Zero additional maintenance stormwater biofilters: from laboratory testing to field implementation

Climate change, rapid urbanisation, and urban densification are significantly changing rainfall runoff in our cities, increasing both the frequency and severity of flooding, as well as pollution in receiving rivers and bays. The traditional approach to managing urban runoff, that is, to convey stormwater away from people and infrastructure as fast as possible is not working any more, and we are seeing the increase in ‘at-source’ stormwater treatment through decentralised solutions (Iftekhar & Pannell 2022). For this purpose, green, nature-based technologies have been praised for their efficiency and multifunctionality to solve not just stormwater management issues, but also reduce urban heat and increase amenity (Zhang et al. 2020).

Biofilters (also known as raingardens, or bioretention systems) are one of the most widely used nature-based solutions for urban stormwater management, especially across Europe and Australia (Oral et al. 2020; Kuller et al. 2021), due to their flexible sizing and efficacy in managing water quantity and quality (Zinger et al. 2021). In the last 20 years, biofilters have been extensively studied for their pollutant removal performance with respect to nutrients (Bratieres et al. 2008; Lynn et al. 2015; Glaister et al. 2017; Zinger et al. 2021), heavy metals (Blecken et al. 2009a; Hermawan et al. 2020; Lange et al. 2020; Fang et al. 2021), microorganisms (Li et al. 2014; Galbraith et al. 2019; Shen et al. 2020), and, more recently, organic pollutants (Spahr et al. 2020). Optimisation of their operational conditions has also been studied (Wu et al. 2017; Hermawan et al. 2020; Zinger et al. 2021), as has their ability to treat other urban water sources, such as greywater and groundwater (Fowdar et al. 2017b; Barron et al. 2019; Zhang et al. 2021a). However, studies that examine longevity and performance of stormwater biofilters in field conditions are limited, and it is even rarer to see data on failures in the field and maintenance needs and challenges. Zhang et al. (2021a) recently studied the pollutant removal performance of two field-based biofilters, but did not consider long-term performance or maintenance requirements. Al-Ameri et al. (2018) showed that a large number of biofilters constructed in major Australian cities are still showing high efficiency for heavy metal uptake, even after more than 16 years of operation, suggesting that well-designed biofilters in residential catchments might never exceed quality guidelines for heavy metal accumulation. However, this study still highlights high-performing field biofilters, and it is likely that in Australia (and across the world), there are a large number of biofilters that have failed due to design or maintenance issues (Blecken et al. 2017; Al-Ameri et al. 2018; Beryani et al. 2021). Lucke & Nichols (2015) showed highly variable pollutant removal performance of 10-year-old field biofilters, and Beryani et al. (2021) showed that at least 40% of tested field biofilters in Sweden need significant maintenance. Maintenance of biofilters still remains a major challenge and obstacle for widespread biofilter implementation, and is generally considered unsatisfactory by experts (Hawken et al. 2021). Poor maintenance is usually attributed to asset owners and managers (mostly local governments) lacking financial and human capital to conduct rigorous assessments and repairs (Blecken et al. 2017; Hawken et al. 2021). Biofilters that are not maintained are more prone to failure by clogging or vegetation damage (die-off, weed ingress), both of which result in reduced pollutant removal performance (Payne et al. 2015; Blecken et al. 2017).

To address maintenance challenges and increase confidence in biofilters as stormwater management assets, this work proposes a new concept called zero additional maintenance (ZAM) biofilters. The concept of ZAM biofilters was first developed by Manningham Council, a local government area in Melbourne, Australia, and aims to ensure that maintenance implications for asset owners and managers are minimised and comparable to maintenance of common nature strips, i.e. the area of public land situated between a private property boundary and the road kerb, also known as a road verge. This would reduce the need for regular expert control, and better align ZAM biofilter maintenance with regular streetscape maintenance activities compared with traditional biofilter designs (Blecken et al. 2017), thus adding benefits of stormwater management to the urban streetscape without increasing the overall maintenance cost. To create ZAM biofilters, three common maintenance challenges need to be addressed: (1) surface clogging of filter media, (2) selection of low-maintenance plants, and (3) ZAM biofilter field design.

