Analysis procedures using rainfall-runoff models are discussed briefly in Section 5.9 of  Chapter 5 of Book 2 of ARR 2019. Urban stormwater drainage is not specifically considered, and an example provided in Section 5.9.10 relates to a rural catchment in the Northern Territory. In Book 9 of ARR 2019, dealing with runoff in urban areas, there are mentions of rainfall ensembles, but no specific procedures or examples are provided.

The 2019 procedure uses sets or ensembles of rainfall patterns. These are 'bursts', extracted from longer sequences of rainfalls, rather than complete storms.

Overall, ARR 2019 recommends that ensembles be used, rather than single design storms, and ensembles of 10 patterns are the only temporal rainfall data supplied in ARR 2019. 10 storms are considered by the ARR 2019 authors to provide an appropriate balance between accuracy and amount of calculations.

To understand ensembles, consider that analyses or simulations of a drainage system using rainfall data that is converted into runoff by a rainfall-runoff model. The best situation would be to have complete knowledge of all the rainfalls that would occur over the life of the system. Since we do not know this, we might use:

• recorded rainfall data at the site, assuming that the past situation applies to the future life of the project, with possible adjustments for climate change;

• temporal burst patterns developed from intensity-frequency-data (I-F-D) relationships (obtained from statistical analysis of recorded rainfalls and interpolation between stations) combined with temporal patterns based on recorded storm data, (as in ARR 1987);
  

• sequences of many rainfall bursts or storms obtained by Monte Carlo analysis, involving fitting of statistical distributions to rainfall data  and the creation of samples using random numbers;

• selection of storm patterns with required characteristics from rainfall records.

The ARR 2019 ensembles are developed using the last process. Monte Carlo methods are mentioned in ARR 2019 and a Monte Carlo Simulation Discussion Paper, but no specific examples are given for urban areas.

For the design of drainage systems, determining pipe, channel or detention basin sizes to meet certain criteria, there are broadly two sources of design flows:

• peak design flowrates based on I-F-D relationships, in the rational method,

• hydrographs calculated from rainfall temporal patterns or ensembles using a hydrograph-producing program such as Horton (ILSAX), ERM, the ARR 2019 IL-CL model, storage routing models (RORB, RAFTS and WBNM) and other rainfall-runoff models.
  

Until the finalisation of ARR 2019, hydrograph calculations were performed using temporal rainfall patterns from ARR 1987, with one pattern per duration. These patterns represented the 'likely' or 'representative'  storms that would impact the drainage system during its life.  Developing a system to cope with these involves setting sizes of components so that the peak flows and water levels are within safe limits for minor and major flows.  This was done in DRAINS using automatic pipe design and trial-and-error methods that are still available in the program.

With ensembles of 10 storms, the events are equally likely for the given AEP and duration, so it is necessary to analyse or design for the 'average' storm, which might be the median or mean one. With an even number of storms it is not possible to select a particular storm as the 'median', so most designers will want to be slightly conservative, and choose the storm that gives the flow or water level that is ranked sixth. This is refered to as the upper median or upper mean.

To handle these procedures in DRAINS, it is necessary to:

• specify that ARR 2019 ensemble rainfall data is to be used in the Project Options property sheet in DRAINS,
  

• obtain rainfall data from the BOM I-F-D site and the ARR Data Hub, as .csv (comma-separated variable) files, readable on spreadsheets

• input this rainfall into DRAINS,

• select major and minor annual exceedance probabilities (AEPs) and design storm durations,

• run DRAINS in the same way as usual, making a Design run where pipe sizes are to be determined, and Analysis runs to define flowrates and hydraulic grade line levels,
  

• examine Analysis results for each component, reviewing peak flowrates or HGL levels from the (no. of durations x 10) storms modelled.

For each duration, a flowrates or HGL level from a representative upper median (or upper mean) storm is selected based on the option selected in Project Options. These are the pink values shown above. From these, DRAINS selects the highest (red) value, and the results for this are presented after a run.  More-detailed results for a number of chosen storms can be obtained by specifying and running individual storms.

With ensembles, DRAINS only provides full hydrographs for the highest median value for the various durations (the red one in the diagram above). However, graphical and tabular results can be obtained for any of the storms by selecting Individual storms in the Project Options sheet and choosing storms from the Select Major/Minor Storms in the Project menu.

Assessing results in this way is more complex than with the 1987 data, since more events must be examined, and the target or focus is the average event for each duration, rather than one representative event. For each duration, the system should fail in four out of 10 events, sometimes involving large exceedances of design requirements. Designers will need to accept this. It is incorrect to design and analyse using the worst result from the 10 storms for a given duration.

The critical storm can be different for various components. Overflows will be more variable than sub-catchment and pipe flows, or HGL levels. The design of detention basins may be quite complex.