EU-MRV: an analysis of 2018’s Ro-Pax CO2data
Gianandrea Mannarini, Lorenzo Carelli, Amal Salhi
Ocean Predictions and Applications Division
Fondazione CMCC (Centro Euro-Mediterraneo sui Cambiamenti Climatici)
Abstract—This paper analyses the EU-MRV dataset containing
CO2emission data from vessels calling at European harbours.
The data is provided in an aggregated form for 2018 and we
decided to focus on ferries. We augmented the original dataset
with data from the Copernicus Marine Environment Monitor-
ing Service, the IMO Global Integrated Shipping Information
System, and an open repository of ferry data. An analysis of
various energy efﬁciency indicators revealed some clustering in
the vessel population and the key factors were year of build,
vessel length, service speed, and fuel type. Georeferencing data
provided additional information on the continental patterns of
the Ro-Pax emissions.
Index Terms—Global warming, Environmental monitoring,
For decades the Intergovernamental Panel on Climate
Change1has been warning about the current and future
impacts of climate change, at a global, regional, and local
scale, including air and marine heat waves, alterations in rain
patterns and the availability of fresh water, ocean acidiﬁcation
and sea level rise, habitat and biodiversity loss, and a decrease
of both ocean and crop productivity2. The ultimate driver
of such rapid, unprecedented changes is the concentration of
greenhouse gases (GHG) in the atmosphere, emitted through
the combustion of fossil fuels, such as coal, oil, and natural
gas. Carbon dioxide (CO2) is the GHG responsible for most
of the warming observed since the beginnings of the industrial
The European Union (EU) has set ambitious targets to
curb CO2emissions and thus limit global warming and its
detrimental consequences. In 2015 the European Parliament
(EP) and the Council approved regulation number 757 on
Monitoring, Reporting, and Veriﬁcation of CO2emissions
from vessels calling at harbours not only from EU Member
States (MS) but within the European Economic Area (EEA)
, . The regulation regards vessels exceeding 5,000 gross
tonnes, prescribes consistent and comparable CO2monitor-
ing methods, and deﬁnes procedures for CO2reporting and
the veriﬁcation of reports. Since 2018, annually-aggregated
emissions reports have been mandatory for each category of
vessels, and these have been provided through a dedicated
system supported by the European Maritime Safety Agency
Funding through GUTTA project (contract no 10043587), part of IT-HR
Interreg programme, is acknowledged.
(EMSA): THETIS-MRV. These data were ﬁrst published as
scheduled on June 30, 2019.
Two other systems have come into existence since the
EU-MRV has been in place: the Data Collection System
(DCS) by the International Maritime Organization (IMO) 
and the Chinese Regulation on Data Collection for Energy
Consumption of Ships3. They differ in terms of the parameters
monitored, type of reporting, and veriﬁcation protocols. A
legislative process is currently ongoing at the EP in order to
align EU-MRV with IMO-DCS.
The THETIS-MRV dataset includes nearly 12,000 vessels,
of various types including: container ships, tankers, chemical
tankers, oil tankers, bulk carriers, and vehicle carriers. Ro-Pax
ships, also known as “Ro-Ro passenger” vessels or simply
“ferries”, are also part of the EU-MRV ﬂeet. According to
Interferry, ferries globally transport over two billion passengers
every year4. In addition, they are generally considered a cost-
efﬁcient means of displacing heavy-duty vehicles in coastal
areas. Furthermore, in THETIS-MRV, ferries represent 3% of
all vessels while accounting for 10% of all CO2emissions5.
The latter observation raises the question of carbon intensity
of vessels of this class: Can it be related to the navigational
domain? Or perhaps to the characteristics of the vessels? If so,
which characteristics and how? All this motivates a closer look
at the Ro-Pax data available in the ﬁrst EU-MRV report with
reference to emissions from 2018. These data will represent
the baseline of all future reports.
Our methodology is presented in Sect. II. The results are
provided in Sect. III, and discussed in Sect. IV.
