Global mass of buoyant marine plastics dominated by large long-lived debris

The study addresses the long-standing discrepancy between large estimated inputs of plastic to the ocean and the much smaller amounts observed floating at the surface. Prior estimates suggest 800–2,400 kilotonnes per year from rivers and 4,800–23,000 kilotonnes per year from coastal regions, while only ~250 kilotonnes are observed afloat. The research question is whether the apparent misalignment arises from overestimated inputs, unaccounted sinks (for example, biofouling-driven sinking, beaching, fragmentation), or other processes, and what the true global mass and size distribution of buoyant marine plastics are. The purpose is to build a three-dimensional, observation-constrained global mass budget across reservoirs (surface ocean, subsurface, beaches, sediments) and sizes (0.1–1,600 mm) over decadal timescales (1980–2020), to quantify standing stock, inputs, fluxes between reservoirs, and residence times, thereby clarifying if a ‘missing sink’ is required and which sizes dominate the total mass.

The paper builds on and re-evaluates earlier estimates and conceptual mass budgets of marine plastics. Prior work estimated riverine inputs on the order of 800–2,700 kt yr−1 and coastal mismanaged plastic waste inputs of 4,800–23,000 kt yr−1, and reported ~250 kt of floating plastics at the surface. Previous mass-balance studies highlighted potential sinks including biofouling-induced sinking to depth, beaching, fragmentation to micro- and nanoplastics, and vertical mixing. Observations show accumulation in subtropical gyres and significant plastic presence in deep ocean sediments, including low-density polymers, suggesting biofouling is critical to export. Earlier studies often used broad size bins and mean masses that may bias mass estimates low for larger, sparsely sampled items. Recent modeling and observations suggest lower inputs from rivers and coastlines than earlier top-down estimates, and that a substantial fraction of open-ocean macrolitter originates from marine sources (for example, fisheries). This study integrates and updates these lines of evidence with a data-assimilative, size-resolved approach.

