1: %% ****** Start of file template.aps ****** %
2: %%
3: %%
4: %% This file is part of the APS files in the REVTeX 4 distribution.
5: %% Version 4.0 of REVTeX, August 2001
6: %%
7: %%
8: %% Copyright (c) 2001 The American Physical Society.
9: %%
10: %% See the REVTeX 4 README file for restrictions and more information.
11: %%
12: %
13: % This is a template for producing manuscripts for use with REVTEX 4.0
14: % Copy this file to another name and then work on that file.
15: % That way, you always have this original template file to use.
16: %
17: % Group addresses by affiliation; use superscriptaddress for long
18: % author lists, or if there are many overlapping affiliations.
19: % For Phys. Rev. appearance, change preprint to twocolumn.
20: % Choose pra, prb, prc, prd, pre, prl, prstab, or rmp for journal
21: % Add 'draft' option to mark overfull boxes with black boxes
22: % Add 'showpacs' option to make PACS codes appear
23: % Add 'showkeys' option to make keywords appear
24:
25: \documentclass[aps,pre,twocolumn,superscriptaddress,showpacs]{revtex4}
26: %\documentclass[aps,prl,twocolumn,groupedaddress]{revtex4}
27: \usepackage{dcolumn}
28: \usepackage{graphicx}
29: \usepackage[dvips]{hyperref}
30: \usepackage[TS1,OT1,T1]{fontenc}
31:
32: % You should use BibTeX and apsrev.bst for references
33: % Choosing a journal automatically selects the correct APS
34: % BibTeX style file (bst file), so only uncomment the line
35: % below if necessary.
36: % \bibliographystyle{apsrev}
37:
38: \begin{document}
39:
40: \newtheorem{thm}{Theorm}
41: \newtheorem{defn}{Definition}
42: \newtheorem{cor}{collory}
43: \newtheorem{lem}{lemma}
44: \newtheorem{prop}{proposition}
45:
46:
47: % Use the \preprint command to place your local institutional report
48: % number in the upper righthand corner of the title page in preprint mode.
49: % Multiple \preprint commands are allowed.
50: % Use the 'preprintnumbers' class option to override journal defaults
51: % to display numbers if necessary
52: %\preprint{}
53:
54: %Title of paper
55: \title{Empirical analysis of the worldwide maritime transportation network}
56:
57: % repeat the \author .. \affiliation etc. as needed
58: % \email, \thanks, \homepage, \altaffiliation all apply to the current
59: % author. Explanatory text should go in the []'s, actual e-mail
60: % address or url should go in the {}'s for \email and \homepage.
61: % Please use the appropriate macro foreach each type of information
62:
63: % \affiliation command applies to all authors since the last
64: % \affiliation command. The \affiliation command should follow the
65: % other information
66: % \affiliation can be followed by \email, \homepage, \thanks as well.
67:
68: \author{Yihong Hu}
69: \email{051025007@fudan.edu.cn} \affiliation{School of management,
70: Fudan University, Shanghai 200433, China}
71: \author{Daoli Zhu}
72: \affiliation{School of management, Fudan University, Shanghai
73: 200433, China} \affiliation{Shanghai logistics Institute, Shanghai
74: 200433, China}
75: %Collaboration name if desired (requires use of superscriptaddress
76: %option in \documentclass). \noaffiliation is required (may also be
77: %used with the \author command).
78: %\collaboration can be followed by \email, \homepage, \thanks as well.
79: %\collaboration{}
80: %\noaffiliation
81:
82: \date{\today}
83:
84: \begin{abstract}
85: In this paper we present an empirical study of the worldwide
86: maritime transportation network (WMN) in which the nodes are ports
87: and links are container liners connecting the ports. Using the
88: different representations of network topology namely the space $L$
89: and $P$, we study the statistical properties of WMN including degree
90: distribution, degree correlations, weight distribution, strength
91: distribution, average shortest path length, line length distribution
92: and centrality measures. We find that WMN is a small-world network
93: with power law behavior. Important nodes are identified based on
94: different centrality measures. Through analyzing weighted cluster
95: coefficient and weighted average nearest neighbors degree, we reveal
96: the hierarchy structure and rich-club phenomenon in the network.
97: \end{abstract}
98:
99: % insert suggested PACS numbers in braces on next line
100: \pacs{89.75.Da,89.75.Dd,89.75.-k}
101: % insert suggested keywords - APS authors don't need to do this
102: %\keywords{}
103:
104: %\maketitle must follow title, authors, abstract, \pacs, and \keywords
105: \maketitle
106:
107: % body of paper here - Use proper section commands
108: % References should be done using the \cite, \ref, and \label commands
109: \section{Introduction}
110: \label{introduction} The recent few years have witnessed a great
111: devotion to exploration and understanding of underlying mechanism of
112: complex systems as diverse as the Internet \cite{Barabasi}, social
113: networks \cite{Newman1} and biological networks \cite{Jeong}. As
114: critical infrastructure, transportation networks are widely studied.
115: Examples include airline \cite{Amaral,GuimeraPNAS,Wli, chiliping,
116: Guida}, ship \cite{Xinping1}, bus \cite{Sienkiewicz, Xinping2,
117: liping, Ferber}, subway \cite{Latorasubway} and railway \cite{Sen,
118: wli2} networks.
119:
120: Maritime transportation plays an important role in the world
121: merchandize trade and economics development. Most of the large
122: volume cargo between countries like crude oil, iron ore, grain, and
123: lumber are carried by ocean vessels. According to the statistics
124: from United Nations \cite{maritimereport}, the international
125: seaborne trade continuously increased to 7.4 billion tons in 2006
126: with a robust annual growth rate of $4.3$ per cent. And over 70 per
127: cent of the value of world international seaborne trade is being
128: moved in containers.
