Sensitivity of a satellite algorithm for harmful algal bloom discrimination to the use of laboratory bio-optical data for training
Martinez-Vicente, V.; Kurekin, A.; Sá, C.; Brotas, V.; Amorim, A.; Veloso, V.; Lin, J.; Miller, P.I. (2020). Sensitivity of a satellite algorithm for harmful algal bloom discrimination to the use of laboratory bio-optical data for training. Front. Mar. Sci. 7: 582960. https://dx.doi.org/10.3389/fmars.2020.582960
In: Frontiers in Marine Science. Frontiers Media: Lausanne. e-ISSN 2296-7745, meer
|   |
| Trefwoorden |
Karenia mikimotoi (Miyake & Kominami ex Oda) Gert Hansen & Moestrup, 2000 [WoRMS]
Marien/Kust |
| Author keywords |
MERIS; optical backscattering; Karenia mikimotoi; harmful algal blooms; ocean color |
| Auteurs | | Top |
- Martinez-Vicente, V., meer
- Kurekin, A.
- Sá, C.
- Brotas, V.
|
- Amorim, A.
- Veloso, V.
- Lin, J.
- Miller, P.I.
|
|
| Abstract |
Early detection of dense harmful algal blooms (HABs) is possible using ocean colour remote sensing. Some algorithms require a training dataset, usually constructed from satellite images with a priori knowledge of the existence of the bloom. This approach can be limited if there is a lack of in situ observations, coincident with satellite images. A laboratory experiment collected biological and bio-optical data from a culture of Karenia mikimotoi, a harmful phytoplankton dinoflagellate. These data showed characteristic signals in chlorophyll-specific absorption and backscattering coefficients. The bio-optical data from the culture and a bio-optical model were used to construct a training dataset for an existing statistical classifier. MERIS imagery over the European continental shelf were processed with the classifier using different training datasets. The differences in positive rates of detection of K. mikimotoi between using an algorithm trained with purely manually selected areas on satellite images and using laboratory data as training was overall <1%. The difference was higher, <15%, when using modeled optical data rather than laboratory data, with potential for improvement if local average chlorophyll concentrations are used. Using a laboratory-derived training dataset improved the ability of the algorithm to distinguish high turbidity from high chlorophyll concentrations. However, additional in situ observations of non-harmful high chlorophyll blooms in the area would improve testing of the ability to distinguish harmful from non-harmful high chlorophyll blooms. This approach can be expanded to use additional wavelengths, different satellite sensors and different phytoplankton genera. |
|
