Dataset: ZooSCAN output MOCNESS BATS: M3 to M13

Methodology:

To obtain samples, a 1 m Multiple Opening/Closing Net and Environmental Sensing System  (MOCNESS; Wiebe et al., 1985) equipped with 150 m nets was deployed during the mid-day and mid-night on cruises carried out in the vicinity of the Bermuda Atlantic Time-series Study site (BATS) in July of 2016, 2017, and 2018 as well as October 2018 (eight casts in total, 63 discrete nets; Table 1). The net sampling plan used adaptive profiling, modifying the depth of closure to capture eight distinct ecological zones from surface to 1000 m depth: e.g. the thermocline, deep chl-a maximum (DCM), above the O2 minimum, within the O2 minimum core, and below the O2 minimum (Maas et al., 2014, Steinberg et al., 2008a). The actual tow depths were set on a cruise-by-cruise basis, using CTD profiles to determine the depths of each ecological zone. Upon retrieval, the catch from each of the eight discrete nets were divided into splits. Half were preserved in 95% undenatured ethanol and the remainder preserved in buffered 4% formalin in seawater.

A representative subsample of the zooplankton community from each net were imaged from the formalin-preserved samples, then measured, using a ZooSCAN ver. 3 at 4,800 dpi (following the methods in: Gorsky et al., 2010, Vandromme et al., 2012). In order to better represent all size classes in the images, the original sample was divided into three size categories. All individuals larger than 2 cm were selected by eye and scanned separately from all the others. The remainder of the sample was sieved through a 1-mm mesh sieve, and both size fractions were individually scanned. From these smaller size fractions, at least 1500 particles were scanned after subsampling using a Motoda splitter (Motoda, 1959), requiring generation of two separate scans for both size classes. This resulted in a total of five images per net. Raw images were then processed in ZooProcess (Gorsky et al., 2010, Vandromme et al., 2012), and the resulting vignettes, along with a measurement and metadata file, were uploaded to EcoTaxa (https://ecotaxa.obs-vlfr.fr/; Picheral et al., 2017) for machine-assisted identification, using a training set developed by the authors. The taxonomy of one paired set of tows was completely validated (July 2017). For the remaining tows taxonomy is based on the Ecotaxa prediction using this 2017 training set.

Object_id and imagenames: