Toward a water-sensitive precinct with stormwater harvesting: a case study in South Africa

South Africa is a water-scarce country with inadequate freshwater resources due to unevenly distributed rainfall. With the rapid increase in water demand due to population growth, it has been reported that there would be a gap between water demand and water supply of some 17% by 2030 (Okedi 2019). Furthermore, there is evidence that climate change will continue to increase temperatures especially in urban areas such as Cape Town, South Africa and reduce mean annual rainfall (Okedi 2019). Unless there is a meaningful change in water supply and use patterns, the impact on water demand, for example, for irrigation, industry and the services sector, will directly and adversely affect the economy. An example of the impact of variable water supply in an urban area was when the City of Cape Town faced the possibility of taps running dry in 2018 due to a prolonged drought that commenced in 2015. To address these challenges, a more holistic, resilient, and sustainable water management approach is required to integrate the utilization of various water sources such as groundwater, rainwater, stormwater, and wastewater. Various studies have shown that with rapid urbanization, the role of stormwater becomes crucial for the sustainability of urban environments (Okedi 2019; Schmit et al. 2020). Urbanization is known to restrict infiltration, thus altering flow regimes of groundwater and surface waters resulting in ‘flash’ floods and larger runoff volumes characterized by higher velocities and increased peak discharges (Dietz & Clausen 2008; McGrane 2016). The changes result in a loss of biodiversity and aquatic habitats as well as increased erosion and sediment loads (Jackson & Booth 1997; Vietz et al. 2015).

While stormwater infrastructure is largely static, the ability of dynamic weirs, gates and valves to more closely achieve desired hydraulic conditions have long been recognized. The use of manually adjustable valves can be applied for dredging purposes, emergency facility drawdown, or other maintenance activities and is common practice in the design and maintenance of extended detention ponds (Nashville & Davidson County 2009; ODOT, 2014; USEPA 2009). It is feasible to add automatic controls to previously manually or passive controlled stormwater facilities due to recent advancements and reduced cost of internet connected sensors and actuators (Kerkez et al. 2016; Bartos et al. 2017). A heuristic or rule-based control (RBC) algorithm is commonly applied as an automation of RTC (Jacopin et al. 2001; Middleton & Barrett 2008; Gaborit et al. 2013; Goodman & Quigley 2015). Here, control rules are manually programed typically in the style of an ‘if-then-else’ logic structure before the RTC system is on-line (Schmit et al. 2020). Design and implementation of RBC typically require expert knowledge of the urban drainage system (Vitasovic 2006).

In this study, an investigation was undertaken to determine the extent to which stormwater harvesting could mitigate water scarcity in a highly urbanized area. Increasing the physical capacity of storage components may be required for the multi-functional use of stormwater infrastructure for water harvesting and to mitigate flood risk. Physical increases of storage capacity in an urban area may require the demolition of infrastructure such as buildings and roads (García et al. 2015). To minimize the need to demolish existing physical infrastructure in an urban area, some studies such as Borsanyi et al. (2008); Okedi (2019) and Mogano & Okedi (2023) have shown that Real-Time Control (RTC) can provide the required additional storage capacity within existing stormwater detention facilities. This can be achieved through dynamic management of the existing storage within the stormwater system by applying RTC operational rules to minimize redundancy in the detention facility (Colas et al. 2004; USEPA 2006). In the studies in the literature, operational rules have been based on water levels with limited integration of inflow data. Some case studies are discussed as follows.

In Québec city, Canada, RTC was implemented of a grassy, dry, on-line stormwater detention basin located at an outlet of a residential catchment. The catchment area was 15.3 ha with an imperviousness area of 33% (Gaborit et al. 2013). The catchment had a dual drainage and sewer system for stormwater that utilized the streets for major flow conveyance and an underground sewer for the conveyance of frequent flows (the minor system). The benefits of RTC implementation included reducing overflows while maximizing the detention time. Runoff retention enabled a smooth drawdown during discharge to minimize TSS re-suspension and minimize hydraulic shocks on receiving water bodies. Overall TSS removal increased from 46 to 90%.

