A flash flood is a flood that arrives within hours of its cause, most often intense rain over a small area. Three things set the hazard at a site: how hard rain falls over a few minutes, how quickly the ground delivers that rain to the site, and how fast the water rises and recedes. Daily rainfall records and end-of-event flood maps both smooth these out.
A flash flood risk assessment asks two questions about a site: how deep the water gets, and how quickly it gets there. The worked example below answers both at Msida, Malta, a valley street with a long record of flooding. It runs 15 design storms through a rain-on-grid model and states what the model leaves out.
Flash flood risk assessment in four points
- Daily rainfall data averages short bursts away, so the assessment works from design storms of minutes to hours, read from intensity-duration-frequency curves.
- Test several storm durations. At Msida the longest storm gave the highest water level, and the shortest storms reached 0.5 m first.
- Record depth through time. With no drainage modelled, the water at Msida passed 0.5 m between 8.5 and 37.5 minutes after the rain began, depending on the storm.
- Rate danger on depth and velocity together, and read any model without drainage as an upper bound.
The rain and the catchment set how fast a flash flood arrives
The US National Weather Service defines a flash flood as one caused by heavy or excessive rainfall in a short period of time, generally less than 6 hours. Flash floods can also occur without rain, after a levee or dam fails or when water held behind a debris or ice jam is released. An ordinary flood is a longer event that can last days or weeks.
Other common flood terms describe where the water comes from: pluvial flooding is rain ponding and flowing over the ground surface, surface water flooding is the usual UK term for the same thing, and fluvial flooding is a river or stream leaving its channel. A flash flood can be pluvial, as when a cloudburst floods a valley road, or fluvial, as when a small steep stream rises within an hour.
The speed of a flash flood depends on the rain and on the catchment, the area of land that drains to the site. Small, steep and paved catchments respond within minutes to an hour. Hydrologists measure the response with the time of concentration, the time runoff takes to travel from the farthest point of the catchment to its outlet. Malta’s Energy and Water Agency gives 76 minutes for the Birkirkara to Msida catchment, which covers 11.8 km² and drains a 4.6 km valley from Naxxar down to the Msida waterfront.
Short, intense rainfall drives most flash floods, and daily data averages it away
Rainfall intensity is a rate, a depth of rain per unit of time, and it only has meaning with a duration attached. The same 30 mm of rain is 120 mm per hour if it falls in 15 minutes and 1.25 mm per hour if it falls across a day. Drains, gutters and soils are overwhelmed by the rate of rainfall, so a flash flood risk assessment works at the durations that matter: minutes to a few hours.
Short bursts of rain are also intensifying as the climate warms. A 2021 review in Nature Reviews Earth & Environment (Fowler et al.) found that rainfall extremes lasting 1 to 3 hours are intensifying at about 7% per degree of warming, the rate expected from the extra moisture in warmer air, and faster than that in some regions. The review concludes that this has likely increased flash flooding at local scales. An earlier analysis of a 99-year record of hourly rainfall at De Bilt in the Netherlands (Lenderink and van Meijgaard, 2008) found one-hour extremes rising about twice as fast with temperature, about 14% per degree, on days with a mean temperature above 12 °C. The IPCC’s Sixth Assessment Report projects the heaviest daily rainfall to intensify by about 7% per degree of further warming.
Climate projections used for screening are usually daily values, and the reanalysis rainfall commonly attached to flood records is a daily total from ERA5 on a grid about 25 to 30 km across. Both average these short bursts away.
A global inventory of news-reported floods (Google Groundsource) holds 88 flood episodes within 10 km of the Msida site between 2000 and 2026, most in September to November. Taking the wettest day in a window from 3 days before each episode began to 1 day after it ended, the median ERA5 rainfall was 10.2 mm, and for a quarter of episodes that wettest day had 1.5 mm or less. The 1 in 5 year 15-minute storm on Malta’s national curves delivers 13.6 mm by itself.
Floods can look dry in daily data for two reasons: news reports give approximate dates and places, and a daily value on a 25 to 30 km cell spreads a short local cloudburst across a whole day and region. A flash flood risk assessment therefore starts from design storms.
Rainfall intensity-duration-frequency curves turn a return period into a design storm
A rainfall intensity-duration-frequency (IDF) curve gives the average rainfall intensity for every storm duration at one return period. The return period expresses the annual chance: a 1 in 50 year storm has a 2% chance of being matched or exceeded in any year. Engineers read design storms from these curves, for example the 30-minute storm at the 1 in 50 year return period.
Malta’s Energy and Water Agency publishes one equation per return period, i = a × t^-0.667, with i in millimetres per hour and t in hours. The constant a is 21.58 for the 1 in 5 year storm, 36.53 for the 1 in 50 year storm and 45.21 for the 1 in 200 year storm. The curves fall steeply: the 1 in 50 year storm averages 92.1 mm per hour over 15 minutes and 36.5 mm per hour over an hour. Total rain still grows with duration, from 23.0 mm in 15 minutes to 36.5 mm in an hour, but more slowly than the duration itself.
| Storm | 1 in 5 year | 1 in 50 year | 1 in 200 year |
|---|---|---|---|
| 10 min (extrapolated) | 11.9 mm (71.3 mm/h) | 20.1 mm (120.7 mm/h) | 24.9 mm (149.4 mm/h) |
| 15 min | 13.6 mm (54.4 mm/h) | 23.0 mm (92.1 mm/h) | 28.5 mm (114.0 mm/h) |
| 30 min | 17.1 mm (34.3 mm/h) | 29.0 mm (58.0 mm/h) | 35.9 mm (71.8 mm/h) |
| 45 min | 19.6 mm (26.1 mm/h) | 33.2 mm (44.3 mm/h) | 41.1 mm (54.8 mm/h) |
| 1 hour | 21.6 mm (21.6 mm/h) | 36.5 mm (36.5 mm/h) | 45.2 mm (45.2 mm/h) |
Total rain over the storm, with the average intensity in brackets. Source: Energy and Water Agency (2019), Preliminary Flood Risk Assessment for the Malta River Basin District, section 4.4.
Before using a curve, check how it was built, the range it covers and any adjustment already applied. Malta’s curves were fitted to the heaviest 24-hour rainfall of each year at daily gauges, so values for storms of minutes come from the shape of the fitted equation. The agency plots 15 minutes to about 8 hours; the 10-minute storm in the table lies outside it and is an extrapolation. The agency also raised the gauge rainfall depths by 20% for climate change before fitting, so these curves carry that allowance.
A design storm also needs a shape in time. The simplest is a uniform block, the same intensity from start to finish, which is what the worked example uses. Alternatives distribute the rain within the storm, with a burst near the peak. A comparison of six design storms in a small Croatian catchment (Krvavica and Rubinić, 2020) found that both the storm shape and its duration changed the modelled flood substantially, and that storms built only from IDF curves overestimated flooding compared with historical storms.
Test several storm durations, because the one that floods a site hardest depends on the site
A short storm is more intense; a long storm delivers more water. Which one produces the highest flood at a site depends on how quickly the catchment delivers runoff. Viglione and Blöschl (2009) showed that the ratio of storm duration to catchment response time is the main control on how a rainfall return period translates into a flood, and that the critical storm duration, the one that maximises the flood, depends mainly on the catchment response time. A single design storm of one duration can therefore miss the worst case.
The worked example tests 15 storms: 10, 15, 30, 45 and 60 minutes at the 1 in 5, 1 in 50 and 1 in 200 year return periods.
In the model, with no drainage, the water level at the Msida cell rises with storm duration at every return period, from 0.58 m for the 10-minute 1 in 5 year storm to 1.24 m for the 1-hour 1 in 200 year storm. The cell sits in a hollow that collects runoff from the valley, so the level follows the total rain, and the longest storm tested gives the highest level. The agency’s time of concentration for the catchment is 76 minutes, longer than any storm in the set, so a longer storm may fill the hollow further.
The time the water takes to reach 0.5 m, the depth above which Australian guidance rates floodwater unsafe for vehicles, children and the elderly, is shortest for the short, intense storms. At the 1 in 50 and 1 in 200 year return periods the 10-minute storm gets there first, in 9.5 and 8.5 minutes, before its rain has stopped; the 1-hour storms take 24.5 and 20.5 minutes. At the 1 in 5 year return period the 15-minute storm is quickest, at 18.5 minutes, and the 10-minute storm reaches 0.5 m 10 minutes after its rain ends.
Model the depth through time and record how fast it reaches a dangerous level
A rain-on-grid model applies the design storm to every cell of a terrain grid and routes the water downslope. The worked example uses the Continuuiti Pluvial Flood Solver. It integrates the local-inertial shallow water equations of Bates, Horritt and Fewtrell (2010), the formulation behind the University of Bristol’s LISFLOOD-FP model, with the flux weighting of de Almeida et al. (2012) for numerical stability. Friction follows Manning’s equation (n = 0.05, mixed urban surfaces), and buildings stand as raised walls within their footprints.
At Msida, the terrain is a 6 km by 6 km box of 30 m bare-earth elevations (GEDTM30), resampled to cells of about 8 by 10 m, with 13,019 mapped building footprints standing as 5 m walls. The box is centred on the valley and holds 9.9 km² of the area that drains to the site, about 84% of the agency’s 11.8 km². Depths are read at a model cell on Msida Valley Road (Triq il-Wied tal-Imsida) chosen before any run by a fixed rule: the lowest ground cell on that road that is not a building and not beside the sea. Its ground elevation is 2.0 m. Each run lasts 2 hours of simulated time.
For the 1 in 200 year storms, the water at the Msida cell passes 0.3 m between 6 and 14 minutes after the rain starts and 0.5 m between 8.5 and 20.5 minutes. Every one of these storms takes the water past 0.5 m before its rain stops. After the rain ends, the level holds. The cell lies about 1.7 m below a rise in the ground some 30 m to the east, and the model has no drainage, so the water stays; over the last 30 minutes of each run the level changed by 3 to 13 mm. Every depth in this example leaves out the flood relief tunnels described below.
Where water flows through a site, the depth rises and falls, and a single snapshot, such as the depth at the end of a run, can miss the peak. Where water collects, as it does at this cell, the depth rises and holds, and the time taken to reach a dangerous depth tells occupants how long they have to move people and vehicles.
Run a Storm on Real Terrain
The pluvial flood simulator runs the same solver used in this worked example. Pick one of the sample sites, set a rainfall depth and a storm length from 30 minutes, and see where the water collects.
Depth and velocity together set the danger to people and vehicles
Fast, shallow water can knock people over and move cars, so flash flood danger is rated on depth and velocity together. The UK method (Defra and Environment Agency, FD2321, 2006) computes a hazard rating HR = d × (v + 0.5) + DF, with depth d in metres, velocity v in metres per second and a debris factor DF of 0, 0.5 or 1. Its classes on d × (v + 0.5) are: below 0.75, low hazard (caution); 0.75 to 1.25, dangerous for some, such as children; 1.25 to 2.5, dangerous for most people; above 2.5, dangerous for all.
The Australian guideline (Australian Institute for Disaster Resilience, Guideline 7-3, 2014) uses six classes on the product of depth and velocity, each with a depth limit. H1, up to 0.3 m²/s and 0.3 m deep, is generally safe for vehicles, people and buildings. H2, up to 0.6 m²/s and 0.5 m, is unsafe for small vehicles. H3, up to 0.6 m²/s and 1.2 m, is unsafe for vehicles, children and the elderly. H6, above 4 m²/s, is unsafe for vehicles and people, and every building type is considered vulnerable to failure.
Velocity comes from the same model run, as the flow per unit width divided by the depth. It is sensitive to surface roughness and to how well the terrain captures kerbs and channels, so it needs more care than depth. The worked example reports depth only.
Drainage capacity decides how much of the modelled flood reaches the street
A rain-on-grid model without a drainage network turns every millimetre of rain into surface runoff. Its depths are an upper bound wherever drains, soakaways or soil take water away. Msida has large drainage works. Malta’s National Flood Relief Project, completed in 2015 at a cost of about €51.7 million, was designed for rainfall events with a 5-year return period. In the Birkirkara to Msida catchment it added an 11 km storm drainage system whose culverts feed tunnels 3 to 7 m in diameter, buried 8 to 52 m deep, discharging to the sea at Ta’ Xbiex. The agency notes that downstream of this infrastructure “the possible flood area is limited and negligible”.
The worked example models no drainage, so its depths describe the catchment’s surface response without the tunnels: the flood the relief system was built to remove. A site assessment that informs a decision has to represent drainage capacity, either by coupling the surface model to a pipe and tunnel network or by removing the design capacity from the rainfall before it reaches the surface.
Check the model against the observed flood record
Observed records show where floods happen and in which season. Inventories built from news reports place each event at the centre of the area the report names, so a recorded position can be several hundred metres from where water stood, and a place with no reports may have had no coverage. At Msida, that record confirms the hazard and its September to November season. Continuuiti’s Flood History Explorer shows the same kind of record within 10 km of any location. Checking a modelled depth needs a measured one, from a survey, a gauge or a dated photograph with a known scale.
A screening assessment leaves out factors that decide site-level outcomes
The method above ranks and compares sites, storms and durations. It is a screening method, and it leaves out:
- infiltration and soil saturation, which reduce runoff from gardens, fields and unpaved ground;
- drainage networks, culverts and tunnels, which remove water up to their capacity;
- blockage and debris, which can raise water levels behind culverts and bridges;
- sediment and the change in flow behaviour when water carries mud and debris;
- compound flooding, such as a high sea level at the outlet holding water back;
- detail finer than the terrain grid: kerbs, underpasses, walls and doorways;
- field validation against depths measured in a recorded flood (the solver itself is verified against analytic test cases and conserves water to machine precision).
Site-specific hydraulic modelling by qualified engineers is needed when a decision depends on the depth at a door, the hazard rating on an access route, or the design of a flood defence. Our guide to flood risk assessment covers how those studies differ from a flood zone lookup.
Assumptions behind the Msida worked example
| Item | Value |
|---|---|
| Location | Msida Valley Road, Malta; model cell at 35.8961° N, 14.4881° E, ground 2.0 m |
| Rainfall | Energy and Water Agency (2019) national curves, i = a × t^-0.667; 10, 15, 30, 45 and 60 min at 1 in 5, 50 and 200 years; uniform rain over each storm; curves include the agency’s 20% climate allowance |
| Terrain | GEDTM30 bare earth, 30 m, 6 km box resampled to about 8 by 10 m cells; 13,019 building footprints (Microsoft GlobalMLBuildingFootprints) as 5 m walls |
| Solver | local-inertial shallow water equations (Bates et al. 2010; de Almeida et al. 2012); Manning n = 0.05; open boundaries; 2 h simulated |
| Not modelled | infiltration, drainage and the 2015 flood relief tunnels, velocity, inflow from beyond the box (84% of the catchment inside) |
| Reported value | maximum depth at the model cell in each 30 s interval; level reached within 2 h; minutes until the depth reaches 0.3 m and 0.5 m |
| Observed record | Google Groundsource flood reports within 10 km (2000 to 2026); wettest ERA5 day (Open-Meteo) in a window from 3 days before to 1 day after each episode |
Sources
- US National Weather Service, Flood and flash flood definitions.
- Fowler, H.J. et al. (2021). Anthropogenic intensification of short-duration rainfall extremes. Nature Reviews Earth & Environment 2, 107-122.
- Lenderink, G. and van Meijgaard, E. (2008). Increase in hourly precipitation extremes beyond expectations from temperature changes. Nature Geoscience 1, 511-514.
- IPCC (2021). Climate Change 2021: The Physical Science Basis, Summary for Policymakers.
- Viglione, A. and Blöschl, G. (2009). On the role of storm duration in the mapping of rainfall to flood return periods. Hydrology and Earth System Sciences 13, 205-216.
- Krvavica, N. and Rubinić, J. (2020). Evaluation of Design Storms and Critical Rainfall Durations for Flood Prediction in Partially Urbanized Catchments. Water 12, 2044.
- Bates, P.D., Horritt, M.S. and Fewtrell, T.J. (2010). A simple inertial formulation of the shallow water equations for efficient two-dimensional flood inundation modelling. Journal of Hydrology 387, 33-45.
- de Almeida, G.A.M., Bates, P., Freer, J.E. and Souvignet, M. (2012). Improving the stability of a simple formulation of the shallow water equations for 2-D flood modeling. Water Resources Research 48, W05528.
- Defra and Environment Agency (2006). Flood Risks to People Phase 2, FD2321/TR1: The Flood Risks to People Methodology.
- Australian Institute for Disaster Resilience (2014). Australian Disaster Resilience Guideline 7-3: Technical flood risk management guideline: Flood hazard.
- Energy and Water Agency, Malta (2019). Preliminary Flood Risk Assessment for the Malta River Basin District.
- Terrain: GEDTM30, OpenGeoHub Foundation (CC BY 4.0), described in Ho, Hengl et al. (2025), PeerJ 13, e19673. Building footprints: Microsoft GlobalMLBuildingFootprints (CDLA Permissive 2.0).
- Mayo, R. et al. (2026). Groundsource: A Dataset of Flood Events from News. EarthArXiv preprint. Rainfall around each episode: ERA5 reanalysis via Open-Meteo.
Frequently asked questions
How do you do a flash flood risk assessment?
Read design storms for several short durations from a rainfall intensity-duration-frequency curve, model the depth through time with a rain-on-grid model over the catchment that drains to the site, report the depth reached and how quickly it gets there, rate danger using depth and velocity, and check the result against the observed flood record. State what the model leaves out, drainage above all.
Which storm duration should a flash flood risk assessment use?
Several. A short storm is more intense and a long storm delivers more water, and which one floods a site hardest depends on how quickly the catchment delivers runoff. At Msida the 1-hour storm gave the highest water level and the 10- and 15-minute storms reached 0.5 m first, so a single duration would have missed one of those answers.
Why are flood depths from a model without drainage an upper bound?
Without drains, culverts or soil to take water away, every millimetre of rain becomes surface runoff. Real drainage removes water up to its capacity, so modelled depths are the high end wherever drainage exists. Msida’s flood relief tunnels were designed for rainfall up to the 1 in 5 year storm, the size of storm the model shows filling the hollow with no drainage.
How is flood danger to people rated?
On depth and velocity together. The UK method computes a hazard rating from depth, velocity and a debris factor, with classes from low hazard to dangerous for all. Australian guidance uses six classes on depth times velocity, each with a depth limit: above 0.3 m floodwater is unsafe for small vehicles, and above 0.5 m it is unsafe for vehicles, children and the elderly.
Can climate models predict flash floods?
Climate projections used for screening are mostly daily rainfall on cells tens of kilometres across, which spreads a short burst over a whole day. Evidence on how short-duration rainfall is changing comes from rain gauge records and kilometre-scale models, which show it intensifying at about 7% per degree of warming, and faster in some regions.
