Two nearby neighbourhoods receive the same measured storm total, but one experiences rapidly rising drains while the other does not. Was the rainfall more intense for a few minutes? Were the soils already wet? Did a drainage route concentrate runoff? Or did one gauge miss the heaviest part of the storm? Hydrometeorology connects weather evidence with the actual movement of water so that these questions can be tested rather than guessed.
Hydrometeorology studies the connections between the atmosphere and water systems, especially how precipitation, evaporation, land conditions and weather forecasts influence catchment response, streamflow and flood risk. Its gold standard is a traceable explanation from observed or forecast weather through a physically appropriate water model to the measured or anticipated consequence, with uncertainty made visible.
This educational root belongs to The Gold Standard, where the term expresses an editorial commitment to verifiable reasoning rather than a certification. It connects the existing Meteorology and Hydrology guides, while following the wider principle that good learning transfers into life.
Quick information: the answer in one minute
- Atmospheric input: measure or forecast precipitation, air conditions and relevant evapotranspiration drivers.
- Catchment response: determine infiltration, storage, flow pathways, drains, streamflow and possible downstream constraints.
- Time and place: compare measurements with matching units, area, accumulation interval and timestamps.
- Prediction: keep rainfall estimates, rainfall forecasts and water-level or discharge forecasts distinct.
- Verification: test real forecasts against later observations, including errors, misses and false alarms.
- Public safety: rely on current responsible-agency warnings for live events, not historical examples or a general educational model.
Find your next learning step
| Your question | Start here | Check next |
|---|---|---|
| How much rain fell, and when? | Rain gauges, radar and satellites | Chapters 5–10 |
| Why did an area respond differently? | Catchment boundaries and wetness | Chapters 11–15 |
| What can be predicted? | QPE, QPF, models and ensembles | Chapters 16–18 |
| How serious is a flood possibility? | Hazard, exposure and probability | Chapters 19–21 |
| What happens in Singapore? | PUB data, drainage and response | Chapters 22–23 |
Full chapter index
- 1. Start with the connection between weather and water
- 2. Distinguish hydrometeorology from meteorology and hydrology
- 3. Follow water and energy through one connected cycle
- 4. Define the catchment, outlet and observation period
- 5. Measure rainfall with gauges and quality controls
- 6. Separate rainfall depth, intensity, duration and volume
- 7. Read weather radar as an estimate with blind spots
- 8. Use satellite precipitation without inventing fine detail
- 9. Build a coherent quantitative precipitation estimate
- 10. Reconstruct storm motion and the order of rainfall
- 11. Account for antecedent soil moisture and storage
- 12. Distinguish infiltration, overland flow and urban drainage
- 13. Interpret the hydrograph and catchment travel time
- 14. Keep water level distinct from river discharge
- 15. Consider coastlines, tidal backwater and control structures
- 16. Distinguish QPE, QPF and nowcasting
- 17. Build a rainfall-runoff model without overfitting
- 18. Use ensemble forecasts as a range, not a vote
- 19. Keep flood hazard, exposure and risk separate
- 20. Verify the forecast that was actually issued
- 21. Use return periods without promising a safe interval
- 22. Understand climate context without confusing it with a forecast
- 23. Learn from Singapore’s rainfall and flood operations
- 24. Protect data lineage, timestamps and correction records
- 25. Transfer the method to another catchment and decision
1. Start with the connection between weather and water
Hydrometeorology explains how atmospheric conditions produce water inputs to landscapes and how those inputs interact with catchments, rivers, groundwater, drainage and human activities. The practical question is rarely just whether it will rain. It is how much precipitation will fall, where and when, how the landscape will respond, and whether that response matters to people or infrastructure. Meteorology supplies evidence about the atmosphere; hydrology supplies evidence about water movement and storage. The connection between them is where many important prediction and interpretation errors occur.
Imagine a storm cloud crossing two neighbouring catchments. One has wet soils, connected impervious surfaces and a short route to a drain. The other has deeper storage, different topography and a longer flow path. Even when their total rain is similar, their flood responses may differ substantially. A rain forecast alone is not a flood forecast, just as a water-level reading at one site is not a complete explanation of rainfall.
The gold standard is a traceable chain of inference: define the water-related decision, measure the atmospheric input, describe the catchment response, quantify uncertainty, evaluate predictions and communicate what a recipient should rely on. For learners, this is a powerful way to understand why disciplines must be connected without being confused.
2. Distinguish hydrometeorology from meteorology and hydrology
Meteorology studies atmospheric conditions and processes: moisture, winds, heat, cloud formation and precipitation. Hydrology examines water in the environment, including where it is stored, how it moves and how its quantity changes. Hydrometeorology studies the links that carry atmospheric forcing into hydrologic systems, especially rainfall, snowfall, evaporation, land-surface energy and the timing of water responses. These fields overlap, but none is a mere replacement for the others.
A weather analyst might estimate the chance of 30 millimetres of rain in an hour. A hydrologist might estimate the discharge resulting from a storm in a particular basin. A hydrometeorological analysis asks whether the rainfall estimate is suitable for that basin, which surface conditions control the conversion to runoff, and whether the forecast can be translated into meaningful hydrologic consequences. This changes the observations and models required.
The World Meteorological Organization’s operational-hydrology definition includes regular hydrometeorological observations, models, forecasts and warnings. Those products are meant to support decisions over different spatial and temporal scales. The important standard is therefore integration of observations and reasoning, not the number of disciplines named in a report.
3. Follow water and energy through one connected cycle
Water moves through evaporation, condensation, precipitation, interception, infiltration, soil storage, groundwater, streams and oceans. Energy also moves: evaporating liquid water requires energy and condensation releases latent heat into the atmosphere. The water and energy cycles cannot be separated completely because atmospheric moisture and land-surface conditions influence each other. A description that counts only rainfall misses important stores and feedbacks.
A hot day does not imply every landscape loses water at the same rate. Evapotranspiration depends on available energy, air demand, vegetation and access to water. A dry soil can limit actual evapotranspiration even when the atmosphere could support a higher potential rate. Rainfall can also be intercepted on leaves and evaporate before reaching the ground, changing the water available for infiltration and runoff.
For a hydrometeorological study, state which processes matter at the timescale of interest. Over 20 minutes, infiltration, surface storage and short flow paths may dominate a rapid response. Over seasons, vegetation, aquifer recharge and evapotranspiration become important. The gold standard chooses a connected description appropriate to the question rather than pretending one process dominates everywhere.
4. Define the catchment, outlet and observation period
A catchment or watershed is the region contributing water to a stated outlet under an applicable drainage interpretation. The outlet controls which land areas and pathways belong in the analysis. In urban settings, pipes, pumps, diversions and constructed drainage may alter the flow directions implied by topography alone. In complex terrain, groundwater and surface-water contributing areas can differ. A catchment boundary must therefore be justified.
The time window matters just as much. A 24-hour total, a one-hour accumulation and a five-minute peak are not interchangeable. A storm crossing the catchment may produce most rain at its upstream end before a gauge near the outlet observes a shower. If the chosen window excludes early rainfall or a slow post-storm response, the calculated balance can become misleading.
Record area, projection and units, outlet coordinates or asset name, time zone, event start and end, sampling resolution and any exclusions. If the study spans a changing drainage system, record the infrastructure configuration for the period being modelled. A map without these details may invite a convincing but invalid calculation.
5. Measure rainfall with gauges and quality controls
A rain gauge provides a local observation of precipitation received at its collection point. Tipping-bucket, weighing and other gauge systems differ in operation, resolution and error sources. Wind exposure, splash, obstructions, instrument malfunction, clock drift and intense-rain undercatch can affect observations. A series of smooth-looking numbers does not establish that the site represents the surrounding catchment.
Inspect the gauge’s position, maintenance records, calibration, collection interval and missing data. A gauge beside a roof or tree may not represent unobstructed rainfall. Instruments with different sampling intervals require careful aggregation before comparing peaks. When a device records cumulative rainfall, the analyst should inspect resets, counter rollovers and delayed uploads rather than interpreting every jump as a true storm burst.
For educational practice, prepare two hypothetical gauges: one near an open field and one in a constrained street. Ask what their differing totals could mean before asserting that one is wrong. Spatial variability is real, and so are measurement errors. A sound analysis considers both. WMO hydrological-practice resources describe the observation-to-information process that makes these checks necessary.
6. Separate rainfall depth, intensity, duration and volume
Depth describes the height of accumulated water over an area, commonly in millimetres. Intensity expresses depth per unit time, commonly millimetres per hour. Duration describes how long an event or sub-interval lasts; volume converts depth across an area into cubic metres. These quantities answer different questions. High total depth is not always the same as high short-duration intensity.
For a teaching catchment with an area of ten square kilometres, 48 millimetres of spatially uniform rainfall corresponds to 0.048 metres multiplied by 10,000,000 square metres, or 480,000 cubic metres of gross precipitation. That calculation is a water input, not observed flood runoff. Some rain may infiltrate, evaporate, remain in storage or drain through slower pathways.
Suppose the same 48 millimetres arrives uniformly in four hours or in half an hour. Average intensity is respectively 12 or 96 millimetres per hour. A catchment’s response can differ even though both storm totals equal 48 millimetres. Always show the accumulation period next to an intensity claim and never turn a rain-depth figure directly into a flood-volume claim without a hydrological accounting model.
7. Read weather radar as an estimate with blind spots
Weather radar samples reflected electromagnetic signals from hydrometeors in the atmosphere. Converting those observations into rainfall estimates depends on drop-size relationships, beam geometry, elevation, calibration and processing. Rain measured aloft may change before reaching the ground. Radar coverage can be blocked, weakened or geometrically broadened, and intense convection may vary over short distances.
A radar product can be valuable because it estimates spatial patterns across areas with few gauges. It is not an exact map of ground-truth rainfall at every pixel. Products may incorporate gauge adjustment, hydrometeor classification and quality-control methods. Note the product’s resolution, accumulation window, processing version and whether it estimates present or past precipitation.
When gauge and radar totals disagree, compare equivalent periods and locations first. The gauge measures a tiny collecting area; radar represents a sampling volume and derived surface estimate. Examine missing data, local weather, clutter, bright-band effects where relevant and uncertainty before adjusting a rainfall-runoff model to force agreement. A better model cannot rescue an unidentified observation error by silently absorbing it as a parameter.
8. Use satellite precipitation without inventing fine detail
Satellite observations help estimate rainfall and snowfall in regions where surface observations are scarce. NASA and JAXA’s Global Precipitation Measurement mission combines direct and indirect remotely sensed information. The IMERG family merges several satellite sources with other inputs to create precipitation estimates over wide areas. This offers valuable coverage while still requiring interpretation and validation.
NASA documents commonly used IMERG products on a 0.1-degree grid with half-hourly time resolution. A half-hourly gridded estimate does not contain a record of every five-minute convective peak within that interval. It would be misleading to interpolate it into a five-minute urban-drainage forcing series and claim the resulting fine-scale peaks were measured. The spatial resolution is also far coarser than many small urban catchments.
Choose Early or Late near-real-time products when that timing is appropriate and understand how they differ from a later Final research product. NASA’s August 2026 Version 07 algorithm documentation explains the multiple stages of data merging and revision. For a critical local flood study, satellite data may complement rather than replace representative gauges and radar. Declare the version, run type, calibration choices and limits of downscaling.
9. Build a coherent quantitative precipitation estimate
Quantitative precipitation estimation, often abbreviated QPE, seeks to describe precipitation that occurred over a defined place and time. Gauge, radar, satellite and model-related inputs may each contribute, depending on the product. The goal is not to select whichever number looks largest. It is to construct or choose an estimate with a documented method and uncertainty appropriate to the intended hydrologic application.
Compare all inputs using common units, start and end times, geographic support and missing-data conventions. A gauge may report at a point, while radar or satellite gives an area average. Regridding and spatial interpolation change the scale of the claim. A point’s high short burst should not be smoothed into a catchment-average map without clearly distinguishing its original measurement.
Preserve individual data sources alongside any merged field. When a later gauge correction changes the rain estimate, retain the earlier version if it was used in a real-time forecast. That distinction matters for evaluating how well a forecasting system performed with information genuinely available when the forecast was issued. Historical reconstruction and operational QPE are different evidential products.
10. Reconstruct storm motion and the order of rainfall
A storm does not arrive everywhere in a basin at once. Movement relative to the drainage network can change the timing at which water from different parts of a catchment meets. Even similar rain totals can create different runoff peaks when the most intense rainfall falls near the outlet or propagates in the same direction as the drainage response.
A hyetograph displays rainfall depth or intensity over time, while a spatial rainfall field shows where rain is estimated. Combining them helps identify the event’s sequence. For example, if the first heavy band is upstream and a second band later strikes downstream, the resulting hydrograph may reflect overlapping contributions. A static storm-total map cannot reconstruct this sequence by itself.
Use time-stamped gauge observations and suitable radar or satellite products to describe storm evolution. Where rapid convection outruns model resolution, acknowledge the uncertainty rather than writing a precise retrospective trajectory from a few images. A robust investigation documents which movement claims were observed and which came from interpolation or a model.
11. Account for antecedent soil moisture and storage
Rainfall falling on a dry, permeable landscape can produce a different response from similar rainfall on already wet or saturated soil. Interception, soil pore space, subsurface pathways, groundwater conditions and surface depressions influence how much water moves quickly toward a stream or drain. Recent rainfall is an indicator of wetness, but it is not a complete measurement of available storage.
A common shortcut is to assign one runoff fraction to all storms. That may be useful for a clearly labelled teaching demonstration, but it conceals physical variability. A site can also respond differently after land-cover change, construction or maintenance work. If a model has an initial soil-water storage parameter, state how it is estimated and whether observations constrain it.
Compare competing initial conditions rather than choosing one merely because it reproduces a single event. Where available, incorporate soil-moisture observations, stream baseflow and other indicators. A good hydrometeorological forecast carries uncertainty in the initial land state forward instead of hiding it behind a single deterministic rainfall number.
12. Distinguish infiltration, overland flow and urban drainage
Infiltration moves water from the surface into the soil, but infiltration is not automatically groundwater recharge. Some water remains in shallow storage, is taken up by vegetation or moves laterally before reaching deeper aquifers. Surface runoff can arise when rainfall exceeds infiltration capacity or when storage is already saturated, among other mechanisms. The relevant process differs by soil, topography and rainfall sequence.
Urban landscapes add roofs, roads, drains, storage tanks, pipes, culverts and control structures. Impervious surfaces can accelerate water delivery, although disconnected impervious patches or retention features may behave differently from fully connected surfaces. A simplified percentage of impervious cover should not be mistaken for a direct measurement of runoff at the outlet.
In applied planning, map the actual drainage connections and management features. Determine whether a rain garden changes infiltration and detention under the local soil conditions. The gold standard follows the pathway from where rain falls to where people or assets are exposed, rather than using a generic runoff coefficient as an assurance that every site behaves the same way.
13. Interpret the hydrograph and catchment travel time
A hydrograph describes how stage or streamflow changes over time at a location. In many catchments it rises after rainfall begins, peaks and then recedes as faster and slower pathways contribute water. The shape carries information about storage, routing and channel behaviour, but it is not a direct fingerprint of one mechanism.
Catchment lag may be defined differently in different models and texts. Some definitions compare the centroid of effective rainfall with the hydrograph peak; others compare rainfall and runoff centroids. NOAA’s glossary distinguishes these conventions. If two analysts use different reference points, they can obtain different lag times from the same event without either arithmetic calculation being wrong.
Do not identify the peak as a property of rainfall alone. Reservoir control, tides, tributary timing, soil conditions and upstream drainage interventions may all matter. Check whether the hydrograph shows discharge or stage, how it was sampled and whether data gaps mask the maximum. For transfer learning, ask what other event could yield a similar-looking hydrograph and what additional observations would distinguish them.
14. Keep water level distinct from river discharge
Water level or stage is the height of a water surface relative to a stated local reference. Discharge is a flow volume per unit time. They are related through channel and hydraulic conditions but are not synonyms. A sensor can reliably report rising stage without directly measuring how many cubic metres per second pass a cross-section.
A rating curve estimates discharge from stage using calibrated observations and assumptions. Channel change, sediment deposition, vegetation, backwater and unsteady flow may alter the relationship. During a rapidly changing flood, the same stage can sometimes correspond to different discharges on the rising and falling limbs, a phenomenon known as rating-curve hysteresis.
When evaluating a rainfall-runoff model, identify whether the target observation is stage, discharge or a derived volume. Preserve any rating curve version and its range of support. Avoid claiming a precise discharge from a level sensor without the necessary hydraulic conversion and uncertainty. USGS’s streamflow teaching materials and WMO’s measurement guidance illustrate why this distinction matters.
15. Consider coastlines, tidal backwater and control structures
In coastal or low-gradient catchments, a river or drain may respond to rainfall while downstream water levels are also influenced by tide, storm surge, sea-level conditions or controlled gates and pumps. A single upstream rain total cannot explain every stage observation when the downstream hydraulic boundary is changing. The result can be backwater, delayed drainage or an altered hydraulic gradient.
When comparing events, document downstream boundary conditions and relevant infrastructure status. A rainfall-runoff model without an appropriate hydraulic boundary may reproduce volume reasonably but miss local water levels. A tide gauge, outfall gate record and pump operation log may be as important as a meteorological input for explaining a difficult flood event.
This does not mean every Singapore flash flood is a coastal flooding problem. It means the correct physical mechanism must be investigated for the specific catchment. State what is observed and which mechanisms are possible before generalising from a neighbouring waterway.
16. Distinguish QPE, QPF and nowcasting
Quantitative precipitation estimation (QPE) describes how much precipitation is estimated to have fallen. Quantitative precipitation forecasting (QPF) describes future precipitation over a specified area and interval. Nowcasting generally focuses on near-term evolution using recent observations and suitable prediction methods. A retrospective radar accumulation should not be labelled as a forecast simply because it is shown beside future panels.
NOAA’s National Weather Service glossary defines QPF as an estimate of future precipitation amounts over a given region and period. The analyst should preserve forecast issuance time, valid time, accumulation interval, resolution, lead time and whether the information was revised. This distinction is essential for honest verification: an analysis based on data acquired later is not what an operator knew at the earlier decision point.
A hydrological forecast may accept QPF as input and predict a different quantity, such as water level, discharge or probability of exceeding a threshold. It has inherited rainfall uncertainty plus uncertainty from catchment conditions and the model. Never equate a 70% rain probability with a 70% flood probability without an explicit model and outcome definition.
17. Build a rainfall-runoff model without overfitting
Rainfall-runoff models range from simple conceptual water stores to detailed distributed numerical descriptions. Their complexity should follow the decision, data availability and required lead time. A model can represent rainfall partitioning, surface and subsurface routing, channel flow and human controls, but not every uncertain process is resolved by more parameters.
Calibration uses observations to identify model choices or parameters; validation tests performance on data not used for fitting where possible. A model that reproduces one famous storm after extensive tuning may perform poorly on common storms, wet starting conditions or changed drainage systems. Parameter equifinality means different parameter sets can produce similar fits, and this uncertainty should be examined.
Record the model version, catchment geometry, data transformations, spin-up and initial conditions. Check water balance and whether plausible parameters were obtained for physically defensible reasons. Evaluate timing as well as peak and volume. Report when model outputs are scenario estimates rather than measurements. The gold standard resists describing an attractive simulation as evidence that the underlying mechanism must be correct.
18. Use ensemble forecasts as a range, not a vote
An ensemble represents multiple plausible predictions arising from alternative inputs, initial conditions, parameterisations or models. It can help describe uncertainty, but members are not necessarily independent or equally reliable. If ten ensemble tracks suggest different rainfall positions, averaging them into one smooth rainfall map may remove the extremes most relevant to a small catchment.
For operational thresholds, ask whether the ensemble is calibrated and whether the predicted probability agrees with outcomes over many comparable cases. A simple fraction of ensemble members crossing a threshold is not automatically an accurate exceedance probability without considering member design and validation. Keep the specific event, location and time window clear.
A useful forecast describes potential outcomes with enough detail for decision-makers to understand urgency and uncertainty. If high-impact low-probability scenarios are important, communicate their plausibility without presenting the worst member as the most likely. An honest uncertainty interval can be more useful than a single high-precision number.
19. Keep flood hazard, exposure and risk separate
Flood hazard concerns the physical possibility and severity of inundation or dangerous water conditions at a place. Exposure describes people, buildings, roads and assets that may be affected. Vulnerability concerns how susceptible those exposed elements are and what consequences arise. A map of heavy rainfall alone does not identify who will be harmed, and a map of a high river level alone is not a complete loss forecast.
A 50-millimetre storm may create relatively low consequences in one landscape and serious disruption in another. Timing, drainage capacity, property elevation, traffic patterns, access routes and flood barriers can alter the outcome. Hydrometeorology contributes essential evidence, but full risk interpretation may require engineering, land-use, social and emergency-management information.
Avoid promising that any area is flood-proof because a model did not show inundation in selected scenarios. Explain the scenario boundary, data limitations and whether the products are intended for long-term planning or immediate public guidance. For actual dangerous weather and floods, rely on authorised local warnings, not a teaching article or a historic model output.
20. Verify the forecast that was actually issued
Forecast quality cannot be inferred from a polished dashboard or from a correct retrospective explanation of one event. Verification compares forecasts as originally available to later observations using declared targets and methods. The World Meteorological Organization released Guidelines on the Verification of Hydrological Forecasts (WMO-No. 1364) in 2025, emphasising accuracy, bias, reliability, resolution and sharpness as distinct aspects of quality.
For a continuous discharge forecast, errors in peak amount, peak timing, total volume and low flows may matter differently. For a categorical threshold warning, assess hits, misses, false alarms and correct non-events while stating the underlying number of opportunities. For probabilistic forecasts, calibration and proper scoring rules test more than whether the most dramatic case happened.
Verification must be stratified where appropriate by season, basin, flow regime and lead time. A model that performs well at long-term monthly totals may miss short violent events. Do not choose only forecast cases that look favourable. Use a baseline such as persistence or climatology where meaningful, and explain when sparse observations prevent confident performance claims.
21. Use return periods without promising a safe interval
A ‘100-year’ rainfall or flood level is often used to describe an annual exceedance probability near 1% under a specified statistical model and conditions. It does not mean that a similar event must wait another century after one occurs. Independent annual events with a constant 1% probability would have about a 9.56% chance of at least one exceedance in ten years: one minus 0.99 raised to the power of ten.
Real hazards may violate assumptions of stationary probabilities because climate, land use, drainage and observation methods can change. Statistical uncertainty also increases for the rarest quantiles. Distinguish a design storm’s rainfall intensity-duration-frequency estimate from a flood level predicted using a catchment and hydraulic model. Similar annual probabilities do not establish identical impacts.
When referring to national or local design criteria, use current authoritative guidance and the specific duration and location. Avoid describing a design value as a guarantee that an asset cannot flood below that frequency. The gold standard is a transparent risk description with stated assumptions and explicit consequences for planning.
22. Understand climate context without confusing it with a forecast
Climate trends, seasonal outlooks and next-hour weather forecasts answer different questions. A projected increase in intense-rainfall risk may justify reviewing infrastructure design and emergency plans, but it does not provide the timing or location of tomorrow’s thunderstorm. Likewise, one intense storm does not by itself quantify the contribution of long-term climate change to that event.
WMO’s climate and hydrology resources place precipitation alongside temperature, humidity, evaporation, soil moisture and other variables as connected parts of the system. Long-term assessments should consider changes in the distribution of weather inputs as well as land use and surface conditions. A historical calibration period may not be a safe assumption for all future scenarios.
Explain which part of a claim is observational, scenario-based, statistical or operational. For a school or public reader, distinguish ‘this location could face more intense rain under some projections’ from ‘it will flood tomorrow’. Responsible hydrometeorology makes that boundary clear rather than trading in certainty that its tools cannot deliver.
23. Learn from Singapore’s rainfall and flood operations
Singapore illustrates why rainfall monitoring, drainage engineering and public warnings must work together. PUB’s flood-resilience programme combines rainfall information, water-level sensors, cameras and field response. Its November 2025 release reported more than 1,000 water-level sensors and over 500 CCTV cameras, with 650 additional water-level sensors planned for installation by the end of 2028. These are figures from that dated announcement, not a claimed live count today.
Singapore’s National Climate Change Secretariat describes PUB’s Source–Pathway–Receptor approach: manage stormwater where it falls, through its drainage pathways and where it could harm people or buildings. On-site detention, drainage upgrades, protective levels and flood response play different roles. A rainfall estimate can help initiate the chain, but water levels and actual site conditions are needed to understand emerging consequences.
The myENV application provides relevant weather and flood information and alerts. A reader should use current agency information during real events. Hydrometeorological education should explain the system and its uncertainties, not substitute a generic model or historical article for PUB’s operational warnings or public-safety instructions.
24. Protect data lineage, timestamps and correction records
Hydrometeorological systems combine instruments, telecommunications, forecasts, algorithms, spatial boundaries and operators. A time-series file can look complete while hiding clock drift, missing readings, duplicated intervals or a changed sensor datum. A flood-forecast dashboard can display a number correctly even though its input has silently switched from provisional radar estimates to corrected post-event totals.
Preserve original gauge and model identifiers, data version, event and receipt time, time-zone basis, quality flags, calibration and all transformations. Record any clipping, interpolation, smoothing, resampling or bias correction. When corrected data are released, retain the operational version needed to fairly assess the original forecast. A new high-quality product is valuable, but must not rewrite the history of what forecasters saw.
Publish the method in a form another analyst can reconstruct. Include a deliberately difficult test event such as a brief intense storm or two gauges that disagree. If an anomaly is repaired, describe how the correction changes the conclusion rather than only swapping the figure in a chart.
25. Transfer the method to another catchment and decision
A method that works for a broad temperate river may fail when applied unchanged to a short Singapore drain, a tidal estuary or a mountain snowmelt catchment. The atmosphere, soil, slope, infrastructure, flow paths and observation density can differ greatly. Transfer requires rebuilding the chain of evidence, not copying a runoff coefficient or an attractive chart.
Start by defining the actual decision and threshold; then check source observations, geographic support, time intervals, land state and routing process. Select an appropriate model, produce honest uncertainty and test it against independent events. If the expected failure is high-consequence, consider additional independent verification and explicit emergency procedures overseen by the relevant authority.
The most useful learning outcome is a disciplined explanation. A reader should be able to say what fell, where it fell, how water moved, what was predicted, what happened, why the two differed and what evidence would change the next decision. That is the gold standard of hydrometeorology as an evidence-based practice.
Worked case A: one rainfall total, very different intensities
An invented catchment covers 10 square kilometres, or 10,000,000 square metres. Both teaching storms produce a spatially uniform total of 48 millimetres (0.048 metres), so both supply 480,000 cubic metres of gross rainfall over the defined area. The total alone does not tell us how fast water was delivered.
| Scenario | Rainfall total | Duration | Average intensity | Gross rain volume |
|---|---|---|---|---|
| Storm A | 48 mm | 4 hours | 12 mm/hour | 480,000 m³ |
| Storm B | 48 mm | 30 minutes | 96 mm/hour | 480,000 m³ |
Now suppose a deliberately simplified hypothetical event-runoff fraction of 25% is used for one initial condition and 60% for another. Under those assumptions, total event runoff would be 120,000 and 288,000 cubic metres, respectively. This is not a prediction of either storm’s real response: the fraction depends on soils, storage, drainage, land use and time. Nor can a runoff volume alone tell us the flood peak. The case shows why a water balance must not be confused with a forecast of peak discharge.
Worked case B: fifteen-minute rainfall intervals hide the peak
A fictional gauge reports 12 millimetres from 14:00–14:15, 24 millimetres from 14:15–14:30 and 12 millimetres from 14:30–14:45. Converting each quarter-hour total to an hourly rate multiplies it by four, producing 48, 96 and 48 millimetres per hour. The event total is 48 millimetres over 45 minutes, for a period-average rate of 64 millimetres per hour. This average is not the same as the 15-minute peak of 96 millimetres per hour.
| Quarter-hour interval | Accumulation | Equivalent rate |
|---|---|---|
| 14:00–14:15 | 12 mm | 48 mm/hour |
| 14:15–14:30 | 24 mm | 96 mm/hour |
| 14:30–14:45 | 12 mm | 48 mm/hour |
If the data were exported only as a one-hour accumulation, the exact 15-minute peak could be hidden. If a downstream water-level sensor reached its maximum at 15:05, that is 35 minutes after the end of the strongest rain interval, not necessarily the catchment’s formally defined basin lag. A complete lag analysis needs a declared rainfall reference and the appropriate hydrograph quantity and definition. All figures are invented and are not Singapore rainfall or water-level observations.
Worked case C: verify probabilities without cherry-picking
Imagine four hypothetical forecasts of a defined streamflow threshold being exceeded: 0.80, 0.60, 0.40 and 0.20. The observed outcomes are yes, no, yes and no, encoded as 1, 0, 1 and 0. The Brier score is the mean of squared forecast-probability minus outcome differences: [(0.80−1)² + (0.60−0)² + (0.40−1)² + (0.20−0)²] ÷ 4 = (0.04 + 0.36 + 0.36 + 0.04) ÷ 4 = 0.20. For a comparison predicting 0.50 each time, this four-case example yields 0.25.
A lower Brier score is better for the declared binary forecast task, and the small sample is entirely fictional. Four outcomes cannot establish whether the forecasting system is well calibrated, dependable under rare floods or superior over realistic seasons and locations. Forecast verification should include the complete event inventory, lead times, event definitions, meaningful baselines and uncertainty. The WMO’s 2025 guidance provides the appropriate broader framework.
Rapid diagnostic: how to repair a misleading flood analysis
- Confirm the observed or forecast quantity: rainfall depth, intensity, water level, discharge, volume, inundation or warning threshold.
- State catchment and outlet boundaries, time zone, sampling interval, instrument datum and spatial support.
- Check gauge calibration, radar and satellite assumptions, gaps, data versions and matching accumulation periods.
- Account for initial soil and channel storage, urban drainage and any tidal or controlled downstream conditions.
- Compare predictions with a physically and statistically suitable observed target rather than convenient proxy values.
- Report both a typical outcome and uncertainties that could materially change the public decision.
- Explain what was corrected and preserve the information available when an operational forecast was originally issued.
Frequently asked questions about hydrometeorology
What is hydrometeorology?
It studies the connections between atmospheric processes such as precipitation and evaporation and hydrologic processes such as infiltration, runoff, streamflow and floods. It includes observation, modelling and forecasting.
How is hydrometeorology different from meteorology?
Meteorology focuses on the atmosphere; hydrometeorology connects atmospheric inputs to water movement and storage and evaluates resulting water-related decisions.
How is it different from hydrology?
Hydrology studies water across the environment. Hydrometeorology particularly examines atmospheric forcing, the land response and how weather information informs hydrologic estimates and forecasts.
Can rainfall forecasts predict flash floods?
They contribute important evidence but do not determine flood outcomes alone. Catchment wetness, drainage connectivity, duration, location, downstream conditions and model uncertainty matter.
What is the difference between QPE and QPF?
QPE estimates precipitation that has fallen; QPF forecasts precipitation that may fall in a specified area and future period.
Is satellite rainfall as accurate as a local rain gauge?
Their errors and spatial supports differ. Satellite estimates offer coverage; a local gauge provides a point observation. Comparing them requires matching time and scale, not assuming one product is always correct.
Why can two equal rainfall totals produce different floods?
The storms may have different short-duration peaks and spatial patterns, and their catchments may differ in antecedent wetness, infiltration, routing and hydraulic controls.
Does a 100-year flood happen only once every 100 years?
No. Under a stationary interpretation it commonly refers to a 1% annual exceedance probability. An event does not reset a fixed century-long clock.
What is a hydrograph?
It is a time series showing streamflow or water level at a location, often used alongside rainfall and catchment information to understand response.
Why does river water level not equal discharge?
Level is a height relative to a datum; discharge is a volume per unit time. Their relationship depends on channel conditions and may change during floods.
What is forecast verification?
It is the documented comparison of forecasts issued in advance with observed results, including timing, errors, reliability, missed events and false alerts.
What is an ensemble flood forecast?
It is a set of plausible outcomes generated using multiple model states, rainfall scenarios or model structures. Its probability interpretations require calibration and evaluation.
Why does Singapore use both rain and water-level information?
Rainfall informs likely inputs and near-term risk, while water-level measurements and ground observations help show how actual drains and waterways are responding.
Can rain radar see exact rainfall at street level?
No. Radar samples the atmosphere and converts returns into estimated rainfall fields, subject to geometric, physical and processing limitations.
What should students learn before hydrometeorology?
Basic graphs, units, proportional reasoning, conservation of mass, the water cycle and introductory weather and hydrology are useful foundations.
What makes a hydrometeorological result trustworthy?
Clear boundaries and time windows, checked observations, appropriate models, calibrated uncertainty, comparison with measured outcomes and transparent corrections.
Is this page a flood warning for Singapore?
No. It is educational. During severe weather, consult current warnings and safety guidance from the appropriate Singapore agencies, including PUB and NEA.
The connected Gold Standard learning network
These subjects are different lenses on the same water system. Begin with Meteorology and Rainfall to understand atmospheric inputs; Hydrology and Urban Hydrology to examine water response; Hydraulics and Urban Drainage to investigate flow paths; and Flood Early Warning to understand how uncertain evidence becomes public action. The specialist studies demonstrate how the same evidence can be challenged in a more demanding setting.
- The Gold Standard of Hydrology
- The Gold Standard of Meteorology
- The Gold Standard of Rainfall
- The Gold Standard of Weather
- The Gold Standard of Weather Forecasting
- The Gold Standard of Hydraulics
- The Gold Standard of Hydrogeology
- The Gold Standard of Hydrochemistry
- The Gold Standard of Urban Hydrology
- The Gold Standard of Remote Sensing
- The Gold Standard of Geospatial Analysis
- The Gold Standard of Weather Radar Rainfall Estimation
- The Gold Standard of Flood Early Warning
- The Gold Standard of Urban Drainage
- Flood Risk Engineering | Catchments, Exposure and Resilient Decisions
- Hydrology | Rainfall Time Steps and Runoff
- Hydrology | Rating-Curve Hysteresis and Unsteady Flow
- Case Study: Hundred-Year Flood Risk
- Case Study: A Flood Warning With an Outdated Map
- The Gold Standard hub
- The Gold Standard of Education | Learning Transfers Into Life
- How X Works hub
Authoritative sources, dates and interpretation limits
- WMO: Definition of Operational Hydrology (updated June 2026)
- WMO: Hydrology Publications and Guide to Hydrological Practices
- WMO: 2025 Guidelines on the Verification of Hydrological Forecasts
- US Geological Survey: Streamflow and the Water Cycle (September 2026)
- NASA GPM: IMERG Data and Precipitation Estimates
- NASA GPM: Data Questions, Early/Late/Final Runs and Resolution
- NASA GPM: IMERG Version 07 Algorithm Information (August 2026)
- NOAA NWS: Glossary, Quantitative Precipitation Forecast
- NOAA NWS: Glossary, Basin Lag
- Singapore PUB: November 2025 Flood Resilience Campaign and Sensor Data
- Singapore NCCS: Drainage and Flood Prevention (July 2026)
- Singapore NEA: myENV Weather and Flood Information
References checked on 24 October 2026. The WMO’s 2025 hydrological forecast-verification guidelines are not replaced by the 2011 manual on flood forecasting and warning; they address different purposes. NASA’s 2026 IMERG Version 07 documentation is algorithm information, not a guarantee that a gridded product is accurate at the scale of an individual street. PUB’s published November 2025 sensor quantities and 2028 deployment plan are dated operational context, not a claim to a current sensor inventory.
The three worked cases, time intervals, catchment areas, rain totals and model outputs are original fictional exercises. None represents a live flood, a PUB operational forecast, an actual rainfall gauge record or a hydrological design certification. Where real property, infrastructure or safety decisions are involved, use current authoritative instructions and suitably qualified professional assessment.