Surface clogging of biofilters from excessive stormwater sediment and debris deposition within biofilter media is a known challenge that interferes with hydraulic and pollutant removal processes (Le Coustumer et al. 2012; Kandra et al. 2014a; Beryani et al. 2021). Clogging occurs mostly in the top layer of biofilter media (top 100 mm, Kandra et al. (2014a)) and filter media configuration, particularly particle size, is an important factor that influences the rapidity and extent to which clogging occurs (Kandra et al. 2014b). However, it is still not clear what top layer biofilter media configuration would be optimal for delaying, and possibly even preventing, clogging. While studies have shown that plants could help with maintaining an appropriate infiltration rate for a longer period (because root growth and die-off helps to maintain porosity (Glaister et al. 2017)), it is not clear if the optimal media configuration determined in laboratory conditions will translate to zero maintenance in the field. Some studies have managed to confirm laboratory findings with field experiments (e.g. Pitt et al. 2021), however, more research is needed.

When considering ZAM biofilters, vegetation design for minimising maintenance is critical. Firstly, biofiltration plants need to seamlessly blend into the local streetscape so that biofilters do not stand out next to nature strip vegetation. Secondly, common plant types need to be used, with no skill requirement for maintenance, so that mowing and general maintenance can be undertaken by local residents, as is the current practice in Australia. While standard biofilter design guidelines typically recommend sedges (Payne et al. 2015), lawn grasses are the most common and simplest plant for asset managers and general community to maintain. Some biofiltration studies have included one or two lawn grass species in plant experiments as a point of comparison to other plants (e.g. Payne et al. 2018; Fowdar et al. 2022), however, there are no dedicated studies of different lawn grass species for their pollutant removal effectiveness in stormwater biofilters. Little is known about how different lawn grass species affect hydraulic and pollutant removal performance. Additionally, field validation is missing, especially with respect to the maintenance needs of such biofilters planted with lawn grasses (longevity, growth rate, robustness, etc.).

Appropriate selection of anti-clogging media is critical for biofilter longevity, but coarse sediment deposition on top of the biofilter (before it enters the media) is still a significant challenge for maintenance (Beryani et al. 2021). Passive sediment pre-removal should be incorporated in the ZAM design to protect the biofilter. Additionally, steep slopes and other safety features (e.g. curbs, fences, gates, etc. (Payne et al. 2015)) could be challenging for maintenance, so they should be minimised in ZAM design. This would help to blend biofilters into nature strips. Such exposed design is yet to be trialled in the field for its longevity and pollutant removal performance.

This study, to the best of the authors' knowledge, is the first study that aims to test the development and implementation of the novel ZAM stormwater biofilter concept through a comprehensive series of laboratory experiments and monitoring of field systems. By providing robust experimental design and critical field-scale evidence, this work contributes to stormwater biofilter development through the following research objectives: (1) understand optimal anti-clogging media configuration for ZAM biofilters; (2) identify vegetation that facilitates effective hydraulic and pollutant removal performance with no additional maintenance requirements for nature strips; and (3) validate stormwater ZAM biofilter nutrient and heavy metal pollutant removal performance of field-scale systems with novel elements (passive sedimentation and kerb design). This study further provides understanding of practical implications for the design and operation of ZAM biofilter.

This work validates the ZAM biofilter concept by investigating performance across three studies. Study 1 utilises a vegetated, laboratory-scale biofilter column experiment to explore how different filter media configurations affect physical clogging from suspended solids in stormwater. The outcomes of this experiment informed Study 2, a 12-month vegetated column experiment that explores the hydraulic and pollutant removal performance of six lawn grass species. Study 3 investigates the treatment performance of three field-scale ZAM biofilters whose design was based on the outcomes of Studies 1 and 2. Detailed methodologies for each study are provided below.

All columns were planted with lawn grass (Soft Leaf Buffalo). Prior to planting, the plant roots were washed to remove the soil in which the grass was grown, since this soil would have constrained the infiltration capacity of the system. It should be noted that, although it is proposed to incorporate a saturated zone in the field-scale prototypes (Zinger et al. 2021), only the upper, unsaturated layers of filter media were constructed for this study since the focus was on physical clogging. In total, 18 columns (6 designs × 3 replicates) were constructed.

The biofilter columns were allowed to stabilise for one week following construction. During this time, they were irrigated periodically with tap water. Following this establishment period, the columns were dosed with semi-synthetic stormwater using an accelerated dosing approach. Semi-synthetic stormwater was prepared by mixing a slurry of sediment collected from the sediment pond of a local constructed wetland with tap water in a tank to achieve a target total suspended solids (TSS) concentration of 150 mg/L typical for stormwater (Zhang et al. 2021a).

The aim of the stormwater dosing regime was to load the biofilter columns with the equivalent of 18 months of inflow over 15 days. To achieve this, the biofilter columns needed to be dosed with 16 L of stormwater per day (2 L was added at hourly intervals eight times per day). The volumetric inflow equivalency was calculated using typical biofilter design (2% of the catchment area, Payne et al. (2015)) and rainfall characteristics for Melbourne (effective annual rainfall = 550 mm).

To assess the development of clogging, outflow rates were manually measured three times a day. Inflow samples were also collected on a daily basis and analysed using a standard method (APHA/AWWA/WPCF 1998) to check the inflow sediment concentration. It should be noted that there were some differences in cumulative inflow volumes across column replicates because, as the columns began to clog, some did not fully drain between dosing intervals. When this occurred, the hourly dosing volume was the volume of water that filled the ponding space.

Six lawn grass species (Figure 1) were tested for their nutrient removal ability and hydraulic performance in a laboratory-scale column study. Their selection was based on their physical characteristics, as well as their sun and humidity tolerance, both of which were tested in a separate, preliminary field trial. Palmetto Soft Leaf Buffalo was found to be suitable for shady and moist sites while Kenda Kikuyu showed excellent survival in sunny sites (defined as <20% shade). Santa Ana Couch was also likely to be tolerant of sunny sites. Empire Zoysia showed very good survival but was expected to be better suited to sites with part shade. Nara Native Zoysia may be suitable for areas with high environmental values because it is less invasive than Couch and Kikuyu, and part shade. Kenda Kikuyu and Village Green Kikuyu offer good wear tolerance, while Nara Native Zoysia and Empire Zoysia are suitable for sites with low-to-medium pedestrian traffic.

Following construction, plants were allowed to establish 6 weeks. For the first 2 weeks, the grasses were watered with approximately 2 L of tap water three times per week. The watering frequency was then reduced to twice per week. Lawn grasses were mowed as required during the trial period.

In early June 2017, dosing of the columns with 9.4 L of semi-synthetic stormwater twice weekly commenced. This dosing regime was based on an annual average effective rainfall of 540 mm for Melbourne and using a biofilter sized to 2.5% of its contributing catchment. Use of semi-synthetic stormwater allowed us to minimise variations in inflow concentration while maintaining realistic composition. It contained sediment from a local stormwater wetland, sieved to 1 mm and mixed with dechlorinated tap water to achieve the target TSS concentration. Laboratory chemicals (potassium nitrate, ammonium chloride, nicotinic acid, potassium sulphate, and sodium thiosulphate) were added to match any deficit in targeted nutrient concentrations, as detailed in Bratieres et al. (2008) and Zinger et al. (2021). Target nutrient concentrations are shown in Supplementary Material, Table S2.

