Modelling Vessel Route Choice Behaviour During Lock Disruptions in Inland Waterway Networks

Abstract

Inland waterway transport is an important mode of freight transport because of its low transport costs, large cargo capacity, and relatively low carbon emissions compared with other inland transport modes. At the same time, its performance strongly depends on the availability of the waterway network, in which locks form critical bottlenecks. As a large proportion of Dutch lock infrastructure approaches the end of its technical service life, maintenance and renewal activities are expected to become more frequent. These activities can temporarily reduce lock capacity, increase waiting times, and cause vessels to reroute via alternative waterways. To minimise the hindrance to inland waterway traffic, insight is therefore required into how vessel operators adapt their route choices during such disruptions.

Existing inland waterway routing models generally determine routes by assuming that vessels select the shortest or least-cost alternative. However, transport behaviour literature shows that decision makers have heterogeneous preferences and may therefore respond differently to identical network conditions. In addition, congestion effects resulting from the route choices of other vessels are generally not incorporated into network-wide route choice decisions. This research therefore addresses the following main research question: how can route choice behaviour of heterogeneous inland waterway vessel operators during lock disruptions be modelled while accounting for congestion effects?
A simulation framework is developed and validated using the Rotterdam--Antwerp corridor as a case study, focusing on maintenance works at one chamber of the Kreekrak Locks. Historical IVS data from 2024 and 2025 are used to analyse vessel movements, lock operations, waiting times, travel times, vessel characteristics, and observed route choice behaviour. The inland waterway network is represented using EuRIS network data.

The proposed framework combines deterministic route determination with discrete choice modelling. The route determination component calculates feasible routes and travel times, including sailing time, lock passage time, and lock waiting time. Congestion effects are incorporated through empirical intensity--waiting time relationships derived from historical IVS observations. Route choice is subsequently modelled using Multinomial Logit and Panel Mixed Logit specifications. The Panel Mixed Logit model represents heterogeneity in travel-time sensitivity between vessel operators, while a disruption-specific coefficient is used to capture context-dependent route choice behaviour during maintenance works.

The historical analysis shows that the 2025 maintenance works substantially affected route choice behaviour. Under normal operating conditions, 95.35% of vessels travelled via the Kreekrak Locks, while this share decreased to 83.16% during the disruption. Over the same period, the average waiting time at the Kreekrak Locks increased from 22 to 206 minutes. Deterministic route determination alone did not reproduce the observed rerouting response. The best-performing specification was the Panel Mixed Logit model including a disruption-specific coefficient, predicting that 17.5% of vessels selected the alternative route during the disruption, compared with 16.8% observed in the IVS data. The results indicate that realised travel time alone is insufficient to explain rerouting behaviour. The disruption-specific effect shows that vessel operators also respond to the disruption context itself, while the significant variation in the travel-time coefficient confirms heterogeneity in travel-time sensitivity.

Several limitations affect the interpretation of the results. IVS waiting times were found to underestimate delays during severe congestion, which limited the influence of the congestion mechanism. In addition, the model was estimated and validated using highly comparable disruption cases, and the explanatory variables were constrained by the information available in IVS data.

It is concluded that route choice behaviour during lock disruptions can be modelled by combining deterministic route determination with a Panel Mixed Logit route choice model. Congestion effects can be incorporated through empirical intensity--waiting time relationships, although their effectiveness depends on the quality of the available waiting time observations. Future research should improve waiting-time estimation using AIS data, include additional explanatory variables, and validate the modelling framework under different disruption scenarios. The developed components can also be integrated into existing inland waterway routing models to improve their representation of congestion effects and heterogeneous route choice behaviour.