Marine snow morphology illuminates the evolution of phytoplankton blooms and determines their subsequent vertical export

The study addresses how the morphology of individual marine snow particles (beyond size alone) governs their vertical export and how these morphologies evolve during different phases of Arctic phytoplankton blooms. The authors hypothesize that (1) marine snow can be objectively classified into a small set of ecologically meaningful morphotypes using image-derived traits, and (2) the concentration, vertical distribution, and attenuation of these morphotypes vary systematically with bloom phase and phytoplankton community composition, with implications for carbon export. Using in situ imaging, they aim to move beyond traditional size-based particle metrics to a trait-based classification and link these morphotypes to bloom dynamics in Baffin Bay and Fram Strait.

Previous work on the biological carbon pump often modeled particles with simplified geometries (spheres or fixed fractal dimensions) and fixed mass–volume relationships due to observational constraints, aggregating all detrital particles into single size-resolved compartments. Although particle size is commonly used, size alone is not consistently predictive of sinking velocity, and other morphological attributes (shape, porosity, internal structure, brightness) are rarely quantified. Advances in optical imaging (e.g., UVP, VPR) have vastly increased observations of particles and plankton, yet standardized, quantitative categories for marine snow remain lacking. Prior studies using traps, floc catchers, and counters improved understanding of aggregates but with limited resolution and without structural characterization. Given that plankton community structure (diatoms vs. flagellates such as Phaeocystis) influences aggregate formation and export efficiency, a morphotype-based approach for marine snow could reveal new mechanistic links between bloom phenology and carbon export.

Study areas and sampling: Two Arctic regions were sampled in June–July 2016. Baffin Bay (GreenEdge project, CCGS Amundsen) included 164 stations aligned along ice retreat. Fram Strait (FRAM/LTER Hausgarten, RV Polarstern) included 26 stations across contrasting water masses (West Spitsbergen Current, East Greenland Current). Stations were grouped into bloom phases using a time-for-space framework based on sea ice conditions. In Baffin Bay, Open Water Days (OWD) quantified time since/til ice-free (negative under ice, positive after breakup), strongly correlated with distance to the ice edge (R=0.88). Phases: ice-covered (<−10 OWD, 54 stations), ice break-up (−10 to +10 OWD, 24 stations), and ice-free (>+10 OWD, 42 stations). Fram Strait phases: ice-covered (5 stations, EGC domain), ice break-up (7 stations in WSC during ice edge, June 25–29), ice-free (same stations sampled July 6–7). The potential bias due to differing water masses (Arctic vs. Atlantic) was assessed; morphotype composition was similar whether using joint data or Atlantic-only (Supplementary Fig. 7).

Imaging system and data acquisition: The Underwater Vision Profiler UVP 5hd detected and counted particles >~100 µm within ~1 L illuminated volume and saved vignettes for objects >80 pixels (~0.5–200 mm). The UVP was mounted on a CTD rosette (SBE 911plus), acquisition up to 20 Hz, descent up to 1 m s−1.

Image curation and trait extraction: Vignettes were classified in EcoTaxa (https://ecotaxa.obs-vlfr.fr/) with ML assistance, then manually validated to separate non-living marine snow from zooplankton. For each marine snow object, 24 morphological properties were computed capturing size (area, perimeter), shade intensity (mean/median gray), shape (symmetry, elongation, circularity), and structural complexity/heterogeneity (gray-level variability). The dataset encompassed >1 million marine snow images.

Preprocessing and dimensionality reduction: The 0.1% most extreme values per variable were trimmed to remove outliers; selected variables were log-transformed to reduce skewness. Principal Component Analysis (PCA; R packages FactoMineR, factoextra) summarized trait variation. The first four PCs explained 87% of total variance.

