Red Sea crisis impacts on maritime shipping networks

Abstract

The Red Sea crisis (RSC) is an ongoing military conflict affecting maritime trade. Based on automatic identification system data, this study analyzed the changes in vessel behavior in the Red Sea using the seasonal decomposition method. The number and speed of vessels decreased significantly after the RSC, and many vessels chose to sail away from the Red Sea rather than anchor at its shores. Additionally, the impact of the RSC on maritime shipping networks was analyzed based on metrics in complex networks. Although shipping network structures were changed by the RSC, they still exhibited resilience. Finally, this study analyzed the RSC future impacts on maritime shipping networks based on node attacks and the Markov method. Because of their resilience, only ports in the top 20 % needed to be considered. This study provides valuable insights for maritime transportation authorities to formulate port strategies and route layouts in response to the RSC.

1. Introduction

The Red Sea Crisis (RSC) is an ongoing military conflict that began on October 19, 2023. The Red Sea occupies a pivotal geographical position and serves as a natural link between the continents of Asia, Africa, and Europe [[1], [2], [3], [4]]. Its strategic importance is further underscored by numerous critical trade routes that traverse its waters, particularly those connecting Middle Eastern oil-producing nations with Europe and Asia. These routes facilitate the flow of large volumes of energy products, including oil and natural gas, and a diverse array of raw materials and finished goods, thereby notably contributing to global trade and economic prosperity [5,6].

The maritime shipping network is the most important form of transportation in international trade, accounting for ∼90 % of global trade volume [7,8]. With increasing globalization of trade, the maritime industry is gradually expanding [[9], [10], [11]]. The numerous events that have occurred during the RSC have the potential to markedly impact vessel traffic and movement in the Red Sea and thus, maritime trade. For example, on November 14th, 2023, the Houthi rebels declared their intention to commence strikes against vessels on the Red Sea deemed to be associated with Israel [12]. On December 19th, 2023, they reiterated their commitment by vowing to prevent ships bound to Israel from traversing the waterways. Therefore, researching and quantitatively exploring the impact of the RSC on vessel behavior and maritime shipping networks is important for optimizing the global shipping port layout, building a resilient global shipping logistics network with strong anti-risk interference capabilities, and evaluating dynamic trends in global shipping freight rates. Currently, there are two important research challenges in analyzing the impact of the RSC from the perspective of maritime shipping networks.

Challenge (1): How can the RSC impact on vessel behavior and maritime shipping networks be explored? As an important link connecting regions worldwide, exploring this transmitted from the Red Sea to global maritime shipping networks can help estimate the crisis losses.

Challenge (2): How can the future impact on shipping networks evolve with the RSC? Research on the RSC is ongoing, and exploring future impact patterns on shipping networks is helpful for port planning, route adjustments, and freight rate estimations.

Automatic identification systems (AIS) [[13], [14], [15], [16]] can help address these challenges. The AIS records the time, position, speed over ground, course over ground, bow, ship type, and origin and destination (OD) of the ship during a voyage. Therefore AIS data could support research at different spatial-temporal scales [17,18]. Thus, this study aimed to analyze the RSC impact on maritime shipping networks from the perspective of vessel behavior and maritime shipping networks. The contributions of this study are as follows. First, utilizing AIS data, this study examined the changes and trends in vessel behavior with RSC evolution, aiming to demonstrate the maritime influence of the RSC. Subsequently, we analyzed the structural and parameter changes in maritime shipping networks before and after the RSC to explore specific impacts. Finally, through simulated node attacks, we investigated potential future changes in maritime shipping networks as the RSC continues to evolve. These contributions provide valuable insights for maritime transportation authorities to formulate port strategies and route layouts in response to the RSC.

The remainder of this paper is organized as follows. Section 2 introduces the data and methods used; section 3 presents the impact and future trends of the RSC on vessel behavior and maritime shipping networks; section 4 discusses suggestions for future research; and section 5 summarizes the contributions and limitations of this study.

2. Data and materials

2.1. AIS data and processing

The data used here comprised the global AIS data collected during the period surrounding the onset of the RSC, specifically from 1st October to December 31, 2023. The data were acquired from Orbcomm and Spire Global satellite data. According to maritime regulations, the AIS regularly broadcasts vessel positions and additional information, such as mobile service identification (MMSI)—a unique identification number for each vessel—longitude, latitude, speed over ground (SOG), course over ground (COG), OD, and status to nearby ships and coastal stations, which can reflect vessel behavior information. AIS data can be used to extract the trajectory of a ship and show its route, speed changes, and other navigation-related information. This study used dynamic information (SOG, COG, MMSI, and status) to assess vessel behavior and static information (OD) to construct maritime shipping networks.

Before using raw AIS data from the Red Sea to assess vessel behavior, data quality needed to be controlled and improved [19]. To control AIS trajectory data, the following pre-processing steps were performed.

• (1)Filtering the trajectory locations to the Red Sea. This study focused on vessel behavior changes in the Red Sea during armed conflicts; thus, raw AIS data were filtered to a geographical range of 12.5°–28°N and 24° to 38°E using the longitude and latitude fields. To calculate the distance, filtered AIS data were projected onto the UTM Zone 38N/WGS 1984 coordinate system.
• (2)Excluding invalid and missing data. SOG, COG, MMSI, and positioning time are important indices used to assess vessel behavior. Therefore, trajectory locations with invalid values or missing SOG, COG, MMSI, and positioning-time data were excluded.
• (3)Separating trajectory data. This study used the MMSI and chronological order of positioning timestamps from the filtered AIS data to separate individual vessel trajectories within the Red Sea. Notably, duplicate trajectory data with the same MMSI were removed. If the duplicated trajectory data had identical content, one trajectory was retained and the duplicates removed. If different trajectories had identical MMSIs over the same period, all trajectories were withdrawn because it is difficult to identify the authentic trajectory [20].
• (4)Cleaning the trajectory data. After separating the trajectory data, problems with trajectory gaps and slow reporting rates were encountered. The false-positive rate of destination port information in AIS data can be as high as 40 % [21,22]; thus, it was necessary to remove outliers from the trajectory data. If the length of the trajectory segment was greater than $d_{\mathit{max}}$, it could be calculated using Eq. (1) with the abnormal track point removed from the trajectory [23].

$$d_{\mathit{max}}=\mathit{max}\left({SOG}_t,{SOG}_{t-1}\right)\times t$$ (1)

Trajectory gaps can be identified by sudden changes in latitude and longitude, which cause unreasonable distances within the trajectory. Based on Eq.(1), the unreasonable distances and abnormal track points were removed based on the SOG and time $t$.

(5) Analyzing vessel behavior changes. Subsequently, this study used the SOG, COG, locations, and timestamps to analyze vessel behavior changes [24]. By calculating the average SOG, average COG, and total number of ships in the Red Sea daily, we obtained daily vessel behavior. Based on this, a time series of vessel behavior changes—the change in average SOG, average COG, and total number of vessels—can be derived.

2.2. Measurement of vessel behavior trends

This study used seasonal decomposition [[25], [26], [27]] to calculate vessel behavior trends with RSC evolution. Daily vessel behavior during the RSC was calculated and viewed as a time series. Each time series ($y_t$) was decomposed by additive decomposition into trend ($T_t$), seasonal ($S_t$), and residual ($R_t$) decomposition components as shown in Eq. (2) and realized by utilizing a seasonal decomposition algorithm [27,28]. $T_t$ was used to assess vessel behavior trends.

$$y_t=T_t+S_t+R_t$$ (2)

The trend component reflects the overall development trend of the time-series data over a longer period. It represents the long-term pattern of data change over time. The seasonal component reflects cyclical fluctuations in time-series data with a fixed cycle length. This study was conducted for 7 days.

To measure vessel behavior trends, a discriminant function ($F_T$) was used, as shown in Eqs. (3), (4). The vessel behavior time series with $F_T$ or $F_S$ close to 1 shows a strong trend or weak seasonality. The vessel behavior time series with $F_T$ or $F_S$ close to zero shows a weak trend or seasonality.

$$F_T=\{\begin{array}{c}0,\frac{Var\left(R_t\right)}{Var\left(T_t+R_t\right)}\ge 1\\ 1-\frac{Var\left(R_t\right)}{Var\left(T_t+R_t\right)},\frac{Var\left(R_t\right)}{Var\left(T_t+R_t\right)}⟨1<\end{array}$$ (3)

$$F_S=\{\begin{array}{c}0,\frac{Var\left(R_t\right)}{Var\left(S_t+R_t\right)}\ge 1\\ 1-\frac{Var\left(R_t\right)}{Var\left(S_t+R_t\right)},\frac{Var\left(R_t\right)}{Var\left(S_t+R_t\right)}<1\end{array}$$ (4)

where $T_t$, $S_t\text{,}$ and $R_t$ are the trend, seasonal, and residual decomposition components of $y_t$, respectively. $Var(·)$ is the variance function.

2.3. Maritime shipping network construction and assessment

This study used global AIS data to construct a global maritime shipping network using L-space representation. The original and destination ports were recorded in the raw AIS; therefore, at the port level, this study viewed the port as a node and the port-to-port connections by vessels as edges. This study aggregated port-level networks based on the countries to which the ports belong, thus forming a national-level maritime shipping network.

After constructing the maritime shipping network, this study used the number of nodes, number of edges, in-degree, out-degree, degree, betweenness centrality, average path length, network efficiency, and maximally connected subgraphs to assess the maritime shipping network. The meaning and formula for each metric [[29], [30], [31]] are shown in Table 1. By comparing these metrics before and after the RSC, its impact on the maritime shipping network could be assessed.

Table 1.

Maritime shipping network performance metrics.

Parameters	Meaning	Formula
Nodes	The number of ports (or countries) in the network.	–
Edges	Intra-network connectivity or accessibility of ports (countries).	–
In-degree	The number of edges pointing to a port (country) in a directed network.	$\mathit{i}{\mathit{d}}_{\mathit{i}}=\sum _{\mathit{j}}^{\mathit{N}}{\mathit{x}}_{\mathit{j},\mathit{i}}$ (5) where $\mathit{N}$ is the number of nodes; $\mathit{i}{\mathit{d}}_{\mathit{i}}$ is the in-degree of the port(country) $\mathit{i}$; ${\mathit{x}}_{\mathit{j},\mathit{i}}=\mathbf{0}$ and ${\mathit{x}}_{\mathit{j},\mathit{i}}=\mathbf{1}$ denotes the existence and non-existence of a directed edge from node $\mathit{j}$ to node $\mathit{i}$
Out-degree	The number of edges pointing to a port (country) originating from a directed network.	$\mathit{o}{\mathit{d}}_{\mathit{i}}=\sum _{\mathit{j}}^{\mathit{N}}{\mathit{x}}_{\mathit{i},\mathit{j}}$ (6) where $\mathit{N}$ is the number of nodes; $\mathit{i}{\mathit{d}}_{\mathit{i}}$ is the in-degree of the port (country) $\mathit{i}$; ${\mathit{x}}_{\mathit{i},\mathit{j}}=\mathbf{0}$ and ${\mathit{x}}_{\mathit{i},\mathit{j}}=\mathbf{1}$ denotes the existence and non-existence of a directed edge from node $\mathit{i}$ to node $\mathit{j}$
Degree	The number of all edges connected by a port (country).	${\mathit{d}}_{\mathit{i}}=\sum _{\mathit{j}}^{\mathit{N}}{\mathit{x}}_{\mathit{i},\mathit{j}}+\sum _{\mathit{j}}^{\mathit{N}}{\mathit{x}}_{\mathit{j},\mathit{i}}$ (7) where ${\mathit{d}}_{\mathit{i}}$ is the degree of the port(country) $\mathit{i}$.
Betweenness centrality (BC)	The number of times that the shortest path passes through the node when all nodes are connected. BC measures the influence of a node within a network.	${\mathit{B}\mathit{C}}_{\mathit{i}}=\frac{\mathbf{1}}{\left(\mathit{N}-\mathbf{1}\right)\left(\mathit{N}-\mathbf{2}\right)}\sum _{\mathit{j}\ne \mathit{i}\ne \mathit{k}}\frac{{\mathit{\eta }}_{\mathit{i}\mathit{j}\mathit{k}}}{{\mathit{\eta }}_{\mathit{j}\mathit{k}}}$ (8) where $\mathit{N}$ is the number of nodes, ${\mathit{\eta }}_{\mathit{i}\mathit{j}\mathit{k}}$ is the number of shortest paths between node $\mathit{j}$ and $\mathit{k}$ that pass through node $\mathit{i}$. ${\mathit{\eta }}_{\mathit{j}\mathit{k}}$ is the number of shortest paths between node $\mathit{j}$ and node $\mathit{k}$.
Average path length (APL)	The shortest path represents the path with the least number of edges between nodes.	$\mathit{A}\mathit{P}\mathit{L}=\frac{\mathbf{1}}{\mathbf{2}\mathit{N}\left(\mathit{N}+\mathbf{1}\right)}\sum _{\mathit{i}\ge \mathit{j}}{\mathit{d}}_{\mathit{i},\mathit{j}}$ (9) where $\mathit{N}$ is the number of nodes; ${\mathit{d}}_{\mathit{i},\mathit{j}}$ is the number of the edges in shortest paths between node $\mathit{i}$ and node $\mathit{j}$. In this study, the APL has no unit.
Network efficiency (E)	The sum of the reciprocals of the shortest path lengths between all pairs of nodes. E is commonly used to quantify the efficiency of information exchange between nodes in a network.	$\mathit{E}=\frac{\mathbf{1}}{\mathit{N}\left(\mathit{N}-\mathbf{1}\right)}\sum _{\mathit{i}=\mathbf{1}}^{\mathit{n}}\sum _{\mathit{j}=\mathbf{1},\mathit{j}\ne \mathit{i}}^{\mathit{n}}\frac{\mathbf{1}}{{\mathit{d}}_{\mathit{i},\mathit{j}}}$ (10) where $\mathit{N}$ is the number of nodes; ${\mathit{d}}_{\mathit{i},\mathit{j}}$ is the number of the edges in shortest paths between node $\mathit{i}$ and node $\mathit{j}$. In this study, E has no unit.
Maximal connected subgraph (S)	The ratio of the number of nodes in the largest connected subgraph to the total number of network nodes. S could reflect the damage degree after attacking.	$\mathit{S}=\frac{\mathit{N}\left({\mathit{G}}^{\ast }\right)}{\mathit{N}\left(\mathit{G}\right)}$ (11) where $\mathit{G}$ is the network; ${\mathit{G}}^{\ast }$ is the largest connected subgraph to $\mathit{G}$; $\mathit{N}(·)$ is the nodes number of a network.

2.4. Future impact analysis method based on node attack

Assuming the RSC remains ongoing, this study utilized degree centrality (DC) or betweenness centrality (BC) to assess the future impact of the RSC on maritime shipping networks.

• (1)Filtering ports in the Red Sea

The RSC is a local geopolitical conflict rather than a global explosive crisis. Therefore, based on global port data, we filtered out ports belonging to the Red Sea using the latitude and longitude fields (Fig. 1).

• (2)Attacking nodes with high DC/BC

Fig. 1.

Port dataset used in this study.

For each port, the DC and BC values of the global shipping network were calculated using Eqs. (7) and (8). The ports were sorted in descending order based on their DC/BC values. In the RSC, the failure of ports with high DC or BC values represents two possible future development trends. Ports with high DC values have numerous connections with other ports. Ports with high BC values suggest a greater number of transit tasks undertaken. Given that the RSC was an intentional attack event rather than a random natural disaster, these two attack models (BC and DC attacks) were devised to envision potential future port attack scenarios in the context of the RSC. Thus, this study focused on eliminating nodes that possess high BC (BC attacks) and DC (DC attacks) values, and explored maritime shipping network changes during the development of the RSC by varying the proportion of attacked nodes.

3. Results

3.1. RSC impacts on vessel behavior

On November 14th, 2023, the Houthi militia announced that it would begin striking vessels associated with Israel in the Red Sea. Therefore, this study identified this date as the starting point of the RSC. Fig. 2 shows the sailing trajectories of eight cargo vessel types—liquefied natural gas (LNG), liquefied petroleum gas (LPG), containers, grains, bauxite, iron ore, copper ore, and oil—in the Red Sea before and after the RSC. Following the RSC outbreak, there was a decrease in the tracks of various types of vessels (Fig. 2). LNG, LPG, container, grain, bauxite, iron ore, copper ore, and oil vessels decreased by 31.13, 28.85, 30.00, 29.41, 60.00, 35.74, 55.56, and 30.13 % respectively. To quantitatively assess the RSC impact on vessel behavior, we calculated mean value changes in the number of vessels (N), COG, and SOG for various vessel types as the RSC evolved. From Fig. 3, qualitative trends in vessel behavior change can be identified. Table 2 quantitatively assesses the trend and seasonality degree of various types of vessels within a weekly circle. In total, the number of vessels decreased by 33.69 % after the RSC.

Fig. 2.

Ship trajectory visualizations in the Red Sea. Panel (a)–(h) shows the trajectories of LNG, LPG, container, grain, bauxite, iron Ore, copper ore, and oil vessels, respectively in the Red Sea. Using the RSC commencement as the dividing line, the trajectories are marked as black and yellow for before and after, respectively.

Fig. 3.

Vessel behavior changes according to vessel number, COG, and SOG. Panel (a) statistics of all vessel changes. Panel (b)–(i) statistics of LNG, LPG, container, grain, bauxite, iron ore, copper ore, and oil vessel changes, respectively. The trend of changes is calculated via seasonal decomposition.

Table 2.

Vessel behavior trend measurements.

Type	${\mathit{F}}_{\mathit{T},\mathit{N}}$	${\mathit{F}}_{\mathit{T},\mathit{C}\mathit{O}\mathit{G}}$	${\mathit{F}}_{\mathit{T},\mathit{S}\mathit{O}\mathit{G}}$	${\mathit{F}}_{\mathit{S},\mathit{N}}$	${\mathit{F}}_{\mathit{S},\mathit{C}\mathit{O}\mathit{G}}$	${\mathit{F}}_{\mathit{S},\mathit{S}\mathit{O}\mathit{G}}$
All vessel	0.7262	0.5354	0.7211	0.0650	0.1115	0.1186
LNG	0.6039	0.4509	0.3927	0.0483	0.0758	0.1085
LPG	0.9003	0.4227	0.4263	0.2374	0.0846	0.0777
Container	0.8277	0.4045	0.5395	0.1089	0.0365	0.1610
Grain	0.7665	0.6148	0.5817	0.0504	0.0485	0.0642
Bauxite	0.7894	0.8340	0.7400	0.0828	0.2254	0.2353
Iron Ore	0.6319	0.5177	0.5827	0.0772	0.0526	0.0314
Copper Ore	0.7852	0.5998	0.6262	0.0606	0.0754	0.0884
Oil	0.7645	0.4627	0.4093	0.0443	0.1034	0.0685

As shown in Table 2, $F_T$ was always higher than $F_S$ in terms of COG, SOG, and the number of each vessel type. Thus, it can be concluded that the degree of the trend of each vessel type was stronger than its seasonality. Meanwhile, $F_s$ was always lower than 0.25 and mostly close to 0, indicating that changes in vessel behavior are primarily driven by random fluctuations rather than periodic variations.

From the perspective of vessels numbers, there was a sharp decline during the week after November 14th, 2023, followed by a gradual increase until a sudden drop occurred in mid-December. After the RSC, the daily number of vessels in the Red Sea decreased from ∼400 to 350. In Table 2, the $F_T$ of the number of vessels in the Red Sea was 0.73, showing a notable decrease in the number of various types of vessels with the intensification of the RSC. Among all vessels, the decreasing trend in LPG vessels and containers was the most significant, with $F_T$ reaching 0.90 and 0.83, respectively.

The SOG of all vessels was 8.15 ± 0.35 knots before the RSC. During the week following November 14th, 2023, the SOG decreased to 7.62 ± 0.35 knots and then gradually increased to 8.65 ± 0.30 knots in mid-December. In the final week of December 2023, the SOG suddenly decreased to 6.95 ± 0.19 knots. In Table 2, the SOG $F_T$ in all vessels was 0.72, implying that the SOG behavior of all vessels is in line with the above change pattern as the RSC evolved. It is noteworthy that when analyzing each type of vessel individually, the trend does not appear to be strongly consistent. $F_T$ ranging from 0.39 to 0.74 indicates that each type of vessel did not fully adhere to this trend.

The COG of all vessels was 221.01 ± 4.62° (from northwest to southeast). During the week following November 14th, 2023, the COG changed to 211.28 ± 4.44°. After mid-December, the COG changed to 233.50 ± 2.95° then returned to 215.56 ± 5.67°. The COG trend was related to the number of vessels. Based on the geographical conditions of the Red Sea in Fig. 2 and the COG trend, many vessels chose to sail away from the Red Sea rather than anchoring at its shores. In Table 2, the SOG $F_T$ in all vessels was 0.54, and the $F_T$ of all vessel types was over 0.50, which implies that the COG trend shown in Fig. 3 is general for all vessels as the RSC evolved.

In summary, the RSC reduced the number of vessels in the Red Sea and altered their behavior (i.e., COG and SOG). Consequently, the RSC had a significant impact on maritime trade, as analyzed in Sections 3.2, 3.3.

3.2. RSC impacts on maritime shipping networks

To construct the maritime shipping network related to the Red Sea, this study viewed the port as the node and the port-to-port connections by vessels as the edges at the port level before and after the RSC (Fig. 4a and b). This study aggregated the port-level networks based on the countries to which the ports belong, forming a national maritime shipping network (Fig. 4c and d). Based on the constructed maritime shipping network, this study used the number of nodes, edges, in-degree, out-degree, degree, BC, and average path length (APL) to assess the maritime shipping network. The changes in the evaluation metrics of this network before and after the RSC are shown in Table 3, Table 4, respectively. There were notable changes in all indicators, analyzed in detail later. To strengthen the conclusions of this study, a random geometric graph (RGG), with the same number of nodes and edges, was constructed for comparative analysis (Table 5). From the RGG comparative analysis and the constructed maritime shipping network, this study considered that the observed network is not randomly structured and changed, but exhibits a specific organizational pattern or regularity. These changes warrant further analysis.

Fig. 4.

Maritime shipping network before and after the RSC. Panel (a) and (b) show the maritime shipping network at the port level. The edge size in both panels is related to the vessel number. Panel (c) and (d) show the maritime shipping network at the national level. The edge color in Panel (c) and (d) is related to the vessel number.

Table 3.

Change in complex network parameters before and after the RSC at the port scale.

Network parameters		Before RSC	After RSC	Variation
Nodes	–	1178	1241	5.35 %
Edges	–	5818	6765	16.28 %
Degree	Max	289	360	24.57 %
	Mean	9.8778	10.9025	10.37 %
In-degree	Max	149	181	21.48 %
	Mean	4.9389	5.4212	9.77 %
Out-degree	Max	140	179	27.86 %
	Mean	4.9389	5.4512	10.37 %
BC	Max	0.1884	0.1893	0.48 %
	Mean	0.0018	0.0019	5.56 %
APL	–	0.9916	1.3698	38.14 %

Table 4.

Change in complex network parameters before and after the RSC at the national scale.

Network parameters		Before RSC	After RSC	Variation
Nodes	–	139	137	−1.44 %
Edges	–	1725	1547	−10.32 %
Degree	Max	114	120	5.26 %
	Mean	24.8201	22.5839	−9.01 %
In-degree	Max	55	58	5.45 %
	Mean	12.41	11.292	−9.01 %
Out-degree	Max	59	62	5.08 %
	Mean	12.41	11.292	−9.01 %
BC	Max	0.1041	0.1721	65.32 %
	Mean	0.009	0.0115	27.78 %
APL	–	1.3258	2.4908	87.87 %

Table 5.

RGG metrics with the same nodes and edges.

Network parameters		Before RSC (port scale)	After RSC (port scale)	Before RSC (national scale)	After RSC (national scale)
Nodes	–	1178	1241	139	137
Edges	–	5818	6765	1725	1547
Degree	Max	21	24	37	35
	Mean	9.8778	10.9025	24.8201	22.5839
In-degree	Max	19	19	31	25
	Mean	4.9389	5.4512	12.4101	11.2920
Out-degree	Max	19	20	28	27
	Mean	4.9389	5.4512	12.4101	11.2920
BC	Max	0.0007	0.0011	0.0010	0.0113
	Mean	0.0001	0.0001	0.0030	0.0029
APL	–	7.2582	7.0739	1.8298	1.8498

3.2.1. RSC impacts at the port scale

At the port scale (Table 3), the number of nodes in the network increased from 1178 to 1241, representing a 5.35 % growth, which means that new ports were added to the maritime shipping network transportation with the evolution of the RSC. The number of edges, which represent connections between nodes, increased from 5818 to 6765, with a marked growth of 16.28 %. This increase indicates tighter connectivity within the maritime shipping network, which enhances its overall connectedness.

The maximum degree increased from 289 to 360 (24.57 %) suggesting the emergence of highly connected nodes likely plays a crucial role within the network. Meanwhile, the mean degree also increased from 9.87 to 10.90 (10.37 %), indicating that nodes are now averaging more connections. If the constructed maritime shipping network is viewed as a directed network, both the maximum in-degree and out-degree increase, thereby highlighting the improved ability of certain nodes to receive and transmit information following the RSC. The average in-degree and out-degree also showed similar upward trends.

BC measures the number of shortest paths between nodes that must pass through a given node. Whereas the maximum BC showed a slight increase, the mean BC increased by 5.56 %, indicating that nodes generally play a more critical role in information transfer paths. APL represents the average length of the shortest paths between all the node pairs in the maritime shipping network, and it increased from 0.99 to 1.37 (38.14 %). This suggests a decline in information transmission efficiency within the network, as the nodes now require longer paths to reach each other. This could be due to the expansion of the network or structural changes.

3.2.2. RSC impacts at the national scale

At the national scale (Table 4), the number of maritime shipping network nodes decreased negligibly from 139 to 137. However, the number of edges changed substantially. Prior to the RSC, there were 1725 edges, indicating robust connectivity; however, after the RSC, this number declined 10.32 % to 1547 edges. This reduction suggests a decrease in the overall connectivity of the maritime shipping network, potentially affecting the transportation efficiency of goods and materials on a national scale.

The maximum degree increased from 114 to 120 (5.26 %), suggesting that a few key countries gained more connections, potentially becoming hubs of network activity. Conversely, the average degree decreased by 9.01 %, reflecting an overall reduction in edges and suggesting that ports are now averaging fewer connections. When considering the network as directed, the maximum in-degree and out-degree increased moderately by 5.45 and 5.08 %, respectively. However, both the mean in-degree and out-degree decreased by 9.01 %, following a general trend of reduced connectivity.

BC exhibited notable changes. The maximum BC increased considerably by 65.32 %, indicating that one or more countries became more central to the network, playing a pivotal role in facilitating the movement of goods during the RSC. Similarly, the mean BC increased by 27.78 %, reflecting the increasing importance of ports in maintaining the connectivity of the maritime shipping network.

Finally, the APL increased substantially from 1.33 to 2.49 (87.87 %). This increase suggests that the average distance between ports increased, resulting in longer shipping routes and potentially lower transportation efficiency. This change may be attributed to a reduction in edges, causing ports to be further apart, or to structural changes in the network that have disrupted previously efficient shipping paths.

The aforementioned analysis conclusively shows that the RSC exerted a substantial influence on the maritime shipping network within 45 days. However, the network exhibited remarkable resilience, adapting by integrating new ports and establishing new connections. Consequently, metrics such as node degree and centrality exhibited notable growth. Nevertheless, the elevation in the network APL value underscores the escalation in the cost of alternative shipping routes compared to previous levels.

3.3. RSC future impacts on the maritime shipping network

By April 2024, the RSC remains ongoing; therefore, it is important to utilize vessel behavior data and maritime shipping networks constructed using AIS for future crisis impact assessments. This study used DC and BC to mark port importance with both indicators standardized using the Z-score method. A Z-score >1.5 indicates that the value of this data point deviates from the mean by more than 1.5 standard deviations. In statistics, this is typically interpreted as a data point being more extreme or significant than the other data points in the dataset. Here, a port was viewed as a hub port if its DC and BC Z-scores surpassed the threshold of 1.5 [7,29]. Hub ports boast extensive connecting capabilities, facilitating the establishment of tight links with numerous other ports, thereby playing a pivotal role in maintaining connectivity (Fig. 5a and b). Prior to the RSC, there were 15 hub ports related to the Red Sea shipping network, which increased to 18 after the RSC. Shanghai Port (China), Houston Port (United States), and Zeebrugge Port (Belgium) ceased to be hub ports associated with Red Sea shipping. Meanwhile, Alexandria (Egypt), Constanta (Romania), Djibouti (Djibouti), Novorossiysk (Russia), Busan (South Korea), St. Petersburg (Russia), and Tanah Merah (Singapore) became new hub ports. Changes in hub ports indicate that the RSC has already affected the global maritime shipping network. It can be inferred that the RSC has, to a certain extent, been detrimental to East Asia and the Americas, while representing an opportunity for North Asia and Africa.

Fig. 5.

Hub port and network variations changes under BC and DC attacks during the RSC. Panel (a) and (b) show the Hub port changes. Blue indicates ports that lost their hub status, while red represents ports that have newly become hubs. Panel (c)–(e) shows the changes in E, APL, and S under DC and BC attacks with the RSC evolving.

To quantitatively estimate the future impact of the RSC on maritime shipping networks, we conducted simulated attacks on ports before and after the RSC. The attack strategies employed were divided into two types: BC and DC attacks. Both strategies focused on eliminating nodes that possess high BC and DC values, respectively. DC attacks primarily target a subset of critical nodes, whereas BC attacks focus on disrupting the network connectivity. This study conducted attacks using two strategies on the global maritime shipping network constructed by global ports (Fig. 1a). Because the RSC is a localized geographical event, only ports within the Red Sea (Fig. 1b) were removed when simulating the attacks. This study calculated the changes in Network efficiency (E) (Fig. 5c), APL (Fig. 5d), and Maximal connected subgraph (S) (Fig. 5e) according to BC and DC attacks. On the whole, it could be found that regardless of the presence of the RSC, all three indicators exhibited a gradual decline during the initial phase of the BC and DC attacks as the number of attacked nodes increased. Subsequently, when the proportion of attacked nodes reached ∼20 %, the E, APL, and S values begin to gradually recover. During the recovery process for the DC attacks, the three indicators typically stabilized when the proportion of attack nodes reached ∼30 %. Conversely, for the BC attacks, the indicators continue to rebound until all the nodes have been attacked. The results showed that the maritime shipping network related to the Red Sea was resilient. When the top 20 % of ports are attacked, alternative paths still exist. The experimental results show that the Red Sea network possesses a certain degree of resilience and that even if important ports are attacked, alternative paths will still exist.

After attacking all the nodes related to the Red Sea, the stable values of E, APL, and S were 0.2980, 0.3212, and 0.9719, respectively. Compared with their states before the attack, E, APL, and S experienced decreases of 6.30, 67.61, and 1.81 %, respectively. Compared to the initial state, although the maritime shipping network related to the Red Sea exhibits resilience, it is still necessary to prioritize ports with top 20 % importance, such as Shanghai, Houston, and Zeebrugge. Fig. 5 shows the resilience of the maritime shipping network before the RSC. The stable values of E, APL, and S were 0.3071, 0.1241, and 0.9715, experiencing decreases of 2.97, −158.78, and −0.04 %, respectively. This implies that the resilience level of the maritime shipping network decreased after the RSC, which manifested specifically in reduced efficiency and a longer APL.

This study also used the Markov method [32,33] to predict important ports in any future RSC developments. First, we collected the numbers of incoming and outgoing ships at each port based on weekly statistics, and then classified these numbers into five levels using the natural break method (Table 6). Ports classified in the fourth and fifth levels based on the natural breaks method were deemed to be of significant importance. Finally, we conduct a Markov prediction based on the changes in the classification levels of each port. Based on the Markov method, Fujairah and Jeddah ports can be viewed as significant outgoing ports after the RSC, whereas Fujairah, Aqaba Industrial, Constanta, Houston, and Jeddah ports can be viewed as significant ports after the RSC (hub ports in Fig. 5(a) and (b)). These hub ports will play an important role in future global supply chains after the RSC.

Table 6.

Incoming and outgoing ship quantity classification based on natural breaks.

Level	Incoming number	Outgoing number
1	0∼1	0∼1
2	2∼4	2∼5
3	5∼12	6∼12
4	13∼29	13∼31
5	30∼57	32∼63

4. Discussion

This study sheds light on the RSC impacts on vessel behavior and maritime shipping networks. The number, SOG, and COG of the vessels in the Red Sea changed significantly leading to a change in the structure of the maritime shipping network. These impacts were quantitatively analyzed using various indicators (e.g., E and APL). However, translating these quantified impacts into corresponding responses is more meaningful. Based on the analysis of the results, policy suggestions are as follows.

• (1)For vessel navigation: Optimize sailing routes. Given the COG changes and the increasing number of vessels choosing to depart from the Red Sea, it is recommended that vessels consider safety factors when planning their sailing routes and avoid potentially risky areas. Owing to the decrease in SOG, vessels that must pass through the Red Sea must also consider the cost losses caused by delays. Because the impact analysis of the RSC on maritime shipping networks indicates the existence of alternative routes, vessels could prioritize taking detours as a preferred option.
• (2)For macro-layout of the maritime shipping network management: Establish more backup ports and routes. Given the insufficient efficiency of alternative routes chosen by vessels for detours, it is recommended to establish more backup ports and routes, particularly in high-risk areas, to enable prompt adjustments to sailing routes during the RSC and reduce the cost and time associated with vessel detours.
• (3)For port operation: Strengthen the construction of ports at key nodes. For critical nodes with high node degrees and BC in the network—Shanghai, Houston, and Zeebrugge ports—it is necessary to increase investment in port construction, enhance their handling capacity and service level, and cope with the increase in vessel traffic caused by the crisis.

5. Conclusion

This study utilized AIS data to analyze vessel behavior with the evolution of the RSC, and used complex network metrics to reveal the impact of the RSC on maritime shipping networks. The conclusions are as follows.

• (1)Vessel behavior in the Red Sea is influenced by the RSC. After the RSC onset, the number of vessels decreased significantly. The SOG reduced from 10 to 8.15 ± 0.35 knots to 6.95 ± 0.19 knots. Many vessels chose to sail away from the Red Sea rather than anchoring at its shores.
• (2)Although the structure of the shipping network was changed by the RSC, it still exhibited resilience. After the RSC outbreak, there was an increase in the number of edges, nodes, and BC. Alternative routes to the original Red Sea shipping lines appeared in maritime areas; however, the APL also increased, indicating an increase in transportation costs.
• (3)Only ports in the top 20 % were considered. Maritime shipping networks are inevitably attacked as the RSC situation deteriorates. However, in the network attack experiment, we found that only 20 % of the ports—Shanghai, Houston, and Zeebrugge ports—played a crucial role in the existing network associated with the Red Sea.

This study had some limitations as it primarily relied on preprocessed AIS data to analyze the impact of the RSC on the maritime shipping network. This pre-processing often involves the removal of data from abnormal vessels (equipment failures, signal interference, and data transmission). As these abnormal vessels were excluded from the analysis, the impact of the RSC may have been underestimated. Past research has shown that the false-positive rate of destination port information in AIS data can be as high as 40 % [21,22]. This also leads to an underestimation of the negative impact of the RSC. After publication of the throughput of various maritime cargo, future research will focus on studying the impact of RSC on the transportation of maritime cargo, which will have a positive impact for the loss assessment of the RSC.

CRediT authorship contribution statement

Zhongyuan Wang: Writing – original draft, Methodology. Zhixiang Fang: Writing – review & editing, Conceptualization. Jianing Yu: Visualization. Xiaoyuan Hu: Data curation. Jinqi Gong: Data curation.

Data availability

The data used in this study are available from the corresponding authors upon request. The raw AIS data and port data could be obtained from https://www.aishub.net/.

Funding

This work was supported by the National key Research and Development Plan under Grants 2023YFB3906703.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.