Network analysis of ballast-mediated species transfer reveals important introduction and dispersal patterns in the Arctic

Global trade and transportation networks are major pathways for introducing non-native species, with risk increasing alongside trade volumes and connectivity. Aquatic invasive species pose recognized threats to Arctic ecosystems, and climate change is accelerating sea ice loss, reshaping shipping patterns, and potentially expanding the environmental suitability of Arctic ports. Ship-borne introductions occur primarily via ballast water discharges and biofouling, and depend on factors such as ship type, voyage duration, ballast uptake/discharge volumes and locations, and environmental similarity between source and destination. Prevention is emphasized in international agreements due to the difficulty and cost of eradication. This study applies network analysis and data mining to assess, visualize, and project ballast-mediated species introduction into the Arctic and dispersal within the Arctic, integrating shipping, ecoregions, and environmental data. The approach complements pairwise risk analyses by revealing multi-port, network-level dynamics critical for prioritizing surveillance, prevention, and management. The objectives are to: (1) build a risk assessment network model for shipping into and within the Arctic using global ship movement (LLI), ballast discharge (NBIC), biogeographic ecoregions (MEOW/FEOW), and environmental (temperature, salinity) data; (2) identify high-risk introduction pathways and emerging Arctic shipping hubs over 1997–2012 and project trends; (3) incorporate environmental tolerance groups to illustrate establishment-constrained spread pathways; (4) model within-Arctic dispersal using higher-order networks that capture path-dependent ship movements; and (5) demonstrate targeted management insights via a case study.

Data: Global ship movement data from Lloyd’s List Intelligence (LLI) for 1997, 1999, 2002, 2005, 2008, and 2012 (May 1 to April 30 the following year) comprising 12,723,028 voyage records (9,569,619 after deduplication). Subset to voyages to and between Arctic ports (as defined by the Arctic Council CAFF boundary), resulting in 48,364 voyages via 3,902 introduction pathways (non-Arctic to Arctic) and 4,715 voyages via 1,269 dispersal pathways (within Arctic). Pathways from the same or neighboring ecoregions (MEOW/FEOW) were excluded to remove likely natural dispersal, reducing analyzed introduction pathways to 2,874. Ballast discharge: Used U.S. National Ballast Water Information Clearinghouse (NBIC) records for Alaska (2004–2016), 4,926 records; after quality filtering (remove missing, zero, or exceeding capacity), 1,280 valid records remained. NBIC data were subset to Arctic-relevant voyages to estimate ballast discharge patterns. Environmental data: Annual average temperature and salinity for ports from the Global Ports Database where available, supplemented with World Ocean Atlas data at nearest coastal locations. Introduction risk per trip: For a ship t traveling from port i to port j with duration Δt (days) and ballast discharge D(t), the relative risk P^(t) = (1 − exp(−μ Δt)) × exp(−A D(t)), with μ = 0.02 day−1 (daily mortality) and A = 3.22×10−6. Discharge D^(t) estimated as D^(t) = Z_k × W_k(GWT), where k is ship type (LLI’s 150 types mapped to NBIC’s 9 types); Z_k is fraction of non-zero discharges for type k; W_k(GWT) is predicted discharge volume as a function of gross weight tonnage and ship type using a random forest regression (trained on non-zero NBIC records; 70/30 split; R² = 0.93). Pathway and port risks: Aggregated pathway risk P_ij = 1 − Π (1 − P^(t)) over all trips t from i to j, assuming independence. Aggregated introduction risk for a destination port j is P_j = 1 − Π_i (1 − P_ij). Environmental tolerance filtering: To approximate establishment constraints, introduction pathways were filtered by temperature and salinity differences between source and destination ports for six tolerance groups: Δt ≤ 2.9°C or ≤ 9.7°C; Δs ≤ 0.2 ppt, ≤ 2 ppt, or ≤ 12 ppt, reflecting freshwater stenohaline, marine stenohaline, and euryhaline species. This yields spread (introduction + establishment) pathways per tolerance group. Trend projection and hub emergence: Linear regressions were used to describe temporal trends (1997–2012) in voyages, deadweight tonnage (DWT), average DWT per voyage, number of pathways, recipient ports, and average risk per pathway. Future projections (to 2027) were displayed with 95% confidence intervals for regressions with p < 0.05, acknowledging they are descriptive trends, not forecasts. A preferential attachment-inspired rewiring simulation adjusted network topology to reflect hub emergence: pathways were reassigned with probability proportional to ports’ incoming degree until a target number of recipient ports was reached, combined with projected per-pathway risk increases. Within-Arctic dispersal networks: Two representations were constructed: (1) First-order network of species flow (SF-FON), where edges represent direct movements with weights based on trip counts/risks; and (2) Species-flow Higher-Order Network (SF-HON), which encodes dependencies on previous ports and weights edges by aggregated trip-level introduction risks, incorporating ship type, size, duration, and discharge. SF-HON retains higher-order dependencies only when significantly different from lower order and applies a minimum support threshold in terms of aggregated risk probability. Clustering (Map Equation/Infomap) was used to identify communities. Port risk rankings for indirect dispersal were computed via random walks with resets (PageRank-like), representing the probability a species reaches a port through multi-step, path-dependent dispersal. Case study: Propagation patterns were compared for Murmansk (first-order) and second-order nodes Murmansk|Tromso and Murmansk|Hammerfest to illustrate targeted pathways and management implications across propagation steps.