Clustering into morphotypes: K-means clustering was applied to the 4D PCA coordinates to define five morphotypes (chosen for interpretability and ecological relevance): dark, elongated, flake, fluffy, agglomerated. Robustness was assessed by training on (a) Baffin Bay only, (b) Fram Strait only, (c) combined data, with inter-rater reliability (Cohen’s kappa; R package irr). Agreement was high: 99.3% (k=0.991) when training on Baffin and predicting Fram; 92.9% (k=0.905) when reversed; prediction of independent datasets vs. 2-campaign classification: 92.2% (k=0.898).

Derived metrics: Marine snow concentrations were computed in 5 m bins as counts per sampled volume (avg ~112 L per bin). Morphological diversity (Shannon-Wiener) treated each morphotype as a species; sensitivity tests with 25/50/100 k-means clusters yielded consistent conclusions. Classical particle descriptors were also calculated: total concentration, total equivalent spherical volume, and size spectrum slope a from ln(n)=a ln(d)+b. Vertical attenuation of concentration over 0–500 m used a Martin-type power law with exponent b, referencing concentration at 100 m (n100). Bulk sinking rates for morphotypes were estimated from temporal deepening of concentration peaks.

Phytoplankton and zooplankton context: Phytoplankton group composition (diatoms, Phaeocystis, others) was estimated from HPLC pigments via CHEMTAX using Arctic-appropriate pigment:Chl a ratios and validated by microscopy. Zooplankton signals were taken from the UVP imagery (living forms). Maps and hydrography contextualized sampling relative to ice and water masses.

Data and code availability: UVP datasets are browseable on EcoTaxa (Baffin: project 149; Fram: project 257). Background data available via GreenEdge and PANGAEA repositories. Code to build morphospace and display images is available at https://github.com/jiho/morphr.

• Five robust marine snow morphotypes emerged from k-means in PCA trait space (n≈1,063,576 particles):
  1. Dark: small (~4 mm mean perimeter 50 px), circular, homogeneous, dark.
  2. Elongated: medium (~7 mm, 78 px), elongated, variable brightness.
  3. Flake: small (~4 mm, 46 px), circular, bright.
  4. Fluffy: medium (~7 mm, 75 px), bright, heterogeneous flocs.
  5. Agglomerated: large (~17 mm, 194 px), bright, multi-element aggregates. First four PCs explained 87% variance; cross-region classification agreements 99.3% (k=0.991) and 92.9% (k=0.905); independent predictions 92.2% (k=0.898).
• Baffin Bay spatio-temporal patterns (by Open Water Days, OWD):
  • Marine snow concentration first elevated under ice (~−20 OWD), increased through break-up (0 OWD), peaking ~+20 OWD.
  • Dark compact morphotype peaked between 50–100 m pre-breakup and progressively deepened to ~1000 m; exhibited the lowest vertical attenuation (b), dominating deeper layers late in the bloom.
  • Elongated forms concentrated in upper 50 m with distinct peaks at −10 OWD and 0 OWD; highest attenuation among morphotypes, decreasing after +10 OWD.
  • Flake and fluffy morphotypes became most abundant in upper 200 m of ice-free waters (>0 OWD), but were rare under ice.
  • Agglomerated forms peaked late ice-free (>+10 OWD), mainly at 20–100 m.
  • Morphotype diversity (Shannon) highest in upper 100 m until break-up, later increased at depth (to ~300 m before and ~500 m after +20 OWD).
• Coupling with plankton (Baffin Bay):
  • Under ice: low pigments (<1 mg m−3), diatom-dominated; marine snow low, mainly dark, elongated, flake.
  • At break-up: narrow surface peak (>0.5 ind L−1) of elongated forms at 20–30 m; pigments increased, diatoms still dominant.
  • Ice-free: diatom contribution decreased; Phaeocystis bloomed below ~20 m; marine snow composition shifted, fluffy and agglomerated outnumbered dark/elongated at surface. Marine snow concentrations generally exceeded zooplankton across phases.
• Fram Strait patterns:
  • Under ice: diatom pigments near surface; Phaeocystis below ~15 m. Marine snow low, dominated by elongated in upper 150 m; fluffy/agglomerated rare but present.
  • Break-up: high pigments and zooplankton in upper 50 m without concomitant surface marine snow increase; instead, elevated marine snow (~0.4 ind L−1) below 50 m down to ~500 m, dominated by flake and fluffy types.
  • Ice-free: surface marine snow peak in upper 50 m dominated by dark and elongated (~50%), coincident with zooplankton maximum; a secondary peak at 50–150 m of mixed flake/fluffy/agglomerated.
• Sinking and attenuation:
  • Dark morphotype showed rapid export: inferred maximum sinking ~90 m d−1 (appearance at 900 m 10 days after surface production), with an average sinking estimate ~38 m d−1 from linear regression of peak time vs. depth.
  • Elongated morphotype had low settling speeds (~7.2 m d−1) and remained mostly above 100 m; likely fragments into dark or aggregates into flake/fluffy/agglomerated.
  • Morphotype-specific attenuation b differed substantially: lowest for dark, highest for elongated; total particle profiles masked these differences, demonstrating added value of morphotype resolution.
• Ecological interpretation:
  • Dark morphotype sources include ice algae residues, fecal pellets, fragmented/densified material; dominate deep ocean marine snow late in bloom.
  • Elongated resemble diatom chains and/or fecal pellets; align with diatom presence and zooplankton distribution.
  • Flake/fluffy/agglomerated increase during late bloom and Phaeocystis dominance; Phaeocystis exopolymers likely enhance stickiness, size, heterogeneity, and scavenging while settling.
• Conceptual advance: Morphotype succession mirrors Arctic bloom phenology (under-ice → ice-edge diatoms → late-bloom Phaeocystis) and clarifies how morphological traits, not size alone, govern settling and export.

