Marine litter in the Mediterranean: observation-based Lagrangian particle tracking

The Mediterranean is a major hotspot for floating marine macro-litter (FMML) accumulation, yet spatially explicit, observation-grounded predictions of its transport and coastal deposition remain scarce. We present a data-driven stochastic Lagrangian framework that uses monthly gridded FMML observations from systematic shipboard surveys as simulation initial conditions complementing existing source-based modelling approaches. Observations were collected from August 2023 to December 2024 within the LIFE Conceptu Maris Project using ferries as monitoring platforms under a standardised protocol. A total of 158 surveys detected 4,801 FMML items across approximately 6,200 km² of surveyed sea surface, of which more than 62% were artificial polymers. Observed monthly FMML item counts at 5 × 5 km resolution initialised ensemble Lagrangian simulations (18 members per target month; 17.2–32.5 million virtual particles per seeding window, 5,000 particles per observed item) using Parcels, forced by Copernicus Marine Service Mediterranean reanalysis currents and ERA5 wind fields with a 1% windage coefficient and a 2D random-walk turbulent diffusion scheme. Monthly frequency-of-occurrence maps were computed at the sea surface and on coastlines. The Southern Tyrrhenian and Northern Ligurian seas, including the Pelagos Sanctuary, exhibited in 2024 near-persistent high sea surface FMML occurrence, peaking in late summer–autumn and reaching a seasonal minimum in winter–early spring. The Balearic Sea showed elevated occurrence from late spring through summer (May–September), declining through autumn. The Adriatic displayed intense sub-basin-scale spreading from January to June and again in December, with summer leakage into the Northern Ionian Sea. Coastal deposition frequencies were highest on Corsica, western Sardinia and northern Tunisia, then the south-eastern Tyrrhenian and Ligurian–Tuscan margins, with Valencia–Barcelona largely confined to summer. Integration with loggerhead turtle nesting records illustrates how coastal frequency-of-occurrence maps can support conservation applications. A web-based Particle Analysis Dashboard translates these outputs into an interactive tool with Marine Protected and Conserved Area overlays for scientists and decision-makers. The framework provides a scalable, observation-anchored foundation for MSFD Descriptor 10 monitoring, transferable to other semi-enclosed sea basins. Priority future work includes complementary validation against independent beach-litter monitoring data, integration of land-based source inventories, and linkage with species distribution models.

Over recent decades, plastic pollution and anthropogenic marine litter have increased significantly, with plastics comprising 50–90% of marine debris worldwide (Galgani et al., 2015; Agamuthu et al., 2019). Global syntheses have estimated that 1.1–4.9 million tonnes of plastic are floating in the ocean’s surface layer, with abundance rising rapidly since the mid-2000s (Eriksen et al., 2023), driven by increased production and inadequate waste management. The decay of plastic in the aquatic environment is extremely slow, typically estimated to take between a hundred and several thousand years (Gregory and Andrady, 2003; Barnes et al., 2009). This high persistence makes plastic a significant threat to marine ecosystems.

Marine plastics have been reported to negatively affect marine mammals and sea turtles through entanglement, ingestion, and toxicity. Among the most negative effects are physical wounds, suffocation, gastrointestinal blockages, additional stress, and death. The plastic’s ability to absorb harmful chemicals, such as heavy metals, leads to physiological impacts in marine organisms, including reduced growth, reproductive failure, and altered behaviour (Tuuri and Leterme, 2023; Fossi et al., 2025). Given these threats, effective management strategies for plastic pollution are needed at both regional and global scales.

Among regional seas, the Mediterranean is recognised as a major accumulation hotspot, where semi-enclosed circulation, dense coastal populations, tourism, and intensive shipping are associated with elevated litter concentrations (Cózar et al., 2015; Macias et al., 2019). The Mediterranean’s Blue Economy can be severely damaged by plastic pollution, which affects fisheries and aquaculture, the maritime and tourism industries, port infrastructures, and desalination stations.

Floating Marine Macro-Litter (FMML) represents a particularly relevant component of marine litter pollution because it directly interacts with marine megafauna inhabiting the sea surface and upper water column. Floating litter can persist at or near the sea surface for extended periods and is transported over large spatial scales by currents and winds, accumulating in convergence zones and potential ecological hotspots (Soto-Navarro et al., 2020; van Sebille et al., 2020; Arcangeli et al., 2025). Due to its mobility and heterogeneous distribution, FMML is considered an important indicator for assessing marine litter distribution patterns, identifying accumulation areas, and evaluating potential exposure risk for vulnerable species such as cetaceans, sea turtles, and seabirds (Arcangeli et al., 2018). For this reason, floating litter is included among the primary criteria of Descriptor 10 of the European Marine Strategy Framework Directive (MSFD, 2008/56/EC), which requires Member States to establish monitoring programmes and assess Good Environmental Status (GES) in relation to marine litter abundance, composition, and distribution in the surface layer of the water column. Despite increasing effort, FMML remains difficult to monitor systematically and continuously at large scale. Offshore areas are vast, remote, and logistically demanding, and litter distribution is highly heterogeneous due to the combined influences of currents, winds, maritime activities, and uncertain sources. These factors limit the availability of spatially detailed, large-scale observational datasets and make it challenging to identify persistent or temporary accumulation areas. Dedicated monitoring protocols based on using ferries as observation platforms have been progressively developed in the Mediterranean Sea, allowing repeated observations along fixed offshore transects and the collection of harmonised FMML datasets across large marine areas (Arcangeli et al., 2015, 2018). These systematic observations provide valuable information on the abundance, composition, and spatio-temporal variability of floating litter and represent an important contribution to MSFD Descriptor 10 monitoring activities.

Previous Lagrangian modelling studies have provided important advances in reconstructing macroplastic transport and accumulation in the Mediterranean and European seas, often using estimated land- and sea-based inputs, such as riverine discharge, coastal sources, maritime activities, or fisheries-related litter, to initialise particle releases (Liubartseva et al., 2018; Garcia-Gorriz et al., 2026). These source-based approaches are essential for assessing potential emission pathways and basin-scale accumulation patterns, particularly in offshore areas where observation remains sparse. However, they generally rely on estimated inputs rather than on direct observations of litter detected at sea.

To complement these approaches, recent efforts increasingly promote integrating observing systems that couple in situ surveys, remote sensing, and numerical models within an Integrated Marine Debris Observing System (Maximenko et al., 2019).

Marine litter modelling frameworks commonly identify and quantify inputs from several anthropogenic sources, including coastal populations, riverine discharges, shipping lanes, urban centres, and ports (Politikos et al., 2017; Liubartseva et al., 2018). Litter inputs are often distributed along shorelines proportionally to coastal population density, reflecting the dominant contribution of land-based sources to marine debris (Jambeck et al., 2015; Zambianchi et al., 2017), while river mouths are represented as major conduits for litter transport, in some cases using river runoff data to estimate temporal variability in discharges (Politikos et al., 2017; Christensen et al., 2023; Cloux et al., 2024; Teng et al., 2025; Garcia-Gorriz et al., 2026). Additional inputs are typically associated with major maritime traffic routes to simulate losses from vessels (Liubartseva et al., 2018; Mansui et al., 2015), as well as with large coastal cities, harbours, and marinas that act as important accumulation and release points (Politikos et al., 2017; Liubartseva et al., 2018).