129:
130: Container liners have become the primary transportation mode in
131: maritime transport since 1950's. Liner shipping means the container
132: vessels travel along regular routes with fixed rates according to
133: regular schedules. At present most of the shipping companies adopt
134: hub-and-spoke operating structure which consists of hub ports,
135: lateral ports, main lines and branch lines, forming a complex
136: container transportation network system \cite{Rodrigue}.
137:
138: Compared with other transportation networks, the maritime container
139: liner networks have some distinct features: (1) A great number of
140: the routes of container liners are circular. Container ships call at
141: a series of ports and return to the origin port without revisiting
142: each intermediate port. It's called pendulum service in container
143: transportation. While bus transport networks and railway networks
144: are at the opposite with most of buses or trains running
145: bidirectionally on routes. (2) The network is directed and
146: asymmetric due to circular routes. (3) Lines are divided into main
147: lines and branch lines. Main lines are long haul lines which
148: involves a set of sequential port calls across the oceans. Sometimes
149: long haul lines call at almost 30 ports. Branch lines are short haul
150: lines connecting several ports in one region to serve for main
151: lines.
152:
153: We construct the worldwide maritime transportation network (WMN)
154: using two different network representations and analyze basic
155: topological properties. Our result shows that the degree
156: distribution follows a truncated power-law distribution in the space
157: $L$ and an exponential decay distribution in the space $P$. With
158: small average shortest path length 2.66 and high cluster coefficient
159: 0.7 in the space $P$, we claim that WMN is a small world network. We
160: also check the weighted network and find the network has hierarchy
161: structure and "rich-club" phenomenon. Centrality measures are found
162: to have strong correlations with each other.
163:
164: The rest of the paper is organized as follows: in Section
165: \ref{construction}, we introduce the database and set up the network
166: using two different network representations. In Section
167: \ref{basicproperty} various topological properties are studied
168: including degree distribution, degree correlations, shortest path
169: length, weight distribution and strength distribution etc. Section
170: \ref{structure} discloses the hierarchy structure by studying the
171: weighted and unweighted clustering and degree correlations.
172: Centrality measures correlations and central nodes' geographical
173: distribution are studied in Section \ref{centrality}. Section
174: \ref{conclusion} gives the conclusion.
175: % The Appendices part is started with the command \appendix;
176: % appendix sections are then done as normal sections
177: % \appendix
178:
179: % \section{}
180: % \label{}
181: \section{Construction of the network}
182:
183: \label{construction} We get the original data from a maritime
184: transport business database named CI-online \cite{CIonline} which
185: provides ports and fleet statistics of 434 ship companies in the
186: world. The data includes 878 sea ports and 1802 lines. The ports are
187: distributed in different regions and we list the number of ports in
188: each region in Table \ref{listofports}.
189: \renewcommand{\arraystretch}{1.2}
190: \begin{table}
191: \begin{center}
192: \caption{Number of sea ports by major geographic region
193: \label{listofports}}
194: \begin{tabular}{l c c c c}
195: \hline
196: Region & & & No of sea ports & \\
197: \hline
198: Africa & & & 96 & \\
199: Asia and Middle East & & & 251 & \\
200: Europe & & & 311 &\\
201: North America & & & 61 &\\
202: Latin America & & & 96 &\\
203: Oceania & & & 63 &\\
204: Total & & & 878 &\\
205: \hline
206: \end{tabular}
207: \end{center}
208: \end{table}
209:
210: \begin{figure}[th]
211: \begin{center}
212: \includegraphics[scale=0.35]{space.eps}
213: \caption{Description of the space $L$ and the space $P$. (a) and (b)
214: are the undirected representations in the space $L$ and the space
215: $P$ respectively. (c) and (d) are the directed representations in
216: the space $L$ and the space $P$ respectively. In the space $L$, a
217: link is created between consecutive stops in one route. In the space
218: $P$ all ports that belong to the same route are connected. Line A
219: (solid line) and line B (dashed line) are two different pendulum
220: routes sharing one common node: the port No. 1. \label{spaceLandP}}
221: \end{center}
222: \end{figure}
223:
224:
225: To construct the worldwide maritime transportation network, we have
226: to introduce the concept of spaces $L$ and $P$ as presented in Fig.
227: \ref{spaceLandP}. The idea of spaces $L$ and $P$ is first proposed
228: in a general form in \cite{Sen} and later widely used in the study
229: of public bus transportation networks and railway networks. The
230: space $L$ consists of nodes being ports and links created between
231: consecutive stops in one route. Degree $k$ in the space $L$
232: represents the number of directions passengers or cargoes can travel
233: at a given port. The shortest path length in the space $L$ is the
234: number of stops one has to pass to travel between any two ports. In
235: the space $P$, two arbitrary ports are connected if there is a
236: container line traveling between both ports. Therefore degree $k$ in
237: the space $P$ is the number of nodes which can be reached without
238: changing the line. The shortest path length between any two nodes in
239: the space $P$ represents the transfer time plus one from one node to
240: another and thus is shorter than that in the space $L$.
241:
242:
243: Since WMN is a directed network, we extend the concept of spaces $L$
244: and $P$ to directed networks according to \cite{Xinping1}. See Fig.
245: \ref{spaceLandP}. Line A and B are two different pendulum routes
246: crossing at the port No. 1. (a) and (b) is the undirected network
247: representation. (c) and (d) is the respective directed version.
248:
249:
250: Based on the above concepts we establish the network under two
251: spaces represented by asymmetrical adjacent matrices $A^L$, $A^P$
252: and weight matrices $W^L$, $W^P$. The element $a_{ij}$ of the
253: adjacent matrix $A$ equals to 1 if there is a link from node $i$ to
254: $j$ or 0 otherwise. The element $w_{ij}$ of weight matrix $W$ is the
255: number of container lines traveling from port $i$ to port $j$.
256:
257:
258: We need to define the quantities used in this weighed and directed
259: network. We employ $k^{L}_{in}(i)$ and $k^{L}_{out}(i)$ to denote
260: in-degree, out-degree of node $i$ in the space $L$, and
261: $k^{L}_{un}(i)$ to represent undirected degree in the space $L$.
262: Similarly $k^{P}_{in}(i)$, $k^{P}_{out}(i)$ and $k^{P}_{un}(i)$ are
263: employed in the space $P$. Hence we have
264: \begin{equation}
265: k^{L}_{in}(i)=\sum_{j\neq i}a_{ji}^L
266: \end{equation}
267: \begin{equation}
268: k^{L}_{out}(i)=\sum_{j\neq i} a_{ij}^L
269: \end{equation}
270: \begin{equation}
271: k^{L}_{un}(i)=\sum_{j\neq i} (a_{ij}^L+a_{ji}^L)
272: \end{equation}
273: which also holds for the space $P$.
274:
275: Strength is also divided into in-strength and out-strength. In the
276: space $L$ the in-strength of node $i$ is denoted by $S^{L}_{in}(i)$
277: and out-strength is denoted by $S^{L}_{out}(i)$. Undirected strength
278: (total strength) is represented by $S^{L}_{un}(i)$. They can be
279: calculated according to the following equations:
280: \begin{equation}
281: S^{L}_{in}(i)=\sum_{j\neq i}w_{ji}^L
282: \end{equation}
283: \begin{equation}
284: S^{L}_{out}(i)=\sum_{j\neq i} w_{ij}^L
285: \end{equation}
286: \begin{equation}
287: S^{L}_{un}(i)=\sum_{j\neq i} (w_{ij}^L+w_{ji}^L)
288: \end{equation}
289: which also holds for $S^{P}_{in}(i)$, $S^{P}_{out}(i)$,
290: $S^{P}_{un}(i)$ in the space $P$.
291:
292: Other quantities like clustering coefficient and average nearest
293: neighbors degree also have different versions in directed and
294: weighted WMN. We employ $c^{L}_i$ and $c^{P}_i$ to denote the
295: unweighted clustering coefficient of node $i$ in the space $L$ and
296: $P$ respectively. Analogously $k_{nn,i}^{L}$ and $k_{nn,i}^{P}$ are
297: used to denote the average nearest neighbors degree of node $i$ in
298: the space $L$ and $P$ respectively. For weighted WMN we add
299: superscript $W$ to the above quantities and consequently they become
300: $c^{WL}_i$, $c^{WP}_i$, $k_{nn,i}^{WL}$ and $k_{nn,i}^{WP}$.
301:
302:
303:
304: \begin{figure*}[th]
305: \begin{center}
306: \includegraphics[width=3.1in]{maritimedegreeL.eps}
307: \includegraphics[width=3.1in]{maritimedegreeP.eps}
308: \includegraphics[width=3.1in]{maritimeinandout.eps}
309: \includegraphics[width=3.1in]{maritimeKLvsKP.eps}
310: \caption{(a) Degree distributions in the space $L$ obey truncated
311: power-law distributions with almost the same exponents. The turning
312: points are at $k=20$ and $k=30$ respectively. (b) Degree
313: distributions in the space $P$ all follow exponential distributions.
314: (c) Positive correlation between in-degree and out-degree. (d)
315: Unweighted degree under two spaces. Degree in the space $P$ is
316: larger than in the space $L$. \label{maritimedegree}}
317: \end{center}
318: \end{figure*}
319: \section{Topological properties}
320: \label{basicproperty}
321: \renewcommand{\arraystretch}{1.2}
322: \renewcommand{\tabcolsep}{9pt}
323: \begin{table}
324: \caption{Basic parameters for spaces $L$ and $P$. $n$ is the number
325: of nodes and $m$ is the number of links. $\langle k\rangle$ is the
326: average undirected degree. $\langle C\rangle$ is the average
327: unweighted cluster coefficient. $\langle l\rangle$ is the average
328: shortest path length.\label{maritimebasicprop}}
329: \begin{tabular}{l c c c c c }
330: \hline
331: Space & $n$ & $m$ & $\langle k_{un}\rangle$ & $\langle C\rangle$ & $\langle l\rangle$ \\
332: \hline
333: Space $L$ & 878 & 7955 & 9.04 & 0.4002 & 3.60 \\
334: Space $P$ & 878 & 24967 & 28.44 & 0.7061 & 2.66 \\
335: \hline
336: \end{tabular}
337: \end{table}
338:
339: \subsection{degree distribution and degree correlations}
340: First we examine the degree distributions in two spaces. Fig.
341: \ref{maritimedegree} shows that in-degree, out-degree and undirected
342: degree distributions in the space $L$ all follow truncated power-law
343: distributions with nearly the same exponents. In-degree and
344: out-degree obey the function $P(k)\sim k^{-1.7}$ before $k=20$. When
345: $k>20$ their distribution curves bend down to the function $P(k)\sim
346: k^{-2.95}$. Unweighted degree in the space $L$ has the same
347: exponents of $-1.7$ and $-2.95$ but the critical point becomes
348: $k=30$. Truncated power-law degree distributions are often observed
349: in other transportation networks like the worldwide air
350: transportation network \cite{GuimeraPNAS}, China airport network
351: \cite{Wli}, U.S. airport network \cite{chiliping} and the Italian
352: airport network \cite{Guida}. It is explained in \cite{Amaral} that
353: the connection cost prevents adding new links to large degree nodes.
