Flood resilience is the half of a city's water problem where there is too much of it: the gauges that see it coming, the drains, tunnels and gates that hold it, and the rule that decides what happens when a reading crosses a line. Measuring a flood has become cheap. Owning the lever attached to it has not.
Twenty-nine documented projects span 25 cities and 18 countries, from a tunnel begun in 1975 to an insurance policy signed in 2026. Supply, leakage, metering and quality — the too-little and too-dirty half — belong to the water management guide. The organising question comes from our essay on when a reading actually triggers action; flooding is where its ladder is clearest.
Four layers recur, often bought out of order. Mapping: a 2021 zoning overlay made Boston's 2070 flood maps binding on new buildings, and volunteers in Dar es Salaam put some 450,000 buildings into OpenStreetMap. Measuring: New York's FloodNet reads street-level depth every minute from 265 solar-powered sensors; Metro Manila's EFCOS has telemetered the Pasig–Marikina basin since 1993. Warning: Singapore texts residents at 50, 75, 90 and 100 per cent of canal depth; Jakarta mounts loudspeakers on poles in the neighbourhood itself. Acting — moving water, or closing something — is a capital works programme, and the layer usually missing.
| City | Instrument | What the reading is permitted to do |
|---|---|---|
| Osaka | Seven tide gates on five rivers | Close themselves on a national tsunami advisory — no human in the loop, because the signal is binary |
| Lagos | Parametric flood insurance, 2026 | Satellite-observed flood extent past a threshold releases up to US$7.5m — the observation is the decision |
| Rio de Janeiro | 33 rain gauges, 164 sirens | Above 55 mm in an hour, Civil Defence confirm the protocol and sound the sirens remotely |
| Bangkok | Radar and 4 m street-level forecast | Informs a 24-hour control room; crews still work the 200 pumping stations by hand |
| Busan | Risk portal, 201 neighbourhoods | Grades every area safe to danger — and stops there |
Less than the sector implies. FloodNet has peer-reviewed backing for its sensing method — evidence the instrument works, not that flooding fell. The only peer-reviewed evaluation of an alerting rule here is Rio's: Calvello et al. (2015) found the thresholds kept missed alerts low but generated many false alarms, one zone's probably set too low.
The rest is operator-reported and labelled as such. Shanghai reports 8 million camera snapshots screened automatically in 2025 and 378 waterlogging incidents detected — the authority's own count. Chicago's TARP reports overflow days falling from roughly 100 a year to 50; no independent evaluation exists. Lagos is sharpest: the trigger threshold and the premium are both undisclosed, and no payout had been reported by late July 2026.
Almost never at the sensor. Busan built its portal after three drivers drowned in the Choryang underpass in July 2020, and warned at launch that its own readings might be wrong: too few instruments, an untested algorithm. Four years after Choryang, 41 of the city's 57 underpasses still had no automatic barrier. Medellín's SIATA has prompted real evacuations; in January 2023 its cameras captured a fatal flash flood in real time at an underpass nobody closed.
The second failure is capacity. Bengaluru runs about 100 telemetric rain gauges and 105 level sensors across 201 mapped flood points, but the drains hold roughly 80 mm of rain — independent reporting is blunt: the network helps position rescue crews, it does not prevent the flood. On 3 November 2025, 131.5 mm fell on a Bangkok system designed for 60 mm and the roads went under anyway.
The third is the works the gauges warn about. Metro Manila measures its floods in public, minute by minute; the defences became the country's largest corruption case: a 2025 validation flagged 421 flood-control projects nationwide as ghosts — 14 of them confirmed non-existent — and in May 2026 the Ombudsman filed plunder charges over them. Buenos Aires is the quieter version: the World Bank's evaluation arm rated the loan Moderately Unsatisfactory — the tunnels got built, the monitoring network never did.
And sometimes the honest answer is not a sensor. After the July 2011 cloudburst, Copenhagen committed DKK 20 billion over twenty years to more than 300 retention parks, cloudburst streets and tunnels, sized by hydraulic modelling. The modelling was the cheap part.
The practical takeaway: buy the consequence before the sensor. Every strong entry here wires a reading to something that moves — a gate, a siren protocol, a payout. Every disappointing one is a city that can describe its flood in extraordinary detail while the water rises.
A warning system tells people water is coming; a defence stops or stores it. The distinction decides what a budget buys. Bangkok can forecast street-level flood depth at four-metre resolution and still could not pump faster than the 131.5 mm that fell on a 60 mm design capacity in November 2025. Bengaluru runs about 100 telemetric rain gauges and 105 level sensors, but its storm-water drains hold roughly 80 mm of rain, so independent reporting describes the network as useful for positioning rescue crews rather than for preventing floods.
No entry in this atlas demonstrates that on its own. The evidence supports the instruments rather than the outcome: New York's FloodNet has peer-reviewed backing for its street-level sensing method, and the one peer-reviewed evaluation of an alerting rule — Calvello et al. (2015) on Rio de Janeiro's rainfall thresholds — found few missed alerts but a large number of false alarms. Damage reduction in this record comes from works, not readings, and the sensing is what targets and verifies them.
New York's FloodNet is the only entry that publishes a unit price: under about $300 per sensor, with the hardware designs open-source on GitHub. The city committed $7.2 million to expand the network from 31 to 500 monitored locations over five years. The sensors are the affordable half of a flood programme — Copenhagen's cloudburst plan runs to DKK 20 billion, and Miami's own modelling puts its full flood-defence buildout at $3.8–5.4 billion.
Four financing models appear in this record. Ratepayers: Washington DC funds its roughly $3.5 billion of tunnels through a charge on every property's impervious area. Parcel tax: Los Angeles County's Measure W taxes impermeable surface at 2.5 cents per square foot. Borrowing: Miami's 2017 Forever Bond earmarked $192 million for sea-level rise and flooding, and Buenos Aires built the Maldonado tunnels on a World Bank loan. Insurance: since March 2026 Lagos State has held a parametric policy paying up to US$7.5 million per event, though a German donor fund covered 90 per cent of the first-year premium.
Four flood systems, four answers to the question that organises much of this atlas — whether a measurement triggers anything. The ladder runs from a reading that acts entirely by itself down to a network that watches everything and decides nothing.
| Project | Since | Status | Outcome | Evidence | Run by | Cost | What the reading may do |
|---|---|---|---|---|---|---|---|
| Tsunami-Alert Automatic Tide Gate ClosureOsaka, 🇯🇵 | 2018 | Live | Ongoing | Official | Regional government | not public | The reading acts by itself: the tide gates close automatically on the national alert, with no human in the loop. |
| Bangkok Flood Control Centre: Drainage Telemetry & Street-Level ForecastingBangkok, 🇹🇭 | 2021 | Live | Ongoing | Official | City government, Utility or public operator | not public | Staff decide with the data: radars and level sensors feed a control centre staffed around the clock, and crews operate the pumps. |
| Busan Urban Flood Warning System (Busan Safety ON)Busan, 🇰🇷 | 2023 | Live | Ongoing | Official | City government | not public | Warning only: the portal shows risk and alerts residents, and the city itself cautioned at launch that the readings could be wrong. |
| SIATA Early Warning NetworkMedellín, 🇨🇴 | 2010 | Live | Ongoing | Media | Research institution | not public | A research-run regional network that watches everything — including, once, a fatal flash flood in real time that no agency acted on. |
The hub sorts flood systems by what the reading is allowed to do — warn, inform an operator, or move a gate. This shows the hardware at the far end of that chain, which is what Osaka's tide gates close and what Bangkok's crews operate by hand.
Picked from a small list of checked channels because it illustrates something on this page. We do not produce these videos and they are not used as evidence for anything stated here — nothing loads until you press play.