Despite these advances, significant limitations remain in both modelling approaches and validation procedures. Current marine litter models are strongly constrained by emission inventories that primarily rely on land-based inputs while often underrepresenting important sea-based sources such as commercial shipping, recreational boating, and abandoned, lost, or discarded fishing gear (ALDFG) (Jambeck et al., 2015; Christensen et al., 2023; Teng et al., 2025; Garcia-Gorriz et al., 2026). Source datasets are frequently affected by sparse observational coverage, uncertainties in absolute litter quantities and weight-per-item conversion factors, incomplete geographic coverage, and simplified temporal assumptions that fail to capture seasonal variability linked to rainfall events or tourism (Liubartseva et al., 2018; Christensen et al., 2023; Sánchez-García et al., 2024; Garcia-Gorriz et al., 2026). Validation remains equally challenging because of sporadic and non-synoptic field observations, methodological inconsistencies among monitoring platforms, biases in historical drifter datasets, satellite detection limits restricted to surface debris above pixel-scale thresholds, and observer-dependent variability in visual surveys, all of which contribute to uncertainty in simulated litter distributions and transport pathways (Zambianchi et al., 2017; Liubartseva et al., 2018; Christensen et al., 2023; Sánchez-García et al., 2024; Garcia-Gorriz et al., 2026).

To overcome some of these limitations, this study uses ferry-based observations as the primary source of marine litter input. Litter distributions were derived from direct sea surface observations and aggregated over 5 × 5 km grid cells, providing a spatially consistent representation of actual floating litter conditions in the Mediterranean Sea. Unlike approaches based solely on indirect proxies or emission estimates, this methodology relies on real observations of debris already present in the marine environment and potentially available for transport and dispersion. Furthermore, all data were collected under a standardised and systematic monitoring protocol (Arcangeli et al., 2020; MSFD Technical Group on Marine Litter, 2023), ensuring methodological consistency and improving the reliability and comparability of the modelled litter distribution patterns.

Building on this concept, we present a model study based on systematic ferry-based observations of floating marine macro-litter collected in the Western and Central Mediterranean and Adriatic Seas as data-driven initial conditions for Lagrangian particle tracking. Regular ferry-based detection of marine litter in the Mediterranean provides unique field data from areas characterised by complex hydrodynamic conditions and uncertain pollution sources. Numerical modelling complements these observations by simulating transport under realistic physical forcing. Using High Performance Computing (HPC), we tracked the observed FMML items to estimate their surface drift pathways and potential beaching on the coastlines.

The paper is organised as follows: Section 2 describes observational and modelling methodologies. Section 3 presents and discusses the results together, identifying persistent accumulation areas and regions of low modelled occurrence, interpreted jointly with the survey-effort coverage (Supplementary Section S1), and closing with the ecological and conservation applications of the framework. In Section 4, we discuss the strengths and limitations of our methodology and outline ways to move forward. Finally, Section 5 summarises the key findings and outlines directions for future research.

Survey design and data collection

In the Mediterranean Sea, a specific protocol for monitoring floating marine macro-litter using ferries as platforms for observation has been developed and applied since 2013 in the Western, Central Mediterranean, and Adriatic seas (Arcangeli et al., 2015, 2018, 2020) in line with the JRC Guidelines (MSFD Technical Group on Marine Litter, 2023). The protocol offers a broad estimate of the amount, composition, and spatiotemporal patterns of macro-litter floating at the surface, supporting the detection of trends across the Marine Strategy Framework Directive (MSFD) assessment period and the definition of baseline and threshold values for Good Environmental Status (GES) assessment (Arcangeli et al., 2025).

The sampling design is based on fixed offshore routes selected to be representative of different environmental and oceanographic conditions, including areas characterised by different current regimes and anthropogenic-maritime pressures. Fixed transects were repeatedly surveyed across all seasons, with 3–5 surveys per season and a minimum sampling coverage of 25 km² per season, increased to 31–40 km² in areas of low litter density, to ensure statistical representativeness and capture seasonal variability in FMML distribution.

FMML observations were conducted using a fixed strip-width sampling method, with observations conducted within a 50-m strip on one side of the vessel. Surveys were carried out by trained observers under daylight and suitable sea-state conditions (Beaufort ≤ 2). The 20-cm threshold is the size class used for FMML monitoring from large vessels in offshore waters, as recommended in the Joint Research Centre (JRC) protocols for floating macro-litter observation (Vighi et al., 2022). All items larger than 20 cm detected within the strip were georeferenced using GPS waypoints and recorded on standardised datasheets. Items were classified according to material, item type, size class, and other characteristics following the Joint List of Litter Categories for Marine Macrolitter Monitoring (Fleet et al., 2021). Standardised data collection by well-trained teams who strictly adhere to the observational protocol supported the consistency and comparability of the data.

In this work, we used a subset of the offshore FMML observations collected from August 2023 to December 2024 during 158 surveys conducted within the LIFE Conceptu Maris Project (LIFE20 NAT/IT/001371).

Data preparation

Linear effort transects were converted into buffered sampling strips according to the fixed strip width of 50 m and intersected with the European Environment Agency (EEA) 5 × 5 km reference grid (ETRS89, EPSG:3035, European Environment Agency, 2013). For each grid cell and month, the effectively surveyed surface area [km²] was calculated from the intersection between sampling strips and grid cells. The total sampled area per cell was computed as the sum of surveyed sampling strips within each cell. FMML observations were then spatially joined to the same grid, and litter density D was computed as the number of observed items per surveyed area [items km −2 ], as given in Equation 1.where N is the number of observed items, W is the width of the strip [km], and L is the length of the surveyed transect [km] (Arcangeli et al., 2018).

This approach allowed the generation of monthly FMML densities while accounting for variations in sampling effort among surveys. No fixed minimum-effort threshold was applied because densities were estimated at a monthly temporal resolution. Applying a threshold of ≥0.25 km², as used for seasonal data, would have excluded a large proportion of sampled cell-month observations. Survey effort per grid cell is mapped in Supplementary Section S1 (Supplementary Figure S1), together with the effect of hypothetical minimum-effort inclusion criteria on the dataset (Supplementary Table S1). Density data were initially inspected to assess their distribution and variability, including standard error (SE), standard deviation (SD), variance, minimum and maximum values, the 25 th and 75 th percentiles. Potential extreme values were initially explored using the conventional upper boxplot criterion (Q3 + 1.5 × IQR; Arcangeli et al., 2025). Because monthly density distributions were zero-inflated and right-skewed, this criterion was used only as an exploratory diagnostic rather than an automatic exclusion rule. Following inspection of the extreme upper tail, 50 objects km⁻², lying above the 99 th percentile of positive cell-month density observations, was used as a conservative operational threshold to exclude exceptionally high density estimates prior to subsequent analyses. This resulted in the removal of 15 cell-month observations out of 12,286 (0.12%). Most excluded observations were associated with small surveyed areas, resulting in high density estimates from only a few litter observations.