354: Analogous cost constraints also exist in the maritime transport
355: network. Congestion in hub ports often makes ships wait outside for
356: available berth for several days, which can cost ships extremely
357: high expense. Consequently new links are not encouraged to connect
358: to those busy ports.
359:
360: While in the space $P$ three degree distributions all follow
361: exponential distributions $P(k)\sim e^{\alpha k}$. The parameters
362: are estimated to be $\alpha=0.0117$ for in-degree, and
363: $\alpha=0.0085$ for out-degree, $\alpha=0.0086$ for unweighted
364: degree. The property that degrees obey truncated power-law
365: distributions in the space $L$ and exponential distributions in the
366: space $P$ is identical to public transportation networks
367: \cite{Sienkiewicz,Xinping2} and railway networks \cite{Sen}.
368: Particularly the Indian railway network \cite{Sen} has exponential
369: degree distributions with the parameter $0.0085$ almost the same
370: with in-degree and out-degree distribution in WMN.
371:
372: Next, the relation between in-degree and out-degree is studied. Fig.
373: \ref{maritimedegree}(c) is a plot of out-degree $k_{out}$ vs.
374: in-degree $k_{in}$. They have a positive correlation under two
375: spaces. Out-degree climbs when in-degree increases. Evidently the
376: in-out degree correlation is very strong.
377:
378: Finally, we want to find out the relations between degrees under
379: different spaces. Fig. \ref{maritimedegree}(d) shows undirected
380: degree in the space $P$ $k_{un}^P$ vs. undirected degree in the
381: space $L$ $k_{un}^L$. All dots are above the diagonal, indicating
382: the undirected degree in the space $P$ is larger than in the space
383: $L$. It is understandable because of different definition of two
384: spaces topology. In the space $P$ all stops in the same route are
385: connected which surely increases degree of each node. Table
386: \ref{maritimebasicprop} lists the basic properties of WMN in two
387: spaces. Average undirected degree in the space $P$ is $28.44$, much
388: larger than average undirected degree 9.04 in the space $L$.
389:
390:
391:
392: \subsection{Line length}
393: Let's denote line length, i. e. the number of stops in one line, as
394: $l$. In Fig. \ref{maritimelengthdist} the probability distribution
395: of line length $P(l)$ can be approximated as a straight line in the
396: semi-log picture representing an exponential decay distribution
397: $P(l)\sim e^{-\alpha l}$ with the parameter $\alpha=0.13$. It
398: indicates there are much more short haul lines than long haul lines
399: in maritime transportation. Long haul lines use large vessels and
400: travel long distance from one region to another region while short
401: haul lines as branch lines travel between several neighboring ports
402: and provide cargos to main lines. For example, the line consisting
403: of the following ports:
404: Shanghai-Busan-Osaka-Nagoya-Tokyo-Shimizu-Los
405: Angeles-Charleston-Norfolk-New York-Antwerp-Bremerhaven-
406: Thamesport-Rotterdam-Le Havre-New York-Norfolk-Charleston-Colon-Los
407: Angeles-Oakland-Tokyo-Osaka-Shanghai, is a typical long haul line
408: connecting main ports in Asia and Europe, calling at ports for 24
409: times.
410: \begin{figure}[th]
411: \begin{center}
412: \includegraphics[scale=0.8]{maritimelengthdist.eps}
413: \caption{Probability distribution of line length. It can be
414: approximated by a straight line in semi-log plot indicating an
415: exponential decay. \label{maritimelengthdist}}
416: \end{center}
417: \end{figure}
418:
419: \subsection{Shortest path length}
420: The frequency distributions of shortest path lengths $d$ in the
421: spaces $L$ and $P$ are plotted in Fig. \ref{maritimepathlength}. The
422: distribution in the space $L$ has a wider range than in the space
423: $P$. The average shortest path length is 3.6 in the space $L$ and
424: 2.66 in the space $P$ (see Table \ref{maritimebasicprop}). This
425: means generally in the whole world the cargo need to transfer for no
426: more than 2 times to get to the destination. Compared with the
427: network size $N=878$, the shortest path length is relatively small.
428:
429: \begin{figure}[th]
430: \begin{center}
431: \includegraphics[scale=0.8]{maritimepathlength.eps}
432: \caption{Frequency distributions of shortest path length under two
433: spaces. The distribution in the space $L$ has a wider range than in
434: the space $P$.\label{maritimepathlength}}
435: \end{center}
436: \end{figure}
437:
438: \begin{figure*}[th]
439: \begin{center}
440: \includegraphics[width=3.1in]{maritimeweightdist.eps}
441: \includegraphics[width=3.1in]{maritimestrengthdist.eps}
442: \includegraphics[width=3.1in]{maritimeinoutstrength.eps}
443: \includegraphics[width=3.1in]{maritimesvsk.eps}
444: \caption{(a) Weight distributions in two spaces. (b) Undirected
445: strength distributions in two spaces. (c) Relation between
446: in-strength and out-strength in two spaces. (d) Correlations between
447: strength and degree in two spaces. The slopes both equal to 1.3
448: approximately indicating nonlinear relationships in the spaces $L$
449: and $P$. \label{maritimeweightdist}}
450: \end{center}
451: \end{figure*}
452:
453: \subsection{Weight and strength distribution}
454: Usually traffic on the transportation network is not equally
455: distributed. Some links have more traffic flow than others and
456: therefore play a more important role in the functioning of the whole
457: network. Weight should be addressed especially in transportation
458: networks. Here we study four properties of weighted WMN: weight
459: distribution, strength distribution, in-out strength relations and
460: the relations between strength and degree. The results are displayed
461: in Fig. \ref{maritimeweightdist}.
462:
463: First we examine weight distributions. In Fig.