• Overall approach: A 3D global, size-resolved mass budget for initially buoyant plastics (0.1–1,600 mm) is constructed by assimilating multi-reservoir observations into a hybrid Lagrangian–Eulerian transport framework. The model spans decades (1980–2020) and estimates sources, transport, sinks, and fluxes among surface ocean (0–5 m), subsurface (>5 m), beaches, and marine sediments.
• Data assimilation: A Bayesian framework with ensemble smoother with multiple data assimilation (55 members, iterated 8 times) updates 16 model parameters governing sources, transport, and sinks. Priors (Gaussian) reflect literature-based ranges; observation errors are characterized via variograms. Post-assimilation, subgrid variability is estimated from model–data mismatch and propagated using Monte Carlo perturbations to derive uncertainty bounds.
• Observations assimilated: 14,977 surface-water measurements, 7,114 beach measurements, and 120 deep-ocean measurements, including number and mass concentrations; fishing-related fractions from 2,303 beach measurements. Wind-mixing bias in net trawls is corrected using the Kukulka method with ERA5 winds.
• Transport core: Lagrangian OceanParcels simulations forced by Mercator Ocean PSY4 (1/12°) advection and vertical diffusivity produce monthly transition matrices (121k×121k) on an Uber H3 ~60-km grid with four depth bins (0–5 m, 5–50 m, 50–500 m, bottom) and coastline segments. Particles are released on a global hexagonal grid and at 12 log-spaced depths (0.5–5,000 m), repeated monthly over 2015–2019. Six explicit sizes (0.1, 0.4, 1.6, 6.4, 26, 102 mm) are simulated; larger particles up to 1.6 m are assumed to behave similarly to 102 mm due to persistent surface buoyancy. Stokes drift/windage and a separate stochastic diffusion term are not added because transition-matrix construction already introduces effective diffusion and prior tests did not improve skill.
• Vertical behavior scenarios: Four scenarios define vertical motions: (1) positively buoyant with size-dependent rise velocities (baseline spherical particles with density 1,010 kg m−3; linear combination across sizes calibrated during DA to represent a distribution of shapes/densities); (2) oscillatory biofouling/defouling (growth/respiration) using BIOMER4 biogeochemistry; (3) permanent fouling (no respiration), allowing export to seafloor; (4) neutrally buoyant. The model treats the ocean as an assemblage of these behaviors with calibrated fractions fosc, fperm, fn (remainder positively buoyant). Particles that hit the seafloor in the permanent-fouling case are sedimented and removed.
• Sources parameterization: Three source categories: rivers (using Meijer et al. 2021 dataset scaled by a global factor S to encompass high/low literature bounds), coastal mismanaged plastic waste (MPW) proportional to coastal population density within 50 km, scaled by Spop relative to river input, and fishing activity from global fishing hours scaled by Sf relative to river input. Priors allow 2.7–7.3× river input for coastal MPW and 0.2–2.0× for fishing-related sources. Input size distribution is log-normal centered around ~0.2 m, consistent with packaging sizes and riverine observations. Temporal trend in inputs parameterized as exponential growth with a prior centered on plastic production trends, lower bound zero.
• Beaching and removal: Beaching probability parameterized via a beaching timescale tbeach (log10 prior; midpoint ~100 days, lower ~25 d). Resuspension timescales depend on size (experimental constraints). A beach removal probability Premoval (burial/cleanup/UV degradation) uses a prior midpoint of 0.2% per month, varied an order of magnitude to capture 0.8–4.0% per month from literature. A coastline length correction accounts for fractal geometry (dimension ~1.27) and limited coastline fraction per grid cell with a minimum-length parameter lbeach_min.
• Fragmentation: Occurs predominantly on beaches (higher temperature/UV/abrasion); ocean fragmentation is neglected if not exceeding beach rates. A cascading fragmentation model with rate λ and shape factor d (dimensionality 2–3) evolves size distributions; particles smaller than 0.1 mm are treated as a sink and removed. Bounds for λ set on log10 scale based on experiments and prior models; d bounded by observed size–mass relations.
• Size spectrum and masses: Full size spectrum from 0.1 to 1,600 mm in factors of 2. Transition matrices for intermediate sizes are linearly interpolated; >0.1 m assumed surface-bound transport. Particle masses for larger items are extrapolated from length using functions consistent with the fragmentation model; both number and mass distributions are tracked.
• Outputs: The assimilation yields posterior parameter estimates and uncertainty ranges for standing stocks, inputs by source category, inter-reservoir fluxes (for example, sedimentation, fragmentation to <0.1 mm), and size-resolved mass and number distributions for a reference year (2020) and scenarios.

• Global standing stock (2020): 3,200 kt of initially buoyant plastics (95% CI: 3,000–3,400 kt).
• Reservoir partitioning: 59–62% at the ocean surface (≤5 m; ~1,909–2,006 kt), 36–39% below 5 m (~1,165–1,262 kt), and 1.5–1.9% on coastlines (~49–61 kt).
• Annual input (2020): 500 kt yr−1 (95% CI: 470–540 kt yr−1), growing ~4% per year. • Source contributions: coastlines 39–42% (≈190–220 kt yr−1), fishing 45–48% (≈220–260 kt yr−1), rivers 12–13% (≈57–69 kt yr−1). • These inputs are at least an order of magnitude lower than several previous estimates for rivers and coastlines.
• Size dominance: >25 mm items account for 90–98% of buoyant plastic mass (≈2,800–3,300 kt). Microplastics (<5 mm): ~49–53 kt; 5–25 mm: ~150–170 kt.
• Sedimentation/export: ~220 kt yr−1 exported to marine sediments, including ~6 kt yr−1 microplastics; cumulative sedimentation of initially buoyant plastics since 1950 estimated at ~6,200 kt.
• Beach sink and fragmentation: Coastline sink (burial/cleanup/UV) ~3 kt yr−1. Fragmentation to <0.1 mm removes ~73 kt yr−1; ~2.2% of >5 mm plastics fragment to <5 mm per year (close to prior estimates ~3% yr−1).
• Residence times and scenarios: After a hypothetical cessation of inputs in 2025, only ~10% of marine plastic mass would be removed within 2 years; loss rates decline as plastics accumulate in subtropical gyres with low beaching and sinking. Under business-as-usual (~4% annual input growth), the standing stock could double within two decades.
• Model–data agreement: Posterior concentrations match observed number and mass across reservoirs and sizes (scatter comparisons within expected measurement error bands), supporting the inferred lower inputs and larger standing stocks without requiring an unobserved sink.
• Implication for sampling: Large, sparse items dominate mass and are undersampled; broad size bins and mean-mass assumptions in prior studies likely led to underestimation of total mass for 5–200 mm and >200 mm classes.

