Remote Sensing Techniques Applying Neural Networks for Effective Retrieval of Harmful Algal Blooms in the West Florida Shelf from VIIRS Satellite Observations, without the need for a Fluorescence Channel, and their comparisons with other Techniques.

Remote sensing approaches using neural networks (NN), are described that make use of the Ocean Color Remote Sensing Reflectances (OC Rrs) available from Visible Infrared Imaging Radiometer Suite (VIIRS) satellite bands at 486, 551 and 671nm to detect and retrieve Karenia brevis (KB) Harmful Algal Blooms (HABs) that plague West Florida Shelf (WFS) coasts impacting the environment and tourism. This approach is necessitated because VIIRS, unfortunately, unlike the Moderate Resolution Imaging Spectroradiometer (MODIS), does not have a 678 nm chlorophyll-a fluorescence channel that is normally effectively used with the normalized fluorescence height (nFLH) algorithm for detecting KB HABs in the WFS. We describe here the application of neural networks (NNs) previously reported by us for Chesapeake Bay chlorophyll retrievals, for the retrieval of phytoplankton absorption at 443 nm (a_ph443) in VIIRS images of the WFS using the existing VIIRS 486, 551 and 671nm bands. These NN retrieved a_ph443 values, are then converted to equivalent [Chla] values using known empirical relationships between them for the WFS. Now waters compatible with KB HABS in the WFS are known to be characterized by a minimum permissible [Chla] and a maximum permissible particulate backscatter bb_p at 551nm (and therefore a maximum permissible VIIRS Rrs 551 nm). These two limiting criteria are then used to create exclusion masks which are then consecutively applied as filters to retrieved VIIRS Rrs551nm images and then to VIIRS [Chla] images (obtained from equivalent NN retrieved a_ph443 images). The residual [Chla] image after application of the filters then shows values compatible with KB HABS in the WFS having satisfied both maximum Rrs551nm and minimum [Chla] criteria. The residual images of KB compatible [Chla] values are then used to identify, delineate and quantify the existing KB HABS. Comparisons with in-situ measurements and other techniques, including nFLH with MODIS, are presented, and confirm viability of both the NN and the filtering approaches devised for KB HABS retrievals in the WFS.