464: \ref{maritimeweightdist} (a) two weight distribution curves are
465: approximately straight declining lines before $w=40$. The power-law
466: distributions are estimated to be $P(w)\sim w^{-0.95}$ in the space
467: $P$ and $P(w)\sim w^{-0.92}$ in the space $L$.
468:
469: Next, Fig. \ref{maritimeweightdist} (b) shows the undirected
470: strength distributions under two spaces both obey power-law behavior
471: with the same parameter. The functions are estimated to be $P(s)\sim
472: s^{-1.3}$.
473:
474: And we also analyze the relation between in-strength and
475: out-strength in two spaces. As we can see from Fig.
476: \ref{maritimeweightdist} (c), in-out strength relation under the
477: space $L$ is plotted as almost a straight line while the relation
478: under the space $P$ has a slight departure from the linear behavior
479: at large $s_{in}$ values. But clearly they are positively
480: correlated.
481:
482: Finally an important feature of weighted WMN, the relations between
483: strength and degree, is investigated. Under two spaces the relations
484: between undirected strength and undirected degree are both unlinear
485: with the slope of the line approximately $1.3$, which means the
486: strength increase quicker than the increase of degree. This is often
487: occured in the transportation networks and has its implication in
488: the reality. It's easy for the port with many container lines to
489: attract more lines to connect the port and thus to increase the
490: traffic more quickly.
491:
492: \section{Hierarchy structure}
493: \label{structure}
494:
495:
496:
497:
498:
499:
500: In this section, we explore the network structure of WMN through
501: studying both the weighed and unweighted versions of cluster
502: coefficient and average nearest neighbors degree. Hierarchy
503: structure and "rich club" phenomenon are unveiled. We conjecture
504: this kind of structure is related to ship companies' optimal
505: behavior to minimize the transportation cost known as the
506: hub-and-spoke model in transportation industry.
507:
508:
509:
510: \subsection{Clustering}
511: \begin{figure}[th]
512: \begin{center}
513: \includegraphics[width=3.1in]{maritimeCK.eps}
514: \caption{Cluster coefficients. $C(k)$ under two spaces both exhibit
515: nontrivial behavior with decay curves as functions of degree $k$.
516: And weighted versions of cluster coefficient are larger than
517: unweighted cluster coefficient, indicating high traffic edges
518: between interconnected vertices.\label{maritimeck}}
519: \end{center}
520: \end{figure}
521:
522:
523: Cluster coefficient $c_i$ is used to measure local cohesiveness of
524: the network in the neighborhood of the vertex. It indicates to what
525: extent two individuals with a common friend are likely to know each
526: other. And $C(k)$ is defined as cluster coefficient averaged over
527: all vertices with degree $k$.
528:
529: We plot $C(k)$ in Fig. \ref{maritimeck}. Either in the space $L$ or
530: in the space $P$, $C(k)$ exhibits a highly nontrivial behavior with
531: a decay curve as a function of degree $k$, signaling a hierarchy
532: structure in which low degrees belong generally to well
533: interconnected communities (high clustering coefficient), while hubs
534: connect many vertices that are not directly connected (small
535: clustering coefficient).
536:
537: $C^P(k)$ lies above $C^L(k)$ and the average cluster coefficient of
538: the network in the space $P$ is 0.7 larger than 0.4 in the space
539: $L$. This can be explained by the fact that in the space $P$ each
540: route gives rise to a fully connected subgraph. With high cluster
541: coefficient 0.7 and small average shortest path length 2.66 in the
542: space $P$, we conclude that the WMN, as expected, has the
543: small-world property.
544:
545:
546: Weighed quantities for clustering and assortativity measures are
547: first proposed in \cite{Barrat3}. Through the case study of WAN and
548: SCN, \cite{Barrat3} demonstrates that the inclusion of weight and
549: their correlations can provide deeper understanding of the
550: hierarchical organization of complex networks. The weighted cluster
551: coefficient is defined as:
552: \begin{equation}
553: c^w_i=\frac{1}{s_i(k_i-1)}\sum_{j,h}\frac{w_{ij}+w_{ih}}{2}a_{ij}a_{ih}a_{jh}
554: \end{equation}
555: which takes into account the importance of the traffic or
556: interaction intensity on the local triplets. And we define $C^w(k)$
557: as the weighted cluster coefficient averaged over all vertices with
558: degree $k$. In real weighted network we may have two opposite cases
559: of the relation between $C^w(k)$ and $C(k)$. If $C^w(k)>C(k)$ in the
560: network, interconnected triplets are more likely formed by the edges
561: with larger weights. If $C^w(k)<C(k)$ the largest interactions or
562: traffic is occurring on edges not belonging to interconnected
563: triplets.
564:
565: In Fig. \ref{maritimeck} we report weighted cluster coefficient
566: under two spaces. Evidently the weighted cluster coefficient
567: $C^{wP}(k)$ and $C^{wL}(k)$ are both above the corresponding
568: unweighted cluster coefficient, i. e. $C^{wP}(k)\geq C^P(k)$, and
569: $C^{wL}(k)\geq C^L(k)$. This indicates some closely interconnected
570: nodes with large degrees have the edges with larger weights among
571: themselves. In other words, high-degree ports have the tendency to
572: form interconnected groups with high-traffic's links, thus balancing
573: the reduced clustering. This is so called "rich-club" phenomenon.