II. ME TH OD S
Although data fusion presents various challenges related to
data quality and consistency, as the data may have originally
been collected for different purposes , it can greatly increase
the value of information. For example, in  data fusion
of kinematical (automatic identiﬁcation system, AIS) and
environmental data (sea state from model analyses) means that
vessel speed loss in waves can be estimated without needing
speciﬁc information about the vessels.
In order to enhance the information contents of the
THETIS-MRV dataset we fused it with several other datasets
(Sect. II-A) before analyzing it (Sect. II-B).
5In version #160, the total emissions are 141,088,729.6 t CO2.
2020 21st IEEE International Conference on Mobile Data Management (MDM)
2375-0324/20/$31.00 ©2020 IEEE
The datasets used in this work are all open-access:
THETIS-MRV: this corresponds to version #160 (down-
loaded from EMSA6on February 28, 2020). Of all the vessels
in the dataset, just the “Ro-pax ships” were extracted, resulting
in 356 database entries. Each entry includes 61 ﬁelds, some
of which are lacking some values. The ﬁrst ﬁeld is the
IMO identiﬁcation number of the vessel, which was used
as primary key when fusing THETIS-MRV with the other
datasets described below.
CMEMS The Copernicus Marine Environment Monitoring
System7provides sea state analysis ﬁelds from numerical
models run operationally for the various seas in Europe.
Related domains and spatial resolutions are speciﬁed in Tab. I.
Hourly-instantaneous signiﬁcant wave height ﬁelds (Hs) for
year 2018 were downloaded (a volume of about 0.1 TB) and
aggregated on an annual basis. Tab. I also reports the counts of
the Ro-Pax vessels based on their average location (see method
explained in Sect. II-B1). Wherever a vessel’s location belongs
to a region with overlapping model outputs (cf. Fig. 1), it is
attributed to the one with the highest resolution (i.e., smallest
GISIS is the Global Integrated Shipping Information Sys-
tem8which has been developed by the IMO and includes
details on the ship and shipping company. From the public
version of GISIS, we only extracted the year when the ship
was built from the ”date of build” parameter for each of the
356 THETIS-MRV Ro-Pax vessels.
Ferry-site9is an open repository with an emphasis on north
European ferries. Whenever multiple values are provided for
the parameters - length overall (LOA), service speed vserv,
and number of passengers nP- we took their highest value;
and lowest value for number of beds nBand cars nC. LOA
and vserv were validated against the MS version in the GISIS
database. Ferry-site also includes information about the route,
from which the ports of call were parsed and georeferenced.
Of the 356 THETIS-MRV Ro-Pax ferries, 342 were found also
The dataset resulting from fusing all the data presented in
Sect. II-A is called hereafter the “augmented dataset”. The
data was processed in three steps, as reported below.
1) Georeferencing: For a preliminary classiﬁcation of
THETIS-MRV data with respect to the typical domain of
operation, we associated a single geographical location with
each Ro-Pax vessel. The rationale for this coarse assignment
was to relate it to the temporal coarse (annual average)
emission information available. Finer spatial analysis would
entail having ﬁner temporal emission data, but such data are
not publicly available.
RO-PAX V ESS EL S IN EUR OPE AN S EAS W IT H THE G RI D SIZ E OF SE A STATE
MODEL OUTPUTS (1 NMI = 1852 M)
Domain short name Resolution # vessels
North West Shelf NWS 0.8 89
Baltic Sea Bal 1.0 76
Arctic Sea Arc 1.6 0
Black Sea BS 1.7 2
Mediterranean Sea Med 2.5 164
Iberian-Biscay-Irish Seas IBI 3.0 11
outliers f4of Sect. II-B2 14
total (THETIS-MRV) 356
The representative location was computed as follows: the
ferry ports of call were retrieved from Ferry-site, they were
then georeferenced10, and for each vessel the average position
of all combinations of legs among those ports was computed.
