Port Arrival Forecast Accuracy (PAFA)

GMI-MAR-030Maritime IndustriesPilot-Ready

Definition

Mean absolute error between AI-forecasted and actual vessel arrival at the planning decision horizon.

The Measurement Gap

What This Metric Measures

PAFA measures the mean absolute error between the AI-forecasted vessel arrival time used for berth and resource planning and the actual arrival time. The measurement is taken at the decision-relevant horizon: the lead time at which terminal resource allocation decisions are made, typically 24 to 72 hours before arrival.

The Decision It Supports

Are vessel arrival forecasts accurate enough to support the berth allocation and resource planning decisions being made based on them? The terminal commits cranes, trucks, yard slots, and labor based on the AI's arrival forecast. If the forecast is wrong by 6 hours, those resources are either idle or unavailable when the vessel actually arrives. PAFA quantifies the reliability of the forecast at the moment the terminal makes its commitment.

Why Legacy Metrics Cannot See This

Legacy port operations track actual vessel waiting time and berth utilization but do not measure the accuracy of the arrival forecast that drove the resource allocation. A terminal might report excellent berth utilization while consistently misallocating resources because the AI forecast was systematically 4 hours early: the berth was ready but empty for 4 hours, then scrambled to accommodate the actual arrival. The waiting time metric shows zero because the berth was available, but the resource allocation was wrong for the entire 4-hour window. PAFA exposes the forecast quality that drives downstream resource decisions, tracked at multiple horizons so the terminal knows when to commit firm allocations versus tentative plans.

Formula

Primary Formula

PAFA = (1/N) × Σ |Arrival_forecast_i − Arrival_actual_i|

Variable Definitions

  • Arrival_forecast_i = AI-forecasted arrival time at decision horizon for vessel i (datetime)
  • Arrival_actual_i = actual arrival time (pilot boarding or first line ashore) for vessel i (datetime)
  • N = number of vessel arrivals in the measurement period

Units: Hours. Lower is better. Track at multiple horizons (72-hour, 48-hour, 24-hour, and 12-hour) to see how accuracy improves with proximity to the arrival event. This multi-horizon view reveals when the forecast becomes reliable enough to commit firm resources.

Edge Cases

Vessels that cancel or divert should be excluded from PAFA calculation since they represent external events, not forecast failures. Congestion-induced waiting time requires separating pilotage arrival from berth arrival, a vessel arriving at anchorage on time but waiting for a berth is a port capacity issue, not a forecast accuracy issue. Minimum sample of 50 vessel arrivals per measurement period is needed for statistical reliability at each horizon.

Proposed formula, pending validation pilots.

At a Glance

Unit
Time (hours)
Trend Direction
Lower
Reporting Frequency
Rolling monthly
Category
Port & Terminal Optimization Integrity
← Berth Cascade Amplification Factor (BCAF)Stowage Deviation Stability Index (SDSI) →

Assess Your Organization

The AI-GPS Assessment evaluates your AI governance maturity and recommends KPIs like this one for your implementation roadmap.

Take the Free AI-GPS Assessment