574:
575: \subsection{Assortativity}
576: \begin{figure}[th]
577: \begin{center}
578: \includegraphics[width=3.1in]{maritimeknni.eps}
579: \caption{Average degree of the nearest neighbors as functions of
580: $k$. Inclusion of weight changes the behavior of $k_{nn}(k)$:
581: assortative behavior in the small degree range but disassortative
582: behavior in large degree range, to definite assortative behavior of
583: $k^w_{nn}(k)$ in the whole $k$ spectrum. \label{maritimeknni}}
584: \end{center}
585: \end{figure}
586:
587:
588: There is another important quantity to probe the networks'
589: architecture: the average degree of nearest neighbors, $k_{nn}(k)$,
590: for vertices of degree $k$. Average nearest neighbors degree of a
591: node $i$ is defined as:
592: \begin{equation}
593: k_{nn,i}=\frac{1}{k_i}\sum_{j}a_{ij}k_j
594: \end{equation}
595: Using $k_{nn,i}$, one can calculate the average degree of the
596: nearest neighbors of nodes with degree $k$, denoted as $k_{nn}(k)$.
597: The networks are called assortative if $k_{nn}(k)$ is an increasing
598: function of $k$, whereas they are referred to as disassortative when
599: $k_{nn}(k)$ is a decreasing function of $k$. As suggested in
600: \cite{Barrat3}, weighted version of average degree of nearest
601: neighbors is calculated by:
602: \begin{equation}
603: k^w_{nn,i}=\frac{1}{s_i}\sum_{j}a_{ij}w_{ij}k_{j}
604: \end{equation}
605: From this definition we can infer that $k^w_{nn,i}>k_{nn,i}$ if the
606: links with the larger weights are pointing to the neighbors with
607: larger degrees and $k^w_{nn,i}<k_{nn,i}$ in the opposite case.
608:
609: Both the weighted and unweighted average degree of nearest neighbors
610: under two spaces are plotted in Fig. \ref{maritimeknni}. The curve
611: of $k^P_{nn}(k)$ lies above the curve of $k^L_{nn}(k)$. The
612: $k^L_{nn}(k)$ and $k^P_{nn}(k)$ grow with the increase of degrees at
613: small degrees but decline when degrees are large. The unweighted
614: network exhibits assortative behavior in the small degree range but
615: disassortative behavior in large range.
616:
617: When we turn to the $k^{wL}_{nn}(k)$ and $k^{wP}_{nn}(k)$, the
618: weighted analysis provides us a different picture. We can see that
619: under two spaces the weighted average degree of nearest neighbors
620: exhibits a pronounced assortative behavior in the whole $k$
621: spectrum. Since the number of WMN nodes is 878, this conforms with
622: the empirical finding in \cite{Sienkiewicz} that public transport
623: networks are assortative when the number of nodes in the network
624: $N>500$ and disassortative when $N<500$.
625:
626:
627: From the above discussion we can see that the inclusion of weight
628: changes the behavior of cluster coefficient and average degree of
629: nearest neighbors. This property is identical to the worldwide
630: airport network \cite{Barrat3} and North America airport network
631: \cite{Barrat4}. In both the airline transportation and maritime
632: transportation networks, high traffic is associated to hubs and
633: high-degree ports (airports) tend to form cliques with other large
634: ports (airports). Their similar organization structure may have a
635: similar underlying mechanisms. We conjecture that this is related to
636: the hub-and-spoke structure which is widely adopted in practice by
637: airline companies or ship companies to achieve the objective of
638: minimizing the total transportation cost \cite{OKelly, Robinson,
639: Moura,Bryan}.
640:
641: Fig. \ref{maritimehubandspoke} describes a typical hub-and-spoke
642: structure which consists of three interconnected hubs and other
643: nodes allocated to a single hub. In maritime transportation main
644: liners travel between hubs handling large traffic while branch
645: liners visit the hub's neighboring ports to provide cargo for the
646: main lines. This structure allows the carriers to consolidate the
647: cargo in larger vessels to lower the transportation cost. This
648: simple structure has the similar property of
649: $C^w(k)>C(k),k^w_{nn}(k)>k_{nn}(k)$ with that of WMN. And it also
650: displays rich-club phenomenon. We think it worths investigating the
651: relations between ship companies' optimal behavior and the real
652: transportation network's hierarchy structure and rich-club property.
653:
654: \begin{figure}[th]
655: \begin{center}
656: \includegraphics[scale=0.4]{hubandspokemodel.eps}
657: \caption{A classical hub-and-spoke structure in maritime
658: transportation. There are three central vertices having very strong
659: links (high traffic) with each other and several nodes having weak
660: links (low traffic) with hubs. It has the same property of
661: $C^w(k)>C(k),k^w_{nn}(k)>k_{nn}(k)$ with that of WMN.
662: \label{maritimehubandspoke}}
663: \end{center}
664: \end{figure}
665:
666:
667:
668: \section{Centrality measures}
669:
670:
671: \begin{figure}[th]
672: \begin{center}
673: \includegraphics[width=3.1in]{maritimebcdist.eps}
674: \includegraphics[width=3.1in]{maritimebcvsk.eps}
675: \caption{(a) Probability distribution of betweenness obeys truncated
676: power-law distribution. (b) Betweenness is a straight line in
677: log-log picture indicating a power-law relation with degree. The
678: exponent in the space $L$ is estimated to be 1.66 and that in the
679: space $P$ is estimated to be 1.93. \label{maritimecentrality}}
680: \end{center}
681: \end{figure}
682:
683:
684:
685: In this section we analyze two centrality measures in social network
686: analysis \cite{Wasserman}: degree and betweenness. The most
687: intuitive topological measure of centrality is given by the degree:
688: more connected nodes are more important. The distribution and
689: correlations of degree has been discussed in Section IV.
690:
691: Betweenness centrality is defined as the proportion of the shortest
692: paths between every pair of vertices that pass through the given
693: vertex $v$ towards all the shortest paths. It is based on the idea
694: that a vertex is central if it lies between many other vertices, in
695: the sense that it is traversed by many of the shortest paths
696: connection couples of vertices. Hence we have
697: \begin{equation}
698: C_i^B=\frac{1}{(N-1)(N-2)}\sum_{j,k} \frac{n_{jk}(i)}{n_{jk}}
699: \end{equation}
700: where $n_{jk}$ is the number of shortest paths between $j$ and $k$,
701: and $n_{jk}(i)$ is the number of shortest paths between $j$ and $k$
702: that contain node $i$.