Although our procedure is quite simple, it is appropriate
for Ro-Pax vessels since their routes (unlike for example
cruise ship routes) are for reasonably short distances and tend
to be straight, though obviously depending on topological
constraints and meteo-marine conditions. However, our pro-
cedure poses an issue for longer-range routes and for convex
coastlines. Moreover, if a vessel operates between spatially-
distinct clusters of ports of call, with only occasional transfer
voyages between clusters, the representativeness of a mean
annual location computed this way is questionable.
2) Filtering: The augmented dataset was ﬁltered to prune
entries (i.e., vessels) that included inconsistencies, that is:
f1) zero total CO2emissions - these entries do not add any
information to the current analysis;
f2) emissions from MS ports greater than total emissions -
this is an inconsistency in the data, as total emissions
include by deﬁnition those among MS;
f3) off range carbon emission factor CF>4- this cannot
be true for most common fuel types, Tab. IV;
f4) no ports of call available or mean annual vessel location
out of the EEA - this is not accepted as data in THETIS-
MRV all refer to Ro-Pax ships calling at EEA ports.
f5) just a single port of call retrieved or average leg distance
greater than 450 nmi (corresponding to 15 h navigation at
a speed of 30 kts) - this implies no reliable identiﬁcation
of the place of operation.
Criteria f1)-f4) include 4.2% of the emissions from the Ro-
Pax population and are applied throughout. Instead, f5) is
applied for speciﬁc purposes only. The number of vessels and
emissions ascribed to each criterium is reported in Tab. II.
Finally, there were some gaps in the THETIS-MRV dataset:
g1) no monitoring method reported - this is a mandatory ﬁeld
for EU-MRV and ships not reporting it are not compliant
with the regulation;
g2) non null CO2emissions reported, but no total fuel con-
sumption value - both data are needed in order to infer
the fuel type (see Sect. II-B3).
OUTLIERS ACCORDING TO THE CRITERIA OF SECT. II-B2
criterium # vessels CO2emissions
f11 0.00 0.0
f23 60,242.61 0.4
f31 75,880.91 0.5
f414 448,148.27 3.0
f578 2,610,077.09 18.6
g131 1,258,810.47 9.0
g27 293,298.06 2.1
3) Derived quantities: Fuel type was inferred from the fuel
emission factor CF. The latter was obtained from THETIS-
MRV data as the ratio of total CO2emissions to total fuel
consumption. The closest value among the emission factors11
reported in Tab. IV was used for assigning the fuel type.
An effective velocity veﬀ can be computed by taking the
ratio of the total CO2emissions (C) not at berth (Cb) to the
total time spent at sea Tsand dividing by the CO2emissions
per distance dC/dx:
In the EU-MRV revision proposal12, “hours underway” re-
places “time at sea”, as in the IMO-DCS. In fact, “hours
underway” provides a more accurate estimation of veﬀ .
Ship energy efﬁciency is a crucial indicator for any regula-
tory framework, as absolute emissions should be compared
to distance sailed, capacity, transport work, or some other
normalization parameter. A brief account of the use of var-
ious indicators in the ongoing debate on the assessment of
energy efﬁciency is provided in . IMO adopted the Energy
Efﬁciency Design Index (EEDI) for new ships in 2011 .
EU-MRV mandates report several ﬁelds that can be used to
assess energy efﬁciency. The emissions per distance sailed per
passenger is a form of Energy Efﬁciency Operational Indicator,
here called “EEOIpax”. Finally, the Energy Efﬁciency per
Service Hour (EESH) is given by the ratio of total emissions
Cby total time at sea Ts.
III. RES ULTS
This section presents the results of applying the methodol-
ogy outlined in Sect. II. As mentioned in Sect. II-B1, some
of the mean annual locations are weakly identiﬁed, and this
may occasionally (less than 8% of the ships) mean that they
are near the coast but actually on land.
Fig. 1. CO2monitoring methods (cf. Tab. III). The dashed boxes correspond
to the domains of Tab. I.
