Marine Pollution Bulletin 103 (2016) 115–127

Contents lists available at ScienceDirect

Marine Pollution Bulletin

journal homepage: www.elsevier.com/locate/marpolbul

Regional approach to modeling the transport of floating plastic debris in

the Adriatic Sea
S. Liubartseva a,⁎, G. Coppini b, R. Lecci b, S. Creti b
a
Centro EuroMediterraneo sui Cambiamenti Climatici, Bologna, Italy
b
Centro EuroMediterraneo sui Cambiamenti Climatici, Lecce, Italy

a r t i c l e i n f o a b s t r a c t

Article history: Sea surface concentrations of plastics and their fluxes onto coastlines are simulated over 2009–2015. Calculations
Received 12 November 2015 incorporate combinations of terrestrial and maritime litter inputs, the Lagrangian model MEDSLIK-II forced by
Received in revised form 18 December 2015 AFS ocean current simulations, and ECMWF wind analyses. With a relatively short particle half-life of
Accepted 21 December 2015
43.7 days, the Adriatic Sea is defined as a highly dissipative basin where the shoreline is, by construction, the
Available online 8 January 2016
main sink of floating debris. Our model results show that the coastline of the Po Delta receives a plastic flux of
Keywords:
approximately 70 kg(km day)-1. The most polluted sea surface area (N10 g km-2 floating debris) is represented
Plastic debris inputs by an elongated band shifted to the Italian coastline and narrowed from northwest to southeast. Evident season-
Lagrangian model ality is found in the calculated plastic concentration fields and the coastline fluxes. Complex source–receptor re-
Markov chain lationships among the basin's subregions are quantified in impact matrices.
Plastic fluxes onto coastline © 2015 Elsevier Ltd. All rights reserved.
Impact matrices

1. Introduction Remarkable progress toward flexibility in plastic litter modeling was

achieved by Maximenko et al. (2012), who suggested a Markov chain
Plastic pollution in the marine environment is of increasing concern model that represents transporting properties of the upper mixed
due to the great threat to human health and the stability of marine eco- layer, allowing the separation of the input distributions from the dy-
systems, and adverse economic impacts on coastal communities namics of the upper mixed layer. Once the Markov chain model was
(Eriksen et al., 2014; Thompson et al., 2009). built, the evolution of particle concentration from any source could be
The spatial and temporal distributions of plastics in the marine envi- calculated, and various hypotheses of input distributions could be tested
ronment depend on input locations and the time-varying intensity of efficiently. A global set of historical trajectories of drifting buoys de-
sources, which are highly uncertain. However, ocean currents, waves, ployed in the Surface Velocity Program and Global Drifter Program
and wind control the transport of plastics, redistributing them at sea (1979–2007) was used for the calculation of a transition matrix. The
until they eventually wash ashore or sink. The high complexity and methodology was improved by van Sebille et al. (2012), who extended
multiscale versatility of the dynamics of the upper mixed layer of the the buoy dataset, introduced the seasonal transition matrices, and im-
ocean, where the majority of plastics float, must be taken into account. posed marine litter inputs related to the coastal population density.
In order to identify the pathways of floating marine litter under such Using directly integrated particle trajectories, Lebreton et al. (2012)
uncertain conditions, several numerical simulations were performed for took into account terrestrial sources from rivers and cities, and marine
different geographical areas varying from global to regional scales. Con- inputs from major shipping lanes. Long-term drift of floating debris in
tributions of geostrophic currents, Ekman drift, Stokes drift, and their the world's oceans was simulated assuming an increase in input
combinations in the North Pacific were simulated by Kubota (1994). De- intensity.
veloping Kubota's approach, Martinez et al. (2009) demonstrated the Recently, marine plastic modeling was carried out on a regional scale
appropriateness of using a high-resolution current field to determine to explain local source–receptor relationships in the southern North Sea
the impact of mesoscale activity on the trajectories of particles. Yoon (Neumann et al., 2014). In the Mediterranean Sea, marine litter drift
et al. (2010) enriched the methodology, switching from homogeneous was simulated in an effort to find permanent accumulation structures
source distribution to more realistic inputs into the Japan Sea from the such as so-called garbage patches (Mansui et al., 2015). No permanent
largest cities and rivers, applying output from the Japan Sea Forecasting sea surface structures able to retain floating items in the long-term
System. were found. However, some relevant coastal features were obtained at
a basin scale. For example, the coastline between Tunisia and Syria
⁎ Corresponding author. was found to be the most littered with plastics, while the western Med-
E-mail address: svitlana.liubartseva@cmcc.it (S. Liubartseva). iterranean demonstrated rather low coastal impact.

http://dx.doi.org/10.1016/j.marpolbul.2015.12.031
0025-326X/© 2015 Elsevier Ltd. All rights reserved.
116 S. Liubartseva et al. / Marine Pollution Bulletin 103 (2016) 115–127

Assuming that marine litter particles can be considered passive La- simulate the plastic concentrations at the sea surface and fluxes onto
grangian tracers, it is important to mention Pizzigalli et al. (2007), the coastline that originated from terrestrial and maritime inputs;
who, for the first time, built a Markov chain model for passive tracers (2) identify source–receptor relationships among the subregions of
in the Mediterranean Sea using the Lagrangian model coupled with the Adriatic Basin solving both the direct and inverse problems; and
the Mediterranean Forecasting System (Pinardi et al., 2003). They fo- (3) present the results in terms of impact matrices.
cused on seasonality in calculated statistics and introduced coastal- The manuscript is organized as follows: in Section 2 the data on
approach-maps to find coastlines that are at risk of pollution originating sources of floating debris in the Adriatic Sea, the Lagrangian model,
from the sea. and the ocean forecasting system are presented; Section 3 contains de-
To a certain extent, transport of plastic marine litter is similar to the scriptions of the Markov chain model; and Section 4 presents results
transport of satellite-tracked Lagrangian drifters, which have been in- and discussion. Finally in Section 5, we draw conclusions.
tensively deployed in the Adriatic Sea. The results obtained in the drifter
experiments conducted by Falco et al. (2000), Poulain (2001) Lacorata 2. Data and models
et al. (2001), Ursella et al. (2006), Veneziani et al. (2007), Poulain and
Hariri (2013) were invaluable for verification of model results on distri- 2.1. Identification of floating debris inputs into the Adriatic Sea
bution of floating debris in the Adriatic Sea.
Focusing on a key role of uncertainty in the plastic debris inputs, According to recent estimations by Jambeck et al. (2015), the total
Isobe et al. (2009) tried to reconstruct sources of plastic debris solving annual input of plastic in the Adriatic Sea was 10,000–250,000 tons in
an inverse problem (backtracking). The main complication of this prob- 2010. In an effort to be more consistent with the previous estimates of
lem arose from the irreversibility of diffusion computed using the the mass of floating plastic debris cited in Jambeck et al. (2015) we
random-walk technique (Csanady, 1973). Recently, when micro- use a lower limit of 10,000 ton year -1 in the present work. Following
plastics became widely recognized as an acute problem, 3D modeling Lebreton et al. (2012), we assume that 40% of the marine litter enters
was developed for meso- and micro-plastics (Isobe et al., 2014). Some the basin through rivers; 40% through coastal urban populations; and
relevant parameters in floating debris modeling, which are comparable the remaining 20% is derived from shipping lanes. They yield 4000,
with those we use in our calculations, are extracted from the literature 4000, and 2000 ton year -1, respectively.
cited and summarized in Table 1. Average annual discharges into the Adriatic Basin from the 62 largest
In the present work, for the first time we (1) develop the Markov rivers are extracted from a database implemented by Verri et al. (2014).
chain model based on coupling the Lagrangian MEDSLIK-II model (De The discharge of the Po River is distributed among its nine greatest dis-
Dominicis et al., 2013a,b) with the Adriatic Forecasting System (AFS) tributaries in proportion to their respective runoffs. Eight rivers entering
ocean currents simulations and ECMWF surface wind analyses to the north Ionian Sea are included in order to take into consideration

Table 1
Relevant parameters in floating plastic debris modeling.