Modelling

The observed FMML item counts, from which the monthly gridded densities of Equation 1 are derived, provided the initial conditions for the Lagrangian virtual particle tracking we performed (Figure 1). The stochastic ensemble approach allows us to identify the most and least frequent areas where FMML tends to occur at the sea surface and on the coastlines.

Modelling tool

Lagrangian particle tracking was conducted using the well-established Parcels framework (Delandmeter and van Sebille, 2019) implemented through the plasticparcels Python package (Denes and van Sebille, 2024). The code was designed specifically to model two-dimensional plastic transport.

At the sea surface, each virtual particle moved due to the sea surface currents. In addition, the wind-induced drift (windage) was modelled as a term proportional to the 10-m wind components, using a constant windage coefficient of 0.01 (Kubota, 1994; Allen and Plourde, 1999; Allen, 2005; De Dominicis et al., 2013; van Sebille et al., 2020; Rial-Osorio et al., 2026). Sub-grid scale turbulence was parameterised using a two-dimensional random walk scheme. Isotropic horizontal diffusivity with a coefficient of 10 m 2 s −1 was used (Liubartseva et al., 2018). Gridded advective velocities extracted from Eulerian oceanographic models are interpolated bilinearly in space and linearly in time to the particle positions. In the near-coast zone, where the sea surface kinematics is not calculated by the Eulerian ocean circulation model, we apply the so-called sea-over-land procedure: a kinematic extrapolation of the current field into the unresolved near-coast cells, preventing artificially vanishing velocities at the last wet grid points, originally formulated for atmospheric variables at land–sea boundaries (Kara et al., 2007) and subsequently applied to ocean currents in the MEDSLIK-II oil-spill model (De Dominicis et al., 2013). Particle trajectories were computed with a fourth-order Runge–Kutta integration scheme at a 30-minute time step. The criterion for choosing the time step was the stability condition equivalent to the Courant–Friedrichs–Lewy condition in Eulerian modelling (Nordam et al., 2017). Deterministic absorbing boundary conditions were imposed on the coastlines using a landmask field. Particles intersecting land grid cells were classified as beached and removed from further transport. This procedure is expected to cause an overestimation of the number of beached particles in comparison to those in Liubartseva et al. (2018), while reducing the computational cost significantly.

Meteo-oceanographic datasets

Particle advection was forced by daily sea surface currents from the Copernicus Marine Service (CMS) daily reanalysis in the Mediterranean with a horizontal resolution of 1/24° (~4.5 km) (Escudier et al., 2021). We also employed the 10-m wind datasets provided by the ECMWF ERA5 reanalysis (0.25° horizontal resolution) aggregated to daily means from hourly frequency sampling (Copernicus Climate Change Service, 2024; Hersbach et al., 2020).

Simulation design

Virtual particles were seeded in direct proportion to the observed item counts: every FMML item detected in a given month generated 5,000 virtual particles, released at the centre of its 5 × 5 km observation cell (n = 5,000 × items per cell). Because each monthly density field seeds three such releases, every observed item contributes 15,000 particles in total, and the totals of Table 1 are exactly 15,000 × the number of items observed in each target month’s seeding window. This preserves the relative spatial gradients of the observations, and an item contributes the same number of particles regardless of how small a fraction of its cell was surveyed — densities extrapolated from partial coverage are never amplified; the dependence of the total monthly seeding on survey coverage and litter loading is addressed by the fractional thresholds defined below. The coupling is therefore a monthly observation-based initialisation: the observations set the initial condition of each release and are not assimilated into the trajectories thereafter. Each release is a point source at the cell centre: with a horizontal diffusivity of 10 m² s −1 the released cloud spreads over the scale of its source cell within the first two days of drift (root-mean-square displacement of about 1.9 km after one day), and the sub-cell placement of the release is below the resolution of the 0.05° diagnostics, which share the scale of the observation cells.

To account for variability in atmospheric and oceanographic conditions, an ensemble strategy was adopted. Virtual particles were released on the 5 th, 15 th, and 25 th of each month from August 2023 to December 2024, across a 150-day rolling window. For each 2024 target month, the ensemble comprised the releases whose 150-day trajectories overlapped that month — those from the target month and the five preceding months — giving 18 members (Table 1). The 150-day integration window corresponds approximately to twice the 80-day surface half-life reported for maritime-source plastics in the Mediterranean (Liubartseva et al., 2018). The age composition of the composites, including the per-lag attrition, is quantified in Supplementary Section S3 (Supplementary Table S2).

To represent the spatiotemporal persistence of floating marine macro-litter, monthly frequency-of-occurrence metrics were computed from the ensemble simulations. The complementary diagnostics describe (i) sea surface distribution and (ii) coastal deposition. Both metrics were computed from the 18-member ensemble simulations using a consistent spatial framework.

The model domain was discretised into regular 0.05° × 0.05° grid cells (~5 km in the Mediterranean), and particle positions were aggregated daily: at the sea surface, C 1 (x, y, t) is the sum of the 24 hourly particle counts in each cell during day t (particle-hours per cell per day); on the coastline, C 2 (x, y, t) is the number of particles newly beached in the cell during day t, each particle counted once, on its day of beaching. A grid cell was considered polluted on a given day when C i exceeded a threshold.

The exceedance thresholds THR 1 = 3000 particle-hours per cell per day (sea surface) and THR 2 = 10 newly beached particles per cell per day (coastline) are operational definitions of substantial accumulation, applied uniformly in space and rescaled by month as set out below; they are not physically derived detection limits. The two diagnostics threshold different quantities, a standing stock (hourly presence summed over the day) and a flux (first arrivals per day), so their values are not meant to coincide. In item-equivalent units (n = 5,000 particles per observed item), they correspond to ~0.025 items continuously present and ~0.002 items arriving per day — units of survey-recorded items, an unquantified fraction of the true loading, so the thresholds rank relative accumulation rather than absolute abundance. The particle multiplicity n affects only the statistical counting precision of the discretisation at these thresholds (~9% for THR 1 and ~32% for THR 2 on a single day, with coastal frequencies aggregating at least 28 days per month). All quantities are kept in units of observed items and are not converted to mass, so that the weight-per-item conversion factors that add uncertainty to source-based inventories do not enter the diagnostics. Under the fixed threshold, the monthly mean surface frequency covaried with the monthly seeding (Spearman ρ = +0.60, member-bootstrap 95% CI +0.38 to +0.77; Supplementary Table S3), confirming that part of the apparent seasonality tracked the released population rather than transport; under the fractional definition adopted below this correlation falls to ρ = +0.44 (95% CI +0.03 to +0.62) at the surface and to ρ = +0.01 for the coastal diagnostic, so the dependence is attenuated rather than removed. Because the monthly seeding varies with the observations (17.2–32.5 million particles per seeding window), we apply both thresholds fractionally (Equation 2):where N m is the number of particles released in the seeding window of target month m (Table 1) and N ¯ is its annual mean. Exceeding the fractional threshold is algebraically identical to applying the fixed threshold to the per-capita field (c o u n t s × N ¯ / N m): normalising the threshold and normalising the field are the same operation. Setting THR = 0, by contrast, reduces the metric to presence/absence, which saturates by construction (monthly means of 0.81–0.89 over occupied cells) and loses dynamic range. Averaged over occupied cells, the summer-to-winter ratio rises from 1.01 at THR = 0 through 1.27 at THR = 1000 to a plateau of 1.56–1.59 at THR = 3000–10000, placing the adopted value in the stable regime of the sensitivity curve. The coastal frequency ranking (Corsica first, the Adriatic last) is stable across an order of magnitude of threshold choices. The late-summer surface maximum is likewise stable across the adopted range (3000–10000) and both normalisation variants, although under equal-mass reweighting August and December are nearly tied. The month of the minimum, however, is definition-dependent, falling between December and April across the variants; we therefore report a winter–early-spring minimum rather than a specific month (Supplementary Section S4).

