Stormwater harvesting (SWH) is an important, water-sensitive urban design (WSUD) approach. This article shows you a new SWH conceptual design modelling framework. It handles the optimal placement and sizing of stormwater harvesting infrastructure within an urban development.

I will be presenting this model at the 2016 Water, Infrastructure and the Environment Symposium on 28 Nov – 2 Dec 2016.

How can we optimise multiple stormwater harvesting benefits?

Stormwater harvesting system designs must account for trade-offs between several objectives. Important trade-offs may include those between:

  • lifecycle cost
  • harvested volume
  • water quality improvement performance.

How a system performs in each objective is dependent on complex interdependencies between design decisions. These include the type, size, and spatial distribution of best management practices (such as wetlands, bioretention, and ponds) and transfer infrastructure.

My goal was to develop a framework that showed decision-makers how design decisions impact the trade-offs between objectives. This assisted with the selection, sizing, and placement of harvesting infrastructure within an urban catchment.

The resulting framework produces preliminary stormwater harvesting system designs that represent optimal design trade-offs.  It advances previous work in stormwater harvesting optimisation studies in two key ways. Previous studies 1) did not  consider water quality as a formal objective, and 2) did not integrate distributed best management practices and transfer infrastructure networks.

Optimiser linked with eWater MUSIC

To evaluate and identify optimal designs, the framework included an integrated urban stormwater model (eWater MUSIC) and a lifecycle cost (LCC) model. These were linked with a multi-objective genetic algorithm (NSGA-II). The framework was used to solve a mathematical optimization problem consisting of decision variables, objectives and constraints, as follows.

Decision variables

The decision variables included the type and surface area of best management practices:

  • biofilters
  • wetlands
  • sediment basins
  • open storage ponds at locations in a drainage network.


The objectives included:

  • the life cycle cost of best management practices and transfer infrastructure
  • average annual reliability of supplying a specified demand
  • average annual total suspended solids (TSS) reduction.


The optimisation constraints included:

  • checks for acceptable combinations of best management practices
  • minimum targets for total suspended solids, total nitrogen, and total phosphorous reduction.

Applying the modelling framework to a case study

The framework was applied to a case study for a ‘green fields’ housing development in Adelaide’s Northern suburbs.

The case study considered three potential irrigation demand scenarios:

  1. Low (61 ML/year)
  2. High (122 ML/year)
  3. High demand plus 40 ML/year export (162 ML/year).

Figure 1 shows how an initial set of random stormwater harvesting system solutions evolved. It shows how the NSGA-II algorithm improved the solutions. And it shows how candidate designs were simulated in MUSIC for one scenario loop.

The NSGA-II used MUSIC simulation results to calculate harvesting system design objective ‘fitness’, to iteratively improve the designs.

There were impractical solutions generated due to random operations in the optimisation process. We pre-emptively screened these out so they were not simulated in MUSIC. Doing this saved approximately 90% of computer run-time.


The case study optimization process shows what in creating the initial population you need to choose decision variable values for each location. The process then moves through (2) evaluate solutions; 3 Selection; 4 Crossover; 5 Mutation; 6 Evaluate solutions; 7 New population. During this process each candidate solution has a subsidary process. The process excludes impractical solutions by screening them ahead of time. It does this by 1 checking practical constraints; 2 set MUSIC model inputs; 3 run the MUSIC model; 4 Check constraints; 5 compute and apply penalties. Model pre-emption moves straight from checking practical constaints into computing and applying penalties. The optimization process returns from New Population back to Selection and runs again.

Figure 1. Details of case study optimisation process. (Adapted from Marchi et al (2016), Di Matteo et al 2016.)

Optimisation results show multi-objective trade-offs

The results showed that we can achieve better harvested stormwater supply reliability and water quality by treating and storing closer to runoff sources. This is because the model distributes capture, treatment, and storage best management practices where space at the catchment outlet is limited.

Figure 2 shows the Pareto-efficient solutions for the three demand scenarios. The figure shows trade-offs between cost, supply reliability, and water quality improvement.

For each scenario, the trade-off pattern shows solutions in a low-cost/high-reliability ‘knee-region’. The knee region represents the best compromise in these objectives. However, for a given cost, there are other solutions available that are away from the knee region and have  a much higher total suspended solids reduction (darker blue). They come with a slightly lower reliability.

What this shows is that, for the same cost, a small compromise in reliability can considerably increase total suspended solids reduction.


This figure is a chart that shows the Pareto trade-off in stormwater harvesting systems.

Figure 2 Trade-offs between objectives for the Pareto optimal stormwater harvesting system designs found by the framework (adapted from Di Matteo et al 2016)

Exploring the impact of design decisions

Table 1 shows the values of the best management practice type and surface area decision variables of six selected Pareto-efficient stormwater harvesting systems. The best management practice numbers (eg BMP 2) refer to the location of the sub-catchment.

Based on MUSIC modelling, biofilters (B) in locations with high run-off in-flows (ie ‘central’ and catchment outlet locations) were  preferred in Pareto-efficient solutions.

In supplying higher demand (scenario 3), placing biofilters in sub-catchments with the highest impervious fraction gives us the best return on investment for improving reliability and TSS reduction.

This table shows decision variable values of selected solutions.

Table 1 Decision variable values of selected solutions (adapted from Di Matteo et al 2016)

The framework can be used to identify preliminary designs for efficient stormwater harvesting systems at catchment or sub-catchment scale. It can be used for any new development. It is a general framework, which means you can substitute stormwater models or evolutionary algorithms as needed.

Upcoming conference

This work will be presented at the 2016 Water, Infrastructure and the Environment Symposium. The conference will be held in Queenstown, New Zealand from 28 Nov – 2 December 2016. Visit the website for more information.



Di Matteo, M., Dandy, G., and Maier, H. (2016). “Multi-objective optimization of distributed stormwater harvesting systems.” J. Water Resources Planning and Management, in press.

Marchi, A., Dandy, G., and Maier, H. (2016). “Integrated Approach for Optimizing the Design of Aquifer Storage and Recovery Stormwater Harvesting Schemes Accounting for Externalities and Climate Change.” J. Water Resources Planning and Management, 10.1061/(ASCE)WR.1943-5452.0000628,04016002.