Domain, Meteo-oceanographic data Distribution of inputs Interaction with Number of Time of

reference coastline Lagrangian particles integration
Spatial Temporal

North Pacific COADS climatology (1970–1979) 10°×10° homogeneous grid Instantaneous No-slip 50 5 years
(Kubota, release conditions
1994)
South Pacific OSCAR based currents (1993–2001) 1°×1° homogeneous grid Instantaneous No-slip Large number, not 8 years
(Martinez release conditions specified (1993–2001)
et al., 2009)
East China Sea Quick scatterometer wind and POM based Two point sources Instantaneous Beaching after 20000 76 days
(Isobe et al., currents release 12-h stagnation
2009)
Japan Sea (Yoon Currents based on Japan Sea Forecasting Sys- Inputs from largest basin'sEvery month Beaching 47676 particles a 4 years
et al., 2010) tem (2003–2006) rivers, cities and, the release year (2003–2006)
Tsushima Strait (2003–2006)
Mediterranean Lagrangian model dataset based on currents 1/8°×1/8° homogeneous grid Every week No-slip In the Markov chain 28 days
Sea (Pizzigalli provided by Mediterranean Forecasting System release conditions model: 400000
et al., 2007) (2000–2004) (2000–2004) particles a week
Global ocean Lagrangian dataset of satellite-tracked drifting 1/2°×1/2° homogeneous grid Instantaneous Beaching after In the Markov chain 10 years
(Maximenko buoys (1979–2007) release 5-day stagnation model: large number,
et al., 2012) not specified
Global ocean Lagrangian dataset of drogued and Inputs from coastal urban 6 pulse releases No-slip In the Markov chain 1000 years
(van Sebille non-drogued drifting buoys (198x–20xx) population a year conditions model: large number,
et al., 2012) not specified
Global ocean 6-year dataset provided by US Navy's Global Inputs from impervious Releases evenly Beaching Over 9.6 × 106 30 years
(Lebreton Atmospheric Prediction System and surface area, coastal distributed over
et al., 2012) HYCOM/NCODA ocean circulation model population and shipping each year
lanes
Southern North BSHcmod: operational weather and ocean Two local clusters with One release in No-slip 200 90 days
Sea (Neumann circulation models homogeneous distributions 28 h conditions
et al., 2014) inside (2000–2008)
Mediterranean NEMO based currents 10×10 km homogeneous Everyday Beaching, 3287 Two runs:
Sea (Mansui grid release stagnation, one year and
et al., 2015) (2001–2009) recirculation off 3 months
shore
Adriatic Sea Lagrangian model dataset based on ECMWF Inputs from largest Adriatic One release in Beaching after In the Markov chain 6 years
(present wind and AFS currents (2009–2015) rivers, cities and shipping 10 days 10-day model: over 6 × 1010 (2009–2015)
work) lanes (2009–2015) stagnation
 S. Liubartseva et al. / Marine Pollution Bulletin 103 (2016) 115–127 117

Fig. 1. Spatial distribution of the floating debris inputs into the Adriatic Basin: shipping lanes (gray scatter plot); rivers (open blue diamonds), the largest rivers (closed blue circles); cities
(open red circles); and the largest cities (closed red circles).

their contributions to plastics entering the Adriatic Basin through the averaged traffic density map provided by Automatic Identification Sys-
Otranto Strait. Total river-born plastic input of 4000 ton year -1 is dis- tem (AIS, 2015). Each lane is described by six parameters: the geograph-
tributed to be proportional to the average annual rivers' runoffs. We ical coordinates of the two end-points (longitudes and latitudes), the
apply annual averaging in order to be consistent with the annually aver- average traffic intensity, and the width. Applying spatial randomization
aged data on urban population and shipping lanes described below. to these parameters, the possible locations of the marine inputs are gen-
Urban population data are taken from a database implemented by erated (Liubartseva et al., 2015). The combined distribution of the in-
Brinkhoff (2010). The 46 largest cities located inside a 10 km coastal puts described above is presented in Fig. 1. The intensities of the top
belt, which contained more than 20,000 inhabitants in 2010, are consid- 10 marine litter inputs into the Adriatic Sea are listed in Table 2. As
ered. Three cities situated on the north Ionian Sea coast are included due shown in Table 2, 20% of the total input comes from the shipping
to their possible contribution to the floating debris flux through the lanes, followed by the Po River (13.5%). Bari is a greatest contributor
Otranto Strait. Total coastal urban population input of 4000 ton year -1 (3.5%) among cities. In general, seven main sources are responsible
is distributed to be proportional to the number of each city's inhabitants. for more than 50% of total marine litter input, namely shipping
To distribute the total plastic input of 2000 ton year -1 along shipping lanes (20%), the Po River (13.5%), the Buna/Bojana River (5.8%),
lanes, we extract 60 of the most congested lanes from an annually Bari (3.5%), Venice (2.9%), the Neretva River (2.8%), and the city of
Trieste (2.3%).
In should be noted that we keep all the inputs to be constant from
Table 2 year to year assuming that they do not change substantially over
Top 10 marine litter inputs into the Adriatic Basin. 2009–2015. Due to a lack of information, we also do not apply any sea-
Input Intensity (ton year -1) % of total inputs sonality to the inputs.
1 Shipping lanes 2000 20.0
2 Po River 1349 13.5
3 Buna/Bojana River 575 5.8 2.2. Lagrangian model: MEDSLIK-II
4 Bari 351 3.5
5 Venice 291 2.9 Similar to the majority of Lagrangian models (Fernandes et al., 2013;
6 Neretva River 283 2.8
7 Trieste 229 2.3
Zodiatis et al., 2012), the oil spill model code MEDSLIK-II (De Dominicis
8 Split 198 2.0 et al., 2013a,b) is applicable to the case of surface passive tracers that
9 Adige River 191 1.9 could simulate the behavior of floating plastic debris at a first approxi-
10 Ravenna 183 1.8 mation. In MEDSLIK-II (http://medslikii.bo.ingv.it), each Lagrangian
Other rivers (58) and cities (41) 4350 43.5
particle moves due to currents, winds, and waves, and follows a
118 S. Liubartseva et al. / Marine Pollution Bulletin 103 (2016) 115–127

trajectory calculated using advective and diffusive displacements: Markov chain model to overcome the limitation on the integration time
of 10 days imposed by the Lagrangian model MEDSLIK-II.
0
dxðt Þ ¼ U ðx; y; t Þdt þ dx ðt Þ
0
dyðt Þ ¼ V ðx; y; t Þdt þ dy ðt Þ; ð1Þ
3.1. Calculation of the transition matrices

In order to compose a model dataset, the suites of 10,000 virtual

where dx ' (t), dy ' (t) are the random walk displacements due to
floating particles are released instantaneously from each cell of the
horizontal turbulent diffusion. Two components of the flow velocity,
AFS grid (number of grid cells is 27,850 including 3622 coastal cells),
U=(U,V) are provided by the currents from the external basin-scale or
and tracked by means of MEDSLIK-II. This multiple release is reset
regional oceanographic model. In addition, a Stokes drift-equivalent cor-
after every 10 days over 12 January 2009–30 April 2015 (230 times),
rection term is taken to be simply proportional to the wind components
which yields a total of more than 6 × 1010 released particles
multiplied by 0.01, in accordance with the work of Kubota (1994).
(27 , 850 × 230 × 10 , 000). Particles do not interact with each other;
If a particle arrives on the coast, the model is able to simulate the ad-
therefore, the combined concentration field is calculated using a linear
sorption of particles into the coastal environment, taking into account a
superposition of concentrations resulting from all the releases over the
probability that particles may be washed back into the water. As out-
basin. After that, concentration fields from each multiple release are
puts, MEDSLIK-II provides the particle concentrations in two states: at
re-gridded onto the AFS grid and stored on a daily basis.
the surface and on the coast. Particle concentrations at the surface and
The integration time for Eq. (1) is chosen to be 10 days for the fol-
coastline are calculated on a finer grid than the one used in the hydro-
lowing reasons. Re-gridding induces the effect of finite-range diffusion
dynamic model (De Dominicis et al., 2013a); in this work we used a res-
with range of the order of the grid cell length (Froyland et al., 2007).
olution of 150 m. MEDSLIK-II simulates particle concentration on the
Consequently, the integration time should be large enough for advective
coast by using a GSHHG coastline (NOAA National Geophysical Data
transport to dominate any discretization-induced diffusive transport. In
Center, 2014) discretized into segments with a resolution appropriate
other words, time of integration should be large enough so that most
for each segment, which varies from a few meters to 100 m for an al-
trajectories leave their initial grid cell. In the calculations reported
most straight coastal segment. The Lagrangian time step is taken to be
here, at the characteristic velocity of 2 cm s -1 in the Adriatic Sea
30 min in MEDSLIK-II.
(Lacorata et al., 2001; Veneziani et al., 2007), particles travel a distance
of 17 km for 10 days, which significantly exceeds the grid cell length of
2.3. Meteo-oceanographic data
2.2 km. However, the Markov chain approach imposes a limitation
caused by within-grid-cell correlation between entry and exit of particle
As detailed above, MEDSLIK-II requires the input of data about atmo-
(van Sebille et al., 2011; Maximenko et al., 2012). As a consequence, we
spheric winds and sea currents. The atmospheric wind data are provid-
have underestimated the advection of floating debris by the weak cur-
ed by the European Centre for Medium-Range Weather Forecasts grid cell length -1
(ECMWF) analyses, 0.25° horizontal and 6-h temporal resolutions. The rents of magnitude less than integration time  0.3 cm s .

