Storm Surge: Close the Last-Mile Gap Between Cyclone Warning and Coastal Evacuation

Diagnosis

The curated note is blunt: storm surge is "cyclone-driven coastal surge, often the killer in BD." That single line carries the whole problem. In a landfalling cyclone, it is rarely the wind that kills at scale on the Bangladesh coast; it is the wall of seawater pushed onto low-lying delta land. Yet this hazard sits in the platform as a tier-1, nowcast-horizon risk with current_state null and data_status set to needs_collector. In plain terms: the country treats the deadliest part of a cyclone as the part it currently cannot measure or forecast operationally inside this system. That is the gap that turns a forecastable event into a body count.

Why now: surge is a nowcast hazard, meaning the decision window is hours, not weeks. A warning that arrives late, that quantifies wind but not water, or that does not translate cleanly into a "leave this village now" instruction, is a warning that fails exactly when it matters. The lead responsible body is the Ministry of Disaster Management and Relief (MoDMR), and the absence of a live surge indicator means MoDMR is effectively flying the most lethal phase of the response on instinct.

Recommended actions

1. Stand up a surge nowcast feed (owner: MoDMR). Mechanism: commission, under MoDMR, an operational storm-surge forecast product that converts an incoming cyclone track into predicted water height and inundation extent by coastal upazila, refreshed on the nowcast cadence. Observable signal that it is working: the platform's current_state for storm surge stops being null and updates automatically when a system enters the Bay, replacing needs_collector with a live feed.
2. Separate the surge warning from the wind warning (owner: MoDMR). Mechanism: a standing MoDMR circular that requires every cyclone bulletin to issue a distinct surge alert level keyed to predicted water height, not folded into a single storm category. Signal: coastal warnings name expected surge depth and the upazilas at risk, and evacuation orders cite that depth.
3. Pre-bind warning levels to evacuation triggers (owner: MoDMR). Mechanism: a published trigger table in MoDMR standing orders that maps each surge alert level to a mandatory action (open which shelters, evacuate which zones, by which hour before landfall). Signal: evacuation begins on the trigger automatically, before landfall, rather than after a separate human deliberation each event.
4. Run an off-season surge drill on the live feed (owner: MoDMR). Mechanism: a MoDMR-led table-top and field drill using a simulated track piped through the new nowcast, testing whether the alert reaches the last mile and whether shelters open on the trigger. Signal: documented time from "feed fires" to "shelter doors open," tracked and shortened drill over drill.
5. Tie the feed to shelter readiness (owner: MoDMR). Mechanism: link the surge nowcast to a MoDMR shelter-status register so that when a surge level fires, the system flags which designated shelters in the at-risk zone are confirmed open and reachable. Signal: every fired alert resolves to a named, confirmed-open shelter per at-risk community.

Sequencing (first 12 months)

First, close the data hole: action 1, the surge nowcast feed, is the keystone because nothing downstream can be automated while current_state is null. Standing that feed up unlocks everything else, the distinct surge alert (action 2), which in turn makes the pre-bound evacuation triggers (action 3) meaningful. Once the feed and triggers exist, the off-season drill (action 4) is the cheap, high-value test that proves the chain end to end before a real cyclone arrives, and it surfaces the shelter-readiness gaps that action 5 then closes. Sequence the drill into the calm window before the next cyclone season so the live feed is battle-tested, not debuted, under a real storm.

Risks and constraints

The binding constraint is institutional, not technical: surge forecasting, warning issuance, and evacuation authority can sit across different bodies, and a feed that MoDMR cannot translate into a binding evacuation order changes nothing. The needs_collector status is itself a warning that sustained data-pipeline ownership has not been funded; a one-off build that no one maintains will silently rot back to null. Fiscally, the recurring cost is the running feed and the drills, not the initial software, and recurring lines are the first cut in a tight budget. Politically, mandatory pre-bound evacuation triggers remove discretion from local officials, which invites resistance unless the trigger table is published and backed at the ministry level.

Bottom line

Storm surge is the lethal core of every Bangladesh cyclone, and it is the one part of the event this platform currently cannot see, with current_state null and the hazard still flagged needs_collector. MoDMR's first move is to give surge its own live nowcast and its own warning level, then pre-bind that warning to automatic evacuation, so the deadliest hours stop depending on improvisation.