Identification and Assessment of Critical Waterways in Water Network Areas from a Community Detection Perspective

Abstract

Inland water transport, a critical component of integrated transportation systems, relies on the unobstructed status of critical waterways to ensure network efficiency. Firstly, a weighted topological network was constructed based on waterway class and length. The Leiden algorithm was then employed to divide the inland waterway network into communities, with community bridges identified as critical waterways. Finally, attack simulation experiments were conducted to verify the methodology. Results revealed that the Jiangsu inland waterway network exhibits a distinct community structure, and the regional division is closely aligned with the actual river system. The rapid performance degradation under community bridge attacks confirmed the validity of the critical waterway identification method. Furthermore, a recommended method for waterway class assignment was explored in the inland waterway weighting network. The innovative identification and assessment of critical waterways from the perspective of community detection breaks through the limitations of traditional methods that rely on betweenness centrality and waterway class. Vessel traffic flow across different waterway classes was analyzed using the Automatic Identification System (AIS) data, enabling tailored management strategies for critical waterways. This research provides theoretical support for an in-depth understanding of the structure and function of the inland waterway network, guiding policymaking and promoting the efficiency and security of inland water transport.

1. Introduction

Inland waterway networks, composed of natural rivers, artificial canals, and navigable channels, form a continuous transportation system [1]. By providing low-cost, low-carbon bulk cargo transportation services, these networks serve as critical enablers of regional economic collaboration. A prominent example is the Rhine waterway system: spanning 1232.7 km with a basin area exceeding 185,000 km², it connects nine European countries and handles over one-third of Western Europe’s freight traffic [2]. However, the overall efficiency of inland waterway networks is also constrained by a pronounced “bucket effect”, where their resilience is governed by bottleneck segments. Disruptions in critical waterways caused by siltation, accidents, or extreme weather can trigger cascading network degradation [3], leading to substantial economic losses. The 2021 Suez Canal blockage starkly illustrated this vulnerability: a single vessel grounding halted traffic for six days, diverting 12% of global trade via the Cape of Good Hope, increasing average voyage durations by nine days, and causing daily cargo delays valued at $9.6 billion [4]. This incident underscores the urgent need for the precise identification of critical waterways to mitigate systemic risks.

Complex network theory [5] offers a methodological foundation for identifying critical waterways. In transportation research, this framework has been extensively applied to analyze topological characteristics and resilience in road, aviation, rail, and urban transit networks [6,7]. For instance, Martín [8] compared road network resilience in Valencia and Sardinia under climate change through accessibility analysis. Bešinović [9] developed a passenger-centric resilience assessment model for Dutch railways by integrating operational dynamics and recovery strategies. Wei [10] explored potential factors shaping naturally formed travel communities in various groups and geographical areas in Beijing through the Leiden algorithm. Recent studies have extended resilience analysis to waterborne transport networks [11,12]. Wang [13] analyzed the resilience of waterway transportation systems through simulation, using vessel load, ship delay, and recovery cost as evaluation metrics. In maritime contexts, He [14] established a resilience framework for container shipping networks by evaluating network metric shifts under random or targeted attacks. Wan [15] quantified the efficacy of recovery strategies for liner shipping networks using risk-based resilience metrics. Qin [16] proposed a three-dimensional model to assess port resilience from structural, functional, and locational perspectives. Liu [17] revealed low resilience in European port networks due to overdependence on hub ports. Wu [18] detected 27 practical relevant port communities within the directed-weighted global container liner transport network using the Infomap algorithm.

For inland waterways, Baroud [19] quantified vulnerabilities in the Mississippi River system and optimized post-disruption recovery priorities. Hosseini [20] decomposed resilience into absorption, adaptation, and recovery capacities using Bayesian networks. Akkermann [21] modeled cascading failure risks in German canal networks, confirming hub waterways’ dominance over connectivity. Han [22] identified critical nodes in the Yangtze River network through percolation theory and node removal simulations. Among them, betweenness centrality is usually used to evaluate the topological statistical description of inland waterway networks [23]. Betweenness centrality in complex network theory quantifies the importance of a node (or edge) based on its role in facilitating communication between other nodes in the network. Nodes with high betweenness centrality are critical for maintaining efficient information flow and are often considered bottlenecks or key connectors. Despite these advances, criticality assessment for inland waterways remains challenging. Current approaches predominantly rely on edge betweenness and waterway class, which face two limitations: (1) betweenness centrality assumes optimal route adherence, neglecting real-world multi-path diversions induced by tidal cycles, port efficiency, and other operational constraints; and (2) conventional physical metrics inadequately quantify waterways’ integrated network functionality.

Novel perspectives emerge from complex network concepts of bridges [24] and communities [25]. Bridges, defined as edges whose removal disconnects the network, are structural linchpins. Communities represent densely interconnected node clusters with high redundancy, while inter-community connections are sparse. Zhang [26] demonstrated the superiority of community bridges in transit network analysis using efficiency, travel time, and correlation metrics. Community bridges [27], serving as cross-cluster conduits, are irreplaceable for regional synergy; their disruption can sever inter-community connectivity and induce cascading congestion. Wang [28] and Ge [29] further validated community structures in multimodal and maritime networks.

In inland waterway networks, inter-community waterways serve dual roles, both enabling regional resource exchange and maintaining global connectivity. Typically situated at multi-path junctions, their load-bearing capacity directly governs systemic robustness. This study thus defines critical waterways as those linking distinct communities and ensuring network-wide connectivity. By constructing a weighted topological network and analyzing community attributes, we propose a community detection-based framework for critical waterway identification. The innovative identification and assessment of critical waterways from the perspective of community detection breaks through the limitations of the traditional method of relying on betweenness centrality and waterway class and provides new ideas and methods for the study of inland waterways, which helps to deeply understand the structure and function of the inland waterway network.

This study contributes fourfold to the literature, as follows: First, it reveals the inherent community structure of inland waterway networks, offering new insights for regional management. Second, it explores and proposes a method for assigning waterway classes in a weighted inland waterway network. Third, it identifies and validates critical waterways through community analysis. Fourth, it develops differentiated management strategies integrating topological features and operational demands, providing actionable decision support for policymakers.

The paper is organized as follows: Section 2 details data sources and processing. Section 3 presents the community detection-based criticality assessment model. Section 4 validates the framework using Jiangsu Province’s inland waterway network. Section 5 and Section 6 are the Discussion and Conclusions, respectively.

2. Data

2.1. Data Sources

Jiangsu Province, located in eastern China, serves as a core component of the high-class waterway network in the Yangtze River Delta, China. With a well-developed water transport infrastructure, the province offers significant advantages in inland navigation. By the end of 2022, its total inland waterway length reached 24,400 km, including 8813 km of classified waterways. These figures account for approximately 20% and 14.3% of China’s total waterway lengths, respectively. The waterway network density stands at 24.75 km per 100 km², ranking first among all Chinese provinces in both scale and density [30]. This study focuses on Jiangsu’s inland waterway system, with the study area’s geographical location illustrated in Figure 1.

Two datasets were employed: Geographic Information System (GIS) data of Jiangsu’s inland waterways and AIS records. The GIS dataset contains waterway names, spatial coordinates (WGS-84), waterway class, and waterway length (

Table 1. Navigational attributes of Jiangsu’s inland waterways.

Waterway Name	Source	Target	Class	Design Vessel Load (ton)	Design Water Depth (m)	Length (km)
Section 1, Yangtze River	121.59, 31.74	121.95, 31.67	I	3000	3.5–4.0	35.74
Gaoyou Section, Beijing-Hangzhou Grand Canal	118.31, 33.98	118.65, 33.71	II	2000	2.6–3.0	52.68
Lianshui Section, Salt Canal	119.32, 34.12	118.95, 33.60	III	1000	2.0–2.4	89.57
Wuzhong Section, Sushen Outer Port Canal	120.73, 31.23	120.65, 31.26	IV	500	1.6–1.9	9.00
Chongchuan Section, Tongzhou Bay Port Access	120.76, 32.09	120.88, 32.19	V	300	1.3–1.6	16.80
Tongzhou Section, Tongyang Canal	120.76, 32.08	120.72, 32.33	VI	100	1.0–1.2	28.51
Donghai County Section, Huashu New River	118.83, 34.37	118.78, 34.29	VII	50	0.7–0.9	10.35
Yuanjiaqiao Section, Yangtze River	120.42, 31.90	120.38, 31.97	VIII *	<50	<0.7	9.15

Notes: * The classification criteria for inland waterways primarily follow the Inland Waterway Navigation Standards (GB 50139-2014) [31], where Class VIII denotes non-classified waterways.

), with spatial distributions visualized in Figure 2.

The AIS dataset comprises 5,800,431 vessel trajectory records from 1 to 31 July 2022. Each record includes fields such as Maritime Mobile Service Identity (MMSI), speed, heading, longitude, latitude, and timestamp, with a sample data structure shown in

Table 2. Example of AIS data.

MMSI	Speed (knots/h)	Heading	Longitude	Latitude	Create Time
413793342	5.2	101	120.2052	31.64514	2022/7/2 0:09
413791877	5.6	511	119.8354	32.30228	2022/7/2 0:24
413993278	4.5	310	119.8468	32.29357	2022/7/2 1:36
413832505	3.8	303	120.2407	31.63442	2022/7/2 1:54
413772333	4.5	352	119.9042	32.55611	2022/7/2 2:14
413809635	3.7	304	120.2415	31.63395	2022/7/2 2:25
413303620	6.6	305	120.8103	32.00513	2022/7/2 2:38
413984853	4.6	295	119.5679	32.19297	2022/7/2 3:21
413793059	6.9	109	118.6826	33.19103	2022/7/2 4:49

. Figure 3 displays the vessel trajectories on 1 July 2022, which align with Jiangsu’s primary shipping corridors. In practice, not all vessels transmit AIS signals that are detectable by transport management systems. However, mandatory lock transit records are considered to provide complete vessel passage data. The reported AIS capture rate for July 2022 was 94.84%, defined as the ratio of vessels with AIS records to those passing through locks. This high coverage validates the representativeness and reliability of the July vessel traffic analysis, establishing a robust statistical foundation for province-wide assessments.

2.2. Data Processing

The data processing workflow comprised three sequential phases:

• Data Cleaning: Data beyond the geographical boundaries of Jiangsu Province were removed, and threshold filtering was applied to abnormal navigation parameters, including the removal of data records with abnormal speeds, unreasonable heading changes, and invalid MMSI [32];
• Spatial Matching: Vessel trajectories were geospatially joined with waterways using Python’s Geopandas library (v0.12.2), establishing spatial topology relationships between navigation paths and waterway networks;
• Traffic Quantification: Unique MMSI counts per waterway were calculated as vessel traffic indicators. The waterway traffic flow distribution map was generated in ascending order, as shown in Figure 4. The analysis reveals pronounced heterogeneous traffic patterns across Jiangsu’s inland waterway network: approximately 68% of waterways exhibited monthly traffic volumes below the network-wide mean (350 vessels), while maximum traffic intensity reached 4304 vessels.

3. Methods

3.1. Construction of Weighted Topology Network

To determine critical waterways in the network, a topological structure was first constructed using the primal graph method [33]. The network is defined as a weighted graph G = (V, E, W), where V represents the set of all nodes, corresponding to intersections between waterways or between waterways and jurisdictional boundaries, denoted as V = {v_i ∣ i = 1,2, …, n}, with n being the total number of nodes; E denotes the set of all edges, representing waterway segments, expressed as E = {(v_i, v_j) ∣ v_i ∈ V, v_j ∈ V, i≠j}; W represents the set of all edge weights, jointly determined by waterway length L (km) and waterway class D, expressed as W = { w_ij ∣ (v_i, v_j) ∈ E }. Waterway length L reflects geographical proximity, where longer segments reduce connectivity efficiency. Waterway class D indicates shipping accessibility, with higher class offering greater navigability. The weight w_ij for edge (v_i, v_j) is calculated as

$$w_{\mathit{ij}}={\displaystyle \frac{1}{L_{\mathit{ij}}\times D_{\mathit{ij}}}}$$

(1)

where D_ij ∈ D is the waterway class coefficient of edge (v_i, v_j), discretized such that Class I corresponds to D = 1, Class II to D = 2, Class III to D = 3, and so forth; L_ij ∈ L is the geographical length (in kilometers) of the waterway segment represented by edge (v_i, v_j).