The results reconcile the apparent mismatch between high estimated inputs and relatively low observed floating mass by jointly lowering plausible inputs (especially from rivers and coastal MPW) and increasing the estimated standing stock, dominated by long-lived macroplastics. Assimilation of multi-reservoir, size-resolved observations with a mechanistic transport and sink framework shows that a ‘missing sink’ is not required to close the buoyant plastic budget. Biofouling-driven export, beaching, and fragmentation at empirically constrained rates suffice to explain observed distributions, including significant mass below the surface and on beaches. The dominance of >25 mm items in mass explains why number-based sampling (biased toward small particles and with variable lower detection limits) can misrepresent total mass unless size-resolved and mass-weighted properly. Longer residence times than previously modeled imply persistent accumulation in subtropical gyres and a slow decline even if inputs cease, underscoring the urgency of mitigation and the likely increasing ecological impacts under continued input growth. The partitioning of sources indicates a major role for at-sea inputs (fisheries) alongside coastal leakage, guiding targeted interventions (for example, fishing gear management, coastal waste management).

This study provides the first observation-constrained, 3D, size-resolved global mass budget for initially buoyant marine plastics, estimating ~3,200 kt in 2020 with >95% of mass in >25 mm items and annual inputs of ~500 kt yr−1 growing at ~4% yr−1. The combined lower inputs and higher standing stock resolve the perceived need for a missing sink and indicate much longer residence times than previously thought. Key contributions include: (1) unified assimilation of multi-reservoir, size-resolved observations; (2) mechanistic inclusion of biofouling-driven vertical motions, beaching/resuspension, and beach-dominated fragmentation; and (3) robust quantification of source contributions and inter-reservoir fluxes. Future research should: (a) improve size-resolved sampling of large, sparse items and report both number and mass with strict lower detection limits; (b) develop continuous size-spectrum methods in field campaigns and models; (c) better constrain fragmentation rates and shapes across environments; (d) expand to include non-buoyant polymers and mixed-density items; and (e) refine sedimentation and nearshore processes, especially close to sources. Policy and mitigation targeting at-sea sources (for example, fishing gear) and coastal leakage, combined with cleanup strategies, are needed to curb growth in the standing stock.

• Scope limited to initially buoyant plastics; denser polymers and negatively buoyant items are excluded, likely underestimating sedimentation near source regions and total marine plastic mass.
• Vertical biofouling oscillations are modeled but not yet empirically verified; parameterizations (fractions of behaviors) carry uncertainty.
• Fragmentation is assumed dominant on beaches and negligible in the ocean; if ocean fragmentation is faster locally, losses to <0.1 mm could be underestimated or misallocated.
• Particle properties simplified: baseline spherical shapes and calibrated rise velocities approximate a distribution of shapes/densities; large-item transport assumed similar to 102 mm class.
• Transition matrices introduce effective diffusion and rely on 30-day windows and 2015–2019 forcing; submesoscale variability and interannual changes may be imperfectly captured.
• Source inputs rely on global datasets scaled by factors (S, Spop, Sf) to address biases; spatial/temporal heterogeneity and unreported sources (for example, illegal dumping) may not be fully represented.
• Observational datasets have heterogeneous methods and detection limits (particularly visual beach surveys), potentially biasing assimilation despite applied corrections.
• The model removes particles <0.1 mm as a sink; fate and transport of sub-100 µm plastics are not resolved.
• Stokes drift/windage terms are omitted following preliminary tests; their localized effects could matter regionally.