119
1 INTRODUCTION
From among the many factors that scholars have
considered as affecting factors on the likelihood of
grounding accident when they have modeled this
type of accident, traffic is one of the factors that are
presentedinalmostalloftheexistingmodelsrelated
to grounding accident; see e.g. [1‐6]. Similar to the
correlation of the ship t
raffic and frequency of ship‐
shipcollisionthatismerelynotedintheliterature[7],
it seems that this effect in grounding accident is
mostlyacceptedbytheexpertsasacommonsenseor
simply by just generalizing the ship‐ship collision
cases to grounding accidents. There is no av
ailable
studyontheactualcausallinkbetweenthetrafficand
groundingaccidentintheliterature.Itcanbeargued
that oneof the reasons behind the common belief of
the existence of causal link between the traffic and
grounding accident is that people assume when the
t
raffic is more dense, the likelihood that the ships
have to alter their courses to avoid collision and
eventually ending up grounded is higher; however
thereisnostatisticalanalysisonthiscommonbeliefto
either support or reject it. One of the problems that
may rise when such doubtful beliefs become
commonly accepted by the researchers is tha
t the
modelsthataredevelopedtoanalyzeaphenomenon
will not be accurate enough and may not represent
the reality; and if the model is used for risk
managementpurposesitmightresultinineffectiveor
evenwrongriskcontroloptions.
Inthi
spaper,ithasbeentriedtotestthiscommon
beliefbyfindingthepossiblecorrelationbetweenthe
two variables, using the statistical data of the actual
groundingaccidentshappenedintheGulfofFinland
(GOF)within22years(1989‐2010).
Correlation between the Ship Grounding Accident
and the Ship Traffic – A Case Study Based on the
Statistics of the Gulf of Finland
A.Mazaheri,J.Montewka&P.Kujala
A
altoUniversity,DepartmentofAppliedMechanics,Espoo,Finland
ABSTRACT:Shipt
rafficisoneofthefactorsthatispresentedinalmostalloftheexistinggroundingmodels,
and is considered as one of the affecting factors on the likelihood of grounding accident. This effect in
grounding accident is mostly accepted bythe experts asa common sense or simplyby just generalizing the
ship‐ship collision cases to grounding accidents. There is no av
ailable research on the actual causal link
between the ship traffic and grounding accident in the literature. In this paper, authors have utilized the
statisticalanalysisonhistoricalgroundingaccidentdataintheGulfofFinlandbetweentheyears1989and2010
andthe AIS dataof thesame area inyear 2010, asthe source of ship t
rafficdata, toinvestigatethe possible
existenceofanycorrelationbetweentheshiptrafficandthegroundingaccident.Theresultsshowthatforthe
studiedarea(GulfofFinland)thereisnocorrelationbetweenthet
rafficdensity andthegroundingaccident.
However, the possibility of the existence of minor relation between the traffic distribution and grounding
accidentisshownbytheresult.Thisfinding,however,needsfurtherinvestigationformoreclarification.
http://www.transnav.eu
the International Journal
on Marine Navigation
and Safety of Sea Transportation
Volume 7
Number 1
March 2013
DOI:10.12716/1001.07.01.16
120
The remainder of this paper is organized as
follows: The data that is used for the analysis is
presented in the next Chapter. The research
methodology and the implemented algorithms for
data analysis are explained and presented in
Chapter3; the results of the data analysis are
presentedinChapter
4;followedbythediscussionof
the results in Chapter 5. The paper is concluded in
Chapter6.
2 DATA
In order to analyze the possible correlation between
thetrafficof ships andthe grounding accidents,two
different sources of historical data are used as: 1‐
HELCOM(HelsinkiCommission) database
regarding
the ship accidents happened in the Baltic Sea area
within the years of 1989 and 2010; 2‐ HELCOM AIS
(Automatic Identification System) data on marine
trafficintheGOFinyear2010.
2.1 HELCOMAccidentDatabase
The first dataset that is utilized in this paper is
HELCOM database regarding the
ship accidents
previously occurred in the Baltic Sea area, including
Gulfof Finland, between theyears of1989 and2010.
