MEPS 284:77-96 (2004)  -  doi:10.3354/meps284077

ABSTRACT: A fundamental problem in limnology and oceanography is the inability to measure the taxonomic composition of plankton quickly over a broad range of scales. Traditional sampling with bottles and nets provides critical data at the species-level, but has limited spatio-temporal coverage and can destroy ubiquitous delicate forms. To augment traditional sampling, recent advances in bio-acoustics and non-imaging optics provide real-time high-resolution data on biomass abundance and size composition. New optical imaging approaches provide coarse taxonomic composition but require manual image identification, preventing real-time observation. Here, we describe a method of optical sampling and analysis, using the Video Plankton Recorder, to automatically identify plankton to major taxa (and to species in some cases) and observe their distributions at sea in real time. We present a detailed assessment of classifier accuracy, including a comparison of machine- and hand-classification of images. The automated classification method was found to be sufficiently accurate for estimating abundance patterns of dominant taxa but not for less abundant taxa. The range in overall accuracy was 60 to 70% for 7 taxa and 79 to 82% for 2 taxa, with accuracies for individual taxa ranging between 45 and 91%. Classification error was small relative to natural variability in abundance of dominant taxa. For a given taxon, the error in the abundance estimate was low in regions of high relative abundance. A manual correction step can be used in areas of low relative abundance to obtain accurate abundance estimates. Example data from 2 cruises are presented to illustrate the utility of real-time taxa-specific data collection. These data represent the first real-time automatic identification and mapping of plankton taxa at sea. This methodology represents an intermediate step towards the ultimate goal of real-time identification of plankton to the level of species and life stage. At present, optical imaging methods cannot replace net and bottle surveys, but can be used to obtain coarse taxonomic composition of plankton (including fragile forms) with an identification accuracy that is high enough to produce quantitative high-resolution maps of abundant taxa in real time.

KEY WORDS: Plankton · Video · Sampling · Pattern recognition · Image analysis · Real time