A. CO2monitoring method
In accordance with the EU-MRV Regulation , for each
vessel one of the four CO2monitoring methods in Tab. III
must be selected. Method A corresponds to the top-down ap-
proach also used in the Third IMO GHG Study . However, a
few Ro-Pax vessels in the THETIS-MRV dataset do not have a
value for the monitoring method (set g1of Sect. II-B2). Some
vessels also reported more than one method. Fig. 1 highlights
that method D (direct CO2measurement) was not used by
any Ro-Pax ships in 2018. Method C was primarily used in
the North West Shelf (NWS) and Baltic seas. However, there
are some “Cs” in the northern Tyrrhenian Sea, two “B&Cs”
in the Mediterranean, and several “As” in both the NWS and
the Baltic. No ferries were identiﬁed in the vicinity of Iceland,
though Iceland is part of the EEA.
In Fig. 2 the effective speed veﬀ from Eq. 1 is compared
to the service speed vserv. While both are well correlated, on
average veﬀ is about 5 kts lower than vserv. This might not
just correspond to a slow-steaming practice, but also to the
overestimation of travel time in veﬀ due to use of Tsin Eq. 1.
We introduce the Froude number Fn=vserv /√g·LOA
for relating speed to LOA13, which exhibits a large spread
in the ferry population (cf. Fig. 8). Furthermore, at least for
13g= 9.8m/s2is the gravity acceleration.
EU-MRV CO2MONITORING METHODS
D Direct CO2emissions measurement
C Flow meters for applicable combustion processes
B Bunker fuel tank monitoring on-board
ABunker fuel delivery notes
and periodic stocktakes of fuel tanks
Fig. 2. Effective speed veﬀ from Eq. 1 vs. service speed vserv, with marker
colors relating to the year of build. High-speed Ro-Pax are shown as crosses;
the 1:1 line is black and dashed.
mono-hulls, Fn>0.4corresponds to the transition to a semi-
displacing regime . Fig. 2 shows that such “high-speed”
Ro-Pax vessels were all built around 2000 or later.
C. Energy efﬁciency
The mean annual CO2emissions per distance sailed per
passenger (EEOIpax) is georeferenced in Fig. 3. Some of the
most efﬁcient vessels with respect to this parameter operate in
the British Isles, but also in the western Mediterranean and in
the Adriatic sea.
EEOIpax is a quite sensitive indicator as, in the THETIS-
MRV vessel population, it varies over ﬁve orders of magnitude.
This spread likely reﬂects not just the various ferry sizes, but
also propulsion features, age, and the various combinations of
vehicles and passenger services (number of cabins, vehicles,
entertainment facilities) in their payloads .
Fig. 4 shows how the EEOIpax relates to service speed, fuel
type and the ratio of the number of beds to passengers (this
ratio may exceed one, possibly because crew members are not
included in the passenger count). There is some evidence that
the logarithm of emission intensity increases with speed. The
ﬁgure also suggests that there are at least three subpopulations
(EEOIpax units are g CO2/(pax·nmi) and vserv kts):
i) 102<EEOIpax <104and vserv <30 : here most vessels
with nB/nP≈1are found;
ii) 103<EEOIpax <104and vserv >35 : these are the
high speed ferries;
iii) EEOIpax <101and vserv <25 : these vessels tend to
Fuel type can be any, apart from in cluster ii) which consists
entirely of vessels using MDO as a bunker.
Year of build is shown in Fig. 5. There is no clear divide
between northern and southern European sea vessels, with
Fig. 3. CO2emissions per passenger per distance sailed (EEOIpax).
EMI SSI ON FAC TOR S (CO2BY FU EL MA SS )USE D IN T HIS W ORK
full name CF(g/g)
MDO Marine Diesel Oil 3.206
LFO Light Fuel Oil 3.151
HFO Heavy Fuel Oil 3.114
LNG Liqueﬁed Natural Gas 2.750
some of the oldest Ro-Pax in the northern Tyrrhenian and in
the Adriatic sea, but also in the Baltic.
The upper histogram of Fig. 6 shows the distribution of
the year of build. Given that the oldest ferries in service
in 2018 were nearly 50 years old, the statistics indicate
an expansive age pyramid, with a median age of about 20
years. Furthermore, the average vessel size (lower histogram
in Fig. 6) shows that larger ferries are now being built.