Thedataconsistofinputssuchasthedateandtimeof
theaccident,geographicalcoordinateoftheaccident,
typeoftheaccident,flagstatesoftheinvolvedships,
name
of the involved ships, whether the accident
caused any pollution, and type and amount of the
possiblepollution.Fromamongalltheavailablefields
inthedatabase,theonlyinputvaluesthathavebeen
utilizedinthisresearcharethetypeandthelocation
of the accident. Although the database was
not
flawless, especially regarding the ship and cargo
properties and crew competences, when it comes to
thedatausefulforthepurposeofthisstudy,theonly
problemthatwasneededtobeaddressedseemstobe
the wrong recorded location of the accident that in
somecaseswerereported
inlandareas.
The data were first filtered regarding the type of
the accident, in which all the accidents that were
recorded as grounding were chosen. Before utilizing
the database, the database was filtered to limit the
data to those accidents that have happened in the
GOF,meanslimitingthecoordination
to21.63
o
Eand
30.31
o
E longitude, and to 58.90
o
N and 60.89
o
N
latitude.Therewereintotal616recordsofgrounding
accidentsinHELCOMdatabasefortheyearsof1989‐
2010,inwhich123ofthemwereoccurredintheGulf
of Finland based on the above geographical
limitation.Next,the databasewas filteredto remove
thoseaccidentsthathavereported
asoccurredinland
area.Todoso,theaccidentpointswerevisualizedin
GISsoftwareoverthemapoftheGOFandthenthose
accident points that have located in the land area
were spotted and manually removed from the
database (Figure 1). From among123 grounding
accident records in
the GOF, 11 were found as
registered by wrong coordination. As the result, 112
grounding cases spotted as happened in the GOF
betweenyears1989and2010,andhavebeenusedin
thestatisticaldataanalysisforthispaper.Hereinthis
paper, the location of these 112 grounding accidents
arereferredasgroundingpoints.
Figure1.Thelocationofthegroundingaccidentshappened
intheGulfofFinlandbetween1989‐2010
2.2 HELCOMAISDatabase
Thesecondsourceofhistoricaldatathatisutilizedin
thisstudyistheAISdataofthemarine trafficinthe
GOFin2010.TheAISdataoftheshiptraffichasbeen
used to extract the recent traffic properties of ships
navigating in the GOF.
The AIS data of the ships
navigatingintheBalticSeaareawereallreportedand
stored in the database of HELCOM. The database
haveinputvaluessuchasIMOandMMSInumbersof
thevessel,timestampoftheAISrecord,geographical
coordinate of the AIS record as latitude and
longitude, main dimension of the vessel, speed over
ground, course over ground, and rate of turn.
However,theonlyinputvalues that were utilizedin
this research were timestamps and geographical
location of the ships. The AIS data received from
HELCOM were filtered and sorted using the
methodologiessuggestedinthe
literature[8,9].
3 METHODOLOGY
Themethodologyadoptedinthis paper isintegrated
into two ma in phases. The first phase is to define
algorithms to detect the traffic definitions as Traffic
Density and Traffic Distribution in the utilized AIS
data; and the second phase was implementing
statisticalhypothesistestinginorder
tofindwhether
there is any correlation between the traffic and the
historicallocationofthegroundingaccidentusingthe
definedalgorithms.
3.1 AlgorithmDefinitions
Intheliterature,thetrafficoftheshipswasutilizedin
twowaysofdefinitionsasTrafficDensityandTraffic
Distribution(lateraldistributionoftheships
alongthe
path). Traffic density is defined as the number of
shipsperunitareaofthewaterwaywithinadesired
time window [10, 11].Since we are dealing withthe
AISdataofyear2010,thetimewindowforthisstudy
is defined as a year. The algorithms to
estimate the
trafficdensity from theAIS databased on theabove
definitionisshowninFigure2.Ingeneral,theGulfof
Finland is divided into grid cells of size one by one
nauticalmile.Thereafter,theannualtrafficdensityin
thecellsthat has at least a groundingpoint
inside is
estimated.Theannualtrafficdensityiscountedasthe
121
number of the ship tracks that passed through such
cellsoverayear.Bythiswayofalgorithmdefinition,
the speeds of the vessels do not affect the results as
the linear interpolation between the available AIS
datapointswillremovetheeffectfromtheintervalof
theAISpointstha
tarevarieddependsontheinstant
speedofthevessel.Anothervariableisdefinedinthe
algorithm to keep the numbers of grounding points
that are located inside the grid cell that the traffic
densityisestimated.Thisvariablehaslaterbeenused
to find the correlation between traffic density and
numberofgroundings.