Over the 10-month study period (12 months from planting), both wet and dry weather conditions were simulated, with 8 months of wet conditions, when columns were dosed twice weekly with semi-synthetic stormwater, and 2 months of dry conditions, when columns were dosed once per fortnight. A total of six water quality sampling runs were conducted (four during wet conditions and two after 1 month of dry conditions), along with three infiltration tests during wet events (experimental timeline in Supplementary Material, Figure S1). To assess the infiltration rate of each column, the drop in ponded water level (i.e. the water level above the filter media surface) was measured every 60 s for at least 15 min, depending on the rate of drainage of the column. The infiltration rate was calculated as the average decrease in water level over the measurement time. During each sampling run, inflow and outflow water samples were collected for pollutant analysis. A composite outflow sample was taken after the column finished draining. This was sub-sampled into a 1 L bottle. All samples were analysed in a NATA (National Association of Testing Authorities, Australia) accredited laboratory according to standard methods (APHA/AWWA/WPCF 1998). All sampling runs were analysed for Total Nitrogen (TN), ammonia (NH₃), and oxidised nitrogen (NO_x). Total Dissolved Nitrogen (TDN) was measured for the first four sampling runs and Dissolved Organic Nitrogen (DON) was calculated as the difference between TDN and NH₃ + NO_x. In the Week 26 (dry) and Week 30 (wet) sampling runs, water quality samples were also analysed for Total Phosphorus (TP), Total Dissolved Phosphorus (TDP), and Filterable Reactive Phosphorus (FRP, a measure of orthophosphate, ⁠). The difference between TP and TDP is a measure of particulate phosphorus.

To reduce the amount of sediment deposits on top of the biofilter and thus delay the onset of surface clogging, sedimentation grooves were installed in the concrete channel upstream of each system (Figure 3(b)). These sediment grooves were designed to be cleaned by street sweepers every 5–6 weeks as part of Manningham Council's regular sweeping programme. Square bars were installed in the inlet with 18–20 mm gaps between bars to prevent large litter (>20 mm in size) from entering the ZAM biofilter and allow collection by a street sweeper or disposal into a downstream side-entry pit (Figure 3(c)).

Beginning in September 2017, the pollutant removal performance was evaluated at each site using simulated inflow events. Each system was dosed with 500 L of semi-synthetic stormwater per event, which is equal to two pore volumes (PVs) of the filter media. Semi-synthetic stormwater was prepared following the method described for Study 2, with the addition of heavy metal dosing (see Supplementary Material, Table S2 for further details).

Five inflow events were simulated at each site over a period of 8 months (in September and December 2017, and January, February, and April 2018). The first sampling point was upstream of the sediment grooves, to characterise the untreated stormwater. The second one was between the sediment grooves and biofilter inlet, to assess the removal efficacy of the sediment grooves. The last sampling location was the biofilter's outflow pipe inside the side-entry pit (Figure 3(d)). Complete outflow volume was measured at the outflow to understand volumetric reduction within the systems. Discrete samples were collected from the outlet at 50 L intervals, then combined to create a composite sample. Three composite water samples were collected into 1 L bottles during each event. All samples were analysed for copper (Cu), zinc (Zn), TSS, TN, and TP by a NATA (National Association of Testing Authorities, Australia) accredited laboratory according to standard methods. In addition, two filter media samples (surface and subsurface) and plant leaf and root samples were collected during each event at all three sites and analysed for Cu and Zn.

Box plots were used to present metals concentrations and removal in the water quality samples, filter media, and the different lawn parts. Shapiro–Wilk W test was performed to check the normality of data. Univariate analysis of variance (ANOVA) test was used to assess the statistical significance of differences between inflow and outflow metal concentrations as well as metal concentration in the filter media and plant parts. Welch and Brown-Forsythe were used when the homogeneity was rejected to be confident that the data were statistically significant. Log-transformed data was used to perform the analysis when the data were not normally distributed. Statistical significance was accepted when the p-value was <0.05. Box plots and statistical tests were produced using IBM SPSS Statistics (Version 25).