The exceedance test is written as a Heaviside function applied to each grid cell (Equation 3):where i = 1 for the sea surface and i = 2 for the coastline, and THR i,m is the monthly fractional threshold of Equation 2.

The hourly counts were accumulated per member and summed across the 18 ensemble members before threshold evaluation, reducing the sensitivity of the monthly diagnostics to any individual release date. By definition, the monthly occurrence frequencies were computed as a fraction of the days during which this threshold was exceeded (Equation 4):where M is the number of days in a month. The frequencies ranged between 0 and 1, where 1 denotes exceedance of the threshold on every day of the month, and 0 indicates that the threshold was never exceeded.

Coastal deposition was evaluated using a beaching module embedded in the Lagrangian framework. Particles intersecting land grid cells were classified as beached and removed from further transport. Monthly binning of beached particles was done using the same spatial grid applied in the sea surface analysis.

This design allowed the construction of probabilistic spatial metrics rather than single deterministic trajectories. It also incorporated the releases of the five preceding months, so that each monthly diagnostic includes litter with ages of up to 150 days rather than only freshly released particles; the age composition of the composites is quantified in Supplementary Section S3 (Supplementary Table S2, Supplementary Figure S3).

FMML transport drivers in the study area

The Mediterranean Sea is influenced by climatological winds (Barbariol et al., 2021). In the study domain, the Mistral is a cold, dry wind from the north/northwest over the western and Tyrrhenian sub-basins, occasionally reaching the African coast. The Vendaval blows in from the Atlantic Ocean in the basin’s westernmost part. In the Adriatic Sea, a dominant wind is the Bora, a cold, dry northeasterly wind that triggers ventilation over the sea after interacting with local orography. This climatological basin-scale wind system is evident in the time-averaged map from August 2023 to December 2024 (Figure 2A).

In the Western Mediterranean, the main sea surface basin-scale current systems (Figure 2B) are represented by the Atlantic Water Current (AWC), Algerian Current (AC), Atlantic-Ionian Stream (AIS), Bifurcation Tyrrhenian Current (BTC), Strait of Sicily Tunisian Current (SSTC), South-Western Tyrrhenian Gyre (SWTG), Eastern Corsica Current (ECC), Northern Tyrrhenian Gyre (NTG), Northern Current (NC), Western Corsica Current (WCC), Western Mid-Mediterranean Current (WMMC), and Western Alboran Gyre (WAG). In the Adriatic Sea, the Western Adriatic Coastal Current (WAC), Eastern Adriatic Current (EAC), Middle Adriatic Gyre (MAG), and South Adriatic Gyre (SAG) are clearly visible (Pinardi et al., 2015). More mesoscale gyres, meanders, and filaments can be identified in the study domain on monthly averaged maps (Figures 3A–L). For example, the cyclonic South-Eastern Tyrrhenian Gyre was fully developed in December 2024 (Figure 3L).

Limitations of the method and outlook for further advancement

While considerable potential lies in incorporating new gridded observational datasets, the developed method has limitations that should be addressed in future studies.

Some limitations stem from the observational data used as initial conditions. Visual ferry-based surveys are conducted during daylight and under low sea states (Beaufort ≤ 2), within a 50-m strip and above a 20-cm size threshold. The Beaufort ≤ 2 restriction is intended to minimise sea-state-related detection bias and maintain consistent detectability of floating litter. Nevertheless, other observational constraints, including the minimum size threshold and the limited detectability of low-buoyancy or submerged items, may lead to underestimation of surface densities (Arcangeli et al., 2020; Compa et al., 2023; Jang et al., 2024).

Furthermore, because observations are restricted to Beaufort ≤ 2, the dataset preferentially samples the litter field under relatively calm sea states, when buoyant items are less affected by wind-induced mixing; the seeding, and hence the modelled fields, inherit this systematic conditioning on sea state. Overcoming these constraints will require complementary observing techniques. Thermal infrared imaging records emitted rather than reflected radiation and therefore operates at night as well as during the day (Goddijn-Murphy et al., 2022). The sea-state constraint is harder to relax from a ship, because visual detectability necessarily limits the range of conditions under which surveys can be conducted. Moreover, the occurrence of suitable survey conditions is not spatially uniform across the Mediterranean, particularly in regions characterised by stronger wind regimes, such as the Gulf of Lion and the waters west of Crete, where the Mistral and the Etesians dominate, respectively (Soukissian et al., 2018). Consequently, ferry-based observations preferentially represent the relatively calm and visually favourable portion of the sea-state distribution, with the observational window itself varying across the basin. Aerial, satellite and subsurface remote sensing cover different spatial scales and are sensitive to different size classes and components of the litter field, and their integration with in situ protocols offers a route to broader and more representative coverage (Kako et al., 2026).

A related caveat concerns the monthly re-seeding itself: because each month is seeded from the current observations while cohorts from previous months are still being tracked, a persistent item that was observed twice would be represented twice. The scope for such double counting is bounded by the probability of re-observation. With the 50-m survey strip, a monthly cohort survival of approximately 0.7 (Supplementary Section S3, Supplementary Table S2) and only a few per cent of a cohort’s surface presence remaining within the surveyed corridor cells one month after release (Supplementary Section S3), the expected fraction of re-sighted items in a monthly survey is of the order of 1% or below. This is consistent with the short surface residence reported for the basin, half-lives of 7–80 days depending on source (Liubartseva et al., 2018), and with GPS-tracked litter items in the northern Tyrrhenian travelling 16–24 km per day along their tracks, retained by submesoscale recirculation for only a few days at a time (Merlino et al., 2023). Surface retention in the north-western Mediterranean and the Tyrrhenian operates transiently and at sub-basin scale, hundreds of kilometres (Mansui et al., 2015; Zambianchi et al., 2017).