Figure2. Algorithm to extract the traffic density from the
AISdata
Lateral distribution of the ships or ship track
distributionistheotherwayofdefiningthetrafficof
the ships when the probability of grounding is
estimated in the literature; see e.g. [2‐4, 12‐16].
Nevertheless,thereisnouniquedefinitionregarding
where exactly along a path the distribution of the
ships should be extra
cted. Obviously the extracted
lateral distributions of the ships would be different
depends on where they have been extracted; the
closertotheshoal,themorecorrectedcoursesbythe
ships, thus the narrower or skewer distributions. In
order to avoid confusion and also to simplify the
algorit
hm for extracting the data, a definition of the
shipdistancefromagroundingpointisdefinedhere.
The utilized algorithm is shown in Figure 3. In
general,theGulfofFinlandisdividedintogridcells
of size five by five nautical miles. The five nautical
milesdistanceisesti
matedtobethedistancethatcan
betravelledbymostofthemerchantships(excluding
High Speed Light Crafts) in half an hour using the
average speed of the ships navigating in the GOF.
Thereafter, the distribution of the ship traffic in the
cells that has at least a grounding point inside is
esti
mated. The distribution is estimated in this way
that when the track of a ship is passed through the
defined grid cell, the distance of the ship on the
border of the grid cell from the grounding point is
calculatedand stored inavariable. Anothervariable
is also counting the number of grounding points
inside the defined grid cell. Later on, the correlation
between the distributions of the calculated distances
andthenumberofthecountedgroundingpointsare
analyzed ba
sed on the first and second moments of
the obtained distributions. This means that it is
assumedtha
tadistancedistributionoftheshipsfrom
a grounding point can be acceptably described by
meanandvarianceofthedistribution.
Figure3. Algorithm to extract the traffic distribution from
theAISdata
3.2 StatisticalAnalysis
One of the methods to find the statistical
dependencies between two random variables is to
find if any correlation exists between the two
variables.Knowingaboutthecorrelationbetweentwo
variables is specifically useful as it indicates a
predictive relationship between the two variables,
whichcan be exploitedin pra
ctice. Nevertheless, the
existence of statistical dependencies as correlation
between two variables does not necessarily imply
causal relation between the two variables. For more
information regarding the correlation and causal
relationthereadersarereferto[17].Thereareseveral
coefficients that represent correlation dependencies
between two random variables, which in thi
s paper
twocoefficientsasPearson andSpearmanhave been
utilized. Pearson coefficient (r) of two variables is
definedas thecovarianceof the variables divided by
the product of theirstandard deviations. Pearson
coefficientgivesavaluebetween1and‐1,whichthe
exactvalueof1and‐1mea
nsthereisaperfectlinear
correlation between the variables. The value of zero
meansthatthereisnolinearcorrelationbetweenthe
two variables. Nevertheless, “no linear correlation”
cannot be interpreted as “absolutely no correlation”
between the two variables; and still some sort of
correlationintheformofnonlinearcorrelation,might
be existing between the two va
riables. Although,
thereisstillnouniquemethodorcoefficientthatcan
revealabsolutecorrelationbetweenthetwovariables,
122
there are some coefficients like Spearman that can
reveal some level of nonlinear correlation between
two variables. Spearman coefficient(ρ) reveals if the
relationofthetwovariablescanbedescribedusinga
monotonicfunction;thusitcangraspadegreeofnon‐
linear correlation between the two variables.
Spearman coefficient gives va
lue between 1 and‐1,
whichtheexactvalueof1and‐1meansthatthereare
norepeateddatavalues,andeachvaria bleisaperfect
monotonicfunctionoftheother.
Although Spearman and Pearson coefficients can
somehow disclose the possible statistical
dependenciesbetween the two variables, one should
beawareab
outtheirlimitationsandassumptions.For
instance, Pearson coefficient is defined assuming the
data are normally distributed, so in other cases it
might be misleading; or Spearman coefficient is
recommended when both variables are ordinal
variables,oroneisordinalandtheotheriscontinues
variable [17]. Therefore, since these assumptions are
not perfectly ma
tched with the limitation of our
variables,someothermethodslikeMutualInformation
testshouldbeusedadditionally.However,othertests
likemutualinformationarenotutilizedinthispaper
andremainedforthefutureresearch.
