Typhoons Saudel and Dolphin snarl East China ports; Shanghai waits hit ~9 days

The Analysis

Put the map of the Yangtze delta on the table and the distances are almost embarrassing. Shanghai and Ningbo sit a couple of hundred kilometres apart, under the same sky, on the same stretch of coast, drawing on the same pool of pilots and lashers and drivers. Two typhoons crossed that coast in quick succession, Saudel and then Dolphin, and when the waiting times were counted, Shanghai was showing roughly nine days and Ningbo roughly four.

Same weather, same fortnight, a gap of more than a factor of two. That gap is the part of this story worth reading, and it is not a weather story. Ports do not fail because wind blows. They fail because the shape of the network gives the water nowhere else to go.

What follows treats the delta as a network rather than a list of casualties: what the two storms actually took out, why two neighbouring gateways absorbed the same blow so differently, where the delay travels once it leaves the quay, when an importer feels it, and what to have in place before the middle of September.

The facts as given are thin and sharp. Shanghai vessel waiting around nine days. Ningbo around four. Singapore, which no typhoon touched, holding transshipment waits near two days on feeder variability. Shenzhen taking diversions from the northern gateways. Rolled cargo expected through mid-September on both the Asia-Europe and transpacific trades. Five numbers, and the interesting one is not the nine.

A typhoon does not stop a port the way a switch stops a light. It closes an envelope: pilot boarding, tug availability, the height at which a crane can safely work a stack, the truck gates on the landside approach. Then it passes, and the port restarts, and here is what gets lost in the reporting, the restart is not rate-limited by demand. It is rate-limited by the physical sequence of berthing, working, discharging, and clearing the yard behind it. Every hour of closure becomes a queue, and a queue is worked off at the port's normal rate, never at a catch-up rate.

One storm produces a queue that clears. Two storms produce something else, because the second arrives before the first queue has drained. Queue arithmetic is unforgiving in that way: remove service capacity for a period and the backlog does not dissipate at the speed it accumulated. The delay from two storms inside one month is not twice the delay from one. It is worse than twice, and the extra is not in the wind, it is in the interval between them.

Which brings the two numbers back. Shanghai at nine and Ningbo at four, in the same weather, is a statement about topology. A gateway's recovery speed is set by how many independent paths exist out of it. Shanghai's deep-water box capacity sits largely offshore, tied to the mainland by one bridge corridor and by barge, so when the quay reopens the cargo still has to cross water a second time to reach the factory that wants it. The berth restarts and the node is still half-closed, because the node has a second gate and one road through it. Ningbo's berth groups are strung along a longer coast with a denser short-sea and barge pattern feeding them, and more of its volume can leave by more than one door.

Read that as capacity and you conclude Shanghai is less efficient. Read it as topology and you get the useful version: the same storm costs more at the node with fewer exits, and no amount of crane productivity repairs a missing road.

The delay does not stay at the quay. Shenzhen absorbing diversions sounds like relief, and for the shipper who gets the slot it is. It is not relief for the network. A diversion moves congestion from a node that is full to a node that was running its own proforma, and the receiving node pays in its own connections. Shenzhen's boxes do not all end in Shenzhen; a share of them feeds the intra-Asia relay, so a squeeze at the delta exports itself southward along the feeder chain.

The Singapore number is the cleanest evidence that this is one system and not a set of separate ports. Nothing happened in Singapore. No wind, no closure. Yet transshipment waits sat near two days on feeder variability alone, which is the sound of the delta's problem arriving two thousand miles away, carried by the small ships that shuttle boxes between hubs on rotations measured in hours. A hub is only as punctual as its least punctual feeder.

For an importer the sequence runs roughly like this. The first signal is not a delay but a missed proforma window, and it lands within the first few days. The second is rolled cargo on the sailing you believed you were on, which you discover when the booking is re-confirmed, or when it is not. The third is a tightening booking book for the following two sailings, as everyone who got rolled competes for the same slots. The fourth arrives late and is less obvious: empties that should have been discharged, turned and reloaded are sitting in the queue instead, so export bookings start failing on equipment rather than on space, somewhere in the third or fourth week. That last one outlives the storm by a month.

Here is the piece the waiting-time figures hide, and it is the piece I would want a client to hold before making any decision this month. Schedule recovery is quantised. A weekly string does not recover in days; it recovers in cycles of seven. Nine days of waiting does not add nine days to your cargo. It pushes the cargo past one entire proforma window, so on the water the effective addition is a full cycle, and the next available slot is the next rotation. Ningbo's four days does not cross that boundary; it costs seven. Shanghai's nine crosses it; it costs fourteen.

The real spread between those two gateways in the first half of September is therefore not five days. It is seven, and the shipper who moved part of a book from one to the other gained twice what the headline suggested. That is the figure to carry into a planning meeting, and it is available to anyone who knows a string's frequency.

There is a second sleight of hand in the same number. Nine days is an anchorage dwell for the vessel, not a clock for your cargo. Your clock starts at discharge, and discharge is followed by yard dwell, and yard dwell runs longer than usual precisely because the yard was full when the gates reopened. The shipper's number is always worse than the port's number, by an amount nobody publishes.

A small piece of arithmetic, with the assumptions flagged because they are mine and not from the source. Assume an importer holding twenty-one days of safety cover on a product line, shipping weekly on an Asia-Europe string, with cargo due to load in the week after the closure. Assume the wait crosses the seven-day boundary, which the Shanghai figure does. The cargo lands fourteen days late against plan. Twenty-one minus fourteen leaves seven, so one weather event consumes two-thirds of the reserve. An importer holding fourteen days of cover is out of stock before the vessel berths, and no amount of expediting at destination recovers that.

Run it twice. Two events inside six weeks, which is what a bad typhoon season looks like, and a twenty-one day cover is gone entirely. The useful question is therefore not whether to add buffer. Everyone will say add buffer. The useful question is which fifth of the SKUs earns the buffer, because the arithmetic does not stretch to all of them.

Had the two storms been three weeks apart instead of days, the queue would have drained, Shanghai's number would have looked like Ningbo's, and nobody would have written this article. The damage lives in the interval, not in the wind. Which is the ten-year reading of a week like this: the binding constraint is not berth capacity, it is recovery time as a share of the sailing calendar. As volume concentrates into fewer, larger gateways, that share shrinks, and the same storm does more damage each decade without growing any stronger. Ten years ago the delta's problem was depth and cranes. Those were bought. What is scarce now is redundancy, and redundancy is the one thing a consolidating network produces less of every year.

What to do about it, in the order I would do it. Before the second week of September, pull every shipment with a proforma window between the seventh and the twentieth and mark the ones whose waiting time could cross a multiple of seven; those get re-planned, the rest can wait. Cap any single product line at about a third of its monthly volume on one delta gateway for the next four weeks, and be honest that the second gateway has to be a different string, not the same string berthing at a different terminal, or nothing has been bought.

Ask for the vessel's actual anchorage date instead of its ETA, and ask for destination yard dwell before promising a delivery date to anyone. Keep a Friday decision point running to the end of September: while the delta wait stays above seven days, the following month's allocation rolls forward rather than waiting to be rescued.

The trap in all of this is the obvious one. Diversion is an advantage only while it is uncommon. If a third of the market moves to the same second gateway in the same fortnight, its four days becomes nine, and the money has bought a relocated queue rather than a shorter one. The delta did not run out of water this month. It ran out of exits, and exits are shared.

↑ Back to the key points

— By Tomás Ferreira