The monthly fractional thresholds (Equation 2) make the frequency diagnostics comparable across different released populations, but they rescale a single monthly total and do not reweight cells within a month: because seeding is proportional to observed item counts, a heavily surveyed corridor contributes more particles than a lightly surveyed one at the same litter density. The model is in addition initialised from observations collected along fixed ferry transects, so the seasonal patterns remain conditional on the survey coverage documented in Supplementary Section S1, and the indices should be read as relative, coverage-conditional quantities rather than as basin-wide estimates of FMML occurrence. This spatial effort weighting does not, however, order the seasonal structure on which the sub-basin results rest: the most heavily surveyed box (the Adriatic, median 135 surveyed cells per month; Supplementary Table S7) has the lowest summer-to-winter contrast of the six (0.78), while the highest (1.92, Balearic) occurs in the box ranked third by survey effort (Table 2). Considering the best coverage of the observation zone, the recently launched Automated Debris Imaging System (De Vries et al., 2026) appears very promising.

Most importantly, the model outputs are conditioned on offshore observational data collected along fixed ferry transects. Therefore, our beaching results should be interpreted as the final coastal destination of the observed offshore litter, rather than as a complete prediction of where FMML accumulates along the Mediterranean shores. Future studies should integrate land-based source monitoring (coastal and riverine inventories) with the offshore observations to close this methodological gap.

Simplified boundary conditions may bias the beaching diagnostic in ways that are not quantified here. Like any gridded model, the diagnostic depends on the landmask adopted, an unavoidable consequence of approximating a continuous coastline by discrete cells. To simulate FMML transport and fate more realistically, the Lagrangian particle tracking model should use state-of-the-art algorithms for particle beaching (e.g., Mørk et al., 2026) and sedimentation (e.g., Alvarez Barrantes et al., 2025).

When conducting long-term simulations, plastic fragmentation (e.g., Kaandorp et al., 2021), interactions with biota, and embedding into the biogeochemical cycles (e.g., Berezina et al., 2021) should also be accounted for.

Wind plays a crucial role in redistributing FMML, especially at the sea surface. Windage refers to the direct, additional drag exerted by surface winds on the parts of floating debris above water. It depends strongly on the characteristics and polymer types of litter. For example, fishing gear is generally recognised as uninfluenced by winds and regarded as low-windage, while expanded polystyrene foam and empty plastic bottles are high-windage. In this work, we used a constant windage coefficient of 0.01 for simplicity. Variable values are recommended depending on FMML characteristics (Ko et al., 2020). The ensemble spread reported here samples release dates and the stochastic closure, not the model parameters, and it should not be read as a full uncertainty estimate on the frequency fields. The sensitivity of both diagnostics to the windage coefficient, the least constrained of the transport parameters, is quantified in Supplementary Section S10 (July 2024): over the plausible range 0.01–0.03 the regional indices change by 2–24% on the coast and by 21–37% at the surface, with the three leading coastal margins unchanged in rank, while switching wind drift off altogether removes 28% of the month’s beaching.

The wave-induced Stokes drift is another of the key mechanisms that influence FMML transport in the sea surface layer (Breivik et al., 2016). Although the Stokes drift might reach notable intensity relative to sea surface currents in the Gulf of Lion and the Balearic Sea (Liubartseva et al., 2018), we did not account for this effect in this work to reduce computational cost. A single-month test including Stokes drift (July 2024, eighteen members) quantifies what the omission costs: the surface occurrence frequency falls in all six sub-basins (by 4.1–37.5%; −20.3% in the Balearic Sea) and, in a separate Gulf of Lion box, by 15.6%, while coastal deposition is redistributed rather than uniformly changed, the Valencia–Barcelona index being unchanged (+0.2%) (Supplementary Section S11). Because this sensitivity test covers July 2024 only, it does not test seasonal or annual patterns. Its inclusion is therefore a priority refinement of the model.

A further simplification concerns the temporal resolution of the forcing: daily-mean currents and winds filter diurnal (sea-land breeze) and near-inertial (~17–18 h at these latitudes) motions, which are therefore absent from the simulated trajectories even though the 30-minute integration step resolves them numerically. This mainly affects nearshore transport in summer, when sea-land breeze circulation is strongest. Adopting hourly forcing, already the native frequency of the atmospheric reanalysis, is a natural refinement for future work.

From a management perspective, macro-litter drifting near the coast and its stranding may be easier to clean up directly from land. Thus, prediction of FMML transport in coastal areas and beaching helps eliminate the pollutant efficiently and reduces the associated societal costs. For downscaling to coastal areas, advanced structured/unstructured grid hydrodynamics is necessary. Indeed, such numerical models can resolve space-time scales, including coastal currents, steric and tidal sea levels simultaneously, at high resolution (Trotta et al., 2021). Extending the integration window, for instance a 300-day sensitivity case for one or two representative target months, would directly test the cohort attrition quantified in Supplementary Section S3, and is a natural next step.

Beyond these methodological refinements, several avenues exist to broaden the spatial scope and operational value of the framework, building on the Particle Analysis Dashboard. First, the modelling could be extended to the full multi-year observational record and enriched with complementary coastal monitoring datasets, such as those collected by the Regional Agencies for Environmental Protection (ARPA) in Italian waters, improving spatial coverage and enabling the characterisation of interannual variability in FMML transport and deposition. The simulated beaching patterns could then be further validated against independent observational data from the MSFD beach-litter monitoring programme, available for the Italian coastline (Arcangeli et al., 2025), providing a direct observational benchmark for the coastal-deposition diagnostics.

The offshore outputs can also support the assessment of exposure risk for vulnerable megafauna, such as cetaceans and sea turtles. So far, such assessments have been based on the co-occurrence of biota and litter recorded along the same survey tracklines, which limits them to the surveyed areas. Combining the modelled distributions with species distribution models (SDMs) would extend this assessment beyond the transects to the whole basin, including areas where direct observations are not available. Integrating these litter and biodiversity layers into a single decision-support tool, together with the Marine Protected and Conserved Area boundaries already included in the Particle Analysis Dashboard, would further support ecosystem-based management and conservation planning.

Observations

During the Conceptu Maris surveys from August 2023 to December 2024, 158 survey transects were conducted. In total, 4,801 FMML items were detected within a ~6,200 km 2 area of the surveyed sea surface (Figure 4). The effectively surveyed area therefore represents approximately 0.25% of the Mediterranean sea surface (~2.5 million km²). The detected FMML items were largely (>62%) composed of artificial polymer materials (plastics). The observed distribution is strongly zero-inflated: 76% of all surveyed cell-months recorded no FMML items.

