Massive Automatic Identification System Sensor Trajectory Data-Based Multi-Layer Linkage Network Dynamics of Maritime Transport along 21st-Century Maritime Silk Road

Automatic Identification System (AIS) data could support ship movement analysis, and maritime network construction and dynamic analysis. This study examines the global maritime network dynamics from multi-layers (bulk, container, and tanker) and multidimensional (e.g., point, link, and network) structure perspectives. A spatial-temporal framework is introduced to construct and analyze the global maritime transportation network dynamics by means of big trajectory data. Transport capacity and stability are exploited to infer spatial-temporal dynamics of system nodes and links. Maritime network structure changes and traffic flow dynamics grouping are then possible to extract. This enables the global maritime network between 2013 and 2016 to be investigated, and the differences between the countries along the 21st-century Maritime Silk Road and other countries, as well as the differences between before and after included by 21st-century Maritime Silk Road to be revealed. Study results indicate that certain countries, such as China, Singapore, Republic of Korea, Australia, and United Arab Emirates, build new corresponding shipping relationships with some ports of countries along the Silk Road and these new linkages carry significant traffic flow. The shipping dynamics exhibit interesting geographical and spatial variations. This study is meaningful to policy formulation, such as cooperation and reorientation among international ports, evaluating the adaptability of a changing traffic flow and navigation environment, and integration of the maritime economy and transportation systems.


Singer layer World
Ducruet [2] Structure and dynamics Multilayer World Zhao et al. [32]; Caschili et al. [33] Structure Container; Sample of world fleet Ducruet [24] Structure and diversity Multilayer world Kaluza et al. [23] Structure Multilayer Gravity model Sample of world fleet Tsiotas and Polyzos [34] Structure and node aggregation Tourism Greece Ducruet et al. [35] Structure container East Asia Kosowska-Stamirowska et al. [16] Structure and evolution Trade Random walk world Liu et al. [21]; Calatayud et al. [22] Structure and robustness Container Borda count Maersk shipping line; Americas Peng et al. [25] Structure and robustness Multilayer world Peng et al. [17] Structure and evolution Crude oil world Yu et al. [

Methodology
Proposed in this paper is an analytical framework for revealing the multi-layer maritime network dynamics of 21C-MSR and other countries, and is illustrated in Figure 1. This framework combines port spatial map and AIS trajectory data to construct a global maritime network based on the origins and destinations of ships using ports. Secondly, the traffic flow characteristics of ports and links can be revealed by means of their shipping capacity and stability. This framework is used to analyze the spatial-temporal dynamics of the maritime network structure and shipping capacity weighted network dynamics for individual countries. Detailed descriptions are provided in the subsections that follow.

Construction of a Maritime Network
This section details construction of a time-varying maritime network based on AIS data. Latitude and longitude (location) information for each vessel is known using the AIS data. An example showing ships 1, 2, and 3 is shown in Figure 2a. The time-series locations of these vessels between ports can be viewed as trajectories. Specifically, trajectories exist among ports AE, EC, CD, and DF for ship 1, among ports FC, CA, AE, ED, and DB for ship 2, and among ports AF, FB, BA, AB, BG, GD, DF, and FG for ship 3. Thus, a time-varying maritime network between ports emerges by connecting port pair trajectories as linkages within a pre-specified time unit, such as days, months, seasons, years, or multi-years. Each linkage includes certain properties, such as voyage number and shipping capacity. Figure 2b shows attributes k1, k2, etc. plotted near its corresponding links. The maritime networks for sampled countries, including those of the 21st-century Maritime Silk Road (21C-MSR) and others, are illustrated in Figure 2c. The maritime network of one specific country is based on the criteria that the connections between ports inside the country and between the ports one inside the country and another located in other countries. The connection between the specific country and other countries can be summarized based on the connection between ports one inside the country and another located in other countries.

Maritime Network Dynamics
The global maritime network can be represented by , where V represents the node set and E represents the link set. This paper proposes a spatial-temporal approach for revealing the multi-component and multi-layer dynamics in the maritime network based on the characteristics of nodes, links, structure, and traffic flow.

Characteristics of Nodes and Links
Transport capacity and stability is very useful for authorities considering strategies for managing maritime traffic, and also for shipping companies to optimize shipping routes. For example, if a port has a highly dynamic transport capacity, this indicates that port authorities need to record and evaluate usage across different time periods in order to improve efficiency. Therefore, this paper

Maritime Network Dynamics
The global maritime network can be represented by G = (V, E), where V represents the node set and E represents the link set. This paper proposes a spatial-temporal approach for revealing the multi-component and multi-layer dynamics in the maritime network based on the characteristics of nodes, links, structure, and traffic flow.

Characteristics of Nodes and Links
Transport capacity and stability is very useful for authorities considering strategies for managing maritime traffic, and also for shipping companies to optimize shipping routes. For example, if a port has a highly dynamic transport capacity, this indicates that port authorities need to record and evaluate usage across different time periods in order to improve efficiency. Therefore, this paper uses the transport capacity and stability to characterize dynamics of nodes and links. The following describes transport capacity [22] for node V k or link L k in time T i : where CP (V k or L k ,T i ) represents the transport capacity in node V k or link L k in , and F (V k or L k ,T i ) n equal the vessel size, vessel number, and sail frequency in individual service n, respectively; and LogCP (V k or L k ,T i ) represents the logarithm of capacity for node V k or link L k in T i .
Capacity stability for nodes and links can be calculated using the model proposed by [36]. Firstly, the monthly traffic flow curves can be divided into segments using crest or trough, and each segment either monotonically increases, decreases, or remains unchanged. This is shown in Figure 3. The stability of one segment takes into account the trend from the start and end points of an individual segment as well as fluctuation along the segment. This can be characterized as follows: where CP s and CP e are the capacity at the beginning time T s and ending time T e of the segment, respectively;pis the number of sample points in SEG i ; ∆F j is the difference between the real transport capacity CP j and calculated trend value y j ; and ∆F represents the average value of ∆F j . Equation (2) describes the mean value of the differences between estimation based on the trend and the real capacity, and Equation (3) measures stability calculated by the trend change and the standard deviation of the differences between the estimated and real capacity. Thus, the stability of the link at T i can be derived from the stability of all segments as follows: where S (V k or L k ) represents the capacity stability of node V k or link L k ; m represents the segment number of node V k or link L k ; and ∆T (V k or L k ,SEG i ) represents the duration for the corresponding SEG i .
represents the logarithm of capacity for node k V or link k L in i T .
Capacity stability for nodes and links can be calculated using the model proposed by [36]. Firstly, the monthly traffic flow curves can be divided into segments using crest or trough, and each segment either monotonically increases, decreases, or remains unchanged. This is shown in Figure 3. The stability of one segment takes into account the trend from the start and end points of an individual segment as well as fluctuation along the segment. This can be characterized as follows: Equation (2) describes the mean value of the differences between estimation based on the trend and the real capacity, and Equation (3) measures stability calculated by the trend change and the standard deviation of the differences between the estimated and real capacity. Thus, the stability of the link at i T can be derived from the stability of all segments as follows: where (

Structure Changes
The timeline method can be employed to evaluate even-related network structure dynamics [10]; thus, this paper compared yearly changes of maritime network structure to reveal the differences between 21C-MSR before and after included by 21st-century Maritime Silk Road, as well as the differences between 21C-MSR and other countries. Consistent with work in this area, considered here is the yearly temporal granularity according to route regularity [37]. The change of one node from time i to i + 1 year can be calculated as follows: where d (T i ,T i+1 ) (v) is the changes in network structure contributed by node v, d (T i ) (v) means the degree of node v at T i ; d (T i+1 ) (v) represents the degree of node v (the count of nodes that have links with node v) at T i+1 ; ad j T i (v) represents the neighbors of node v (the collection of nodes that have links with node v) at T i ; ad j T i+1 (v) denotes the neighbors of node v at T i+1 ; V(Out Node) represents the collection of nodes in the maritime network at T i but not in the maritime network at T i+1 , namely as missing nodes; V(In Node) represents the collection of nodes not in the maritime network at T i , but rather in the maritime network at T i+1 , namely as new nodes; and V(Stable Node) represents the collection of nodes both in the maritime network at T i and T i+1 , namely as stable nodes. The change in network structure can be defined as follows: where σ (T i ,T i+1 ) is the maritime network structure changes from T i to T i+1 after normalization, including the changes for missing nodes after normalization , for new nodes after normalization , and for stable nodes after normalization represents the union of the node sets of maritime network at T i and T i+1 , and can be used to normalize the maritime network structure changes by reducing the differences derived from the network size; is the total changes for missing nodes from T i to T i+1 .

Weighted Structure Changes
The conventional timeline method can capture the yearly node-link connected structure changes. However, yearly maritime networks may have the same structures, and different transport capacity loaded on nodes and links. Transport capacity changes can reflect the efficiency of maritime transportation, and is very important to maritime policy development. Thus, the weighted structure changes are proposed to analyze both structure and flow changes.
The transport capacity is an important component in maritime network; thus, the transport capacity cannot be ignored. This paper also analyzes the transport flow evolution by means of the proposed capacity weighted timeline method. The transport flow change derived from one node from time i to i + 1 can be calculated as follows: Sensors 2019, 19, 4197 indicates the overlap of the capacity of node v between T i and T i+1 ; means the integration of the capacity of node v between T i and T i+1 . V(Out Node) represents the collection of nodes in the maritime network at T i , but not in the maritime network at T i+1 , namely as missing nodes; V(In Node) represents the collection of nodes not in the maritime network at T i , but rather in the maritime network at T i+1 , namely as new nodes; and V(Stable Node) represents the collection of nodes both in the maritime network at T i and T i+1 , namely as stable nodes. The transport flow changes σ (t i ,t i+1 ) from T i to T i+1 can be calculated as follows: is the transport flow change from T i to T i+1 after normalization, including the changes for missing nodes after normalization , for new nodes after , and for stable nodes after normalization ; CP(g T i ) ∪ CP(g T i+1 ) represents the integration of capacity between T i and T i+1 , and can be used to normalize the transport flow changes; ∀v∈V(STABLE) d (T i ,T i+1 ) (v) represents the total transport flow changes for stable nodes from T i to T i+1 ; ∀v∈V(IN) d (T i ,T i+1 ) (v) is the total transport flow changes for new nodes from T i to T i+1 ; and ∀v∈V(OUT) d (T i ,T i+1 ) (v) is the total transport flow changes for missing nodes from T i to T i+1 .

Study Area and Dataset
AIS data from January 1, 2013 and December 31, 2016 (available at: http://www.myships.com/ myships/ and http://www.shipfinder.com/, July 27, 2017) were employed to create an origin-destination (OD) dataset for vessels and connecting ports worldwide. The tanker vessels include transport crude oil, refined oil products, and other chemical oil products. The data categories for each ship are listed in Table 2. All AIS locations for each ship were simplified as a sequence of ports, according to dataset records.
The AIS is compulsory for most commercial ships through the International Convention for Maritime Safety, but loading rates and cargo amount are unavailable [22]. The global maritime network derived from AIS ship data in 2013, 2014, 2015, and 2016 is summarized and can be decomposed by bulk, container, and tanker types to reveal multi-layer dynamics.

Spatial-temporal Dynamics of Nodes and Links
The global shipping networks across the study period by bulk, container, and tanker capacity and stability for ports are displayed in Figure 5    with large bulk, container, and tanker capacity are higher than those in other countries, respectively. There are lower proportions for the ports in 21C-MSR that have low bulk, container, and tanker capacity. The differences between the proportions of ports with high bulk, container, and tanker capacity in 21C-MSR and other countries was narrowed down in 2015 and 2016 compared to 2014 as shown in Figure 6a, which may be related to that newly countries included in 21C-MSR in 2015 and 2016 have a number of ports with lower capacity. Figure 5b,d,f illustrate more ports in 21C-MSR have high changes in bulk, container, and tanker traffic flow in 2014. For example, the proportion of ports with less stable bulk traffic flow in 21C-MSR is higher than that in other countries, but the proportion of ports with more stable bulk flow is lower than that in other countries, the same as container and tanker flow. That indicates some of ports in 21C-MSR present highly dynamics in traffic flow compared to other countries in 2014. Furthermore, the differences between the proportions of ports with high bulk, container, and tanker flow changes in 21C-MSR and other countries was narrowed down in 2015 and 2016 compared to 2014 as shown in Figure 6b, which may be related to that newly countries included in 21C-MSR in 2015 and 2016 have a number of ports with lower flow dynamics.  The ports average capacity and stability in 21C-MSR before and after included by 21st-century Maritime Silk Road are illustrated in Figure 7. Most of the countries have higher average capacity after being included by 21C-MSR than before being included by 21C-MSR. That may be related to more frequent interaction between ports in 21C-MSR after being included by the 21st-century Maritime Silk Road. In more than half of the countries, there exist bigger traffic flow dynamics after being included by 21C-MSR. That indicates that the shipping interaction among ports changes after these countries are included by 21C-MSR. The ports average capacity and stability in 21C-MSR before and after included by 21st-century Maritime Silk Road are illustrated in Figure 7. Most of the countries have higher average capacity after being included by 21C-MSR than before being included by 21C-MSR. That may be related to more frequent interaction between ports in 21C-MSR after being included by the 21st-century Maritime Silk Road. In more than half of the countries, there exist bigger traffic flow dynamics after being included by 21C-MSR. That indicates that the shipping interaction among ports changes after these countries are included by 21C-MSR.         Table 3 illustrates that the links of capacity are continuously increasing across the study period in the bulk and tanker layer. There are more links with continuously increasing capacity in bulk maritime network of 21C-MSR than in other countries. The bulk and tanker links with continuously increasing capacity in 21C-MSR represent higher average increased capacity than in other counties. For example, the average increased capacity of bulk and tanker links in 21C-MSR are 3,125,407.29 Dead Weight Tonnage (DWT) and 1,696,798.74 DWT, respectively, whereas in other countries they are 1,243,252.48 DWT, and 1,408,144.76 DWT.  Table 4 illustrates the links of total ships capacity are continuously increasing across the study period in the container layer. There are more links with continuously increasing capacity in container maritime network of 21C-MSR than in other countries, and these container links in 21C-MSR represent higher average increased capacity than in other counties. For example, the average increased total ships capacity of container links in 21C-MSR is 3,524,387.11 DWT, whereas in other countries it is 2,320,540.44 DWT. 38 of 250 links with continuously increasing capacity in other countries are those connected with US and other countries (not considering Origin-Destination in the US), whereas 88 of 388 links with continuously increasing volume in 21C-MSR are those connected with CN and other countries (not considering Origin-Destination in CN). That indicates that CN has more links than the US with continuously increasing container capacity across the study period.  Figure 9 illustrates the spatial differentiation of links with evident flow dynamics (stability lower than 0.2) in the bulk, container, and tanker maritime network. As indicated in Figure 9a between other countries around the English Channel, the straits in Turkey, Gulf of Mexico, and Panama Canal, as well as the links between 21C-MSR around the Strait of Gibraltar and the Strait of Malacca appeared to have high dynamics in tanker maritime network in 2016. This spatial variation might be driven by the different supply-demand structure and trends for different types of cargo transportation among multiple routes. Additionally, there are continuously increasing links with

Spatial-temporal Dynamics of Maritime Network Structure
Spatial-temporal dynamics of maritime network structures of 21C-MSR and other countries are summarized in Figure 10. The nodes represent different countries with different degree in the maritime network, and the widths of the links are characterized through their increasing accumulative weights derived from total capacity after normalization (divided by the maximum capacity). There are some differences in the overall evolutionary patterns of 21C-MSR in the time periods from 2014 to 2016.  Although there are corresponding countries continuously included by 21C-MSR, there are still many weak connections between 21C-MSR. This indicates it will still take a long tie for 21C-MSR to enhance mutual cooperation in their maritime shipping industry. The highest shipping connection between other countries is relatively low in comparison to 21C-MSR, and the highest increasing accumulative weights is also lower than 21C-MSR, as indicated figure 10. The maritime network structure of other countries consist of two centralities, one for United States (US) connected with Canada (CA), Colombia (CO), Mexico (MX), and Panama (PA), and another for European countries (e.g., United Kingdom (GB), Germany (DE), the Netherlands (NL), Belgium (BE)). There are no significant changes for shipping connections among other countries between 2014 and 2016.
The maritime network can be decomposed into bulk, container, and tanker layers. The analysis will focus on the countries with network structure dynamics located in the top 20%. These countries carry out new business with additional ports (nodes) in 21C-MSR, and these new nodes contribute to the larger dynamics of the maritime network structure than the new nodes of other countries. However, these countries appear to reduce business with fewer ports in 21C-MSR, as missing nodes contribute to smaller dynamics of the maritime network structure than the missing nodes with other countries. This indicates that these countries exhibit evident dynamics in maritime network structures, especially for the shipping structure with 21C-MSR, which may be related to supply-demand shipping structure adjustment and carrying out new business with additional ports in 21C-MSR. It is obvious that there are more countries with a higher dynamic bulk, container, and tanker shipping network in 2016 than in 2014 and 2015. This is maybe related to the fact that the 21st-century Maritime Silk Road is still under construction and in the initial stages in 2014 and 2015, and the effectiveness of MSRI have been gradually presented since 2016. The maritime network structures with obvious dynamics include the bulk, container, and tanker of AU, MY, ID, and Japan (JP), bulk and container of CN, bulk and tanker of AE, bulk of TH, container of IT, PT, SG, and TR, and tanker of SA, as shown in Figure 11 (Note that "MSR_Sta" and "Other_Sta" represent the maritime network dynamics derived from the stable nodes with 21C-MSR countries and other countries, respectively; "MSR_In" and "Other_In" represent the maritime network dynamics derived from the new nodes with 21C-MSR countries and other countries, respectively; moreover, "MSR_Out" and "Other_Out" represent the maritime network dynamics derived from the missing nodes with 21C-MSR countries and other countries, respectively). In addition, JP presents high dynamics in all layers of maritime network structure although it wasn't included by 21C-MSR. Japan has held a skeptical attitude toward MSRI, and developed some policies and measures to maintain competitiveness in the international shipping industry.
contribute to smaller dynamics of the maritime network structure than the missing nodes with other countries. This indicates that these countries exhibit evident dynamics in maritime network structures, especially for the shipping structure with 21C-MSR, which may be related to supply-demand shipping structure adjustment and carrying out new business with additional ports in 21C-MSR. It is obvious that there are more countries with a higher dynamic bulk, container, and tanker shipping network in 2016 than in 2014 and 2015. This is maybe related to the fact that the 21st-century Maritime Silk Road is still under construction and in the initial stages in 2014 and 2015, and the effectiveness of MSRI have been gradually presented since 2016. The maritime network structures with obvious dynamics include the bulk, container, and tanker of AU, MY, ID, and Japan (JP), bulk and container of CN, bulk and tanker of AE, bulk of TH, container of IT, PT, SG, and TR, and tanker of SA, as shown in Figure 11 (Note that "MSR_Sta" and "Other_Sta" represent the maritime network dynamics derived from the stable nodes with 21C-MSR countries and other countries, respectively; "MSR_In" and "Other_In" represent the maritime network dynamics derived from the new nodes with 21C-MSR countries and other countries, respectively; moreover, "MSR_Out" and "Other_Out" represent the maritime network dynamics derived from the missing nodes with 21C-MSR countries and other countries, respectively). In addition, JP presents high dynamics in all layers of maritime network structure although it wasn't included by 21C-MSR. Japan has held a skeptical attitude toward MSRI, and developed some policies and measures to maintain competitiveness in the international shipping industry.

Spatial-temporal Dynamics of Traffic Flow Weighted Maritime Network Structure
The countries that have traffic flow weighted maritime network structure dynamics that ranked in the top 20% are illustrated in Figure 12. Additionally, these countries carry out new business with additional ports (nodes) in 21C-MSR, and these new nodes contribute larger dynamics of the traffic flow weighted maritime network structure than the new nodes with other countries. However, these countries close business with fewer ports in 21C-MSR, and these missing nodes contribute smaller dynamics of the traffic flow weighted maritime network structure than the missing nodes with other

Spatial-temporal Dynamics of Traffic Flow Weighted Maritime Network Structure
The countries that have traffic flow weighted maritime network structure dynamics that ranked in the top 20% are illustrated in Figure 12. Additionally, these countries carry out new business with additional ports (nodes) in 21C-MSR, and these new nodes contribute larger dynamics of the traffic flow weighted maritime network structure than the new nodes with other countries. However, these countries close business with fewer ports in 21C-MSR, and these missing nodes contribute smaller dynamics of the traffic flow weighted maritime network structure than the missing nodes with other countries. This indicates their traffic flow weighted maritime network structures exhibit evident dynamics, especially for the shipping structure and capacity with 21C-MSR, which may be correlated with carrying out additional business with 21C-MSR. The traffic flow weighted maritime network structures emerging obvious dynamics include the bulk, container, and tanker of AU, MY, ID, and JP, bulk and container of CN, bulk and tanker of AE, container of EG, IT, PT, SG, and TR, and tanker of SA.
The bulk and container of CN, bulk container, and tanker of AU, MY, ID, and JP, bulk and tanker of AE, container of IT, PT, SG, and TR, and tanker of SA all exhibit evident dynamics in both the maritime network structure and traffic flow weighted maritime network structure, as illustrated in Figures 11 and 12. This suggests that these countries built corresponding new shipping relationships with the ports in 21C-MSR, and these new linkages carried a significant amount of traffic flow between 2013 and 2016. The bulk of TH exhibit evident dynamics in the maritime network structure, but small dynamics in the traffic flow weighted maritime network structure. Therefore, TH built new bulk shipping linkages with numerous ports in 21C-MSR, but these new linkages carry only a small part of the traffic flow. The container of EG exhibits small dynamics in the maritime network structure but evident dynamics in the traffic flow weighted maritime network structure, which indicates that EG builds new container relationships with certain ports in 21C-MSR, and these new linkages carry significant traffic flow. additional ports (nodes) in 21C-MSR, and these new nodes contribute larger dynamics of the traffic flow weighted maritime network structure than the new nodes with other countries. However, these countries close business with fewer ports in 21C-MSR, and these missing nodes contribute smaller dynamics of the traffic flow weighted maritime network structure than the missing nodes with other countries. This indicates their traffic flow weighted maritime network structures exhibit evident dynamics, especially for the shipping structure and capacity with 21C-MSR, which may be correlated with carrying out additional business with 21C-MSR. The traffic flow weighted maritime network structures emerging obvious dynamics include the bulk, container, and tanker of AU, MY, ID, and JP, bulk and container of CN, bulk and tanker of AE, container of EG, IT, PT, SG, and TR, and tanker of SA. The bulk and container of CN, bulk container, and tanker of AU, MY, ID, and JP, bulk and tanker of AE, container of IT, PT, SG, and TR, and tanker of SA all exhibit evident dynamics in both the maritime network structure and traffic flow weighted maritime network structure, as illustrated in Figures 11 and 12. This suggests that these countries built corresponding new shipping relationships with the ports in 21C-MSR, and these new linkages carried a significant amount of traffic flow between 2013 and 2016. The bulk of TH exhibit evident dynamics in the maritime network structure, but small dynamics in the traffic flow weighted maritime network structure. Therefore, TH built new bulk shipping linkages with numerous ports in 21C-MSR, but these new linkages carry only a small part of the traffic flow. The container of EG exhibits small dynamics in the maritime network structure but evident dynamics in the traffic flow weighted maritime network structure, which indicates that EG builds new container relationships with certain ports in 21C-MSR, and these new linkages carry significant traffic flow.

Conclusion
Understanding multi-layer maritime network dynamics is an initial step to predict change trend [1]. In this study, we have proposed a spatial-temporal framework to explore multi-layer maritime network dynamics, implemented the proposed framework using complex network theory and traffic flow stability, and investigated the spatial-temporal dynamics of nodes, links, network structure, and traffic flow between 2013 and 2016. The results are as follows. First, there are more ports in 21C-MSR countries that have high bulk, container, and tanker capacity and high changes in bulk, container, and tanker traffic flow between 2013 and 2016. This indicates in some ports in

Conclusions
Understanding multi-layer maritime network dynamics is an initial step to predict change trend [1]. In this study, we have proposed a spatial-temporal framework to explore multi-layer maritime network dynamics, implemented the proposed framework using complex network theory and traffic flow stability, and investigated the spatial-temporal dynamics of nodes, links, network structure, and traffic flow between 2013 and 2016. The results are as follows. First, there are more ports in 21C-MSR countries that have high bulk, container, and tanker capacity and high changes in bulk, container, and tanker traffic flow between 2013 and 2016. This indicates in some ports in 21C-MSR countries, there exists a high shipping dynamic between 2013 and 2016. Additionally, most of the countries have a higher ports average capacity and stability after being included in 21C-MSR, which may be related to more frequent interaction between ports inside 21C-MSR. Second, there are more links with continuously increasing transport amount in all layers of maritime network in 21C-MSR countries compared to other countries, and these bulk, container, and tanker links in 21C-MSR countries present a higher average increased capacity compared to other counties. This illustrates that some of the linkages between 21C-MSR countries present an incremental shipping capacity between 2013 and 2016. Third, there are more links that have a big flow dynamics in bulk, container, and tanker maritime networks between 21C-MSR countries than between other countries. This indicates there are more links under the 21C-MSR geographic scope existing flow variation between 2013 and 2016. Fourth, the global maritime network dynamics exhibit geographical and spatial variations. For example, there are fewer container trade linkages with high dynamics between 21C-MSR countries around the Strait of Malacca than bulk linkages. Finally, certain countries (CN, SG, AU, and AE) have established new corresponding shipping relationships with some ports in 21C-MSR, and these new linkages carry substantial traffic flow between 2013 and 2016.
Although this research is investigating the spatiotemporal changes of the maritime network, extension may be possible. Geographical heuristics, place, and ship interaction dynamics in maritime transportation management and planning may be informative as would accounting for national shipping transportation strategies taking geopolitics into consideration. This research nevertheless provides policy insights. First, incremental transport amount of some ports and links in 21C-MSR countries between 2013 and 2016 may be relative to the shipping strategy adjustment. However, it is still premature whether 21C-MSR countries will become more competitive than other countries, and possibly hold a better position in the maritime trade. This indicates that the maritime transportation infrastructure, operational efficiency, and shipping routes for 21C-MSR countries can be further improved. The enhanced capacity for some ports maybe have potential effects for the nearby ports due to competitiveness, thus, maritime shipping policy development will need to account for the possibility of benefits conflicts among some ports in 21C-MSR countries. Second, maritime network dynamics are very useful for guiding global maritime shipping network improvements towards better utilization, including reducing friction in maritime trade and network shockwave both in 21C-MSR and other countries. Third, global maritime network dynamics provide some guidance for policy makers and stakeholders in decisions making as complicated maritime transportation markets reflect important structure and traffic flow evolution. For example, the ports or links with incremental transport capacity will cause transit time changes for shipping companies, thus, the adjustment is needed to maximize the benefits.
There are some limitations that should be noted. First, we only examine the global maritime network dynamics between 2013 and 2016, the difference between 21C-MSR and other countries, and the differences between before and after included by 21C-MSR across study period. The differences between before and after the MSRI was announced can be further explored if the long run AIS data are available. The spatial-temporal dynamics can be further evaluated if the complete data source is accessible. Second, this paper only focused on the multi-layer maritime network changes, and some uncertainty and challenge remains regarding implications for the actual impact of the MSRI on the maritime network. Third, there are some differences between shipping capacity and actual cargo amount owing to unknown cargo amount and loading rate. Fourth, this paper could not access the detail classification of goods and fixed importing and exporting countries for different products (e.g., crude oil and refined oil), which can be fulfilled in the future research.
In the future, the proposed approach could be enhanced by combining comprehensive information on maritime natural resource utilization data, social-cultural factors, and economic activities in order to provide a powerful and mutually consistent explanation for the manner in which geopolitical initiatives have different impacts on maritime shipping planning and management [39]. Furthermore, future research should explore a deeper understanding of the mechanisms driving the structural and spatial and regional dynamics in global maritime networks, analyzing the urban transportation contributed by the geopolitical policy, and connecting these changes to the corresponding maritime network types.