|
Preliminary results of previous work had shown a Neural Network (NN) technique developed by us as effective in detecting Karenia brevis Harmful Algal Blooms (KB HABs) plaguing West Florida Shelf (WFS) from VIIRS satellite observations. We extend comparisons of NN retrievals against a data set of near simultaneous in-situ measurements in the WFS spanning the 2012-2016 period for which there was available VIIRS data. Specifically we looked for match ups where the overlap time windows between satellite observations and in-situ measurements were 15 minutes and 100 minutes. We then compare the accuracy of the NN retrievals against the in-situ measurements, with the accuracies achieved with similar of retrievals using OC3, GIOP, QAA and RGCI algorithms. The NN technique exhibited the best retrieval accuracy statistics. The retrievals for all the algorithms very clearly showed the impact of temporal variations of the KB HABS on retrieval accuracies. Thus, retrievals using a 15 minutes overlap window between satellite observations and in-situ measurements yielded much higher accuracies than those with the 100 minutes overlap window. Temporal variabilities were also studied, using consecutive overlapping VIIRS images. These variabilities, as well as the patchiness of KB blooms were also confirmed by a set of in-situ measurements near Sarasota, FL.
|