ocean current data are provided by daily simulation fields produced We use the re-gridded data to calculate the transition matrices ac-
by the Adriatic Forecasting System, AFS (http://oceanlab.cmcc.it/afs). cording to van Sebille et al. (2012), Froyland (2000), Froyland et al.
AFS uses the POM-based Adriatic regional model as a theoretical back- (2014). First of all, we split the interior and coastal grid cells. By defini-
ground (Guarnieri et al., 2010). The model domain covers the geograph- tion, the former are described as the particles floating at the sea surface
ical area of 39–45.82° N and 12.2–20.78° E with a horizontal resolution (srf). The latter represents the plastic particles on the coast (cst). Hence,
of about 1/45°, which is approximately equal to 2.2 km, on 31 sigma- four types of transition matrices can be introduced as follows.
layers. The bottom topography is obtained from the U.S. Navy 1/60° Matrices [Pkt00kt]srf →srf express a chance for a particle to move from its
bathymetric database DBDB1, by bilinear interpolation of the depth initial position k0 at the sea surface (srf) at time t0 to the final position k
data into the model grid. The model is forced by momentum, water, at the sea surface (srf) at time t. For simplicity of notation, two spatial in-
and heat fluxes interactively computed by using model predicted sea dices, i and j, are combined into one spatial index k. The interior matrices
surface temperature and the atmospheric ECMWF data at 0.25° horizon- [Pkt00kt]srf →srf describe the transport of particles between interior grid cells
tal and 6-h temporal resolutions. The river input into the basin has been in probability terms.
implemented through river climatology (Raicich, 1994). Precipitation is Then, [Pkt00kt]srf →cst indicate a chance for a particle to move from its ini-
taken from the global climatological monthly means (Legates and tial position at the sea surface (srf) to a final position at the coast (cst).
Wilmott, 1990). AFS is nested into the Mediterranean Forecasting Sys- Matrices [Pkt00kt]cst→ srf signify a chance to move from the initial posi-
tem, MFS (Oddo et al., 2009; Pinardi et al., 2003; Tonani et al., 2008). tion at the coast (cst) to the final position at the sea surface (srf). In
The nesting technique is specified in Zavatarelli and Pinardi (2003), the present work, we assume that all the elements of [Pkt00kt]cst → srf are
Oddo et al. (2005). Tidal signal is introduced in AFS through the south- equal to zero, and perform this by the removing the particles after
ern lateral open boundary conditions on barotropic velocity. 10 days of stagnation in the coastal grid cells. Maximenko et al. (2012)
performed the same after 5 days of stagnation. It should be emphasized
3. Markov chain model that this assumption plays a key role because it provides dissipativity of
the system due to interaction with the coastline.
Following Maximenko et al. (2012) and van Sebille et al. (2012), To complete the consideration, the unit diagonal matrices [Pkt00tk]cst→cst
concentrations of floating particle debris are calculated in two steps: that express a chance to move from the initial position at the coast (cst)
to the final position at the coast (cst) can be formally introduced.
(1) calculation of the transition matrices by means of the ensemble Following van Sebille et al. (2012), in an attempt to limit random-
runs of MEDSLIK-II coupled to the AFS currents and ECMWF ness of our Markov chain we have used 230 ten-day transition matrices.
wind, and
(2) construction of the Markov chain to simulate the evolution over
time of particle concentrations from terrestrial and maritime in- 3.2. Plastic particles' release experiment
puts as identified in Section 2.1.
The transition matrices described above provide a statistical summary
of the advective-diffusive flow of the Lagrangian particles by the underly-
Although the most logical way is directly aggregating the virtual tra- ing high-resolution atmospheric and oceanographic fields (Froyland,
jectories (Lebreton et al., 2012; Mansui et al., 2015), we have chosen the 2000). Once the transition matrices have been obtained, the evolution
 S. Liubartseva et al. / Marine Pollution Bulletin 103 (2016) 115–127 119

Fig. 2. Averaged 2009–2015 maps (a) of floating debris concentration (g km-2) at the sea surface, and (b) of surface currents (m s -1).

of the particle concentration at the sea surface grid cells cktn+1

srf and in the exceed 0.7% of the total particles in the domain. However, due to influ-
coastal grid cells cktn+1
cst from any source can be iteratively calculated as ence of the southern lateral open boundary conditions we have
follows underestimated the floating debris concentrations and the coastal
h i fluxes in the southern part of the domain, particularly in the Northern
kt t t
csrfnþ1 ¼ cksrf0 t n  Pkn0 knþ1 ð2Þ Ionian Sea.
srf →srf
After the gridding of each marine litter input specified in
h i Section 2.1 onto the AFS grid, and the spreading over the four closest
kt k0 t n t n t nþ1
ccstnþ1 ¼ ckt
cst þ csrf  Pk0 k
n
: ð3Þ
srf →cst grid cells, the instantaneous pulses of particles are initialized period-
ically from each input. As previously detailed in van Sebille et al.
Eq. (3) represents accumulation of particles in the coastal cells. In (2012), these pulse-like releases tend to provide clearer patterns of
order to meet practical needs, we calculate fluxes of plastic debris particles' behaviors over time. In the present calculations, each
onto the coastline segments: pulse is activated on a synoptic time scale of 10 days over the period
" # from January 2009 to April 2015, including a spin-up period that will
k skcst ckt 2 kt 1
cst  ccst be justified in Section 4.1.
f cst ¼ ; ð4Þ
l
k t2  t1 In the framework of the methodology described above, we interpret
cst
virtual floating particles as plastic marine debris. It should be noted that,
where skcst is the area of the coastal cell; lkcst is the length of the coastal some important natural processes have been neglected, including plas-
segment; and ckt kt2 tic that sinks due to a loss of buoyancy caused by biofouling and adher-
cst and ccst are plastic concentrations at time t1 and t2,
1

respectively. ence of inorganic particles (Muthukumar et al., 2011; Woodall et al.,

Quantification of the fluxes (Eq. (4)) allows us to estimate plastic de- 2014). Additionally, breaking into microplastics and ingestion by biota
posits on the coastline. For example, if we know the residence time of (Andrady, 2011; Cole et al., 2011) are not considered. All of these issues
marine plastics on the beach τ (Kataoka and Hinata, 2015), the deposits need to be better understood and quantified, and should be taken into
of plastic particles on coastline can be calculated as account in further modeling.

k k 4. Results and discussion

dcst ¼ f cst τ: ð5Þ

If we define a cleanup period of tcleanup as the time period between reg- 4.1. Half-life time of particles as a measure of dissipativity of the basin
ular cleanups, mean deposit of plastics can be estimated as
Half-life time of floating particles is an important transport charac-
k k
dcst cleanup ¼ f cst t cleanup : ð6Þ teristic of the basin that stems from the basin geometry and dynamics.
We estimate this value by means of a long-time integration of Eqs. (2)
Deposit values calculated by Eqs. (5) and (6) can be used to plan clean- and (3). The mean particle half-life, i.e., the time after release at which
up activities. 50% of the particles still remain at the sea surface, is found to be approx-
To calculate correctly the particle concentration in the Otranto Strait, imately 43.7 days, which is in good agreement with the drifter mean
the southern boundary of the domain is expanded to 39° N, which is a half-life of 40 days observed by Poulain (2001) in the Adriatic Sea. Ac-
southern boundary of AFS. At this boundary, zero inflow of particles is cording to this value, the Adriatic Sea, as a land-locked basin, is defined
imposed. Particles that outflow are removed from the domain as if the as a highly dissipative system with respect to floating plastics, in con-
39° N boundary would serve as an artificial coastline. During the time trast to the global ocean, where the half-life time of particles equals
period of 2009–2015, percentage of outflowing particles does not 19 years (Maximenko et al., 2012).
120 S. Liubartseva et al. / Marine Pollution Bulletin 103 (2016) 115–127

It should be noted that a 6-year integration of Eqs. (2) and (3) does Table 3
not reveal any so-called garbage patches in the Adriatic Sea, which is Worldwide measurement of concentration of plastic debris floating at the sea surface in
mass concentration units.
identical to the conclusion drawn by Mansui et al. (2015) for the Medi-
terranean Sea. However, substantial accumulation of plastics on the Domain Concentration Reference
coastline is obtained, which is logically implied from Eqs. (2) and (3) (g km -2)

as a consequence of the mass conservation law. Thus, in the present Global Ocean 1–10,000 Eriksen et al. (2014)
work, shoreline is, by construction, the main sink of plastic debris. In re- North Pacific Garbage Patch 300–3500 Wong et al. (1974)
46–1210 Day and Shaw (1987)
ality, the sink is mainly partitioned between the shoreline and seafloor
64–30169 Moore et al. (2001)
(Galgani et al., 2000), but we could not calculate in which proportion 3600 Yamashita and Tanimura (2007)
because loss of the plastic buoyancy is still unknown. 70.96 (mean) Eriksen et al. (2014)
To compute correctly the floating debris concentration, a spin-up pe- North Atlantic Garbage Patch 290 (mean) Carpenter and Smith (1972)
riod of 90 days, which exceeds a double half-life time, is taken into ac- South China Sea 16.7 (mean) Zhou et al. (2011)
Mediterranean Sea 22–1934, Cozar et al. (2015)
count. This allows avoidance of any depletion in concentration fields 423 (mean)
at the beginning of calculations.

4.2. Concentration of plastic debris floating at the sea surface and its flux kilometers to several tens of kilometers. We also associate these fea-
onto coastline tures with the areas of relatively weak circulation.
Seasonal averaging the plastic debris concentration is carried out ac-
After integration of Eqs. (2) and (3), daily averaged distributions of cording to the regional climatological seasonality defined by Artegiani
plastic debris concentration are obtained for both the interior and coast- et al. (1997) as follows: winter is January–April, spring is May–June,
al cells. A short animation of the floating debris drift over 2013–2015 is summer is July–October, and autumn is November–December. The sea-
presented to show spatial–temporal variability of the concentration sonally averaged concentrations (Fig. 3a–d) show some general points
fields on a daily scale (http://plastics.cmcc.it/files/DFG_suppl.mov). of agreement among each other and with the total average map
Fluxes of plastics onto the coastline are calculated based on concentra- (Fig. 2a) described above: the elongated band and local coastal lenses.
tions in the coastal cells by means of Eq. (4). At the same time, substantial variability is revealed, which is driven by
Starting our analysis from the averaged 2009–2015 map of debris seasonally varying meteo-oceanographic conditions. As Fig. 3e–h
concentration at the sea surface (Fig. 2), we compare our results with shows, boundary currents and jets representing the Adriatic general cir-
available field observations (Table 3). Unfortunately, the majority of culation tend to vary in strength in different seasons.
the observations count debris items per square kilometer, which does In winter (Fig. 3a), a lateral extension of the area of elevated concen-
not allow a direct comparison with our model results. Taking into ac- trations is found, particularly in the middle and southern Adriatic. The
count the differences in the inputs and dissipativity among various do- reasons this feature appears are rather complex (Fig. 3e), namely: (1) in-
mains with respect to floating debris, it can be concluded that the terconnection of the South Adriatic gyre with a notable inflow of water
calculated concentrations that vary in the range of 0–60 g km-2 from the Otranto Strait, (2) reinforcement of the Middle Adriatic cyclon-
(Fig. 2a) are in reasonable quantitative agreement with the field obser- ic gyre by the Eastern Southern Adriatic Current which is well-
vations, at least as far as the order of magnitude is concerned. developed in winter, and (3) a relative weakness of the outflow through
Turning our focus to spatial variability of the concentration field the Otranto Strait such that it is unable to efficiently ventilate the middle
(Fig. 2a), we note that areas of the highest concentrations of plastics and southern Adriatic.
(N10 g km-2) are represented by an elongated band shifted to the Italian In spring (Fig. 3b), plastic debris tends to be trapped in the northern
coastline narrowing from northwest to southeast. In the northern Adriatic at concentrations of more than 20 g km-2 due to a distinct cyclon-
Adriatic, this area almost covers the waters between the Po Delta and ic circulation deflecting the plastics from the Po Delta and Venice Lagoon
Gulf of Trieste. The band has discontinuities at the Gargano Promontory, toward the Gulf of Trieste and vice versa (Fig. 3f). The area of highest con-
and then recovers in the waters in front of Bari. centrations spreads up to the Istrian Peninsula. Additionally, patchy traces
The floating debris concentration tends to correspond to the spatial of elevated concentrations are found near Bari and in the southeastern
distribution of the plastic debris inputs (Fig. 1); however, some connec- sector of the basin where the South Adriatic gyre is located (Fig. 3f).
tions with the known patterns of the general circulation (Artegiani et al., In summer (Fig. 3c), the band of elevated concentrations tends to
1997; Oddo et al., 2005; Zavatarelli and Pinardi, 2003) are also clearly shrink toward the Italian coastline. In the middle Adriatic, longitudinal
visible (Fig. 2b). The shape of the band follows the climatological gradients of plastic concentrations become sharper. Due to a well-
shape of the intense current jet stretched against the western Adriatic developed outflow from the Otranto Strait (Fig. 3g), the southern
coast, the so-called Western Adriatic Coastal Current. Along the trans- Adriatic is efficiently ventilated, which leads to a cleansing of the plas-
versal line off Bari, the distribution reveals a penetration of plastics to- tics from the southern and middle parts of the basin.
ward the Adriatic interior caused by the so-called South Adriatic gyre In autumn (Fig. 3d), the belt shows a longitudinal widening in the
(Zavatarelli and Pinardi, 2003). Apart from the elongated band, the dis- northern Adriatic. It detaches from the coastline at Conero Promontory
tribution reveals elevated concentrations in a semi-closed area slightly and comes back near Pescara. Dispersed patches in floating debris con-
south of the Buna/Bojana Mouth and near Rijeka. Although the former centrations are visible slightly north of Gargano Promontory, where the
is connected with a high local input from the Buna/Bojana River very strong Middle Adriatic gyre is located (Fig. 3h). In the southeastern
(Table 2), and the latter is caused by the mutual inputs from Rijeka part of the basin, an obvious increase in concentrations up to
and the Rjecina River (Fig. 1), both areas exhibit very steady water 7–10 g km-2 is found, which can be identified as a notable peculiarity of
stagnation (Fig. 2b). Additionally, a great number of local high- the season. This feature can be caused by a tight connection between
concentration lenses are obtained near the coast, particularly along the eastern side of the South Adriatic gyre and a well-defined inflow of
the rugged eastern shoreline. Their typical sizes range from a few water through the Otranto Strait (Fig. 3h).

