A flow-duration curve answers a practical question: how often has a river carried at least a particular amount of water? It rearranges a record of river discharge from highest to lowest and plots flow against the percentage of observations that equal or exceed it. Unlike a hydrograph, it does not preserve the order in which floods and low flows occurred. That difference makes it useful for comparing water availability but unsuitable for answering every question about drought or flood risk. Understanding what the curve keeps—and what it discards—helps students interpret river data without confusing a common low flow with a rare flood, or a historical percentage with a guarantee about tomorrow.
Key takeaways
- A flow-duration curve shows exceedance percentages, not a timeline.
- High flows usually appear on the left and low flows on the right.
- Daily-average curves do not preserve instantaneous flood peaks.
- The same curve can represent very different sequences of dry days.
- Record length, missing observations and changing river management affect interpretation.
How to read a flow-duration curve
Discharge is the volume of water passing a river cross-section per unit time, commonly measured in cubic metres per second. On a conventional flow-duration curve, discharge is plotted vertically and percentage exceedance horizontally. Exceedance means being equalled or exceeded, rather than being below a threshold.
A value labelled Q95 commonly means the discharge equalled or exceeded for 95 per cent of the record. It is therefore usually a relatively low flow. Q5 is commonly a high flow because only 5 per cent of the observations reach or exceed it. Always check the graph’s definition: percentile labels can be used differently in other fields.
Q50 is the median discharge. Half the observations are at or above it under the chosen convention. It is not necessarily the arithmetic mean, because occasional very large flows can pull the mean well above the middle observation.
Building the curve from observations
Start with a series measured at consistent intervals, such as daily mean discharges. Check units, dates, missing values and quality flags. Then sort the valid observations from largest to smallest. Each ranked value receives an exceedance position, and the pairs are plotted.
One common plotting position is 100 multiplied by the rank, divided by the number of observations plus one. Other conventions exist, especially for handling ties and estimating extreme percentiles. The convention should be documented rather than hidden in a spreadsheet.
For a deliberately small example, imagine nine daily flows of 18, 12, 9, 7, 5, 4, 3, 2 and 1 cubic metres per second after sorting. With the rank divided by ten, the plotting positions run from 10 to 90 per cent. The fifth value, 5, occupies the 50 per cent position.
Nine days are nowhere near sufficient to describe a river’s long-term behaviour. This example teaches the calculation only. Its neat percentages are a consequence of the chosen numbers, not evidence about a real catchment.
What sorting removes
Suppose two rivers have the same collection of daily flows. In one record, the lowest values occur on consecutive days. In the other, those values are separated by wetter days. Their flow-duration curves can be identical even though one river experiences an uninterrupted dry spell and the other does not.
This matters for organisms, water treatment and storage. Ten low-flow days in a row can have different consequences from ten scattered days. Sorting removes that distinction. A curve cannot show whether a shortage happened in winter, during a heatwave or at the same time as peak irrigation demand.
To recover timing, use the original hydrograph alongside the curve. Earth Site’s explanation of seasonal river-flow patterns shows why annual summaries should be interpreted with the calendar still in view.
What the shape can suggest about a catchment
A steep curve indicates a large contrast between high and low flows over the plotted range. A flatter section indicates less variation there. Catchments with substantial storage can release water gradually, while other catchments respond rapidly to rainfall and fall quickly afterwards.
Geology, soils, lakes, snow and reservoirs can all influence these patterns. A relatively sustained low-flow tail may be consistent with groundwater support, but the graph alone does not prove the source of water. Reservoir releases or imported water can produce superficially similar effects.
Comparing shapes is therefore a way to develop hypotheses, not a substitute for catchment information. A useful investigation combines the curve with maps, rainfall records and knowledge of abstractions and regulation. The USGS’s classic work on flow-duration curves emphasised their value for understanding storage and geological differences.
Comparing rivers of different sizes
A large catchment often produces more discharge than a small one simply because it receives water over a larger area. Comparing raw flow values can obscure differences in behaviour. One option is to divide discharge by catchment area, producing a measure of flow per unit area.
Another is to express each discharge relative to a characteristic flow, such as the mean. This highlights relative variability rather than absolute water volume. Neither transformation is automatically best: the choice depends on the question.
If the question concerns how much water could physically pass an intake, absolute discharge matters. If it concerns how differently two catchments respond to rainfall, area-normalised or relative curves may help. A graph should state the transformation clearly so that readers do not mistake a dimensionless ratio for cubic metres per second.
Flow duration is not flood frequency
A flood-frequency analysis often uses annual maximum flows or peaks above a threshold to estimate the probability of unusually large events. A flow-duration curve commonly uses all daily mean flows. The datasets and questions are different.
A flow exceeded on 1 per cent of days is not automatically a “100-year flood”. One per cent of daily observations describes time spent above a discharge. A one-in-one-hundred annual exceedance probability describes the chance of an event being exceeded in a year under a statistical model.
Daily averaging creates another difference. A brief sharp peak can be much higher than the day’s mean discharge. Using a daily curve to design a bridge for peak flooding would discard information that the design needs. The timescale must match the decision.
What can the curve tell us about water supply?
A curve can indicate how often historical river flows exceeded an intake’s desired rate. That is a useful starting point, but it is not the same as calculating a sustainable abstraction. Other users, ecological requirements, licences and water quality also affect what can be taken.
Consider a hypothetical intake needing 2 cubic metres per second. If the river exceeded that rate on 90 per cent of recorded days, it does not follow that all 2 cubic metres per second could be withdrawn on those days. Doing so could leave too little water downstream.
Storage adds another complication. A reservoir’s ability to meet demand depends on the sequence of inflows, available capacity, evaporation and operating rules. A curve cannot show whether wet days arrive before or after a prolonged shortage. Supply simulations therefore need chronological data.
Why ecological interpretation needs more than percentiles
Fish movement, spawning and floodplain connection respond to timing and duration as well as magnitude. A particular percentile may serve as one indicator, but no single point on a curve represents the full needs of a river ecosystem.
Earth Site’s guide to environmental flow regimes explains these additional dimensions. Two operating plans could produce similar annual curves while one creates abrupt daily fluctuations and the other preserves a seasonal rise.
Sediment movement also depends on more than the frequency of a discharge. Available material and local channel conditions matter. The background processes are covered in river erosion and deposition. A discharge threshold is useful only when linked to the process being studied.
Record quality and the low-flow tail
Missing observations should not be silently treated as zero flow. A zero is a physical observation or estimate; a missing value means the condition is unknown. Confusing them can make a river appear to dry more often than it did.
Low flows can also be difficult to measure accurately. Small changes in channel conditions may affect the relationship between water level and discharge. At very low values, proportional measurement errors can be large. Quality flags and station notes help readers judge how much confidence to place in the tail.
Logarithmic axes are useful when discharges span a wide range, but zero cannot be plotted on an ordinary logarithmic scale. A graph must explain how zero-flow days were handled. Omitting them without disclosure changes the apparent frequency distribution.
Historical records and a changing future
A curve summarises its stated observation period. A record dominated by wet years may not represent a longer climate history. A dam built halfway through the record may combine two different operating regimes in one distribution.
Season-specific curves can reveal differences hidden by annual analysis. Separate pre- and post-change curves can support investigation, although rainfall differences must also be considered. A visible shift does not, on its own, identify the cause.
Forecasting requires an explicit assumption about whether past conditions remain relevant. Changes in climate, land cover, abstraction or management can alter future flows. The honest wording is “exceeded during this record”, followed by an explanation of any assumptions used to extend that evidence.
A short checklist for reading an unfamiliar graph
First identify the station, observation period and unit. Next check whether the values are instantaneous, hourly or daily averages. Confirm whether the horizontal axis shows exceedance or non-exceedance, and whether discharge has been normalised.
Then examine missing data, zero flows and known management changes. Finally ask whether the question needs chronology. This sequence prevents common mistakes before any sophisticated statistical interpretation is attempted.
Do not confuse observation frequency with sampling frequency
A curve built from daily means gives each valid day a place in the distribution. A collection of occasional field measurements may overrepresent convenient conditions, because teams can sample more easily during ordinary flows than dangerous floods. Sorting those measurements does not remove that sampling bias.
For example, twenty visits made mostly during summer cannot establish the percentage of all days experiencing low flow. They describe the visited days. Before interpreting a curve as time exceedance, check that the underlying observations represent time consistently or that an appropriate weighting method has been used. A mathematically correct graph can still answer the wrong question if its input record is unrepresentative.
Conclusion
Flow-duration curves are powerful because they compress a long record into a clear distribution. Their weakness comes from the same compression: the order of events disappears. Use them to compare frequencies and variability, but retain hydrographs, seasonal information and measurement context when decisions depend on drought sequences, flood peaks or ecological timing.
Frequently asked questions
Is Q95 a high flow?
Under the usual exceedance convention, it is a relatively low flow, reached or exceeded for 95 per cent of the observations. Check the definition printed with the dataset.
Does Q50 equal the average discharge?
It is the median, not necessarily the arithmetic mean. A few large floods can raise the mean without greatly changing the median.
Can two different hydrographs produce the same curve?
Yes. Sorting identical values into the same order removes their original timing, even if the low flows occurred in very different sequences.
How long should the record be?
There is no universal minimum for every use. Longer, representative and well-documented records usually support stronger interpretation than short or incomplete ones.
Can a curve predict tomorrow’s flow?
No. It describes a distribution, not a date-specific forecast. Tomorrow’s discharge requires information about current conditions and the processes affecting the catchment.