The findings validate that a compact, objective morphotype classification derived from image-based traits can track bloom dynamics and predict vertical export behavior better than size-only metrics. Morphotype composition evolved consistently with phytoplankton succession: elongated and dark forms dominated earlier (under ice/ice-edge), while flake, fluffy, and agglomerated forms increased during late bloom and Phaeocystis prevalence. Morphotype-specific sinking and attenuation reveal that compact, dark particles export efficiently to depth whereas elongated particles are retained near the surface, and large agglomerates can form subsurface maxima and extend to depth under certain bloom conditions. These patterns demonstrate that composition, compactness, and density modulate export, complementing and refining classic size-based paradigms and improving interpretation of carbon export processes during Arctic blooms.

This study introduces a scalable, objective classification of marine snow into five ecologically meaningful morphotypes using UVP imagery and multivariate trait analysis. It demonstrates strong, reproducible links between morphotypes, bloom phases, and vertical export dynamics across two Arctic regions. The approach provides improved insight into particle attenuation and sinking behavior relative to size-only metrics, offering a path to better quantify and model the biological carbon pump. Future directions include: expanding training datasets across regions, seasons, and ecosystems to refine/extend morphotype catalogs; integrating image-derived porosity/density to estimate flux and sinking speeds directly; developing generic algorithms to convert morphotype composition to mass and flux; compiling global UVP datasets to assess universality vs. regional specificity; and defining transformation functions among morphotypes to capture particle life cycles. Real-time classification aboard ships and on autonomous platforms could guide adaptive sampling and enable high-frequency monitoring of particle dynamics.

• The typology was developed and validated within Arctic ice-edge bloom contexts; generalization to other ecosystems and seasons requires additional training datasets.
• Morphotypes arise from continuous trait variation; boundaries are operational and may encompass diverse particle origins.
• The origin/content of individual particles cannot be confirmed because particles were imaged in situ and not collected.
• Fluxes were not computed due to unknown density and porosity; bulk sinking rates were inferred indirectly from peak deepening.
• Time-for-space (OWD) framework may introduce biases due to differing water masses/currents; analyses suggest limited impact, but residual effects may remain.
• Short observation windows limited sinking rate estimates for some categories (e.g., agglomerated).
• Zooplankton and particle interactions are inferred from co-occurrence in imagery, not direct process measurements.