Spatiotemporal distributions of FMML abundance (Figure 3A–L) indicated relatively high values (25–50 items km −2), particularly in the Ligurian–Provençal sub-basin, which includes the Specially Protected Area of Mediterranean Importance (SPAMI) ‘Pelagos Sanctuary’. Its highly productive waters are rich in plankton, supporting a remarkable abundance of megafauna: whales, dolphins, sea turtles, large fish, and elasmobranchs. The Pelagos Sanctuary was established to safeguard marine mammals, their habitats and migratory pathways (Fossi et al., 2013). The maximum FMML abundances of 50–70 items km −2 were detected in that sub-basin in April 2024 (Figure 3D). In the Adriatic Sea, abundances were relatively stable, ranging from moderate (7–13 items km −2) to high (25–50 items km −2), with a distinct peak in the centre of the sub-basin in February 2024 (Figure 3B). Slightly lower but persistent abundances, ranging from 4–7 to 25–50 items km −2, were typical of the Southern Tyrrhenian and the Strait of Sicily.

Attribution of the observed abundance patterns to individual coastal sources requires solving the corresponding inverse problem (Liubartseva et al., 2019; Kaandorp et al., 2020), for which nighttime-lights-based source inventories (Liubartseva et al., 2026) and the FMML observations presented here provide complementary constraints.

Modelling

Frequency maps

The frequency-based diagnostics developed in this study provide a spatially explicit measure of the persistence and recurrence of FMML transport and coastal deposition across the Mediterranean basin. Using the data-driven stochastic ensemble approach, the method identifies dynamic hotspots that are likely to represent ecologically meaningful exposure zones.

With a spatial resolution of 0.05° × 0.05° (~5 km in the Mediterranean), monthly FMML occurrence frequencies at the sea surface and on the coastlines are presented in Figures 5, 6.

The monthly frequency fields carry the combined imprint of the observation-driven seeding and of wind- and current-driven transport; their out-of-sample skill, and the degree to which the apparent alignment with the survey transects is built in by the seeding, are quantified in the Results (Out-of-sample skill of the monthly composites). The monthly representations of these transport mechanisms in Figures 3A–L revealed more mesoscale patterns than in Figure 2, as expected from the shorter averaging period. Most of the observed FMML items drift in the Western and Central Mediterranean and Adriatic seas, with seasonal summer leakage into the Ionian Sea. Further east the ensemble reaches the basin only marginally: east of 22°E the modelled surface occurrence frequency never exceeds 0.19, is zero in seven of the twelve months, and the per-country annual-mean coastal deposition indices there are 0.03 or below (monthly peaks 0.07 or below) (Supplementary Table S6). The present maps therefore do not support any claim about Eastern Mediterranean transport, and none is made here. Relatively steady-state patterns of high FMML occurrence frequencies were found in the Southern Tyrrhenian and Northern Ligurian seas, where the monthly sub-basin indices remain at 0.39–0.77 and 0.30–0.79 respectively throughout 2024 (Figures 7A, C; Table 2). In the Pelagos Sanctuary, a near-constant high frequency of FMML occurrences was detected, with a winter–early-spring minimum. Each monthly frequency field integrates the releases of the target month and of the five preceding months (Table 1); the winter–early-spring minimum therefore reflects the low observed abundances in this sub-basin in late autumn and winter — a feature of the observations that the basin-wide fractional normalisation preserves by design — rather than conditions in any single month. Elevated FMML occurrence was found in the Balearic Sea from late spring through early autumn (May–September; Figures 5I, K and 6A, C, E), when the sub-basin index rose to a September maximum of 0.56 (a summer-to-winter JJA/DJF contrast of 1.92; Figure 7B), before falling through autumn to 0.36 in November and, after a December rebound to 0.41, to a winter minimum of 0.16 (Figures 6G, I, K). In the South-Western Mediterranean, the index rose from 0.21 in July to an August maximum of 0.37 (Figures 6A, C) before returning to July levels by October, with the annual maximum reached in December (0.45; Figure 7F; Table 2). In the Adriatic Sea, the intense sub-basin-scale FMML spreading was evident from January to June and again in December 2024 (Figures 5A, C, E, G, I, K, 6K). During July and August 2024, modelled surface occurrence extended into the Northern Ionian Sea, consistent with the summer leakage driven by the strengthening of currents exiting the Adriatic (Figures 6A, C).

Regions of persistently low modelled occurrence were equally identifiable. The western Alboran Sea was the lowest in the basin throughout 2024, with mean frequencies over occupied cells of 0.05–0.20 and no cell above the monthly threshold at all in July and August. The western Tyrrhenian was at its lowest from January to May (0.11–0.31, against 0.37–0.78 from June to December) and the northern Adriatic from July to December (0.12–0.40, against 0.44–0.55 from March to May), while in the Gulf of Lion the ensemble reached only 13–30% of its annual footprint between January and April, against 49–99% over the rest of the year. These are areas the seeded litter did not reach in quantity within the 150-day integration window, and their interpretation, like that of the maxima, is conditional on survey coverage (Supplementary Section S1).

Beaching showed distinct patchiness and seasonality, broadly mirroring the sea-surface patterns described above. In annual mean, coastal deposition frequency was both highest and steadiest on Corsica; the western coast of Sardinia and the northern Tunisian coastline follow, separated by less than 0.005 and reported as a statistical tie, ahead of the south-eastern Tyrrhenian and Ligurian–Tuscan margins. Corsica ranks first and the Adriatic last at every threshold tested (Supplementary Table S4). Deposition along the Corsican and Ligurian–Tuscan margins is nevertheless sustained over much of the year (Figures 5J, L, 6B, D, F, H, J), and the Ligurian–Provençal box index peaks at 0.41 in August (Figure 7A; Table 2), with further persistent deposition along the Sicilian–Calabrian coast around the Strait of Messina (Figures 6B, D, F, H, J), while the Adriatic accumulated the largest total deposition along its considerably longer coastline (Figures 5D, F, H, J, L, 6B, J, L). The northern Tunisian coastline showed the most frequent North African deposition, peaking in January (Figure 5B), July–September (Figures 6B, D, F), and December 2024 (Figure 6L). The Valencia–Barcelona coastline displayed high occurrence frequencies confined mainly to June–August 2024 (Figures 5L, 6B, D). For most of the year, FMML beaching occurred on the south-eastern shores of the Tyrrhenian Sea, with elevated deposition in July, August, November, and December 2024 (Figures 6B, D, J, L). The western coast of Sardinia formed a further persistent depositional margin, with high beaching frequencies maintained for much of the year (Figures 5B, D, F, 6F, L), consistent with Mistral-driven onshore transport along the island’s western façade. In the Adriatic Sea, deposition along the Italian coastline was dominant compared with Slovenia, Croatia, Montenegro, and Albania.