- Network scope: 310 Arctic ports and 7,187 non-Arctic ports considered; 3,902 active introduction pathways identified (1997–2012), reduced to 2,874 after excluding same/neighboring ecoregions. - Rising shipping intensity into Arctic (1997–2012): voyages increased by +128/year (p < 0.05); total DWT increased by +2.52×10^16/year (p < 0.01); average DWT per voyage increased by +253/year (p < 0.05). Number of distinct introduction pathways increased only 2% (p > 0.05), indicating intensified traffic on existing pathways rather than more pathways. - Emergence of hubs: Recipient ports per introduction pathway declined significantly (−2.03/year, p < 0.01). Some ports (e.g., Murmansk) saw marked increases in incoming pathways (from 110 in 1997 to 158 in 2012), evidencing preferential attachment and hub formation. - Per-pathway risk increased: Average P_ij per introduction pathway rose by +9.99×10^−4/year (p < 0.01), reflecting higher frequency, larger ships, and stable pathway counts. - High-risk introduction corridors: Many high-risk pathways originate from Northwestern Europe (e.g., Rotterdam, Hamburg, Amsterdam) to Arctic ports such as Narvik (Norway) and Murmansk (Russia). Some Alaskan ports (Afognak, Kodiak) have high aggregated introduction risk due to many connected pathways despite lower per-pathway risk. - Establishment constraints by environment: For the most sensitive group (Δt ≤ 2.9°C, Δs ≤ 0.2 ppt), Churchill (Canada) has highest aggregated introduction risk P_j = 3.6%. Relaxing temperature tolerance to Δt ≤ 9.7°C with Δs ≤ 0.2 ppt leaves Churchill most vulnerable (P_j unchanged). Relaxing salinity constraint to Δs ≤ 2 ppt with Δt ≤ 2.9°C shifts vulnerability to Afognak (USA, P_j = 5.6%) and Dutch Harbor (USA, P_j = 4.7%) due to compatibility with NE Asia sources. Broadly tolerant species open more pathways throughout the Arctic. - Within-Arctic dispersal (direct, SF-FON): 1,269 dispersal pathways identified; strongest links within Arctic Europe (e.g., Murmansk–Tromso region). Trends show increases in voyages, ship size, and incoming pathways to the largest hub (+0.89/year, p < 0.04). - Higher-order dispersal (SF-HON): Ships’ next destinations depend on prior ports (up to 5th-order dependencies). SF-HON reveals five clusters largely aligned with geography; inter-cluster links are sparse, indicating management leverage points. Port rankings differ between direct (SF-FON) and indirect (SF-HON) dispersal risks: Reykjavik ranks 14th for direct risk but 3rd for indirect risk (cluster center); Dutch Harbor ranks 24th direct but 8th indirect, consistent with observed non-native/cryptogenic species presence. - Case study (Murmansk): First-order propagation highlights many potential targets even at first step, complicating management. Second-order nodes (Murmansk|Hammerfest vs. Murmansk|Tromso) reveal distinct, fewer high-risk next-step targets (e.g., Varandey and Longyearbyen vs. Varandey, Dudinka, Advent Bay, Torshavn, Reykjavik), enabling more precise, cost-effective interventions.

Findings indicate increasing ballast-mediated invasion risk into and within the Arctic due to intensifying shipping concentrated through emerging hubs (e.g., Murmansk, Narvik). Environmental filtering shows that even sensitive species have viable pathways (e.g., to Churchill), while euryhaline and warmer-tolerant taxa can spread broadly from Europe and NE Asia. Within-Arctic dispersal is structured by path-dependent ship movements; SF-HON reveals clusters and inter-cluster bridges that can be targeted to prevent cross-region spread. Ports with modest direct connectivity (e.g., Reykjavik, Dutch Harbor) can be high-risk receivers via indirect dispersal, suggesting that management strategies should incorporate higher-order movement patterns for long-term prevention. Policy implications include augmenting IMO ballast water management with geographically targeted controls at hubs and inter-cluster links, prioritizing surveillance (including eDNA) at high-risk ports, and strategically locating on-shore treatment facilities. Direct-dispersal rankings can guide short-term actions, while indirect-dispersal (SF-HON) rankings inform long-term prevention and network-level resilience.

This study integrates shipping, environmental, and biogeographic data into a unified network framework to assess ballast-mediated introduction and dispersal risks in the Arctic. It documents increasing shipping intensity, the emergence of Arctic hubs concentrating risk, and identifies key introduction corridors and environmentally constrained pathways. By applying higher-order network modeling, it captures path dependencies in within-Arctic dispersal, revealing clusters, inter-cluster bridges, and ports at elevated indirect risk—providing actionable insights for targeted management. The framework supports risk-based prioritization of surveillance, prevention, and response, and can be adapted to species-specific tolerance data and extended to biofouling as data improve. Future work should expand datasets (broader shipping and ballast reporting, especially outside the U.S.), incorporate higher-frequency environmental observations, strengthen biological submodels linking discharge to establishment, integrate real-time monitoring (e.g., AIS, eDNA), and validate predictions through coordinated biodiversity surveys across Arctic ports.

- Risk estimates are relative, not absolute; model parameters and data limitations preclude precise absolute probabilities. - Shipping data (LLI) have limited coverage in some Arctic regions (e.g., northern Canada). Ballast discharge data (NBIC) are U.S.-centric (Alaska), requiring extrapolation to other regions. - Environmental filtering uses coarse annual average temperature and salinity and assumes negligible temperature-salinity interactions, plasticity, and adaptive evolution; establishment tolerance ranges are generalized across taxa. - Linear trend projections are simplistic and likely conservative given complex future climate-shipping dynamics. - Biofouling is not explicitly modeled, though it can be a comparable or greater risk vector. - Validation is constrained by sparse, non-standardized biodiversity surveys across ports and time lags between introduction, establishment, and detection. - Independence assumptions for aggregating trip risks and pathways may not hold in all cases.