Fig. 4. EEOIpax vs. service speed with marker color given by number of
beds per passenger and marker type given by fuel type (cf. Tab. IV).
The height of the bars in Fig. 6 represents the total emissions
and mirrors the distribution of the number of vessels built per
year. The average EEOIpax is used to colour this histogram.
Given the great spread in values, a linear EEOIpax averaging
would be dominated by the largest values. In order to mitigate
this, the following averaging is applied:
EEOIpax = exp log(EEOIpax)(2)
Some of the highest emissions come from newer vessels (built
E. Sea state
The International Maritime Organization (IMO) recom-
mends avoiding “rough seas and head currents”. This is among
the ten measures of the Ship Energy Efﬁciency Management
Plan . Therefore sea state (waves) can be relevant when
considering carbon intensity . Sea state can greatly change
during a year, as the timescale of its variability depends
on meteorological events. In wave spectra, wind waves can
be differentiated from swell14. Fig. 7 reports the wind-wave
component alone and shows an annual average Hs∼1m in
most domains, with higher values in western European seas,
and lower values in the Adriatic and in the Aegean Sea.
The mean annual EESH (Fig. 8) shows a weak dependence
on Hsand this only for the smaller vessels (Pearson’s R =
0.57 for LOA <120 m). More time-resolved data (daily, at
least) should highlight a clearer dependence of EEOI on sea
state and even sea currents . Fig. 8 shows that EESH
increases with LOA, with larger vessels producing higher
unit emissions per time. The emission intensity of high-speed
ferries is comparable to that of much larger ferries.
F. Fuel types
Fuel type can be inferred for most THETIS-MRV Ro-
Pax vessels, but the g2)set of Sect. II-B2. Fig. 9 shows
Fig. 5. Ro-Pax vessels’ year of build.
Fig. 6. Total emissions by year of build of the vessels, with bar color given
by EEOIpax from Eq. 2. The number and cumulative distribution of vessels
built in each year are given in the upper histogram and the average LOA in
the lower one (both as grey columns).
Fig. 7. 2018 mean annual signiﬁcant wave height Hsfrom the sea state
model outputs of Tab. I interpolated at the representative vessel locations. f5
their distribution in the various European sea domains. HFO
is dominant in both the Mediterranean and the NWS. LFO
dominates in the Baltic. LNG is used for NWS and Baltic
ferries only. MDO is relevant in the NWS and in the Baltic
and, to a lesser extent, in the Mediterranean too. Fig. 9 also
highlights that the Mediterranean alone accounts for nearly as
many Ro-Pax emissions (6.9 Mton CO2) as all other European
seas combined (7.1 Mton CO2).
Fig. 8. Energy efﬁciency per service hour (EESH) vs. mean annual Hs, with
marker color given by vessel LOA. Crosses represent high-speed ferries.
Fig. 9. Total Ro-Pax emissions in the various European sea basins; the
fractions due to the various fuel types are highlighted. The total emissions
in 2018 were 14.0 Mton CO2. Colorbar’s N/A is due to the g2set, the N/A
column to the f4set.
This paper reports the ﬁrst ever publication of ship-
distinctive CO2annual emission data (EU-MRV) from Ro-Pax
vessels in the EEA.
Some noisy geographic patterns were found and can be
summarized as: multiple CO2monitoring methods used in
northern European seas, newer vessels in western European
seas, weak dependence on mean annual sea state, and half of
the EEA emissions stemming from the Mediterranean Sea. The
latter fact is in line with ﬁndings from a bottom-up inventory
Although our georeferencing procedure provides a rough
attribution of emissions to speciﬁc sea basins and a visualiza-
tion of continental patterns, it could be improved by integrating
more detailed tracking data, such as the ports of calls of each
voyage or the full AIS data.