Figures 7A–F; Table 2 summarise these patterns quantitatively as monthly sub-basin occurrence-frequency indices: the mean occurrence frequency over occupied cells within each of the six sub-basin boxes used for the out-of-sample validation (Supplementary Section S6, Supplementary Figure S7), computed under the monthly fractional thresholds of Equation 2, with the spatial interquartile range across cells as the measure of spread. The indices make the scale dependence of the seasonality explicit, pronounced at sub-basin scale but weak basin-wide: the summer-to-winter (JJA/DJF) contrast of the surface index reaches 1.80 in the Ligurian–Provençal box and 1.92 in the Balearic box, against 0.78 in the Adriatic, whose index peaks in May in phase opposition to the July maximum of the Northern Ionian. The coastal-deposition index peaks in August (0.41) in the Ligurian–Provençal box and in April (0.28) in the Adriatic. Full monthly values are given in Supplementary Section S7 (Supplementary Table S5), and per-country relative deposition indices in Supplementary Table S6.

Comparing sea surface occurrence frequencies with beaching occurrence frequencies in the adjacent areas suggests that coastal deposition is more closely associated with the near-shore than with the offshore surface field. The Adriatic provides an illustrative comparison. In January 2024 (Figures 5A, B) the surface occurrence frequency averages 0.63 along the basin axis but falls to 0.37 within 15 km of the Italian coast, where coastal deposition averages 0.23 over 218 depositing cells. By June 2024 (Figures 5K, L) the near-shore field has risen to 0.42 and deposition to 0.30 over 390 cells, while the axis changes little (0.63 to 0.67): this example indicates that the change in deposition is consistent with the near-shore field rather than with the offshore maximum.

Beyond this internal consistency, the sub-basin indices of Figure 7 place the comparison with the seasonal structure reported by Soto-Navarro et al. (2020) on an explicit footing: the Adriatic index peaks in May, matching their spring maximum, with its minimum in September, adjacent to the August minimum they report; the Northern Ionian index peaks in July, reproducing the opposite-phase relationship those authors obtained; and the Balearic index is highest in late summer (0.56 in September), consistent with their August maximum. As in their study, basin-scale seasonality is weak whereas sub-basin seasonality is pronounced (Table 2). The concentration of persistent warm-season occurrence in the Ligurian–Provençal basin, the Southern Tyrrhenian and around the Balearic Islands is consistent with the summer hotspot formation described by Macias et al. (2019).

The modelled coastal-deposition patterns are consistent with previous Lagrangian particle tracking in the Mediterranean in a broad qualitative sense. Importantly, our results reflect the distinct nature of the observation-based initial conditions from August 2023 to December 2024. Thus, only qualitative comparison with the previous modelling results is possible. Moreover, we found both a broad qualitative agreement with them and some clear inconsistencies in the modelled patterns. The dominant beaching coastlines in our simulations — the Adriatic, the Ligurian–Tuscan and Corsican margins, the Sicilian–Calabrian coast around the Strait of Messina, and the western façades of Sardinia — coincide with the eastern-basin and westward-facing island coasts identified as preferentially polluted by Macias et al. (2019), and the elevated Adriatic and North-African deposition mirrors the coastal fluxes reported by Liubartseva et al. (2018). The principal divergence is informative rather than contradictory: source-based models that prescribe coastal-urban and riverine inputs rank the Catalan–Barcelona coast and the Po and Rhône mouths among the most heavily beached segments (Liubartseva et al., 2018; Garcia-Gorriz et al., 2026), because in those frameworks each coastline is dominated by its own adjacent terrestrial sources. Recent local-scale research highlighted the complexity of plastic fate modelling in these near-source areas: the Barcelona coast (Hernandez et al., 2026), the Po River Delta (Atwood et al., 2019), and the Rhône River Mouth (Weiss et al., 2024). Seeded instead from litter already observed offshore, our simulations under-represent these source-dominated coasts and preferentially resolve shorelines that intercept litter transported from the open sea, so the two approaches are complementary descriptions of where Mediterranean litter strands. Basin-wide deposition peaks in summer and is lowest in winter, consistent with the seasonal accumulation described by Macias et al. (2019). Finally, an internal out-of-sample assessment of the composites against the ferry observations themselves is reported in the Results (Out-of-sample skill of the monthly composites); the planned complementary validation against the MSFD beach-litter dataset follows the approach of Garcia-Gorriz et al. (2026).

Out-of-sample skill of the monthly composites

Because each monthly composite integrates releases initialised from the observations of its own target month (together with those of the five preceding months), part of the agreement between the composites and the FMML observations is imposed by construction. To quantify how much, we performed an out-of-sample hindcast experiment: for every target month T of 2024 we rebuilt the surface composite using only the three ensemble members released in month T −1 — a one-month-ahead prediction that has never seen the observations of month T — and compared it against the ferry-based densities of month T over all surveyed 5-km cells, zeros included (257–1,183 cells per month). Two references bracket the result: the full in-sample ensemble (18 members, including the releases of month T itself) and observational persistence (the densities observed in month T −1 at the same cells). Agreement is measured with Spearman rank correlations, since the two fields carry different units (particle-hours versus items per km²); for the same reason an absolute bias is not defined for these dimensionless exceedance diagnostics.

At the native 5-km scale the out-of-sample correlation is low (median ρ = +0.06 across the twelve target months) and, once significance is assessed with a transect-cluster bootstrap (surveys are strip transects along 8–31 ferry routes per month, so neighbouring cells are not independent), it is significantly positive in only two months (June and August). The in-sample ensemble reaches a median of ρ = +0.37 over the same cells — six times the out-of-sample value. The age-stratified composites indicate the mechanism: same-month releases contribute only ~19% of the composite mass at basin scale but are several-fold (typically 4–6 times) enriched along the surveyed corridors where they are seeded (age-stratified statistics in Supplementary Section S3), anchoring the in-sample fields to the transects. An in-sample control built from the three same-month members alone reaches a median ρ = +0.50 — higher than the full 18-member ensemble — confirming that the gap reflects this anchoring rather than the smaller ensemble available one month ahead. Observational persistence itself reaches only ρ = +0.12: at 5 km the observed field barely correlates with itself one month later — an empirical benchmark suggesting a low intrinsic predictability limit at this scale, set by litter patchiness and the 50-m strip-sampling geometry.

Predictive information emerges under spatial aggregation (Table 3). Median out-of-sample correlations rise from +0.06 at 0.05° to +0.12–0.13 at 0.5–1.0°, while persistence rises to +0.23–0.27 across 0.25–1.0°, at grid scales neither the composites nor persistence reconstructs the observed field. At the sub-basin scale (six boxes defined in Supplementary Section S6, jointly covering two thirds of the surveyed cell-months) the picture changes, and the comparison depends on the treatment in an informative way. Pooling the 68 basin-months, the composites reach ρ = +0.32 (month-cluster bootstrap 95% CI +0.05 to +0.55) and persistence ρ = +0.35; pooled correlations, however, are inflated for persistence by the seasonal cycle shared across basins. The within-month treatment removes this common cycle and asks the operational question, which basins carry more litter this month: there the composites rank the basins with a median across-basin correlation of +0.46 (95% CI +0.18 to +0.60; permutation p = 0.007) against +0.28 for persistence, ranking ahead of persistence in 5 of the 12 months. Neither predictor therefore dominates consistently; the practical difference is coverage: persistence exists only where a basin happens to have been surveyed the month before, whereas the composites cover every basin without any prior survey. These results are insensitive to the exact box definition (Supplementary Section S6).

