The DWMP blog – Episode 18. The significance of rainfall

This is the latest in a series of blogs discussing the development of Drainage and Wastewater Management Plans (DWMPs).?If you haven’t already seen the earlier episodes in this series, they are all here (https://tinyurl.com/MartinOsborneArticles) I suggest that you start from Episode 1 (https://tinyurl.com/DWMP-blog).

In this episode I consider the importance of correctly representing rainfall and how we have perhaps tied ourselves in knots using different approaches.

Rainfall is the key driver for flood risk and environmental impact of drainage and wastewater systems (except completely separate foul systems).?It is therefore important to consider all aspects of modelling rainfall to provide a sound basis for the modelling.

Different representations of rainfall are used for assessing flood risk and overflow spill.?

Rainfall for flood risk

Assessment of flood risk generally uses synthetic design storms; that is storms that have the correct statistical characteristics for the rainfall climate of the catchment, but have an idealised shape.

In the UK these storms are symmetrical about the centre (although other countries used skewed storms with the peak intensity early or late in the storm).

Traditionally the probability of a storm was described as the return period; that is the average number of years between occurrences.?This could be confusing as it incorrectly implied that once a 50-year return period storm had occurred there wouldn’t be another for 49 years.?Modern practice is to describe the annual probability of occurrence of the storm.?The two are simply related as a 50-year return period storm has an annual probability of 1 in 50.?The terminology 1/50 is used here although others prefer to write this as a 2% probability.

Choice of storm probabilities

The published DWMP guidance suggests that storms with annual probabilities from 1/1 to 1/50 should be used for flood assessment.?In practice the 1/1 probability is not required as flooding is unlikely in these events and would not be tolerated for future performance.

A range of 1/5, 1/10, 1/20, 1/30 and 1/50 is therefore recommended; although even the 1/5 may not produce significant flooding.

Annual maxima / peaks over threshold

The depth of rainfall for a defined duration and return period is derived from a statistical analysis of rainfall records. ?In the UK this can be obtained from the FEH webservice for any location.?There are two statistical approaches.?An analysis using only the largest storm in each year (annual maxima) or analysis of all storms larger than a certain size (peaks over threshold).?The webservice are provides results for both.?Annual maxima is more accurate for large infrequent storms, with POT appropriate for annual probabilities greater than 1/5.?For the range of probabilities recommended here, annual maxima analysis should be chosen.

Seasonal rainfall depth

Both statistical analyses give the storm characteristics irrespective of season.?However, seasonal correction factors are provided to adjust this.?For summer storms the adjustment is very small but for short duration storms in winter the depth may need to be reduced by up to 40%.?As the antecedent catchment wetness will be different for different seasons, we need to analyse the two seasons separately and so need to use these adjustments.?

For most urban drainage systems summer is the critical season because of the greater depth of rainfall although for systems that have greenfield runoff or have large amounts of storage winter conditions may be worse because the catchment is wetter giving more runoff in long duration storms.

Areal reduction factors

The statistical model defines the depth of rainfall at a single point and this will tend to overestimate the rainfall that occurs across a catchment; as the rainfall will not be as intense at all points.?A correction is made for this using areal reduction factors that depend on the storm duration and the area over which the influence of the storm is to be assessed.?However short duration storms only influence the small area that the flow can travel over in the duration of the storm.?Default ARFs are therefore inappropriate and reduction factors for all situations should be between 0.9 and 0.95 (that is 5 to 10% reduction)

The FEH webservice provides a choice of the point rainfall for a 1 km grid square or the catchment averaged rainfall for a large catchment.?The default for the catchment average is to apply an ARF appropriate for the whole catchment area and this is incorrect.?Grid square rainfall should therefore be used.

Durations and compound storms

Each point in the catchment will have a critical duration of storm that gives the worst flood response.?This is related to the travel time of flow to the point from all upstream areas.?To correctly represent the flood risk across the catchment therefore requires analysis of a large range of storm durations.?Typically, durations from 15 minutes to 12 hours will be used although longer durations may be required for very large catchments or ones with large volumes of detention storage.

An alternative approach that has been adopted by some companies is to use a compound storm (“superstorm”) built up from all of the different durations to give one single long duration storm.?This simplifies the analysis and reduces the computational burden.

Climate change

The effect of climate change on rainfall intensities must be taken into account for future planning horizons.?Guidance on this is constantly being reviewed and updated.?The percentage increase will depend on the region of the country and may also depend on the season, storm duration and return period.?The increase will also have a significant uncertainty and this needs to be taken into account in the future scenarios.

Antecedent conditions

There is guidance on the appropriate antecedent conditions for runoff models for use with design storms.?This will vary with the annual rainfall of the catchment and with the season.?They may also vary with storm duration or return period; although this is normally ignored.

No analysis has yet been carried out on climate change effects on antecedent conditions due to changing rainfall and increased summer evaporation.?This is likely to have a relatively small effect but is an area for future research.

Rainfall for spill assessment

For assessing flood risk, we can use storms of idealised shape from two seasons with a range of durations.?Assessing overflow discharges is not as clear cut particularly if we are using water quality modelling to assess the environmental impact.?We are looking at more frequent events so just considering two seasons may not be enough to predict spill frequency and volume.?The pollutant load of the discharge depends on the length of the dry period beforehand and potentially on the skew of the storm.?The impact depends on the combination of sewer and river flow and temperature and other factors.?The approach for spill assessment is therefore to use a timeseries of rainfall events representing the day-to-day variation of conditions.?This is a significant computational burden.?Some things to consider.

Length of timeseries

The required length of the timeseries depends on the environmental standards that we are using and the frequency of events that they define (see Episode 14).?For 99 percentile standards we are excluding the outliers of the worst 88 hours of the year.?Analysing one typical year is therefore adequate.?For FIS standards we exclude only the worst hour of the year.?We therefore need at least ten years of data to include the variation from year to year.

There is a similar argument for spill frequency assessment.?If we are assessing 10 spills a year then one typical year is adequate.?If we are assessing 1 or 2 spills a year then we need at least ten years of data.

Source of timeseries

If there is 20 or more years of recorded hourly rainfall data for a catchment then this can be used as a source of the timeseries.?If not, then a synthetic timeseries can be generated based on recorded daily data or rainfall statistics using the StormPac tool.?This is typically used to generate 40 years of representative rainfall data.?

Selection of appropriate years

Given a long real or synthetic rainfall record, how do we select the years to use for the analysis?

To select an average year, we look for the year closest to the average for annual rainfall, number of dry days, number of very wet days.?Simples.

Selecting a ten-year series is more complicated.

• We should not select individual years and put them together; rather we should select blocks of years.?This avoids the problem of one year having heavy winter rain in the last week of December and another year in the first week of January.
• We should exclude any year with an exceptionally large storm; say greater than a 1:30 year event.
• ?We should not select years that are closest to the average.?The whole point of using a 10-year series rather than an average year is to represent the variation from year to year.?We should therefore select a block of years that overall is similar to the average of the whole series but also has similar standard deviation to the whole series.?

Climate change

Techniques are available and are being improved for modifying a timeseries to allow for climate change.?Ideally these should allow the generation of high, medium and low climate change scenarios.

Antecedent conditions

The timeseries contains the full rainfall record and can therefore be used to generate the antecedent conditions for each part of the record.?Ideally the timeseries should be run as a continuous simulation to keep all catchment conditions continuously updated.

Alternative approaches

The use of long timeseries of rainfall data is required for assessing the impact of discharges on receiving water quality, so that we represent the full range of conditions.?However, we are increasingly moving towards the use of spill frequency rather than environmental impact.

To assess a spill frequency of once a year we currently require the use of a ten- or twenty-year timeseries, but to assess a flood frequency of once a year we would use a limited range of synthetic design storms.?Have we tied ourselves in knots?

Work was carried out 30 years ago (although based on very limited data) to define synthetic design storms with probabilities of multiple times a year.?That definition is still included in InfoWorks ICM although it is not widely used.?Is it time to update that work to allow a catchment specific analysis and also to include climate change??We could then reduce the computational burden of long timeseries and give us more time to think and refine our DWMPs.