On the other hand, the non-spatial information shows that
the energy efﬁciency is quite diverse within the Ro-Pax pop-
ulation, reﬂecting service speed, fuel type, size, and, possibly,
type of payload. High-speed ferries stand out when it comes to
the relationship between energy efﬁciency and speed or size.
Dataset augmentation was key for both the spatial and
non-spatial part of this analysis. Due to the annual temporal
aggregation, it was not possible to clearly assess if sea state
affects the energy efﬁciency. Also, a more reﬁned clustering
may help reﬁne the dependence of the energy efﬁciency on the
other vessel parameters considered here. This would enable
comparisons among similar vessels, assessing their potential
for decarbonization, and facilitate a credibility test of the
reported data. As mentioned in the preamble of  this would
simplify the EU-MRV veriﬁcation process.
 G. Myhre, D. Shindell, F. Br´
eon, W. Collins, J. Fuglestvedt, J. Huang,
D. Koch, J. Lamarque, D. Lee, B. Mendoza et al., “Anthropogenic and
Natural Radiative Forcing, Climate Change 2013: The Physical Science
Basis. Contribution of Working Group I to the Fifth Assessment Report
of the Intergovernmental Panel on Climate Change, 659–740,” 2013.
 Regulation (EU) 2015/757 of the European Parliament and of the
Council of 29 April 2015 on the monitoring, reporting and veriﬁcation
of carbon dioxide emissions from maritime transport, and amending
 Commission Delegated Regulations (EU) 2016/2071 and 2016/2072.
 IMO, “MEPC.278(70) Amendments to MARPOL Annex VI (Data
collection system for fuel oil consumption of ships),” International
Maritime Organization, London, UK, Tech. Rep., 2016.
 E. Camossi and A.-L. Jousselme, “Information and source quality
ontology in support to maritime situational awareness,” in 2018 21st
International Conference on Information Fusion (FUSION). IEEE,
2018, pp. 1–5.
 G. Mannarini, L. Carelli, D. Zissis, G. Spiliopoulos, and K. Chatzikoko-
lakis, “Preliminary inter-comparison of AIS data and optimal ship
tracks,” TransNav, vol. 13, no. 1, pp. 53–61, 2019.
 G. Panagakos, T. d. S. Pessˆ
oa, N. Dessypris, M. B. Barfod, and H. N.
Psaraftis, “Monitoring the carbon footprint of dry bulk shipping in the
EU: An early assessment of the MRV regulation,” Sustainability, vol. 11,
no. 18, p. 5133, 2019.
 IMO, “MEPC.1/Circ.681 Interim guidelines on the method of calculation
of the Energy Efﬁciency Design Index for new ships,” International
Maritime Organization, London, UK, Tech. Rep., 2009.
 ——, “MEPC 67/INF.3 (Annex) ”Third IMO GHG Study 2014”,”
International Maritime Organization, Tech. Rep., 2014.
 G. Grigoropoulos, “Recent advances in the hydrodynamic design of
fast monohulls,” Ship Technology Research, vol. 52, no. 1, pp. 14–33,
2005. [Online]. Available: https://doi.org/10.1179/str.2005.52.1.003
 IMO, “EE-WG 2/INF.4 Information on a study describing a method
for using the energy efﬁciency operational indicator (EEOI) in Ro-Ro
passenger ships,” International Maritime Organization, Tech. Rep., 2011.
 ——, “MEPC 59/INF.10 (Annex) ”Second IMO GHG Study 2009”,”
International Maritime Organization, Tech. Rep., 2009.
 G. Mannarini and L. Carelli, “VISIR-1.b: ocean surface gravity waves
and currents for energy-efﬁcient navigation,” Geoscientiﬁc Model
Development, vol. 12, no. 8, pp. 3449–3480, 2019. [Online]. Available:
 J.-P. Jalkanen, L. Johansson, and J. Kukkonen, “A comprehensive
inventory of ship trafﬁc exhaust emissions in the european sea
areas in 2011,” Atmospheric Chemistry and Physics, vol. 16,
no. 1, pp. 71–84, 2016. [Online]. Available: https://www.atmos-chem-