Port Disruption Watch

Ports to watch

What to do with an alert

• Shippers and forwarders: for cargo due at a flagged port, check the terminal's notices and your carrier's advisories, and hold a booking at one of the listed nearby ports that isn't flagged.
• Planners: add buffer to delivery dates that depend on a flagged port. A 20% chance means about 1 in 5 of these weeks sees calls fall by 30% or more.
• Analysts: "why" shows which signals pushed the chance up. "Calls already falling" is the strongest signal; a storm alert on its own is weaker than it looks.
• Everyone: most disruptions are not foreseeable from this data. Use it to decide where to look first, not as a forecast of everything.

Map

Where the risk is

Expected number of disrupted ports by region: the sum of every port's chance.

Track record

Ships returning to Suez: a forecast test

How this works

A disruption is a week when a port's ship calls fall at least 30% below its usual level (the median of the 13 weeks before) and far enough to stand out from its normal week-to-week swings, and the port didn't drop the same way a year earlier (so regular holiday dips don't count). Every Tuesday IMF PortWatch publishes port calls up to the previous week. Right after, a model trained on 2020 onward (a LightGBM classifier) gives each port a chance of a disruption in three weeks: last week (already over, but its data isn't out), this week and next week. Inputs: recent calls against normal, past drops, time of year and holidays, storm and disaster alerts from GDACS, traffic through the nearest chokepoints, and nearby ports' calls. Alerts are the top 1% of ports each week. The track record uses only predictions made without seeing the outcome: 2023 to the model's launch is a backtest (retrained every 12 weeks on past data only), after that the live weekly runs. Known gap: typhoons often reach ports before GDACS lists them, so typhoon disruptions are mostly missed.