703:
704:
705:
706: Correlations between two centrality measures are presented in Fig.
707: \ref{maritimecentrality}(b). In both the spaces there is a clear
708: tendency to a power-law relation with degree $k$: $C^B(k)\sim
709: k^\alpha$ with $\alpha=1.66$ in the space $L$ and $\alpha=1.93$ in
710: the space $P$. The power-law correlations between degree and
711: betweenness is also found in bus transportaiton networks
712: \cite{Sienkiewicz, Ferber} and ship transport network
713: \cite{Xinping1}. It's worth noting that this power-law relations
714: together with the truncated scale-free behavior of the degree
715: distribution implies that betweenness distribution should follow a
716: truncated power law. This behavior is clearly identified in Fig.
717: \ref{maritimecentrality}(a). We find the betweenness centrality has
718: two-regime power-law behavior $P(C^B)\sim C^{-\alpha}$. For the two
719: spaces, exponents are almost the same: $\alpha=0.14$ at small degree
720: regime and $\alpha=1.0$ at large degree regime.
721:
722: The power-law relations between degree and betweenness suggest that
723: they are consistent with each other. It is proved in the comparison
724: of each port's degree and betweenness. The 25 most connected ports
725: are listed in Table. \ref{20ports}. Singapore is the most busy ports
726: in the world with the largest degree and betweenness. Antwerp and
727: Bushan are the second and third either in degree or in betweenness
728: measures. Only 5 ports in these ports are not listed in the 25 most
729: central ports in betweenness measure. WMN is not like the case of
730: the worldwide airline network \cite{GuimeraPNAS} which has anomalous
731: centrality due to its multicommunity structure. The difference may
732: due to the fact that there are less geographical and political
733: constraints in maritime transportation than in air transportation.
734: Ships can travel longer distance than airplanes. And airports are
735: usually classified into international and domestic airports and
736: international airlines are limited to connect international airports
737: instead of domestic airports. So there are distinct geographically
738: constrained communities in WAN. In the maritime transportation there
739: are no such constraints. Sea ports basically are all international
740: ports with the possibility to connect to any other sea ports in the
741: world.
742:
743: \label{centrality}
744: \renewcommand{\tabcolsep}{2pt}
745: \begin{table}
746: \begin{center}
747: \caption{The 25 most connected ports in the worldwide maritime
748: transportation network.
749: * These ports are not among the 25 most central ports. \label{20ports}}
750: \begin{tabular}{c l c c c}
751: \hline
752: Rank & Ports & Degree & Betweenness & Region\\
753: \hline
754: 1 & Singpore & 120 & 124110.1258 & Asia\\
755: 2 & Antwerp & 102 & 113368.6161 & Europe \\
756: 3 & Bushan & 92& 69094.7490 & Asia\\
757: 4 & Rotterdam & 87 & 78097.8754 &Europe\\
758: 5 & Port Klang & 83 & 62111.6226 & Asia\\
759: 6 & Hongkong & 78 & 46072.9799 & Asia\\
760: 7 & Shanghai & 75 & 37748.4316 & Asia\\
761: 8 & Hamburg & 60 & 40362.4625 & Europe\\
762: 9 & Valencia & 60 & 27346.0956 & Europe\\\
763: 10 & Le Havre & 58 & 48231.1636 & Europe\\\
764: 11 & Gioia Tauro & 55 & 20148.4667 & Europe\\
765: 12 & Yokohama & 54& 19716.7287 &Asia\\
766: 13 & Kaohsiung & 52& 15363.5781 & Asia\\
767: 14 & Port Said* & 52& 16121.8476 & Africa\\
768: 15 & Bremerhaven & 50& 24998.9088 & Europe\\
769: 16 & Colombo* & 48 & 15128.362 & Asia\\
770: 17 & Tanjung Pelepas &46 & 18735.1267 & Asia\\
771: 18 & Jeddah* & 45& 8570.7724 & Middle East\\
772: 19 & Jebel Ali & 44& 20137.5255 & Middle East\\
773: 20 & Ningbo* & 44& 12315.9515 & Asia\\
774: 21 & Algeciras & 43& 18701.2619 & Europe\\
775: 22 & Barcelona* & 43& 15256.3903 & Europe\\
776: 23 & Kobe & 43& 8869.7313 & Asia\\
777: 24 & New York & 42& 25935.6692 & North America\\
778: 25 & Kingston & 41& 23281.3223 & Latin America\\
779: \hline
780: \end{tabular}
781: \end{center}
782: \end{table}
783:
784: \begin{figure*}[th]
785: \begin{center}
786: \includegraphics[scale=0.4]{maritimeworldmap3.eps}
787: \caption{The geographical distribution of 25 most connected ports.
788: They are located along east-west lines, including 13 ports in Asia
789: and Middle East, 9 in Europe, 1 in Africa, 1 in North America and 1
790: in Latin America. \label{maritimeworldmap}}
791: \end{center}
792: \end{figure*}
793:
794: In Fig. \ref{maritimeworldmap} we plot the 25 most connected ports
795: on the world map. They show unbalanced geographical distribution
796: mainly located in Asia and Europe, including 13 ports in Asia and
797: Middle East, 1 in Africa, 9 in Europe, 1 in North America and 1 in
798: Latin America. Particularly they are located along the east-west
799: lines. Lines in maritime transportation are usually divided into
800: east-west lines, north-south lines and south-south lines
801: \cite{Rodrigue,Fremont}. The fact that 25 most connected ports in
802: the world are in east-west trade routes represents rapid growth and
803: large trade volume in Europe-America, Asia-America and Asia-Europe
804: trade \cite{maritimereport}.
