remote sensing images down to the sub-pixel scale. Here, we evaluate these two algorithms by applying them to the same remotely-sensed images to determine PFT proportions within the images. The San Francisco Bay salt ponds were chosen as the study site to compare the two algorithms, given the unique aquatic environment containing many different PFTs, as well as the access to in-situ data to ground-truth the results from both algorithms. The validation of a reliable and accurate algorithm for determining PFT proportions will provide greater confidence when using remote sensing data to estimate phytoplankton composition. Validated use of MESMA will also allow for simultaneous determination of other non-phytoplankton endmembers important for coastal waters, such as colored dissolved organic matter and total suspended sediment, in future applications.