Fig. 3. Seasonally averaged maps of plastic concentration (g km-2) at the sea surface (left panel) and surface currents (m s -1) (right panel): (a), (e) winter; (b), (f) spring; (c), (g) summer;
and (d), (h) autumn.
S. Liubartseva et al. / Marine Pollution Bulletin 103 (2016) 115–127 121
122 S. Liubartseva et al. / Marine Pollution Bulletin 103 (2016) 115–127

Plastic debris fluxes onto coastline are calculated for the 54 selected debris than the western one (from the Marano Lagoon to Crotone).
coastal segments over the 6-year interval (t1 = 1 May 2009, t2 = 30 The distinctive “hot spots” are found in the Po Delta (70.1 kg(km day)-
1
April 2015 in Eq. (4), and are plotted in Fig. 4. Fig. 4a shows a pro- ); Venice (35.8 kg(km day)-1); Chioggia (32.6 kg(km day)-1) and in
nounced asymmetry in plastic debris fluxes between the eastern and the Reno Mouth (25.3 kg(km day)-1). Bari receives a plastic debris flux
western coastlines of the Adriatic Basin (Fig. 4b): the eastern part of 24.2 kg(km day)-1), followed by the western coastline by Marina di
(from Igoumenitsa to the Marano Lagoon) tends to get less plastic Ravenna (21.7 kg(km day)-1), Ancona (16.9 kg(km day)-1), and Pesaro
(15.4 kg(km day)-1). Pula, situated at the southern tip of the Istria Pen-
insula, demonstrates the highest flux (24.9 kg(km day)-1) of the eastern
coastline of the Adriatic Basin, followed by the Island of Mljet
(18.8 kg(km day)-1) and Split (15.7 kg(km day)-1).
Applying the residence time of marine plastics on the beach of
τ = 209 days (Kataoka and Hinata, 2015), we obtain (Eq. (5)) more
than 14 ton km -1 of plastic debris deposit in the Po Delta, if no cleanup
is conducted. If we apply a regular cleanup once every 10 days, we get

Table 4
Seasonal variability of floating debris fluxes kg(km day)-1 onto the Adriatic coastline.

Floating debris flux onto coastal segment

(kg(km day) -1)

Coastal Segment Winter Spring Summer Autumn

(Jan–Apr) (May–Jun) (Jul–Oct) (Nov–Dec)

Igoumenitsa 0.4 1.0 0.9 0.3

Corfu 9.2 7.7 8.6 10.8
Saranda 2.3 2.9 3.1 2.1
Karaburun Peninsula 6.1 9.1 9.1 7.0
Vlorë 3.2 9.9 6.7 2.3
Durrës 5.3 14.8 13.3 2.2
Bar 6.9 10.6 6.7 9.6
Budva 6.5 6.8 1.9 9.7
Dubrovnik 6.3 6.9 2.3 7.6
Mljet 22.9 19.9 12.4 22.7
Lastovo 7.8 4.9 1.8 8.8
Korcula 10.8 6.9 5.5 12.1
Vis 5.9 3.0 1.7 5.9
Hvar 4.1 3.5 2.9 5.0
Split 14.8 15.5 16.5 15.8
Šibenik 8.4 7.3 6.2 8.5
Zadar 6.8 7.4 6.2 5.5
Lošinj 3.3 5.3 3.8 3.4
Rijeka 3.3 4.8 4.1 4.7
Pula 23.2 26.7 23.3 29.5
Rovinj 1.5 6.2 2.8 1.5
Piran 1.5 2.1 3.4 0.9
Koper 1.0 4.0 3.5 0.5
Trieste 5.7 10.5 6.4 4.4
Isonzo Mouth 8.1 9.2 5.7 6.7
Marano Lagoon 3.8 8.1 4.8 2.6
Livenza Mouth 5.3 13.1 6.5 6.7
Venice 28.5 53.2 41.7 20.9
Chioggia 40.7 21.9 33.7 25.0
Po Delta 66.2 63.7 78.5 67.1
Reno Mouth 25.3 20.0 31.0 19.1
Marina di Ravenna 16.8 19.9 28.9 18.9
Rimini 7.1 6.7 11.6 8.7
Pesaro 17.4 11.3 15.4 15.7
Ancona 21.3 12.8 13.5 19.0
San Benedetto del Tronto 6.4 6.7 6.9 3.6
Pescara 9.4 15.0 11.3 6.1
Termoli 17.0 18.6 13.9 8.0
North Gargano 16.1 12.6 9.8 9.5
Gargano Tip 3.8 1.6 1.9 3.0
Manfredonia 4.3 5.3 6.3 4.2
Barletta 8.6 12.3 15.5 5.5
Bari 23.9 29.9 26.6 14.1
Monopoli 14.8 21.6 12.9 8.0
Brindisi 13.3 14.4 11.2 7.5
Otranto 9.1 16.7 12.7 3.1
Salento Tip 4.7 5.4 6.1 3.4
Gallipoli 3.3 4.6 3.6 3.5
Avetrana 4.4 2.5 3.0 6.0
Taranto 15.0 16.3 14.7 14.9
Crati Mouth 1.7 2.3 3.2 1.8
Fig. 4. (a) Plastic debris fluxes (kg(km day)-1) onto the Adriatic coastline (2009–2015).
Cariati 1.6 4.1 6.0 1.5
The maximum flux in the Po Delta is highlighted in red. (b) Locations of the coastal
Crotone 1.6 4.7 5.7 2.4
segments named after the nearby geographic objects.
 S. Liubartseva et al. / Marine Pollution Bulletin 103 (2016) 115–127 123

mean deposition of about 700 ton km -1 (Eq. (6)) for the period between aimed at the practical needs of (1) researchers who implement a strat-
cleanups. egy of plastic litter monitoring at sea; (2) stakeholders involved in
Although there is considerable similarity between the seasonally av- coastal cleanup; and (3) policy makers who develop regulations on dis-
eraged 2009–2015 plastic debris fluxes onto coastlines (Table 4) and charge of garbage in the marine environment. Formally, the methodol-
the flux over the whole 6-year interval of integration (Fig. 4), the tem- ogy is based on forward-in-time (so called direct problem) and
poral variability caused by the meteo-oceanographic conditions is also backwards-in-time (so called inverse problem) integration of Eqs. (2)
visible. The Po Delta receives the maximum fluxes of the basin that and (3) with respect to a selected set of the subregions of interest.
peak in the summer season (July–October) at up to 78.5 kg(km day)-1. Fig. 5 shows the division of the Adriatic basin into the 30 selected
Venice and Chioggia demonstrate the elevated fluxes in all the seasons. subregions comprising 26 coastal and four open-sea subregions. In the
Fluxes to Venice dominate the fluxes to Chioggia in spring (May–June) present work, the division is based on the basin geography and major
and summer (July–October) but fall behind in autumn (November– patterns of the general circulation (Artegiani et al., 1997). Depending
December), and particularly in winter (January–April). There are on the research interest, it can be also conducted on the basis of coastal
coastline segments that receive high and stable fluxes of plastics geomorphology or the administrative structure, economics, geopolitics,
(in a sequence of winter–spring–summer–autumn), namely, on the and so forth.
eastern coastline: Split (14.8–15.5–16.5–15.8 kg(km day) -1) and The solution to the direct and inverse problems will be illustrated
Pula (23.2–26.7–23.3–29.5 kg(km day)-1); on the western coastline: below in an example of a 13th subregion (44.0–44.7° N 12.0–13.0° E),
Reno Mouth (25.3–20.0–31.0–19.1 kg(km day)-1), Marina di Raven- which is a part of the Italian administrative region Emilia Romagna
na (16.8–19.9–28.9–18.9 kg(km day)-1), Pesaro (17.4–11.3–15.4– (Fig. 6). We chose this subregion because recent field observations
15.7 kg(km day)-1), Gargano North (16.1–15.5–9.8–9.5 kg(km day)-1). (Mazziotti et al., 2015) reveal the elevated concentration of plastics
Saranda (2.3–2.9–3.1–2.1 kg(km day)-1), Koper (1.0–4.0–3.5– there which stem from the direct influence of the Po inputs.
0.5 kg(km day)-1), and Gargano Tip (3.8–1.6–1.9–3.0 kg(km day)-1) re-
ceive permanently low fluxes. In contrast, some locations exhibit sea- 4.3.1. Direct problem
sonal contrasts. For example, the coastline near Durres gets As shown in Fig. 6, the inputs from a 13th subregion of Emilia Roma-
approximately six times more flux (14.8 kg(km day)-1) in spring than gna consist of the plastic debris fluxes from the shipping lanes; the riv-
in autumn (2.2 kg(km day)-1). The coastline segment in the vicinity of ers of Marecchia with a mean value of 104 ton year -1, Reno
Bari receive a flux of 29.9 kg(km day)-1 in spring in contrast to (43 ton year -1), Lamone (11 ton year -1), Uniti (11 ton year -1), Savio
14.1 kg(km day)-1 in autumn. The Croatian Island of Mljet collects (11 ton year -1), Bevano (5 ton year -1), Rubicone (5 ton year -1), Uso
22.9 kg(km day)-1 in winter but 12.4 kg(km day)-1 in summer. Hence, (5 ton year -1); and the cities of Ravenna (183 ton year -1), Rimini
a common practice of neglecting cleanups in winter is not suitable for (163 ton year -1), Cervia (32 ton year -1), Cesenatico (29 ton year -1).
Mljet Island. To sum up, Split exhibits the lowest seasonal max/min Forward-in-time integration of Eqs. (2) and (3) is carried out to cal-
ratio of 1.1, while Piran shows the highest seasonal min/max ratio of culate the relative contribution of the floating debris inputs from the
10.2. 13th subregion to each subregion of the Adriatic basin. As Table 5 indi-
cates, the main receptor (36.3%) coincides with the main source: Emilia
4.3. Regional approach to modeling the floating debris transport Romagna subregion. Together with the second receptor of North
Marche (19.8%), they collect more than half of the floating debris from
To study source-receptor interconnections among various subre- Emilia Romagna. The rest is redistributed at sea mainly following the
gions in the Adriatic basin, a regional approach was developed. It is Western Adriatic Coastal Current that tends to dominate along the

Fig. 5. Division of the Adriatic Basin into the subregions named after nearby geographic objects.
124 S. Liubartseva et al. / Marine Pollution Bulletin 103 (2016) 115–127

Table 6
Top 10 source-subregions contributing the floating debris (%) to the 13th subregion of
Emilia Romagna (2009–2015).

Subregion-source (number) % of all subregion-sources

1 Po Delta (12) 46.0

2 Emilia Romagna (13) 35.0
3 Venice (11) 6.8
4 Trieste Gulf (10) 6.4
5 Istria (9) 2.2
6 North Marche (14) 1.3
7 North Adriatic (30) 0.5
8 Montenegro (4) 0.3
9 Abruzzo (16) 0.3
10 South Albania (2) 0.2

city of Ravenna is responsible for more than 50% of the plastic accumu-
lated on the coastline of Emilia Romagna.

4.3.3. Impact matrices

The direct problem described above can be generalized to quantify
the source–receptor relationships among all the Adriatic subregions.
To present the results we have introduced a direct impact matrix
(Fig. 7). The rows of the matrix represent the subregions that are con-
sidered sources of the floating debris. The columns indicate the subre-
gions that serve as the receptors of plastics. The matrix elements are
colored according to their relative contributions in percent. We have in-
troduced the 31st row and column to take into consideration the float-
Fig. 6. The enlarged 13th subregion and the inputs of floating debris labeled as in Fig. 1: ing debris that leaves the domain through the open 39° N boundary.
shipping lanes (gray scatter plot); rivers (open blue diamonds), the largest rivers Then, a sum of all the elements in the row equals 100%. For example,
(closed blue circles); cities (open red circles); and the largest cities (closed red circles). the floating debris from the 3rd subregion of North Albania is distribut-
ed as follows: approximately 30–50% remains in North Albania; 20–30%
goes to Montenegro and South Croatia; 5–10% to South Albania; 1–5% to
North Gargano, Bari, Brindisi, Lecce, Otranto, Gallipoli, and out of the do-
western coastline of the basin. Due to across-basin transport some main. Other subregions receive less than 1%.
minor receptors are found on the eastern coastline of the basin, namely As indicated in Fig. 7, the maximum values tend to be located on the
3.1% in North Croatia, 2.5% in North Central Croatia, 2.0% in South Cen- main diagonal of the impact matrix. This indicates that a major part of
tral Croatia, and 1.9% in South Croatia. In total, the eastern coastline re- floating debris tends to beach onto the coastline of its own subregion-
ceives more than 9% of the floating plastic debris from Emilia Romagna. source. However, there are some exceptions, namely South Marche
and Abruzzo send a majority of their plastics to North Gargano. Keeping
4.3.2. Inverse problem in mind that these features are obtained on the long-term integration
To identify the relative contribution of the floating debris inputs (2009–2015), we can conclude that such a local self-cleansing is provid-
from each subregion to a 13th subregion, the backwards-in-time inte- ed by a stable, long-term Western Adriatic Coastal Current. The appear-
gration of Eqs. (2) and (3) is conducted. As Table 6 indicates, the main ance of the significant matrix elements that are far from the main
polluter of Emilia Romagna with respect to plastic debris is the subre- diagonal of the matrix provide evidence about the across-basin ex-
gion of the Po Delta which accounts for 46.0% of contamination. This change. Such a process is revealed for the eastern subregions of
fraction even exceeds Emilia Romagna's own contribution of 35.0%. Montenegro and South Croatia, and for the western subregions of
Others, including Venice, the Trieste Gulf, Istria, and items 6–10 listed North Gargano, Bari, Brindisi, and Lecce. The inputs from the open-sea
in Table 6 are in total responsible for only 18.0% of the plastics. subregions constitute only off-shore contributions from the shipping
Additionally, we calculate the relative contributions from each float- lanes. As shown in Fig. 7 30–50% of plastics from the ships in the
ing debris input depicted in Fig. 1. Unsurprisingly, the 13th subregion is North and Central Adriatic go to North Gargano, while 20–30% from
mostly influenced by the Po River (Table 7), which together with the the ships in the South Adriatic go to South Croatia. The plastics from

Table 5 Table 7
Top 10 subregion-receptors receiving the floating debris (%) from the 13th subregion Top 10 inputs contributing the floating debris (%) to the 13th subregion-receptor of Emilia
source of Emilia Romagna (2009–2015). Romagna (2009–2015).

Subregion-receptor (number) % of all Subregion-receptors Input % of all inputs

1 Emilia Romagna (13) 36.3 1 Po River 39.1

2 North Marche (14) 19.8 2 Ravenna 14.0
3 North Gargano (17) 12.0 3 Shipping lanes 10.8
4 South Marche (15) 3.7 4 Rimini 6.8
5 North Croatia (8) 3.1 5 Reno River 3.7
6 Abruzzo (16) 2.9 6 Adige River 3.0
7 Bari (19) 2.6 7 Trieste 2.8
8 North Central Croatia (7) 2.5 8 Cervia 2.1
9 South Central Croatia (6) 2.0 9 Cesenatico 1.7
10 South Croatia (5) 1.9 10 Brenta River 1.2
 S. Liubartseva et al. / Marine Pollution Bulletin 103 (2016) 115–127 125

Fig. 7. Direct impact matrix quantified (%) the source–receptor relationships among the subregions in the Adriatic Basin.

the ships in the Ionian Sea tend to leave the domain through the open coastlines receive plastics, that is, receptors, while the columns show
boundary. the sources. A sum of all elements in the row equals 100%. For example,
As far as the inverse problem is concerned, the approach also allows the coastline of the Bari subregion gets 30–50% from the inputs situated
the generalization to quantify the receptor-source relationships among in the home subregion of Bari, 5–10% from North and South Gargano, as
all the Adriatic subregions. Thus, an inverse impact matrix can be calcu- well as 5–10% from the shipping lanes located in the South and Central
lated (Fig. 8). The rows of the matrix represent the subregions whose Adriatic.

Fig. 8. Inverse impact matrix quantified (%) the receptor–source relationships among the subregions in the Adriatic Basin.
126 S. Liubartseva et al. / Marine Pollution Bulletin 103 (2016) 115–127

Some subregions collect the floating debris locally, from the nearest Romagna demonstrates that more than one-third of the plastics are
neighbors (e.g., North Albania and Venice), while others receive it also beached on the coastline from which they originate. About 41.0% go
from the whole basin including remote sources. For instance, South southward, to the Italian coastline from North Marche to Bari, while
Croatia collects its own plastic debris (10–20%), and also receives it 9.5% is eventually washed ashore in Croatia. Inversely, the coastline of
from the nearest locations: Montenegro (20–30%); South Central Emilia-Romania receives 46.0% of its beached plastic debris from the
Croatia (10–20%); North Albania and the South Adriatic (5–10%); and Po Delta subregion, 35.0% of its own floating debris, and 15.4% from
the Central Adriatic (1–5%). A portion of plastics comes from the oppo- Venice, the Gulf of Trieste, and Istria combined.
site locations: Bari (5–10%); and Abruzzo, North Gargano, and South The results obtained can be used to monitor the floating debris, plan
Gargano (1–5%). The rest is received from remote locations including cleanups, and make policy-relevant decisions. The methodology devel-
South Albania (5–10%); and the Po Delta, Emilia Romagna, North oped for the Adriatic Basin can be extended to any other domain, and
Marche, Brindisi, and the North Adriatic (1–5%). could be further improved by including consideration of particles sink-
ing, breaking into microplastics, and being ingested by biota. In addition,
5. Conclusions further investigation of spatial–temporal variability of plastic litter in-
puts is vital. Furthermore, we believe that in the future higher resolution
In the present work, we have shown the first results of modeling the current data will enhance the Lagrangian representation of floating
floating debris concentrations at the sea surface and on the coastlines in plastic debris transport.
the Adriatic Basin over 2009–2015. The calculations are based on com-
bining data of terrestrial and maritime plastic litter inputs with the Mar-
Acknowledgements
kov chain model built by means of the Lagrangian model MEDSLIK-II,
forced by AFS ocean currents simulations and ECMWF wind analyses.
This work has been supported by the DeFishGear (Derelict Fishing
The Markov chain model provides a significant flexibility and computa-
Gear Management System in the Adriatic Region (http://www.
tional efficiency in simulating any configuration of the plastic debris in-
defishgear.net/) IPA Adriatic strategic project 1° str/00010 implemented
puts, which are very uncertain. Not only does the Markov chain model
with co-funding by the European Union, Instrument for Pre-Accession
allow the forward-in-time simulation of plastic concentrations, it also
Assistance (IPA). The authors wish to thank Giogia Verri for developing
gives an opportunity to perform correctly the backwards-in-time
the Adriatic river database; Francois Galgani for thoughtful advice on
simulation.
boundary conditions for floating debris; Andrej Kržan for initiating a rel-
The mean particle half-life in the Adriatic is found to be approxi-
evant discussion about units for presentation of the model results; and
mately 43.7 days, which allows us to define the Adriatic Sea as a highly
Michela De Dominicis for encouraging appreciation of the first results.
dissipative system with respect to floating plastics. By construction, the
coastline is the main sink of floating plastic debris. Further consideration
of sinking plastic debris is necessary to calculate its distribution on the References
seabed.
AIS, 2015. The database on automatic identification system. http://www.marinetraffic.
On long-term time-mean scales, floating debris distributions corre- com (Online accessed: 10-January-2015).
spond to the spatial distributions of plastic debris inputs (rivers, cities, Andrady, A., 2011. Microplastics in the marine environment. Mar. Pollut. Bull. 62,
and shipping lanes), and indicates a tight connection with patterns of 1596–1605.
Artegiani, A., Bregant, D., Paschini, E., Pinardi, N., Raicich, F., Russo, A., 1997. The Adriatic
the general Adriatic circulation. Thus, the elongated band of elevated Sea general circulation. Part II: baroclinic circulation structure. J. Phys. Oceanogr. 27,
concentration at the sea surface (N10 g km-2) follows the climatological 1515–1532.
shape of the Western Adriatic Coastal Current. Penetration of plastics to- Brinkhoff, T., 2010. The database on city population. http://www.citypopulation.de (On-
line accessed: 10-January-2015).
ward the Adriatic interior opposite Bari is revealed as the evidence of Carpenter, E., Smith, K., 1972. Plastic on the Sargasso Sea surface. Science 175, 1240–1241.
transport by the South Adriatic gyre. Inshore areas and a great number Cole, M., Lindeque, P., Halsband, C., Galloway, T., 2011. Microplastics as contaminants in
of local lenses with high plastic concentrations seem to be related to the marine environment: a review. Mar. Pollut. Bull. 62, 2588–2597.
Cozar, A., Sanz-Martin, M., Marti, E., Gonzalez-Gordillo, I., Ubeda, B., Galvez, J., Irigoien, X.,
coastal areas of weak circulation. Duarte, C., 2015. Plastic accumulation in the Mediterranean Sea. PLoS ONE 10, 1–12,
On seasonal time-mean scales, we indicate the winter plastics' ex- e0121762.
pansion into the basin's interior, spring trapping in the northern Csanady, G., 1973. Turbulent Diffusion in the Environment. D. Reidel Publishing Company,
Dordrecht (NLD).
Adriatic, summer cleansing the middle and southern Adriatic and au-
Day, R., Shaw, D., 1987. Patterns in the abundance of pelagic plastic and tar in the North
tumn spreading into the southeastern Adriatic. These seasonal features Pacific Ocean, 1976–1985. Mar. Pollut. Bull. 18, 311–316.
result mainly from the seasonal variability of the basin scale circulation De Dominicis, M., Pinardi, N., Zodiatis, G., Archetti, R., 2013b. MEDSLIK-II, a Lagrangian
including Western Adriatic Coastal Current, the South Adriatic gyre, the marine surface oil spill model for short-term forecasting — part 2: numerical simula-
tions and validations. Geosci. Model Dev. 6, 1871–1888.
Middle Adriatic gyre, and the Eastern Southern Adriatic Current. De Dominicis, M., Pinardi, N., Zodiatis, G., Lardner, R., 2013a. MEDSLIK-II, a Lagrangian ma-
Distinctive coastal “hot spots” are found: on the Po Delta coastline rine surface oil spill model for short term forecasting — part 1: theory. Geosci. Model
that receives a plastic flux of 70.1 kg(km day)-1; Venice: Dev. 6, 1851–1869.
Eriksen, M., Lebreton, L., Carson, H., Thiel, M., Moore, C., Borerro, J., Galgani, F., Ryan, P.,
35.8 kg(km day)-1; Chioggia: 32.6 kg(km day)-1; the Reno Mouth: Reisser, J., 2014. Plastic pollution in the Worlds Oceans: more than 5 trillion plastic
25.3 kg(km day)-1; and Pula, situated at the southern tip of the Istria pieces weighing over 250,000 tons afloat at sea. PLoS ONE 9, 1–15, e111913.
Peninsula: 24.9 kg(km day)-1. As far as seasonality in plastics flux onto Falco, P., Griffa, A., Poulain, P.M., Zambianchi, E., 2000. Transport properties in the Adriatic
Sea as deduced from drifter data. J. Phys. Oceanogr. 30, 2055–2071.
coastlines is concerned, the highest seasonal max/min ratio of 10.2 is Fernandes, R., Neves, R., Viegas, C., 2013. Integration of an oil and inert spill model in a
found in Piran, while Split has the lowest min/max ratio of 1.1. framework for risk management of spills at sea — a case study for the Atlantic area.
To meet practical needs the regional approach was developed as fol- AMOP 36 Technical Seminar, pp. 326–353.
Froyland, G., 2000. Extracting dynamical behaviour via Markov models. Nonlinear Dy-
lows: (1) the basin is divided into the subregions of interest; (2) the rel-
namics and Statistics, pp. 1–33.
ative contribution of the floating debris inputs from each source- Froyland, G., Padberg, K., England, M., Treguier, A., 2007. Detection of coherent oceanic
subregion is calculated by solving the direct problem; (3) for each structures via transfer operators. Phys. Rev. Lett. 98 (224503), 1–4.
Froyland, G., Stuart, R., van Sebille, E., 2014. How well-connected is the surface of the
receptor-subregion, the relative contribution from the source-
global ocean? Chaos 24 (033126), 1–10.
subregion is calculated by solving the inverse problem; and (4) the re- Galgani, F., Leaute, J., Moguedet, P., Souplet, A., Verin, Y., Carpentier, A., Goraguer, H.,
sults are summarized in terms of direct and inverse impact matrices. Latrouite, D., Andral, B., Cadiou, Y., Mahe, J., Poulard, J., Nerisson, P., 2000. Litter on
Impact matrices built for all the subregions reveal complex source– the sea floor along European coasts. Mar. Pollut. Bull. 40, 516–527.
Guarnieri, A., Oddo, P., Pastore, M., Pinardi, N., 2010. The Adriatic Basin Forecasting Sys-
receptor relationships among the various Adriatic subregions. The ex- tem new model and system development. Coastal to Global Operational Oceanogra-
ample of floating debris that originated from the subregion of Emilia phy: Achievements and Challenges, pp. 184–190.
 S. Liubartseva et al. / Marine Pollution Bulletin 103 (2016) 115–127 127

Isobe, A., Kako, S., Chang, P.H., Matsuno, T., 2009. Two-way particle tracking model for Pinardi, N., Allen, I., Demirov, E., De Mey, P., Korres, G., Lascaratos, A., Le Traon, P., Maillard,
specifying sources of drifting objects: application to the East China Sea shelf. C., Manzella, G., Tziavos, C., 2003. The Mediterranean ocean forecasting system: first
J. Atmos. Ocean. Technol. 26, 1059–1064. phase of implementation (1998–2001). Ann. Geophys. 21, 3–20.
Isobe, A., Kubo, K., Tamura, Y., Kako, S., Nakashima, E., Fujii, N., 2014. Selective transport of Pizzigalli, C., Rupolo, V., Lombardi, E., Blanke, B., 2007. Seasonal variability dispersion
microplastics and mesoplastics by drifting in coastal waters. Mar. Pollut. Bull. 89, maps in the Mediterranean Sea obtained from the Mediterranean Forecasting System
324–330. Eulerian velocity fields. Prog. Oceanogr. 22 (C05012), 1–17.
Jambeck, J., Andrady, A., Geyerand, R., Narayan, R., Perryman, M., Siegler, T., Wilcox, C., Poulain, P.M., 2001. Adriatic Sea surface circulation as derived from drifter data between
Lavender Law, K., 2015. Plastic waste inputs from land into the ocean. Science 347, 1990 and 1999. J. Mar. Syst. 29, 3–32.
768–771. Poulain, P.M., Hariri, S., 2013. Transit and residence times in the Adriatic Sea surface as de-
Kataoka, T., Hinata, H., 2015. Evaluation of beach cleanup effects using linear system anal- rived from drifter data and lagrangian numerical simulations. Ocean Sci. 9, 713–720.
ysis. Mar. Pollut. Bull. 91, 73–81. Raicich, F., 1994. Note on flow rates of the Adriatic rivers. Technical Report. CNR Istituto
Kubota, M., 1994. A mechanism of the accumulation of floating marine debris north of Ha- Talassografico Sperimentale, Trieste, pp. 1–8.
waii. J. Phys. Oceanogr. 24, 1059–1064. van Sebille, E., Beal, L., Johns, W., 2011. Advective time scales of Agulhas leakage to the
Lacorata, G., Aurell, E., Vulpiani, A., 2001. Drifter dispersion in the Adriatic Sea: Lagrangian North Atlantic in surface drifter observations and the 3D OFES model. J. Phys.
data and chaotic model. Ann. Geophys. 19, 121–129. Oceanogr. 41, 1026–1034.
Lebreton, L., Greer, S., Borrero, J., 2012. Numerical modeling of floating debris in the van Sebille, E., England, M., Froyland, G., 2012. Origin, dynamics and evolution of ocean
world's oceans. Mar. Pollut. Bull. 64, 653–661. garbage patches from observed surface drifters. Environ. Res. Lett. 7 (044040), 1–6.
Legates, D., Wilmott, C., 1990. Mean seasonal and spatial variability in a gauge corrected Thompson, R., Moore, C., Vom Saal, F., Swan, S., 2009. Plastics, the environment and
global precipitation. Int. J. Climatol. 10, 121–127. human health: current consensus and future trends. Philos. Trans. R. Soc. Lond. Ser.
Liubartseva, S., De Dominicis, M., Oddo, P., Coppini, G., Pinardi, N., Greggio, N., 2015. Oil B Biol. Sci. 364, 2153–2166.
spill hazard from dispersal of oil along shipping lanes in the Southern Adriatic and Tonani, M., Pinardi, N., Dobricic, S., Pujol, I., Fratianni, C., 2008. A high-resolution free-
Northern Ionian Seas. Mar. Pollut. Bull. 90, 259–272. surface model of the Mediterranean Sea. Ocean Sci. 4, 1–14.
Mansui, J., Molcard, A., Ourmieres, Y., 2015. Modelling the transport and accumulation of Ursella, L., Poulain, P.M., Signell, R., 2006. Surface drifter derived circulation in the north-
floating marine debris in the Mediterranean basin. Mar. Pollut. Bull. 91, 249–257. ern and middle Adriatic Sea: response to wind regime and season. J. Geophys. Res.
Martinez, E., Maamaatuaiahutapu, K., Taillandier, V., 2009. Floating marine debris surface 111 (C03S04), 1–14.
drift: convergence and accumulation toward the South Pacific subtropical gyre. Mar. Veneziani, M., Griffa, A., Poulain, P.M., 2007. Historical drifter data and statistical predic-
Pollut. Bull. 58, 1347–1355. tion of particle motion: a case study in the central Adriatic Sea. J. Atmos. Ocean.
Maximenko, N., Hafner, N., Niiler, J., 2012. Pathways of marine debris derived from trajec- Technol. 24, 235–254.
tories of Lagrangian drifters. Mar. Pollut. Bull. 65, 151–162. Verri, G., Pinardi, N., Oddo, P., Ciliberti, S., Coppini, G., 2014. The influence of the river in-
Mazziotti, C., Bertaccini, E., Benzi, M., Martiniand, P., Lera, S., Silvestri, S., Ferrari, C., 2015. flow on the circulation and dynamics of the Adriatic and Northern Ionian Sea. EGU
Sea surface microplastics distribution on Emilia-Romagna coast (Northwestern General Assembly 2014 (pp. EGU2014, 16855).
Adriatic Sea — Italy): DeFishGear Project preliminary results. Proceedings of the Wong, C., Green, D., Cretney, W., 1974. Quantitative tar and plastic waste distributions in
MICRO-2015 Seminar on microplastics issues, p. 27. the Pacific Ocean. Nature 247, 31–32.
Moore, C., Moore, S., Leecaster, M., Weisberg, S., 2001. A comparison of plastic and plank- Woodall, L., Sanchez-Vidal, A., Canals, M., Paterson, G., Coppock, R., Sleight, V., Calafat, A.,
ton in the north Pacific central gyre. Mar. Pollut. Bull. 42, 1297–1300. Rogers, A., Narayanaswamy, B., Thompson, R., 2014. The deep sea is a major sink for
Muthukumar, T., Aravinthan, A., Lakshmi, K., Venkatesan, R., Vedaprakash, L., Doble, M., microplastic debris. R. Soc. Open Sci. 1 (140317), 1–8.
2011. Fouling and stability of polymers and composites in marine environment. Int. Yamashita, R., Tanimura, A., 2007. Floating plastic in the Kuroshio Current area, western
Biodeterior. Biodegrad. 62, 276–284. North Pacific Ocean. Mar. Pollut. Bull. 54, 485–488.
Neumann, D., Callies, U., Matthies, M., 2014. Marine litter ensemble transport simulations Yoon, J.H., Kawano, S., Igawa, S., 2010. Modeling of marine litter drift and beaching in the
in the southern North Sea. Mar. Pollut. Bull. 86, 219–228. Japan Sea. Mar. Pollut. Bull. 60, 448–463.
NOAA National Geophysical Data Center, 2014. Global self-consistent, hierarchical, high- Zavatarelli, M., Pinardi, N., 2003. The Adriatic Sea modelling system: a nested approach.
resolution geography database GSHHG. http://www.ngdc.noaa.gov/mgg/shorelines/ Ann. Geophys. 21, 345–364.
shorelines.html (Online accessed: 10-January-2015). Zhou, P., Huang, C., Fang, H., Cai, W., Dongmei, L., Xiaomin, L., Hansheng, Y., 2011. The
Oddo, P., Adani, M., Pinardi, N., Fratianni, C., Tonani, M., Pettenuzzo, D., 2009. A nested At- abundance, composition and sources of marine debris in coastal seawaters or beaches
lantic–Mediterranean Sea general circulation model for operational forecasting. around the northern South China Sea (China). Mar. Pollut. Bull. 62, 1998–2007.
Ocean Sci. Discuss. 6, 1093–1127. Zodiatis, G., Lardner, R., Solovyov, D., Panayidou, X., De Dominicis, M., 2012. Predictions
Oddo, P., Pinardi, N., Zavatarelli, M., 2005. A numerical study of the interannual variability for oil slicks detected from satellite images using MyOcean forecasting data. Ocean
of the Adriatic Sea (2000–2002). Sci. Total Environ. 353, 39–56. Sci. 8, 1105–1115.