4 RESULTS
After calculating r and ρ coefficients for each sets of
variables,thestatisticalsignificanceoftheresultsare
tested assuming the null hypothesis (H
0) as “no‐
correlation” against alternative hypothesis (H
1) as
“non zero correlation”byχ
2
test in significant levels
of 95% (i.e.α= 0.05). The results are all shown in
Table1.
AsisseeninTable1,H
0cannotberejectedforthe
trafficdensityandalsoforthemeanvalue(mu)ofthe
trafficdistributionin95%significant level. Therefore
the validity of the null hypothesis is consistent with
the resultant data. Thus, the existence of any
correlation between the two variables as traffic
density of the ships and the grounding accident is
questionable. It should be noticed tha
t the defined
null hypothesis is a composite hypothesis; thus the
trueness of the hypothesis cannot strictly verify that
there is absolutely no correlation between the ship
traffic and grounding accident; however it can
stronglyquestionitsexistence.
Table1.Correlationbetweenthenumberofgroundingsand
traffic as density and distribution with two coefficients as
Pearson(r)andSpearman(ρ)
_______________________________________________
Traffic Coefficient Correlation P‐valueAccepted
PropertyValueHypothesis
_______________________________________________
Trafficr0.0045 0.9654 H0
Densityρ0.1102 0.2850 H0
Distribution r‐0.0573 0.5714 H0
(mu) ρ0.0093 0.9272 H0
Distribution r0.2462 0.0135 H1
(std)ρ0.2418 0.0154 H1
_______________________________________________
OneinterestingpointthatcanbeseenfromTable1
isthatthenullhypothesiscanberejectedforstandard
deviation (std) of the traffic distribution in 95%
significant level. This might be the sign of slight
correlation between the distributiveness of the ship
traffic along the path and the grounding accidents.
Although, the existence correlation is not very
significant (less tha
n 0.25), it has the potential for
furtherinvestigation.
5 DISCUSSION
The result of this study shows that there is no
significant correlation between the density of ship
traffic and the grounding accidents, while it shows
slight correlation between the distributiveness of the
t
rafficandgroundingaccident.
Although this is an important and interesting
result by its own as it is a counter claim for the
currently existing common belief in the society, it
should be used by caution and it needs further
research for fully confirmation due to the av
ailable
uncertainties. The three main sources of the
uncertaintiesarethedefinedalgorithmstoextractthe
dataregarding thetrafficdensityandtheshiptraffic
distribution, the issues regarding the utilized data,
andtheusedstatisticalmethods.
The algorithms defined in this research to extract
the data based on the ut
ilized definition of the ship
trafficintheliteraturehaveamainissue,whichisthe
sizeoftheusedgridcells.For thetrafficdensity,the
grid size of 1 by 1 nautical mile and for the traffic
distribution 5 by 5 nautical miles are utilized.
Therefore, the number of the grounding points tha
t
willbecaughtbythese gridsizesmaybechangedif
the grid size is increased or decreased. Besides, the
numberoftheshipsintheareamayalsobechanged
if the size of the grid is changed. Thus, the effect of
the grid size on the resultshould beinvest
igated by
choosing different grid sizes. The question is the
factorsthataffect the grid size. One importantfactor
istheaveragespeedofthetrafficinthearea,whichin
thisstudyisbelievedtobeneutralizedbytheutilized
algorithm. The other factor might be the t
raffic
congestionin the area,which may be neutralizedby
affecting the width of the waterway. However, one
may argue that by affecting the width of the
waterway the result is biased to the location. The
counter argument would be that the nature of such
research is in fact bia
sed to the location of the
previous grounding accidents; and in fact the
grounding accident, in contrary to the collision
accident, is very location dependent as it only may
happenintheshallowwaterareas.
Nevertheless, the existence of any correlation
between the grounding accident and the location
(waterway) has a great potentia
l for further
investigation, especially since the slight correlation
between the grounding accident and the
distributivenessofthetrafficcanbeseenastheeffect
of the width of the waterway on the traffic [18] and
thusonthegroundingaccident.
Furthermore,thesizeoftheshipsnavigatinginthe
areama
yalsoaffecttheresults,asinsomedefinition
of trafficdensity the size of the ships is an affecting
factor[11].Therefore,thealgorithmshouldbefurther
123
modifiedinordertotakeintoaccountthesizeofthe
shippresentedinthetraffic.
One another sourceof uncertainty is that theAIS
data utilized for this research represent the recent
shiptrafficinthearea(year2010),whilethehistorical
accident data were from the years 1989
‐2010. The
current traffic does not necessarily represent the
actual traffic density and distribution in the past. In
fact, the ship traffic in the area has significantly
increasedduringthepastdecades,duetotheopening
of the new ports in the area and also the economy
growthofthe
neighboringnations[19].Therefore,the
resultofthisstudyneedstobeverifiedusingtheAIS
data of different years. Moreover, the AIS data used
forthisstudycoversthewholeyearof2010;however,
thestudiedareaisnormallycoveredwithiceduring
the winter. The icy waterways may affect
the traffic
pattern inthe area, which in this study is neglected.
Thus,theeffectof winter traffic on the resultshould
alsobeinvestigatedlater.
The other matter is related to the hypothesis
testing, where the null hypothesis can never be
provenanditcanonlybe“rejected”or
“failtoreject”.
Failingtorejectanullhypothesisdoesnotmeanthat
thenull hypothesis isalwaystrue, ratheris showing
the null hypothesis is consistent with the resultant
data;meaningthatthereisnoenoughevidenceinthe
historicaldata to prove the opposite. Thus, although
theexistence of
any possible correlationbetween the
maritime traffic and the grounding accident is
doubtedbytheresultofthisstudy,itcertainlycannot
beconcludedasabsolutely“no correlation”. Besides,
the utilized coefficients as Pearson and Spearman
may not fully detect the existence of nonlinear
correlation between the two variables. Therefore,
implementingothermethods,likemutualinformation
test,seemsusefulinordertodecreaseuncertaintyof
theresults.
The last but not the least matter is that whether
“non‐correlation can imply non‐causation”. The
opposite statement as “correlation does not imply
causation” is widely accepted between the
statisticians [20]; however, “non
‐correlation implies
non‐causation” is still being discussed within the
statisticiansand otherscholars [21]. Distinguishinga
truecausalrelationshipisverydifficultandcannotbe
directly resulted from a correlation test. Therefore,
eveniftheresultsofthisresearchcanbeacceptedas
the proof for non‐correlation between
the density of
shiptrafficandgroundingaccident,stillthecausality
link between these two should be investigated and
discussedfurther.
6 CONCLUSION
Itisshowninthisresearchthatthereisnocorrelation
between the ship traffic density and the grounding
accidents,whilethereisslightcorrelationbetweenthe
grounding
accidentandthetrafficdistributiveness.
The results are based on the historical grounding
accidentsthattookplaceintheGulfofFinlandwithin
the years 1989‐2010 and the ship traffic of the same
areain2010.Thus,itisworthtohighlightagainthat
the obtained results are only
valid for the studied
area, and they cannot be generalized over other
locationswithoutfurtherinvestigation.
There are some levels of uncertainty involved in
the presented result,which are mostly related to the
utilized algorithms to extract the required data from
the databases. Some assumptions like the used grid
cells should
be tested against the different sizes in
ordertofindtheeffectofthesizeofthegridcellsin
thefinalresult.Besides,theeffectofthewintertraffic
and size of the ships on the results, which are
neglected in this study, have the potential of further
investigation.
Moreimportantly, the non‐existence of
any causal link between the ship traffic and the
grounding accidents cannot be merely concluded
from the result of this research, and it needs further
researchanddiscussion.
AKNOWLEDGMENT
This study was conducted as a part of “Minimizing
risks of maritime oil transport by holistic
safety
strategies” (MIMIC) project. The MIMIC project is
funded by the European Union and the financing
comes from the European Regional Development
Fund,TheCentralBalticINTERREGIVAProgramme
2007‐2013;theCityofKotka;Kotka‐HaminaRegional
Development Company (Cursor Oy); Centre for
Economic Development, and Transport and
the
EnvironmentofSouthwestFinland(VARELY).
REFERENCES
1. Fujii, Y. and R. Shiobara, The Analysis of Traffic
Accidents.TheJournalofNavigation,1971.24(4):p.534‐
543.
2. Pedersen, P.T. Collision and Grounding Mechanics. in
ProceedingsofWEMTʹ95ʹ.1995.Copenhagen,Denmark:
The Danish Society of Naval Architecture and Marine
Engineering.
3. Simonsen, B.C., Mechanics of Ship
Grounding, in
Department of Naval Architecture and Offshore
Engineering. 1997, Technical University of Denmark:
KongensLyngby.
4. COWI,RiskAnalysisforSeaTrafficintheAreaaround
Bornholm. 2008, Danish Maritime Authority: Kongens,
Lyngby.
5. Kristiansen, S., Traffic‐Based Models, in Maritime
Transportation‐Safety Management and Risk Analysis.
2005.
6.
Mazaheri A., et al.,A Decision Support Tool for VTS
Centers to Detect Grounding Candidates. TransNav‐
International Journal on Marine Navigation and Safety
ofSeaTransportation,Vol.6,No.3,pp.337‐343,2012
7. Judson, B., Collision risk circumstances and traffic
routeingintheapproachestothestraitofJuan
deFuca,
in 7th International VTS Symposium. 1992: Vancouver,
Canada.
8. Goerlandt, F. and P. Kujala, Traffic simulation based
ship collision probability modeling. Reliability
EngineeringandSystemSafety,2011.96(1):p.91‐107.
9. van Dorp, J.R. and J.R.W. Merrick, On a risk
management analysis of oil spill risk using
maritime
transportationsystemsimulation.AnnalsofOperations
Research,2009:p.1‐29.
10.Fujii,Y., etal.,SomeFactorsAffectingtheFrequencyof
AccidentsinMarineTraffic:I‐TheDiameterofEvasion
for Crossing Encounters, II‐ The probability of
124
Stranding,III‐TheEffectof DarknessoftheProbability
of Collision and Stranding. The Journal of Navigation,
1974.27(2):p.239‐247.
11.Mazaheri, A. and J. Ylitalo, Comments on geometrical
modeling of ship grounding, in 5th International
Conference on Collision and Grounding of Ships
(ICCGS).2010:Espoo,Finland.
12.Otto,
S., et al., Elements of Risk Analysis for Collision
and Grounding of a RoRo Passenger Ferry. Marine
Structures,2002.15:p.461‐474.
13.Karlsson,M.,F.M.Rasmussen,andL.Frisk,Verification
of ship collision frequency model, in Ship Collision
Analysis.1998:Balkema,Rotterdam.p.117‐121.
14.Ramboll, Navigational
safety in the Sound between
Denmark and Sweden (Oresund). 2006, Ramboll
DanmarkA/S:Denmark.
15.Gucma, L. The method of navigational risk assessment
onwaterwaysbasedongeneralisedrealtimesimulation
data. in International conference on marine simulation
and ship maneuverability. 2006. Terschelling,
Netherlands.
16.Kaneko, F. A method for estimation
of grounding
frequencybyusingtrajectoriesofshipsandgeometryof
seabed. in International Conference on Collision and
GroundingofShips(ICCGS).2010.Espoo,Finland.
17.Lehman, A., JMP for BasicUnivariate andMultivariate
Statistics: A Step‐by‐stepGuide.2005, Cary, NC, USA:
SASPress.
18.Seong, Y.
C., et al., The relation with width of fairway
andmarinetrafficflow.TransNav‐InternationalJournal
of Marine Navigation and Safety of Sea Transportation
2012.6(3):p.317‐321.
19.Kujala,P.,etal.,AnalysisoftheMarineTrafficSafetyin
theGulfofFinland.ReliabilityEngineeringandSystem
Safety2009.
94(8):p.1349‐1357.
20.Pearl, J., Causal inference in statistics: An overview, R‐
350, Editor. 2009, Computer Science Department,
UniversityofCalifornia:LosAngeles.p.96‐146.
21.Neufeld,E.andS.Kristtorn,Doesnon‐correlationimply
non‐causation. International Journal of Approximate
Reasoning,2007.46:p.257‐273.