3.2. Community Detection Based on Leiden Algorithm

Community detection algorithms were applied to explore structural characteristics and functional partitions within the inland waterway network. Current community detection methodologies primarily fall into five categories: hierarchical clustering [34], modularity optimization [35], spectral clustering [36], statistical inference [37], and deep learning [38]. Modularity optimization, a widely adopted method for identifying community structures, operates by maximizing modularity values. Modularity [39], a key metric for evaluating community detection outcomes, is defined as

$$Q={\displaystyle \frac{1}{2m}} {\sum }_{i,j}\left[w_{\mathit{ij}}-{\displaystyle \frac{s_i{s}_j}{2m}}\right] \delta \left(C_i,C_j\right)$$

(2)

where Q denotes modularity ranging between [0, 1), with higher values indicating more stable community structures; m is the sum of all edge weights in the network; w_ij represents the weight of the edge between nodes v_i and v_j; s_i and s_j are the weighted degree centralities of nodes v_i and v_j, calculated as the sum of weights of edges connected to each node; and C_i denotes the i community containing v_i—δ(C_i,C_j) equals 1 if v_i and v_j belong to the same community, and 0 otherwise.

For the constructed waterway network G = (V, E, W), the Leiden algorithm [40] was employed to partition nodes into communities. The computational workflow (Figure 5) involves three key phases:

• Local Node Movement: Optimize community assignments by calculating modularity gain, relocating nodes to neighboring communities;
• Partition Refinement: Optimize preliminary community structures through hierarchical subdivision of established communities into sub-communities while eliminating suboptimal interconnections;
• Network Aggregation: Abstract communities into super nodes, constructing a higher-level network topology based on inter-community connection weights.

Compared to the traditional Louvain algorithm [41], which has been shown to produce a bad connection in the community and becomes disconnected when executed iteratively [42], the Leiden algorithm addresses loosely connected intra-community structures through its secondary optimization mechanism to resolve the weaknesses. Furthermore, compared with other algorithms, the Leiden algorithm can ensure stable and high-quality partitions in small-scale complex networks [43], such as local inland waterway networks.

3.3. Critical Waterway Identification Based on Community Bridge

Building on the partitioned waterway network, this study introduces the CB metric to evaluate the strategic value of inter-community waterways in the global network. This metric assesses the importance of edges connecting different communities by integrating the degrees of connected nodes and the sizes of their respective communities, reflecting both community scale and node influence. The CB value [26] is calculated as

$$\mathit{CB}=\left\{\begin{array}{c}\sqrt{x_i \times x_j \times k_i \times k_j}, C_i \ne C_j\\ 0, C_i={ C}_j\end{array}\right.$$

(3)

where CB is the community bridge value; x_i and x_j represent the number of nodes in communities C_i and C_j, respectively; and k_i and k_j denote the degree of nodes v_i and v_j. A higher CB value indicates a greater strategic importance of the waterway. If two nodes belong to the same community (C_i = C_j), the CB value is zero.

3.4. Verification of Recognition Effectiveness Based on Attack Simulation

To validate the effectiveness of the critical waterway identification method, attack simulation [44] was employed to assess network degradation under simulated waterway failure scenarios. Three attack strategies were designed:

• Random Attack: Simulates random disruptions (e.g., natural disasters, vessel accidents), with results averaged over 100 independent trials;
• Edge Betweenness Attack: Prioritizes removal of waterways with the highest edge betweenness centrality;
• Community Bridge Attack: Targets inter-community critical waterways identified by the proposed CB metric.

Network performance was quantified using two indicators: Largest Subnetwork Size (S) measures global connectivity retention, and Relative Network Efficiency (NE) reflects retained transport efficiency.

$$S=V\left(G’\right),$$

(4)

$$E={\displaystyle \frac{1}{n\left(n-1\right)}} {\sum }_{i\ne j}{\displaystyle \frac{1}{d_{\mathit{ij}}}},$$

(5)

$$\mathit{NE}={\displaystyle \frac{E\left(G’\right)}{E\left(G\right)}},$$

(6)

where V(G′) is the node count of the largest connected component post-attack; n is the total number of nodes; d_ij is the shortest path edge count between v_i and v_j; and E(G′), E(G) denote post-attack and original network efficiencies, respectively.

4. Results

4.1. Regional Waterway Network Detection

The inland waterway network of Jiangsu Province was constructed using the primal approach, comprising 664 nodes and 805 edges. Applying the Leiden algorithm for community detection, we partitioned the network into 19 regional subnetworks, as shown in Figure 6. The high modularity index (0.845) confirms strong community structure characteristics [45]. Therefore, this strongly confirms that the Jiangsu inland waterway network does indeed have significant characteristics of community structure, which offers new insights for regional inland waterway management.

Geospatial analysis reveals close alignment between community detection and Jiangsu’s four major natural water systems: Yi-Shu-Si River Basin, Huai River Basin, Yangtze River Basin, and Tai Lake Basin. These communities further subdivide the basins into functionally integrated sub-units. For instance, the Yangtze River Basin splits into two communities: Community 0 incorporates the 12.5 m deep-water trunk waterway (navigable for 50,000-ton sea-going vessels) and core port clusters, such as Suzhou Port and Nantong Port, serving as a river–sea intermodal hub. Community 1 centers around inland transfer nodes such as Nanjing Port and Zhenjiang Port, primarily facilitating bulk cargo transport for riverside industrial zones. This indicates that the subnets identified by the Leiden algorithm are not only cohesive in topology but also have strong, realistic explanatory power in terms of geography and function. Therefore, it can be considered that the result of this subnet division is reasonable and effective.

4.2. Critical Waterway Identification

Based on community detection results, 42 critical waterways were identified using the CB metric, as shown in Figure 7 and listed in

Table 3. Critical waterway identification results.

No.	Waterway Name	Class	Length (km)	CB	Traffic Flow (Vessels/Month)
1	Section 2, Yangtze River	I	14.21	327.51	1370
2	Jingjiang Section, Xiashi Port	VI	14.91	327.51	1004
3	Changzhou Section, Sanshan Port	VII	23.29	279.07	686
4	Xiangcheng Section, Wangyu River	V	40.45	271.35	679
5	Jiangyin Section, Xicheng Canal	III	12.14	241.68	1181
6	Xishan Section, Xishiyiwei Canal	V	14.80	241.68	1188
7	Kunshan Section, Suliu Canal	V	19.97	235.00	932
8	Kunshan Section, Shenzhang Canal	V	10.02	235.00	503
9	Jiangdu Section, Beijing-Hangzhou Grand Canal	II	23.45	222.92	3109
10	Danyang Section, Jiuqu River	VI	27.84	222.92	1056
11	Taixing Section, Rutai Canal	VII	11.72	216.48	23
12	Rugao Section, Lianyungang-Shanghai Canal	IV	29.50	202.64	3246
13	Xinghua Section, Jiankou Canal	VI	31.47	199.99	687
14	Jianhu Section 1, Sheyang Port Access Channel	III	50.24	199.99	775
15	Changzhou Section, Desheng River	VI	23.80	193.05	1271
16	Tongyang Canal	III	19.29	187.48	314
17	Rugao Section, Ruhai River	VI	33.66	175.49	713
18	Chongchuan Section, Tongzhou Bay Port Access Channel	V	16.80	175.49	816
19	Tongzhou Section, Tongyang Canal	VI	28.51	175.49	359
20	Tongzhou Section, Tongyang Canal	IV	19.27	175.49	3
21	Xinghua Section, Xingdong Canal	VI	41.03	173.20	1353
22	Jiangyan Section, Taidong Canal	III	20.88	173.20	1706
23	Biancheng-Dongjiatan, Tangda Canal	VII	14.79	173.20	120
24	Huqiu Section, Hu-Guang Canal	VI	25.88	160.44	4304
25	Gaoyou Section 1, Beijing-Hangzhou Grand Canal	II	32.10	149.74	2572
26	Changzhou Section, Beijing-Hangzhou Grand Canal	III	21.06	142.46	1611
27	Changzhou Section, Changyi Canal	VI	13.89	142.46	535
28	Liyang Section 1, Wushen Canal	VI	22.91	142.46	1681
29	Xiangshui Section, Huai River Outbound Channel	IV	21.99	130.49	390
30	Jianhu Section, Yanbao Canal	IV	28.59	121.05	108
31	Lianshui Section, Yanhe River	III	89.57	119.66	1637
32	Dongshan Wharf-Xiaolongwan, Qinhuai Canal	IV	17.59	116.77	171
33	Sihong Section, Xu-Hong Canal	III	57.41	115.41	44
34	Gaoyou Section 2, Beijing-Hangzhou Grand Canal	II	52.68	115.41	2059
35	Jianhu Section 2, Sheyang Port Access Channel	III	33.95	114.32	1139
36	Funing Section, Northern Jiangsu Irrigation Main Channel	V	55.56	98.83	93
37	Dongtai Section, Lianyungang-Shanghai Canal	III	25.20	92.81	1291
38	Shuyang Section, Chaimi River	VI	50.11	83.46	9
39	Sihong Section, Lixi River	V	14.04	81.61	23
40	Shuyang Section, Gupo River	VI	49.83	68.15	89
41	Huaiyin Section, Huai-Shu-Xin River	VI	34.65	63.21	93
42	Liyang Section 2, Wushen Canal	III	33.31	49.60	540

. Among these, 19 critical waterways have a class of IV or higher, exhibiting a 45% spatial coupling rate with Jiangsu’s “Two Vertical, Five Horizontal” trunk shipping corridors. This demonstrates the algorithm’s effectiveness in capturing the topological hub roles of high-class waterways. The top five critical waterways are Section 2 of Yangtze River, the Jingjiang Section of Xiashi Port, the Changzhou Section of Sanshan Port, the Xiangcheng Section of Wangyu River, and the Jiangyin Section of Xicheng Canal. Geographically, these waterways are concentrated in the economically vital Yangtze River coastal zones, serving as bridges connecting the Beijing-Hangzhou Grand Canal, the Yangtze trunk line, and tributary networks.

The identified critical waterways exhibit significant heterogeneity in attributes: waterway classes range from I to VII, with lengths spanning 10 km to 80 km. This underscores that the proposed method does not rely on single attributes for identification. Notably, the identified critical waterways focus on Jiangsu’s internal inland waterway network, a result shaped by the technical scope of the community detection algorithm. In reality, as an integral component of the Yangtze River Delta’s comprehensive shipping network, Jiangsu’s waterways exhibit complex topological linkages with adjacent provincial networks. From a macro perspective of regional coordinated development, boundary waterways not classified as critical in this study may hold significant strategic value in larger-scale network analyses.

4.3. Identification Effectiveness Verification

The inland waterway network was tested to evaluate the structural integrity of the network when it was attacked using the three attack strategies described in the previous text. Specifically, in the deliberate edge betweenness attack, the waterways of each attack are selected from high to low according to the edge betweenness centrality. For the deliberate community bridge attacks, each attack selects the waterways from high to low based on the CB values of the identified critical waterways. The waterways of the random attack are selected randomly, then the 95% confidence interval was calculated as the network performance indicator through 100 independent repeated experiments. The results are shown in Figure 8. To ensure comparability, all strategies were applied over 42 attack steps, matching the number of critical waterways identified by community bridge.

Figure 8 illustrates the performance differences of the inland waterway network in Jiangsu Province under varying attack strategies. After removing 42 random waterways, S and NE declined by only 9.8% (95% confidence interval [9.4%, 10.3%]) and 7.9% (95% confidence interval [7.5%, 8.3%]), respectively, demonstrating the network’s strong network robustness against random failures. The narrow confidence intervals (half-width < 0.5%) confirm the stability of repeated experimental results. In sharp contrast, targeted removal of the 42 waterways with the highest edge betweenness centrality caused S and NE to decline by 57.3% and 47.9%. In contrast, removing the critical waterways based on community bridge triggered a sharp collapse, with S and NE plummeting by 73.0% and 81.7%. A cascading failure threshold emerged at the 25th attack step (S: −52.9%, NE: −59.1%), fragmenting the network into isolated subnetworks.

The rapid performance degradation under community bridge attacks validates that the CB metric effectively pinpoints “bottleneck waterways” critical to global connectivity. Disruptions to these waterways would sever regional coordination, force vessel detours through tributaries, and significantly increase transport costs and delays. These findings confirm the method’s effectiveness, offering quantitative guidance for prioritizing waterway maintenance.

5. Discussion

5.1. Impact Analysis of Waterway Class Assignment Methods

In modeling the weighted topological network of inland waterways, waterway length serves as an objective geographical parameter that can be directly quantified. However, the discretized assignment of waterway class coefficients D relies on subjective scaling rules. This study designs nine waterway class assignment methods (Table 3), with Method 1 as the baseline, to systematically evaluate their impacts on community detection and critical waterway identification. Methods 1–6 employ fixed-interval linear assignments, Method 7 adopts a nonlinear assignment based on design water depth, and Methods 8–9 utilize nonlinear assignments based on design vessel load.

As shown in

Table 4. Results under different waterway class assignment methods.

Assignment Method	Modularity	Number of Communities	Number of Critical Waterways	Number of Critical Waterways Identical to the Baseline Method
1: D = 1, 2, 3, 4, 5, 6, 7, 8	0.845	19	42	-
2: D = 1, 3, 5, 7, 9, 11, 13, 15	0.848	20	43	40
3: D = 1, 4, 7, 10, 13, 16, 19, 22	0.847	21	44	40
4: D = 2, 4, 6, 8, 10, 12, 14, 16	0.859	22	47	42
5: D = 10, 20, 30, 40, 50, 60, 70, 80	0.860	21	45	38
6: D = 11, 12, 13, 14, 15, 16, 17, 18	0.874	23	51	38
7: D = 5, 8, 11, 14, 17, 22, 28, 38	0.869	23	50	38
8: D = 1, 5, 10, 30, 50, 100, 200, 300	0.786	26	58	24
9: D = 10, 50, 100, 300, 500, 1000, 2000, 3000	0.773	26	54	22

Note: The direction of assignment of D is from the Class I waterway to the Class VIII waterway.

, the network modularity remains consistently above 0.7 across all methods, confirming the network’s pronounced community structure and validating the applicability of the critical waterway identification framework. Linear assignment methods (Methods 1–6) demonstrate strong stability: modularity fluctuations are limited to less than 1.7%, and the overlap rate of critical waterways with the baseline exceeds 90%. Although Method 7 employs nonlinear assignment, its gentle inter-class gradient preserves a 90.5% overlap rate. Notably, methods based on design vessel load (Methods 8–9) exhibit significant deviations, particularly Method 9, where the overlap rate with the baseline plummets to 52.4%.

The comparative analysis in Figure 9 demonstrates that the 42 critical waterways identified by the baseline method are predominantly distributed along trunk waterways in Jiangsu Province, aligning with actual shipping patterns and the defining characteristics of critical waterways. Method 6 yields nearly identical critical waterways to the baseline. In contrast, Method 8, which employs large incremental gradients between waterway classes, excessively amplifies the accessibility contributions of high-class waterways. This assignment approach results in two anomalies: isolated nodes are incorrectly classified as standalone communities, and certain branch waterways are misidentified as critical (highlighted by red circles in Figure 9b).

These findings indicate that linear assignment methods preserve balanced weighting relationships across waterway classes, thereby maintaining the structural authenticity of the network and enhancing the reliability of critical waterway identification. Consequently, this study recommends adopting natural number sequences for waterway class assignment.

5.2. Policy Recommendations Based on Waterway Traffic Analysis

The critical waterway identification method proposed in this study evaluates network topological characteristics and infrastructure capacity, with a focus on structural risks induced by connectivity loss upon waterway disruption. However, real-world shipping demand driven by socioeconomic factors may not align with high traffic volumes on critical waterways. Therefore, differentiated management strategies that integrate traffic patterns are essential for targeted governance.

Inland waterways of different classes exhibit distinct design standards (e.g., width, depth, curvature) and navigational capacities, leading to significant variations in vessel sizes. Conversely, waterways within the same class share comparable design specifications, enabling systematic traffic comparisons.

Using July 2022 AIS data, monthly vessel traffic across waterway classes was analyzed, as illustrated in Figure 10. The 85th percentile thresholds for monthly vessel traffic were established as follows: Class I (>1300), Class II (>2700), Class III (>1800), Class IV (>1900), Class V (>1700), Class VI (>1900), and Class VII (>900).

Among the 42 critical waterways identified, 4 exhibited traffic volumes exceeding their class-specific thresholds. For instance, the Huqiu Section of Hu-Guang Canal (Class VI) recorded 4304 vessels/month, representing 2.3 times its threshold, while the Rugao Section of Lianyungang-Shanghai Canal (Class IV) reached 3246 vessels/month, 1.7 times its threshold. Similarly, the Jiangdu Section of Beijing-Hangzhou Grand Canal (Class II) and the Section of Yangtze River (Class I) exceeded their respective thresholds by 15% and 5%.

To address these disparities, optimization strategies must balance topological importance and operational demands through dynamic monitoring and adaptive planning. High-traffic critical waterways, vital for both accessibility and shipping demand, require capacity expansion measures such as channel widening and lock upgrades. Concurrently, implementing traffic early warning systems can mitigate congestion risks and enhance network resilience. In contrast, low-traffic critical waterways, though less utilized, remain essential for network connectivity. These should prioritize cost-effective monitoring to maintain baseline navigability without excessive resource allocation.

Notably, the Master Plan for the Trunk Waterway Network of Jiangsu Province (2023–2035) [46] aligns with these findings, upgrading the Huqiu Section of Hu-Guang Canal and the Rugao Section of Lianyungang-Shanghai Canal from Class VI and V to II. Such policy adjustments validate the practical value of complex network analysis in guiding targeted infrastructure investments, ultimately fostering synergies between economic efficiency and waterway resilience.

6. Conclusions

This study identifies critical waterways in Jiangsu Province’s inland waterway network using a community detection algorithm. The waterways with a high value of CB play irreplaceable roles in maintaining regional connectivity. Targeted attack simulations involving sequential removal of critical waterways reveal the network’s vulnerability. Specifically, removing community bridge-based critical waterways caused S and NE to plummet by 73.0% and 81.7%, respectively. These findings underscore the necessity of protecting strategic waterways to prevent cascade network failures. The results provide actionable insights for future transportation management strategies, enabling targeted monitoring and maintenance of network vulnerabilities to mitigate disruptions caused by natural disasters, maritime accidents, or intentional sabotage, thereby safeguarding cargo transport security and socioeconomic stability.

Methodologically, the critical waterway identification framework exhibits robustness in parameter selection and demonstrates practical feasibility. The network modularity remains consistently above 0.7 across all assignment methods, with a 90% overlap rate between critical waterways identified under linear assignments and the baseline method. These results confirm the structural stability of the algorithm under arithmetic sequence weighting, demonstrating the network topology’s tolerance for hierarchical scaling rules. Analysis of traffic thresholds based on the 85th percentile from AIS data shows a spatial discrepancy between topological importance and shipping demand, indicating the need for differentiated management strategies. Specifically, high-traffic critical waterways require capacity enhancements such as channel widening and lock upgrades, while low-traffic critical waterways necessitate basic maintenance to preserve accessibility.

The proposed “community detection–critical waterway identification–network performance validation” framework exhibits broad applicability. For complex river systems like the Amazon, Rhine, and Mississippi, localized modeling can be achieved by adjusting weighting parameters, offering valuable management insights. Additionally, the methodology extends to multimodal transport networks by integrating road and rail layers to identify cross-modal hubs, supporting comprehensive resilience planning.

Four limitations remain: First, the single-month AIS dataset lacks temporal robustness, and the lack of vessel load data prevents a comprehensive assessment of the actual impact of vessels on shipping lanes and may lead to biased regional flow assessments. Second, the model does not take into account the differences in route preferences between vessel types (e.g., container vessel vs. bulk carrier), which may underestimate the strategic value of a particular shipping lane. Third, although a variety of waterway class assignment values were designed and compared, the nonlinear assignment methods were not explored in depth enough to fully analyze their advantages and disadvantages in different scenarios. This is a theoretical issue that requires special attention in the future. Fourth, the current network scope focuses solely on Jiangsu’s inland waterways, neglecting boundary waterways that may exhibit critical roles in the broader Yangtze River Delta integrated shipping network. Despite these constraints, this study innovatively employs community detection algorithms to inland waterway systems, establishing an analytical framework with dual perspectives of topology and operational demand for critical waterway identification.

Future studies should prioritize expanding temporal–spatial data granularity by integrating multi-year AIS datasets with vessel load records from port authorities and logistics platforms, enabling holistic assessments of shipping lane utilization patterns. Research must advance specific vessel behavioral modeling through historical trajectories to quantify route preference among different vessel categories, thereby improving the strategic value assessment of critical waterways. Substantial theoretical innovation is needed to explore nonlinear optimization, such as deep reinforcement learning for dynamic waterway class arrangement. The geospatial scope should extend beyond provincial boundaries to model the Yangtze River Delta’s integrated network, incorporating critical trans-regional waterways and multimodal interfaces.