Port congestion is a system state, not a polygon
In brief
- Port-congestion studies measure different objects: queues, waiting time, turnaround, capacity, dwell, delay cost, emissions and landside occupancy.
- A fixed-zone metric can report recovery when a holding policy merely moves waiting vessels elsewhere.
- I treat congestion as a latent system state generated by arrivals, service capacity, priorities and disruptions; location is evidence, not the definition.
INTERACTIVE METHOD ARTIFACT
Congestion measures, step by step
A six-week synthetic port forces heterogeneous congestion measures through the same arrival surge, queue build-up, holding-policy change and recovery.
If the interactive embed is blocked by the source, use the link above. The artifact contains the complete visual sequence and paper-by-paper method walk-through.
A queue can move without congestion changing.
The supplied artifact makes this visible by changing only the spatial expression of waiting. An arrival surge first builds a queue in charted anchorages. A later holding rule moves new arrivals beyond a holding line, where they drift rather than anchor. Demand and service pressure remain, but the geometry of the queue changes.
A metric tied to one anchorage polygon can therefore report an improvement when the port has not improved operationally.
Congestion is a latent system state produced by the relationship between arrival demand, service capacity, priorities and disruption. A queue is one observable consequence of that state.
There is no single congestion variable
The methods in the artifact end with different quantities:
| Quantity | What it captures | What it can miss |
|---|---|---|
| Queue size | immediate waiting demand | service speed and priorities |
| Waiting time | vessel-level delay | future queue growth |
| Turnaround | end-to-end port delay | where the delay occurred |
| Operating capacity | service ability under current conditions | transient disruption |
| Ultimate capacity | saturation limit | near-term operational state |
| Dwell / speed indices | local movement degradation | queues outside the monitored zone |
| Delay cost | economic consequence | physical mechanism |
| Emissions-at-anchor | burden of anchored waiting | drifting or virtual queues |
The correct target therefore depends on the decision the model is expected to support.
The moving-queue stress test
The most revealing comparison is what happens after the holding rule moves waiting vessels offshore. Methods that infer waiting events or follow dwell into the new holding area remain informative. Measures defined only inside official anchorage zones can collapse even while ships continue to wait.
This is a general failure mode: when policy changes where congestion is expressed, a location-defined metric can mistake relocation for recovery.
The stress test also exposes a second issue. Some methods split one drifting wait into several short events when speed briefly rises above a fixed threshold. A congestion estimate is then sensitive not only to the port state, but to the segmentation rule used to construct its events.
Queue models are useful—but assumptions matter
Queueing formulations add structure that pure indices lack. Arrival rate, inter-arrival variability, queue length and service rate are meaningful system variables. They can support operating-capacity and saturation analyses.
However, the artifact also shows why validation must be independent. If a service rate is solved specifically to reproduce an observed queue, then using a closely related queue identity as the principal validation statistic can become partly circular. Similarly, a first-come-first-served assumption may be violated by priority policies even when the aggregate queue fit appears reasonable.
The useful question is therefore not only whether a queue model fits, but which operational assumptions are identifiable from the available data.
A world-model formulation
I would separate four layers:
- Arrival process — demand by vessel class, terminal and time.
- Service process — berth availability, cargo operations, tug/pilot constraints and effective service rates.
- Waiting state — anchorage, drifting, holding-area and virtual-arrival waiting.
- System state — latent congestion inferred from the interaction between demand and service.
The model becomes
| **P(congestion_t | arrivals, service process, waiting state, priorities, disruptions, observation quality).** |
A world model should then predict how these variables evolve jointly, not merely extrapolate one congestion index.
From prediction to intervention
Port operations ultimately need a counterfactual quantity:
Had the same arrivals occurred under a different service, berth-priority or holding policy, what would waiting time, fuel use and recovery time have been?
Candidate interventions include Just-In-Time speed advice, berth resequencing, priority changes, temporary capacity changes and alternative holding instructions.
This is where causal modelling becomes operationally relevant. It distinguishes an arrival surge from a service slowdown, and a policy-induced queue relocation from genuine recovery.
Evaluation
I would evaluate at four levels.
State reconstruction
- waiting-event precision/recall;
- queue-size and waiting-time error;
- robustness when the queue relocates.
Forecasting
- probabilistic waiting-time and queue forecasts;
- recovery-time prediction;
- calibration by terminal and vessel class.
Intervention
- treatment-effect estimates for JIT or berth policies where evidence exists;
- sensitivity to unobserved confounding;
- spillover across terminals and vessel classes.
Impact
- waiting hours;
- fuel / CO₂ proxy;
- service reliability;
- congestion beyond the formal port boundary.
The key question is not how many ships are in the anchorage? It is whether demand is persistently exceeding the system’s ability to serve it, and which feasible intervention would change that state.
References and method companion
- Chen et al. — Monitoring and evaluation of ship operation congestion status at container ports based on AIS data, Ocean & Coastal Management (2023)
- IMO / GreenVoyage2050 — Just-In-Time arrival and maritime decarbonisation guidance
- Port Call Optimization Taskforce
- Interactive congestion-method comparison
The interactive comparison is a methodological stress test on a common synthetic port. Its outputs should not be read as a leaderboard of published empirical performance.
The embedded artifact above contains the complete visual sequence from the supplied congestion companion: every paper walkthrough, map, queue trace, capacity illustration and evaluation comparison. The article uses that material to isolate the modelling question rather than reproduce the artifact page by page.