Generally, all designs containing a protective layer delayed the onset of clogging compared with the design with no protective layer. The extent to which clogging was delayed appeared to be partially determined by particle size distribution; designs whose protective layers were comprised of a smaller mean particle size had higher outflow rates, perhaps because they were better able to trap and distribute incoming sediment across the full depth of the protective layer (Figure 4). The average infiltration rates for Configurations 1 (no protective layer), 3 (2 mm + 1 mm mixed), 4 (1 mm + 0.5 mm mixed), and 6 (Daisy's Coarse White Sand) dropped below 100 mm/h after 120–130 L of stormwater application (equivalent to a cumulative hydraulic loading of ∼10 m). This is consistent with the findings of Kandra et al. (2014b) who reported that biofilters tested under an accelerated loading regime, like in this work, experienced a rapid drop in the infiltration rate across the first 10 m of water applied, after which it stabilised. Configurations 2 (2 + 1 mm separate layers) and 5 (Burdetts 20/30 sand) showed somewhat higher resistance to clogging, especially across one outlier replicate in each configuration (Figure 4). This suggests that layers of coarser filter media at the top of the biofilter can help with surface clogging.

In addition to performance, selection of the protective layer configuration needs to consider complexity of construction, cost, and capacity to support healthy plant growth as well as ability to delay the onset of clogging. A single-layer protective system is far more practical to build than a mixed layer system. The cost of the commercially available Burdetts and Daisy's sand products is approximately 20% of the engineered sands, which is significant when considering the scale at which biofilters are built. Finally, the use of a coarser material for the top 100–150 mm of the media profile is unlikely to impact plant growth as even the root systems of shallow-rooted plants, such as lawn grasses, penetrate well beyond this depth. This was also observed in our experiment, where grasses on all configurations remained unaffected. In a separate laboratory study of plant species in biofiltration systems, it was found that 95% of the root system of Soft Leaf Buffalo occupied the entire unsaturated zone (300 mm deep) and some roots were sufficiently long to penetrate well into the saturated zone (>300 mm deep, Payne et al. 2013). Considering these practical criteria, along with sediment removal performance, it is recommended that Configuration 5 (Burdetts 20/30 sand) be used as the protective layer for future ZAM biofilter design.

The average infiltration rates were statistically similar across grass species (p > 0.05) after 15, 21, and 28 weeks of planting, respectively. Interestingly, particularly in November 2017, planted columns had significantly higher infiltration rates than unplanted columns (Figure 5). Vegetation indeed loosens up filter media with roots. Santa Ana Couch and Kenda Kikuyu had the least variation in infiltration rates across replicates while Empire Zoysia was the greatest (Figure 5).

It should be noted the high infiltration rates observed in this trial may not be reflective of field conditions because the biofilter columns were not subjected to vehicular and/or human traffic or litter accumulation. Nevertheless, these results give a clear comparison of the hydraulic behaviour of different lawn grasses in the field.

The two dry weather periods, one imposed during winter and the other in summer, had different effects on nitrogen removal. Week 26 sampling occurred after 4 weeks of dry conditions, and lower TN removal was observed for all grass species except Santa Ana Couch and Empire Zoysia, with Nara Native Zoysia and unvegetated columns leaching TN. Reduced nitrogen removal performance is typical after dry weather spells and is attributed to root die-off, leaching from filter media, and reduced microbial activity as a result of desiccation (Zinger et al. 2021). It is likely that Nara Native Zoysia had not well established at that time (it went into a state of dormancy during winter) which resulted in a decrease in its performance. Interestingly, the two-week drying period in summer had no significant effect on nitrogen removal (during sampling events at weeks 45 and 51 (Figure 6)). It is likely that the more mature plants were less impacted by this dry period because it was shorter than the winter dry period and also because they were able to draw water from the saturated zone and thus sustain nitrogen removal performance. A similar is seen in Zinger et al. (2021) where 2 weeks of drying had no significant effect on biofilter columns with a saturated zone.