805:
806:
807: \section{Conclusion}
808: \label{conclusion} In this paper we have presented an empirical
809: study of the worldwide maritime transportation network (WMN) under
810: different representations of network topology. We study the
811: statistical properties of WMN and find that WMN is a small world
812: network with power law behavior. There are strong correlations in
813: degree-degree, strength-degree and betweenness-degree relations.
814: Central nodes are identified based on different centrality measures.
815: Based on the analysis of weighted cluster coefficient and weighted
816: average nearest neighbors degree, we find that WMN has the same
817: hierarchy structure and "rich-club" phenomenon with WAN. We
818: conjecture that this structure is related to optimal behavior both
819: existing in air transportation and maritime transportation. So our
820: future research direction is the evolution modeling of WMN using
821: optimal behavior to reproduce real properties in WMN.
822:
823: \section*{Acknowledgment}
824: The work was supported by Natural Science Foundation of China and
825: USA ffgg (NSFC 70432001).
826: % Specify following sections are appendices. Use \appendix* if there
827: % only one appendix.
828: %\appendix
829: %\section{}
830:
831: % If you have acknowledgments, this puts in the proper section head.
832: %\begin{acknowledgments}
833: % put your acknowledgments here.
834: %\end{acknowledgments}
835:
836: \appendix
837: \begin{thebibliography}{000}
838:
839: \bibitem{Barabasi} A. -L. Barabasi, R. Albert, Nature 286 (1999) 509.
840:
841: \bibitem{Newman1}
842: M. E. J. Newman, Proc. Natl. Acad. Sci. USA 98 (2001) 404.
843:
844: \bibitem{Jeong}
845: H. Jeong B. Tombor R. Albert Z. N. Oltvai, A.-L. Barabasi, Nature
846: 407 (2000) 651.
847:
848: \bibitem{Amaral}L. A. N. Amaral, A. Scala, M. Barthelemy, H. E. Stanley, Proc. Natl. Acad. Sci. USA 97 (2000) 11149.
849:
850: \bibitem{GuimeraPNAS} R. Guimera, S. Mossa, A. Turtschi and L. A. N. Amaral, Proc. Natl. Acad. Sci. USA 102 (2005) 7794.
851:
852: \bibitem{Wli} W. Li, X. Cai, Phys. Rev. E 69 (2004)
853: 046106.
854:
855: \bibitem{chiliping} L.P. Chi, R. Wang, H. Su, X.P. Xu, J.S. Zhao, W. Li, X. Cai, Chin. Phys. Lett. 20 (8) (2003)
856: 1393.
857:
858: \bibitem{Guida} M. Guida, F. Maria, Chaos soitions and fractals 31 (2007) 527.
859:
860: \bibitem{Xinping1}
861: Xinping Xu, Junhui Hu, Feng Liu, Chaos 17, 023129 (2007).
862:
863: \bibitem{Sienkiewicz}
864: Julian Sienkiewicz, Janusz Holyst, Phys. Rev. E 72 (2005) 046127 .
865:
866: \bibitem{Xinping2}
867: Xinping Xu, Junhui Hu, Feng Liu, Lianshou Liu, Physica A 374 (2007)
868: 441.
869:
870: \bibitem{liping} Liping et al, Chin. Phys. Lett. 23 (12) (2006) 3384.
871:
872: \bibitem{Ferber}
873: C. von Ferber, T. Holovatch, Yu. Holovatch, and V. Palchykov, Arxiv
874: 0803.3514v1, 2008.
875:
876: \bibitem{Latorasubway}
877: V. Latora, M. Marchiori, Physica A 314 (2002) 109.
878:
879: \bibitem{Sen}
880: P. Sen, S. Dasgupta, A. Chatterjee, P.A. Sreeram, G. Mukherjee, S.S.
881: Manna, Phys. Rev. E 67 (2003) 036106.
882:
883: \bibitem{wli2} W. Li, X. Cai, Physica A 382 (2007) 693.
884:
885:
886:
887: \bibitem{maritimereport} United Nation conference on trade and development, Review of maritime transport 2007, http://www.unctad.org/en/docs/rmt2007en.pdf
888:
889: \bibitem{Rodrigue} Jean-Paul Rodrigue, Claude Comtois, The geography of transport systems, New York: Routledge, 2006.
890:
891: \bibitem{CIonline} http://www.ci-online.co.uk/
892:
893: \bibitem{Barrat3} A. Barrat, et al.,
894: Proc. Natl. Acad. Sci. USA 101 (11) (2004) 3747.
895:
896:
897:
898: \bibitem{Barrat4} A. Barrat, M. Barth\'{e}l\'{e}my, A.
899: Vespignani, J. of Stat. Mech. (2005) P05003.
900:
901: \bibitem{OKelly} M. O'Kelly, Journal
902: of transport geography, 6 (3) (1998) 171.
903:
904: \bibitem{Robinson} Ross Robinson, Maritime Policy and Management 25 (1) (1998) 21.
905:
906: \bibitem{Moura} M. C. Moura, M. V. Pato, A. C. Paixa, Maritime Policy and
907: Management 29 (2) (2002) 135.
908:
909: \bibitem{Bryan} Bryan and M. O'Kelly, Journal of regional science,
910: 39 (2) (1999) 275.
911:
912:
913: \bibitem{Wasserman} S. Wasserman, K. Faust, Social networks
914: analysis, Cambridge University Press, Cambridege, 1994.
915:
916: \bibitem{Fremont} A. Fremont, Journal of transportation
917: geography, 15 (2007) 431.
918:
919:
920:
921:
922:
923: \end{thebibliography}
924:
925: \end{document}
926: