Flash floods and river floods: the hydrology of warning time

"Flood" covers two situations that need different engineering. In a river flood the water arrives hours or days after the rain, and the question is how well you can forecast. In a flash flood it arrives within minutes to a few hours, and the question is how fast you can detect and act. This article shows where the difference comes from, how to calculate the warning time a system can really give, and what that means for the design.

The difference is a ratio of times

The behaviour of a flood depends on two times:

• Storm duration, T_rain: how long the heavy rain lasts.
• Time of concentration, tc: how long water from the farthest part of the catchment takes to reach the outlet.

When the rain lasts as long as tc or longer, the whole catchment contributes at once and the peak is highest. Small, steep catchments have a short tc, so a short, intense burst can produce the full peak. The Kirpich estimate, tc = 0.0195 · L^0.77 · S^−0.385 in minutes for channel length L in metres and slope S in m/m [1], gives:

(Kirpich's equation was derived from small agricultural basins, so these are orders of magnitude.) A common operational definition treats a flood as a flash flood when it develops within a few hours, often taken as about six, of the causing rain. That is a statement about the first three rows: the river responds before anyone downstream has had time to look at a forecast. Studies of Himalayan cloudbursts use exactly this relation. The analysis of the August 2010 Leh cloudburst, for example, used the ratio of time to peak to time of concentration to estimate how long the storm lasted [2].

How big, and how fast, is the rise?

The rational method estimates the peak discharge from a storm that lasts at least tc:

Q_peak (m³/s) = 0.278 · C · i · A

with runoff coefficient C (0 to 1), rainfall intensity i in mm/h over the time of concentration, and catchment area A in km². For a 10 km² steep catchment with C = 0.6 and an intense burst of i = 80 mm/h:

Q_peak = 0.278 × 0.6 × 80 × 10 ≈ 133 m³/s

To see what that means at a gauge, take a channel 20 m wide with a mean velocity of 3 m/s. From Q = w · h · v, the depth is 133 / (20 × 3) ≈ 2.2 m. If the stream was nearly dry before and tc is about 30 minutes, the level rose roughly 2 m in half an hour, or about 4 m an hour. This is a rectangular-channel simplification, and real hill channels are irregular, but the order of magnitude is the point: a rate of rise of several metres per hour is normal in a flash flood, and an hourly reading would capture one or two points on it.

How long does the water take to reach people?

The distance a rise travels is covered at the kinematic wave speed, c = (5/3) · v for a wide channel with Manning friction, a little faster than the water itself. The time from a gauge to a community is distance / c:

A river flood in a large lowland basin lives in the right-hand columns and the slow rows: tens of hours between a gauge far upstream and the town. A hill-stream flash flood lives in the left-hand columns and the fast rows: 10 to 30 minutes between a sensor and the village. The same equation applies to both; only the numbers differ.

The lead-time budget

A gauge does not deliver all the travel time. Every step in the chain takes some:

T_lead = T_travel − (T_sample + T_comms + T_decision + T_dissemination)

Compare two situations, each with a gauge upstream of the community:

The river flood tolerates a person in the loop and a slow reporting interval and still leaves almost two hours. The flash flood cannot: with the same slow, manual chain (7.5 + 2 + 20 + 10 = 39.5 minutes), the whole 25 minutes of travel would be gone before the siren sounded. That is why systems for flash floods put the decision on the device, report frequently, and use direct channels such as sirens.

What people need to do with the time sets the requirement. Moving 500 m to higher ground on foot at about 3 km/h takes 10 minutes. Evacuating a settlement with children and elders takes much longer. A lead time shorter than the time people need is no warning at all, so planning starts from the response and works back.

Evidence on what warning is worth

The benefit of early warning has been estimated at global scale. The Global Commission on Adaptation reported in 2019 that just 24 hours' warning of an impending hazard can cut the resulting damage by about 30 %, and estimated that investing in such systems in developing countries would avoid several billion dollars of losses each year [3]. The WMO and UNDRR describe early warning systems as returning more than ten times their cost, and organise them around four pillars: knowing the risk, observing and forecasting, disseminating and communicating the warning, and being prepared to respond [4]. These are averages across hazards and countries, so they say that the investment is worthwhile. They do not predict the result for any one river. For flash floods the effect is carried by the last two pillars: a siren is worth nothing if it does not sound in time, and a warning is worth nothing if people do not know what to do.

Triggers: rainfall, level and rate of rise

Three kinds of trigger are used, usually together:

• Rainfall thresholds. Automated local flood warning systems in the USA, called ALERT systems, trigger notifications from rain gauges when depth in a short window exceeds set values. One county programme uses 0.25, 0.40 and 0.75 inches in 15 minutes, corresponding to intensities of about 25, 41 and 76 mm/h [5]. These give the earliest cue, before the river has responded, but they are blunt because rainfall over a catchment is unevenly spread and soil wetness changes how much runs off.
• Level thresholds. Absolute stage above a mark. Reliable and simple, but late on steep streams.
• Rate of rise. For example, more than 0.3 m in 5 minutes on two consecutive samples. It catches fast events earlier than an absolute threshold, as the 4 m/h example above shows (about 0.33 m in 5 minutes).

River floods lean on forecasts, which combine rainfall, upstream levels and model output. Flash floods lean on observation and on rules that run without human review.

False alarms and missed events

Every threshold is a trade-off between two errors. Define them so they can be measured:

Probability of detection (POD) = hits / (hits + misses)
False alarm ratio (FAR)        = false alarms / (hits + false alarms)

Lower the threshold and you detect more events but you also alarm for events that do not become floods. Communities that hear too many false alarms stop responding. Practical ways to keep both errors low:

1. Two stages. An advisory when conditions are ripe, a warning when the river is rising. The siren is reserved for the second.
2. Confirmation. Require two consecutive samples, or two of three signals (level, rate of rise, rainfall), before a siren sounds.
3. A camera or second sensor at critical sites to confirm what the radar reads.
4. Record every trigger with the readings around it, so thresholds are tuned from real events.
5. Test the chain from sensor to siren before every monsoon and after any change.

Putting the two together

A single river often has both, so a basin network should be designed for the fastest threat in the catchment, and the slower threats then take care of themselves.

Where ClimMaTech fits

For slow river floods, Tarang and Pravaah provide the level and surface velocity readings that feed forecasts. For flash floods, Doordrishti adds camera confirmation and on-device detection at the critical site, and Rakshak sirens warn the community without waiting for a person to press a button. For ClimMaTech systems, forecast lead time ranges from about 30 minutes to 48 hours depending on the conditions upstream. See how a flood early warning system works, river monitoring in hill catchments, and the flood early warning solution.

References

1. Texas Department of Transportation, Hydraulic Design Manual: Time of Concentration (Kirpich method). txdot.gov.
2. Study of cloudburst and flash floods around Leh, India, during August 4-6, 2010, Indian Academy of Sciences repository. repository.ias.ac.in.
3. Global Commission on Adaptation, Adapt Now: A Global Call for Leadership on Climate Resilience (2019), as reported in Grist.
4. WMO, Early Warnings for All: Executive Action Plan 2023 to 2027. public.wmo.int.
5. Coconino County, Arizona, Flood warning and ALERT system rainfall thresholds. coconino.az.gov.

Worked values are computed from the stated relations (Kirpich, rational method, kinematic wave speed). They show the scale of each effect, are simplifications of real channels, and should be replaced with event data from the catchment in question.