Per-basin temporal correlations are heterogeneous (Supplementary Section S8, Supplementary Table S7): positive in the Adriatic, Ligurian–Provençal and Balearic, near zero in the southern Tyrrhenian (+0.06), and negative in the northern Ionian (−0.46) and south-western Mediterranean (−0.21), though outside the Adriatic none is individually distinguishable from zero at n ≤ 12 (permutation p ≥ 0.15). The Adriatic — the basin with the densest and most regular monthly sampling (41–173 surveyed cells in each of 11 months, median 135) — is the only one where the temporal test is informative: there the out-of-sample composite tracks the observed seasonal cycle with ρ = +0.86 (permutation p = 0.001) and exceeds observational persistence (+0.65) in point value, although the difference is not itself testable at n = 11 and month-on-month increments remain largely uncaptured (first-difference ρ = +0.35).

These numbers define how the monthly composites should be read. At the sub-basin and monthly scale at which we interpret the product they carry genuine out-of-sample information; at grid scale they are observation-anchored descriptions of where the observed litter was transported — nowcasts rather than stand-alone forecasts — and the grid-scale maps should accordingly not be read cell by cell for site-level prediction.

Particle analysis dashboard: a web-based application for end users

In contemporary oceanography, web-based applications help scientists, industries, and decision-makers navigate complex datasets and analyse various ‘what-if’ scenarios to safeguard marine environments and mitigate associated hazards.

A dedicated web-based application was developed to visualise and examine the sea surface and beached FMML (see Data Availability Statement). The Graphical User Interface was designed to be user-friendly, making it easy to read and to interact with. The Particle Analysis Dashboard provides users with access to information that represents the fusion of observations and modelling.

Both sea surface and beached FMML occurrence frequencies can be analysed. Figure 8 represents an example for beached FMML occurrence frequencies in the Balearic Sea in April 2024. Users can choose an appropriate threshold to focus on the part of the histogram they are interested in. As an example, we chose five particles per cell (Figure 8). Note that the dashboard applies user-chosen fixed thresholds to the raw daily fields for interactive exploration; the products of Section 3 use the monthly fractional thresholds THR i,m, so a fixed dashboard value x reproduces the paper’s definition for month m when x = THR i · N m / N ¯. The dashboard also offers this fractional mode natively, alongside the fixed-threshold exploration. The application shows how our choice influences the histogram and corresponding statistics. This allowed us to remove from the interactive display the uninformative part of PDF (Probability Density Function). Finally, a map showing the frequency of FMML landings indicates that the most likely coastal areas for such occurrences were on Formentera Island, including the Natural Park of Ses Salines d’Eivissa i Formentera as a protected area of historical salt pans and an important stopover site for migrating flamingos. The north-eastern coasts of Mallorca and Minorca Islands also exhibited substantial FMML occurrence frequency values. The zoom option facilitates examination of coastline patterns at a native horizontal resolution of ~5 km.

An example of the sea surface FMML occurrence frequencies and more details on the Particle Analysis Dashboard can be found in Supplementary Section S9 (Supplementary Figure S8).

Ecological applications and conservation implications

Monthly occurrence frequency maps provide a spatially explicit diagnostic of recurrent FMML occurrence in both offshore and coastal environments. When integrated with spatial distributions of Mediterranean megafauna, including cetaceans and sea turtles, these outputs enable quantitative assessments of exposure and risk. By identifying areas where biota and marine litter might co-occur spatially and temporally, this interdisciplinary application supports ecosystem-based management. A thorough examination of this kind was conducted at a high spatial resolution within the Pelagos Sanctuary in the summer of 2019 (Fossi et al., 2025).

Integrating the modelled FMML coastal deposition patterns with loggerhead turtle nesting data from the Western Mediterranean (Hochscheid et al., 2022) indicates differential modelled deposition frequencies at the three key nesting hotspots identified by the Getis–Ord Gi* hotspot analysis of that study — the Nefza area (northwest Tunisia), Cilento (Campania, Italy) and Capo Carbonara (southeast Sardinia) — during the nesting season (June–September). Within each hotspot, the modelled coastal-deposition diagnostic was evaluated at the model grid cell containing a representative site: Chott Zouaraa beach (cell centre 36.98°N, 8.88°E), Le Saline beach (40.03°N, 15.28°E) and Villasimius (39.13°N, 9.53°E), whose beaches, Porto Giunco and Simius, fall within that same cell. The diagnostic was summarised as the June–September mean of the monthly maximum coastal-deposition frequency within a 3 × 3 grid-cell window (0.15° × 0.15°, approximately 13 km by 17 km at these latitudes) centred on that cell. Mean June–September deposition frequencies reach 0.6 at Chott Zouaraa, peaking at 0.8 in July, against approximately 0.4–0.5 through the season at Villasimius and 0.2 at Le Saline, which reaches a September maximum of 0.5–0.6. These values are relative, threshold-conditioned diagnostics of modelled deposition and no mechanistic link between beached-litter frequency and nesting outcome is implied. The comparison demonstrates how the modelled products can flag spatial and temporal co-occurrence between litter deposition and critical nesting habitat, helping to prioritise monitoring and mitigation effort at these sites.

This study presented a data-driven framework for assessing the spatiotemporal distribution of floating marine macro-litter in the Mediterranean Sea, coupling systematic shipboard observations along fixed ferry transects with ensemble Lagrangian particle tracking. By using monthly gridded FMML observations from 158 surveys and 4,801 detections as direct initial conditions for stochastic ensemble simulations, this approach complements existing source-based Lagrangian modelling strategies and provides frequency-based diagnostics of FMML persistence and coastal deposition.

The frequency-of-occurrence diagnostics revealed distinct spatiotemporal variability in FMML occurrence, jointly controlled by the seasonality of the observed loading that initialises each release and by transport under sea surface currents and winds. The resulting monthly maps provide a reproducible diagnostic framework that can be updated as new observational data accumulate. An out-of-sample hindcast showed that they carry predictive information at the sub-basin scale but not cell by cell, which defines the scale at which they should be read. The Particle Analysis Dashboard translates these outputs into an accessible tool for scientists and decision-makers, integrating Marine Protected and Conserved Area boundaries and enabling direct spatial assessment of the overlap between FMML accumulation hotspots and designated conservation areas. Several methodological limitations nevertheless constrain the present framework and point to productive directions for future work. Coupling the modelled FMML distributions with species distribution models would extend exposure assessments from the survey transects to basin-wide scales.

Overall, this framework advances Mediterranean marine litter modelling by anchoring initial conditions in direct observations rather than estimated source inputs alone, and by delivering probabilistic diagnostics and an interactive decision-support tool within a single reproducible workflow. The methodology is transferable to other sea basins and monitoring programmes, and can contribute to MSFD Descriptor 10 monitoring of floating-litter abundance, composition, and spatial distribution.