Another study in Tehran, Iran applied on-line RTC with controllable and uncontrolled gates and openings to temporarily store excess storm runoff due to a lack of hydraulic capacity in the drainage network (Jafari et al. 2019). The operation focused on reducing flood inundation at downstream end of the system, hence rules on how to regulate gate openings were used to optimize the system performance. It was determined that the efficient use of the regulation capacity was influenced by the ability to regulate all controllable elements in the system. Application of RTC led to the optimal utilization of the reservoir capacity where excess water was temporarily stored. This water can be later used for other purposes such as irrigation of urban green landscape. Additionally, partial control of the system compared with a fully controlled case led to a 17% increase in flood inundation.

Thus far, the use of optimization-based control approaches such as model predictive control and linear-quadratic regulators is a shift that RTC practitioners and researchers have turned toward for prevention of downstream capacity exceedance by urban drainage systems (Marinaki & Papageorgiou 2003; García et al. 2015; Wong & Kerkez 2018). However, operations personnel cannot easily understand and adjust these systems which are computationally intensive and make decisions in a ‘black box’ (Vitasovic, 2006). The implementation of translucent, rule-based or optimization-based control could mitigate the novelty of RTC technology for stormwater management applications that may be a hindrance to its adoption (Schmit et al. 2020).

Rohrer & Armitage (2017) used a Life Cycle Cost Analysis (LCCA) as an economic assessment which revealed that stormwater harvested in a decentralized manner – i.e., harvesting from multiple ponds – tends to maximize the volume harvested, which also increases the overall cost of the system due to increased maintenance, operation and capital costs. By contrast, centralized systems – i.e., harvesting from a single pond – minimizes the infrastructure costs but with reduced yields. Additionally, Fisher-Jeffes (2015) also used an LCCA which noted that subcatchments with higher density populations typically had cheaper SWH systems (when costs were reduced to ZAR per kilolitre), this made sense when considering the cost of the dual reticulation network which is a major component of the cost of a SWH scheme. In general, it was also evident that the higher demand associated with larger population would result in lower cost per kilolitre.

The study described herein simulated an operation approach that integrates RTC into existing stormwater infrastructure through low-cost adaption by adjusting or improving functionality to a changing hydrologic environment as suggested in Kerkez et al. (2016); Okedi (2019). The study applied RTC rule-based approaches on a representative urban catchment at the University of Cape Town (UCT) with the objective of determining the opportunity to transition the urban area toward a water-sensitive precinct. Unlike in previous studies were RTC rules were only linked to water levels, in this study, the rules were applied to the UCT dam operations to initiate pre-storm releases in real time based on forecast rainfall volumes. In addition, the study assesses the application of RTC in relation to harvested stormwater. Four stormwater harvesting configurations were modeled, i.e., two conventional configurations (SC1 and SC2) and two dynamically managed configurations with RTC (RTC-1 and RTC-2). An LCCA was also applied to the four configurations to assess the economic viability of the four harvesting stormwater approaches.

Due to its natural features and topography, the rainfall in the watershed is far higher than the average rainfall for Cape Town ranging from 1200 mm to 1500 mm/year while evaporation varies from 600 to 1500 mm/year (Fisher-Jeffes 2015). The UCT dam and its catchment were selected to represent a typical storage reservoir used for irrigation. The catchment area is 0.33 ha, has a slope the varies from 20 to 30% (average slope of 24.5%) and is 98% pervious, with land use composed of an open greenfield area. The catchment drains into the UCT dam with a maximum depth of 12.5 m and a surface area of 0.82 ha. For modeling purposes, the catchment was divided into 31 subcatchments, with the stormwater conveyance network consisting of open natural channels. A Digital Elevation Model (DEM) was used to define subcatchment boundaries as well as to extract stormwater channel cross sections.

The EPA SWMM 5.1 model is an open-source program for hydraulic and hydrologic studies. It was used to model the rainfall–runoff process in the study area. The EPA SWMM model is commonly used for RTC simulations as control algorithms can be also applied externally via a programing wrapper or programed directly into the model as described in various studies such as Joksimovic & Sander (2016); Gaborit et al. (2013, 2016); Goodman & Quigley (2015); Muschalla et al. (2014); Degrave et al. (2013); Heusch & Ostrowski (2011); Wong & Kerkez (2018).

To model stormwater harvesting, a substantial amount of data was required including a DEM of the catchment, infiltration data, rainfall data, evaporation data, and water demands. The DEM had a 10 m resolution which was adequate for modeling purposes. To estimate infiltration, a Green-Ampt equation was used based on soil parameters, which are widely available in literature. Borehole logs of the University of Cape Town were used to identify the soil conditions (upper soil type zones) within the catchment. The borehole logs were useful when estimating infiltration parameters. Additional data were obtained from the ‘as built’ drawings of the UCT dam including the outlet structure, the spillway properties, and its stage-storage curve. The behavior of the storage during operation was modeled using a depth versus surface area curve as shown in Table 1.

While historically, the UCT dam was able to adequately supply water for irrigation of the sports fields, this has not been the case in recent years. The water supply for irrigation is now being supplemented with municipal potable water which is not ideal due to the problems of water scarcity. The estimates of irrigation demands were based on an ‘irrigation demand model’ and the water demands are presented in Table 2.

The data in Table 2 show that a significant amount of irrigation water (about 50%) was drawn from municipal system. The irrigation demand met by the municipal system accounted for about 26% of the total municipal demand (see Table 2). In this study, the objective was to determine whether management approaches can be applied to the UCT Dam such that the use of municipal water for irrigation can be minimized and allow for irrigation demand to largely met by stormwater harvesting. This study focused on the application of RTC techniques to dynamically manage the available storage in the UCT dam including initiating prestorm releases in real time based on forecast rainfall to increase the capture of stormwater and to reduce overflows from the dam.

In this study, two conventional and static control configurations were applied to the UCT dam namely, Static Control 1 and Static Control 2 denoted as SC1 and SC2 (see Table 3). The clock times are linked to irrigation regimes, i.e., system operations between 08:00–13:00 h.

Various assumptions were made in the application of the RTC rules, i.e., SC1 represents the baseline operational procedure that is currently in practice at the UCT dam without planned periodic maintenance (i.e., no dredging to remove sediments from storage), while SC2 denotes a case with maintenance. SC2 is an improvement to the current practice. It assumes that maintenance is regularly undertaken, i.e., dredging to remove sediments from storage and the catchment is managed adequately and optimally to convey runoff into the UCT dam. The study determined that the sediment deposited in the UCT dam has accumulated to about a 6 m depth, thus depriving the stormwater harvesting system of almost 8% of the dam storage. The SC2 scenario would thus establish the benefit of periodic planned maintenance and account for a situation where dredging would have been undertaken to remove sediments from the UCT dam to maximize its storage. For the modeling of SC1 and SC2 static control simulations, input data consisted of 5-min time-interval historical rainfall time series for the period 2015–2022 obtained from the South African Weather Services (SAWS).

The study then developed two RTC configurations, i.e., RTC-1 and RTC-2 linked to the two static control operations, i.e., SC1 and SC2, respectively. The simulation of the two RTC operations considered actuator settings and control rules to dynamically manage storage available in the two static control operations, i.e., SC1 and SC2 and managed based on rainfall forecast information. Rainfall forecast data used for RTC application in the study was based on Global Forecast System (GFS) that provide predictions over two weeks in a spatial form. A rainfall forecast at 6-h intervals provided by the National Centre for Environmental Prediction (NCEP) was used for this study. The available GFS forecast data were collected for the study area where GFS data were available at latitude 34° 00′ S and longitude 18° 31′ E. The data were extracted for the period 2015–2022 and used for the RTC simulations linked to forecasted rainfall volume to initiate pre-storm releases.

The description of the data used to model the stormwater harvesting system in the study area at the University of Cape Town is summarized in Table 4. The stormwater harvesting model with RTC was developed in EPASWMM 5.1 (Rossman 2015). The software is commonly used for RTC simulations with open-source control algorithms. The dynamic wave routing of conduit lengthening and a routing time step of 15 s was used to represent the hydraulic processes of the system.

To incorporate user-defined controls appropriate for the study area, algorithms were developed and applied externally via PySWMM, which is a Python programming language package that allows for the stepwise observation and modulation of SWMM models. PySWMM tools can observe and manipulate nearly any parameter in SWMM, however, only data that could be feasibly monitored in a real-world application can be used for RTC application to create a realistic scenario (Schmit et al. 2020).

For model development, a 5-min time step was used for the simulation of rainfall–runoff based on historical data. This has previously been determined to provide a realistic interval to receive and transmit data and control commands while providing adequate temporal resolution (Bartos et al. 2017). The GFS of 6-h interval data were used to estimate the forecast rainfall volume, then linked to RTC to initiate pre-storm releases via PySWMM. PySWMM can also be used to interrupt SWMM simulations during processing to retrieve relevant measurements such as precipitation, flow, depth etc., at any interval (Schmit et al. 2020). Based on these studies, control algorithms were developed and applied to guide decisions such as appropriate outlet valve opening for prestorm release and normal operation of the UCT dam. The actuator was set to the required position based on a control algorithm input via PySWMM to account for pre-storm release and normal operation requirements, while SWMM simulations continued until the next time step if no interruption was required.

The results shown in Table 6 include the calibration and validation errors for runoff quantity continuity and flow routing continuity.

The following steps were taken in applying LCCA to the stormwater harvesting system:

• (i) The discount rate was taken as the difference between a government 10-year bond rate and the inflation rate (Consumer Price Index) as shown in Table 7.

• The values in Table 7 are in line with the National Treasury recommendations, which state that ‘For practical purposes, the discount rate is assumed to be the same as the risk adjusted cost of capital to government. The government has been used by some institutions as the discount rate for a particular project over a comparable period. The argument in favor of using the government bond is that it reflects the actual cost to government of raising funds at any given time. This ignores several factors that are difficult to quantify, including: various risk margins relating to increased government borrowing; various tax implications of diverting funds from private to public consumption; and government's time preference of spending’.

• (ii) Life cycle duration – the LCCA was undertaken over a 50-year period that is recommended to represent the lifespan of a stormwater harvesting system. Typical life expectancies of 50 years are recommended for economic analysis of water supply systems (Mackenzie 2010).

• (iii) Annual estimates such as the harvested stormwater volume and the cost of energy were selected to coincide with the period used in model simulations.

• (iv) Life span of components in the stormwater harvesting system – the future capital costs and residual values of each system component was selected to coincide with the end of the life cycle duration.

• (v) The Equivalent Annualized Cost (EAC) was computed using Equation (4), and this value was estimated as the cost per year of ownership and asset operation.

  

  (4)

  where EAC = equivalent annualized cost (in ZAR); n = number of years from present year; i = real discount rate and LCC = life cycle cost (in ZAR).

The assessed average annual stormwater yields are compared in Figure 8. The study determined that RTC improved system performance in terms of average annual harvested stormwater volume by up to 6.3% in comparison to the SC1 scenario (see Figure 8 and Table 8).

For the baseline operational procedure that are currently in practice at the UCT dam (scenario SC1), the results show that application of RTC operation measures increased average annual harvested stormwater volume by approximately 2.1 ml (see Table 6) In the case of scenario SC2 where sediments are removed from the UCT dam to maximize storage and RTC operation measures are applied increased average annual harvested stormwater volume by approximately 1.1 ml (see Table 6). Although the increase seems minimal, the application of RTC-1 and RTC-2 algorithms on the UCT dam provided valuable additional capacity to capture stormwater that is harvested to meet demand. As shown in Figure 10, with application of RTC, harvested stormwater exceeded demand in both RTC 1 and RTC 2, i.e., with or with dredging to increase dam capacity. The study also determined that periodic planned maintenance was essential as it increased stormwater average annual harvested stormwater volume by 1.5 ml. The increase due to maintenance was sufficient to meet irrigation demand.

For each configuration, i.e., SC1, SC2, RTC-1 and RTC-2, the unit costs of harvested stormwater were computed by dividing the EAC with the average annual harvested stormwater volume. The unit cost of harvested stormwater for each configuration is presented in Figure 10.

Water usage by sports bodies in Cape Town are billed at a flat rate (excluding VAT) of 29.55 ZAR/kL (CCT 2022). Static (SC1 and SC2) and Real-Time (RTC-1 and RTC-2) configurations harvested stormwater at approximately four times and three times the tariff, respectively. RTC-1 and RTC-2 harvested stormwater was estimated to be about 1.80 and 1.99 ZAR/kl, i.e., 21 and 25% more expensive than SC1 and SC2, respectively.

In summary, the study determined that the average annual harvested stormwater volume from a stormwater harvesting system can be improved by using RTC. This can be accomplished with prestorm release, i.e., collecting rainfall forecasts in real-time and discharging water from the system prior to the occurrence of a rainfall. The use of pre-storm releases would be based on upcoming storm runoff to provide additional system storage capacity. The upcoming storm runoff would be estimated from rainfall forecast data available from platforms such as GFS. However, it was determined that the GFS data had some differences in magnitude in comparison to observed events especially the time when the peaks occurred. In the study area, some differences were identified such as an event with a peak that was 40% higher than GFS data due to poor quality of the forecast. The disadvantage of an underestimation in the forecast would be no release from the storage (UCT dam) with subsequent occurrence of a flood. The quality of forecast can be improved by implementing a forecast with smaller time-steps (i.e., 5- or 15-h temporal resolution) and forecast revision for better water management. Forecasts can be revised when the difference between the actual conditions and forecast are such that public safety and security are at risk and/or inconveniences the public. In addition, precipitation should be accounted when the chance of precipitation is equal to or greater than 30%. Overall, the GFS data were relatively accurate, but with rainfall depth of forecasts generally lower than real-time which often produced underestimated volume of prestorm release for this study. To mitigate the impact of this observed underestimated, the study implemented prestorm release with additional amount of water on occasion for the RTC system. Hence, this has the possibility of diminishing performance for water supply. This ‘underestimation’ can be improved by using a more accurate and smaller time-interval rainfall forecast data which optimizes the system.

In this study, a continuous simulation was conducted to model the ability of two types of SWH system, namely a conventional (static) and RTC system (dynamic) to deliver non-potable water for irrigation. The study determined that the implementation of RTC can significantly improve the annual yield and volumetric reliability performance of a SWH system. The dynamic management of the UCT dam with RTC-1 and RTC-2 approaches increase yield by 2.1 and 1.1 ml, respectively. Additionally, RTC-1 and RTC-2 approaches increase volumetric reliability by 5.3 and 2.5%, respectively, while maintaining the required level of service of a stormwater harvesting system. SC1 and SC2 results in water savings of up to 21.15 and 21.45 ZAR/kl, respectively, while RTC-1 and RTC-2 could save up to 19.35 and 19.45 ZAR/kl. Thus, static configurations result in water savings approximately 9% in comparison to RTC. In addition, Static configurations harvested stormwater at a relatively lowest unit cost in comparison to RTC configurations. Hence, RTC approaches increase yield and volumetric reliability with relatively low-cost implications. Notwithstanding this finding, the RTC system exhibits a great potential in reshaping the stormwater harvesting system to simultaneously deliver water conservation and stormwater management. The ability of an RTC system to provide centralized control and failure detection, which can be readily adapted to variation of climate and local conditions over both the short and long term, opens possibilities of delivering a system that is more stable and reliable.

This research mainly focused on the benefit of RTC in enhancing stormwater harvesting to transition a precinct in Cape Town toward a water-sensitive future. This study represents only a part of a greater effort to reuse water in an urban area including inter alia harvesting and reuse of stormwater and rainwater, and treatment and reuse of greywater. RTC approaches applied in combination with urban drainage system retrofits and periodic maintenance can pave the way for smarter stormwater management.

M.M.M. undertook the research including the literature review, selection of a suitable method, development of the assessment model, generation of the results, wrote the draft paper, and implemented the revisions from feedback/comments received from co-author and reviewer. J.O. (ORCID https://orcid.org/0000-0001-7707-2721) proposed the research topic, managed, and supervised the study, assisted in the paper writing, formatted the paper for submission and is the corresponding author.

This research received no external funding. The first author only received some funding contribution toward MSc study from the South African Department of Water and Sanitation (DWS), and this has been mentioned in the acknowledgements.

The first author is grateful for financial assistance in form of an MSc study bursary from the South African Department of Water and Sanitation (DWS), and academic support from the University of Cape Town – Department of Civil Engineering and Future Water Institute. The research was linked the larger University of Cape Town (UCT) funded Khusela Ikamva Sustainable Campus Project – a 5-year sustainable campus project where UCT campuses will be used as living laboratories for collective innovation to reduce the ecological footprint. The Khusela Ikamva project aims to address five research areas, i.e., energy/carbon footprint, water, waste, wildlife, and social responsiveness, and will play a major role in addressing the institution's sustainability. The authors also appreciate the City of Cape Town and the South African Weather Services for providing data used in the study.

The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of the editor(s). The editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.