Dataset: Cell abundances of taxonomic groups determined using a custom convolutional neural network from live Imaging FlowCytobot (IFCB) at seven stations sampled during R/V Pt. Sur cruise PS 18-09 in the western Gulf of Mexico, Sept-Oct 2017.

On each of 2 cruise legs 01 and 03, samples were collected at 7 stations (S01, S06, S11, S16, S21, SS, and GI) from 2 depths [surface and chlorophyll maximum depth when possible; see HRR-bottle data]) by CTD-rosette. At each station, triplicate 5-ml samples pre-filtered through 150 µm Nitex were analyzed immediately with an onboard Imaging FlowCytobot.  All image data can be viewed on the TOAST dashboard: https://toast.tamu.edu/timeline?dataset=HRR_cruise.

Image analysis and feature extraction were performed using software developed by Sosik and colleagues which is available on github (https://github.com/hsosik/ifcb-analysis/). The automated classification approach of Sosik & Olson (2007), as modified and described by Anglès et al. (2019), was employed and the automated classification results were then inspected visually and manually corrected into a total of 102 categories that included 35 categories of diatoms, 30 categories of dinoflagellates, 10 categories of ciliates, 10 categories of flagellates, and 17 ‘others’, which included filamentous cyanobacteria, freshwater chlorophytes, coccolithophorids, and small cells that could not be identified taxonomically from images (refer to Fiorendino et al. 2021. for more details).

For comparison with the Texas Observatory for Algal Succession Time-series (TOAST), IFCB images were also classified automatically into one of 112 classes utilizing a custom convolutional neural network (CNN) trained on a curated set of images (Henrichs et al. 2021.). See related dataset.

Biomass for each image was estimated using the algorithm developed by Moberg & Sosik (2012) to calculate cellular volume from the extracted image features and then convert to total carbon per image (Menden-Deuer & Lessard 2000) and summed for each class. See related dataset.

Sampling locations: