It is 6:40 on a July evening and the ward office phone has buzzed four times. One message says heavy rain is likely. One says an alert has been issued. A third, from a junior engineer at the canal, just says "level going up". The ward officer stands by the window with her raincoat half on, watching the road, trying to decide whether this is the evening she pulls crews off the market road and sends them to the underpass.
A flood alert and a flood warning are different instructions, and a city that sends both in the same words has given its people a feeling, not a decision.
The short answer: an alert means conditions are building and teams should get ready; a warning means flooding is likely or under way in named places and teams should act now. In practice an urban flood early warning system for municipal corporations should run three levels (watch, alert, warning), each tied to a ward, a trigger, an owner and a first action. Our guide on how an urban flood early warning system works covers the inputs; this post is about the levels.
An alert means get ready; a warning means act now
The difference is not severity alone. It is what you expect the person reading it to do in the next ten minutes. An alert asks people to check, stage and stand by. A warning asks them to move crews, open diversions and tell citizens.
Confusion starts when every message is called an "alert". After a few false starts, people learn to read all of them as background noise. Then the one message that mattered gets the same shrug as the twelve before it.
| Level | What it means | What the reader does |
|---|---|---|
| Watch | Heavy rain is forecast or has started; the network is coping | Check pumps and gauges, confirm who is on duty |
| Alert | Rain rate and channel trend are moving toward a threshold for named wards | Stage crews and equipment, warn the control room |
| Warning | A threshold is crossed or water is already standing in named wards | Send crews, open diversions, message citizens |
IMD colours stop at the district boundary, and floods do not
Indian city teams already live with a colour code. The India Meteorological Department issues colour-coded weather warnings: green for no action, yellow to stay updated, orange to be prepared and red to take action. These are useful and every engineer knows them.
The gap is scale. A district forecast cannot tell a ward officer which underpass fills first, or whether the channel behind the bus depot is keeping up. Two wards under the same orange forecast can face very different evenings, because one has a silted drain and a pump that has not reported since noon.
So we treat the weather colour as an input to the city's own levels, not a substitute for them. An orange forecast can open a watch for every low-lying ward. Only local readings, rain rate and channel trend should move a ward up to alert or warning.
Three levels are enough if each one has an owner and a first action
Cities sometimes design five or six levels to look thorough. In our view three is the number people can remember at 2 am with water at the door. More than that and nobody recalls what level four was meant to trigger.
What makes a level real is not its colour. It is four facts written down before the monsoon:
- The trigger: which readings move the ward up, for example rain rate plus channel trend, never rainfall alone.
- The area: the wards and drain segments the level covers, not the whole city.
- The owner: one named person who acknowledges it, with a deputy if they do not.
- The first action: one concrete step, such as staging the pump crew at a named junction.
If any of the four is missing, the level is a notification. People will read it and wait for someone else to move.
A worked example: one ward, one channel, fifteen minutes that change the call
Here is an illustrative case with simple numbers; it is not a customer result. Say a ward drains into a channel that overtops a low road when the water reaches 1.8 m. At 6:30 pm the channel stands at 1.2 m and is rising 10 cm every 15 minutes.
The headroom is 1.8 minus 1.2, which is 0.6 m. At 10 cm per 15 minutes that is six steps of 15 minutes, so about 90 minutes before the road goes under. That is an alert: stage the crew, check the pump, no need to close the road yet.
At 6:45 the rain rate jumps and the channel now rises 20 cm every 15 minutes, from 1.3 m. Headroom is 0.5 m, which is two and a half steps of 15 minutes, so about 37 minutes. At the same moment the pump on that channel stops reporting. That combination should move the ward to warning: crews go now, the diversion opens, citizens on the low road get a message.
The rain did not double the danger. Losing nearly an hour of headroom in fifteen minutes did, and only a level tied to the channel trend can see that.
Notice what moved the level. Total rainfall barely mattered. What mattered was the rate of rise, the remaining headroom and a pump that had gone quiet. The ward officer at the window now has a clear call instead of four messages and a hunch.
A silent gauge should raise the level, not lower it
Missing data is the trap most rule sets fall into. If a pump stops sending status, the easiest code path is to show the last good reading. On screen the pump looks fine, and the ward stays at alert when it should be at warning.
We would rather the rule say the opposite: unknown is treated as worse than the last reading. In CityWatch, pumps show as running, offline or "no telemetry", never assumed, and silent sensors are flagged. That flag should count as a reason to step up the level and send someone to look. We make the full case in a silent pump is not a running pump.
The strongest objection: levels breed false alarms
Every alert costs fuel, overtime and goodwill, and if half of them come to nothing the field team stops believing any of them. City engineers raise this for good reason, and it is the real reason many cities keep warnings vague.
Our answer is to separate who is told at each level. A watch goes to the control room only. An alert goes to the ward officer and the crew lead. A warning, the expensive one, goes to crews and citizens. False alarms at watch level cost almost nothing, so you can afford to be cautious there and strict at warning.
The second answer is review. If the system proposes the alert and a person confirms it, every confirmed and rejected alert becomes data. After a few storms the thresholds can be tuned ward by ward, using what actually happened instead of opinion.
Every level change belongs on the record
When a ward moves from alert to warning, the time, the readings and the person who acted should be written down automatically. That record answers the morning question fairly: did we have the minutes, and did we use them?
It also protects the people on the ground. A crew that moved within minutes of a warning is covered by the record, even if the road still flooded because the channel was over capacity. For the wider argument, see reactive vs early-warning flood response, and browse the CityWatch guides for checklists and playbooks.
What to do this week
- List the ten wards that flooded first last monsoon and the one channel or underpass that fills first in each.
- For each, write down the reading at which it starts to overtop, even if it is a rough mark on a wall.
- Agree three levels (watch, alert, warning) with one trigger, one owner and one first action each.
- Decide who is told at each level, and keep citizens for warning only.
- Mark every pump or gauge that has not reported this week as unknown, and send someone to check it.
None of this needs a new building or a new department. It needs the words alert and warning to mean two different things, and a ward officer at a window who knows which one she is holding.