Effluent NH₃ concentrations were low in all cases (Figure 6). NH₃ is mainly removed through adsorption and microbial processing via nitrification which occurred effectively (Zhang et al. 2021b). There was a consistent net reduction in DON concentrations across by all column configurations, both vegetated and unvegetated. In contrast, effluent NO_x concentrations varied across the study period, ranging from leaching to effective removal. Where leaching occurred, it is possible that this was due to net production via nitrification (the microbial conversion of NH₃ to NO_x) and insufficient plant uptake. An overall decrease in effluent NO_x concentration was observed with time, matching plant growth and development of root systems. In fact, TN removal was influenced by the extent of NO_x removal and/or production.

There was a significant difference in TN removal and effluent NO_x concentrations across grass type (p < 0.001) in the initial four sampling runs, with some species performing better than others (Figure 6). This difference could be attributed to the growth pattern of the different lawn grasses. For instance, some species were dormant during winter while others experienced active growth (particularly the two Kikuyu species), leading to higher initial removal performance. Previous studies have found that plant uptake plays a key role in nitrogen removal by stormwater biofilters (Payne et al. 2014b). The growth rate of the lawn grasses is likely another factor influencing nitrogen removal. Nara Native Zoysia is known to be slow growing which explains its lower performance in the first half of the experiment. It also becomes dormant during drought, which explains the poor performance during the first dry period (week 26). Interestingly, after establishment and under more mature state, the difference in performance across lawn grasses became insignificant.

The results indicate that all tested lawn grasses were able to achieve a TN removal efficiency of up to 70–80%, which meets the 45% reduction recommended by Australian best practice guidelines for stormwater management. Species such as Santa Ana Couch and Kenda Kikuyu have shown to be robust, high performers throughout the experimental period.

Design choice of lawn grass species will depend on several criteria, including pollutant removal performance, maturation speed, site conditions, clogging risk, and aesthetics. Because of their faster growth rate and sustained growth during winter, Kenda Kikuyu and Village Green Kikuyu were the best performing lawn species during the initial trial period, and optimal species if fast establishment is required. At maturity, all species were universally effective at pollutant removal. However, it appears that lawn grasses may be more susceptible to seasonal variation in comparison with other species (e.g. native shrubs, sedge, or ornamental plants) (Fowdar et al. 2017a). For example, Nara Native Zoysia, Santa Ana Couch, and Empire Zoysia may go dormant during winter, which might cause some performance issues, particularly in winter rainfall-dominated climates.

Clogging was not found to be an issue, hence the choice of any of the tested lawn species would be suitable for ZAM biofilter design in that respect, although it is acknowledged that the 12-month monitoring period represents only a small proportion of the expected lifespan of stormwater biofilters. However, it is evident that lawn species with poor survivability would result in poor nutrient removal, and particularly nitrogen. Lawn species that are healthy will provide the greatest treatment benefit. Always choose lawn species that will grow well in the particular location (although slow growing species can be expected to initially render lower nutrient removal) for effective nutrient removal. For example, it is likely that installing Kenda Kikuyu in a shady location (which prefers sunny conditions) will facilitate lower removal rates than Palmetto Soft Leaf Buffalo, which is more shade tolerant. From this study, while it can be speculated that nutrient removal of the lawn grasses is a function of grass health and growth rate, plant growth and vegetation mass changes were only visually observed during this study (Supplementary Material, Figure S4). A summary of lawn grass performance and characteristics for consideration in field application is provided in Supplementary Material, Table S4.

In contrast, removal of TN was highly variable, in the range of 28–64% at both Hummel Way sites. Effluent concentrations of TN from the Edwin Road biofilter (ER) were very high and leaching was observed in some cases, resulting in an average removal of 11%. The low removal of TN at the ER biofilter was likely due to the absence of a saturated zone, which is known to sustain denitrification in stormwater biofilters (Igielski et al. 2019) and promote plant removal of NO_x (Payne et al. 2014a). In addition, dense patches of clover were observed at the ER biofilter, which had invaded the raingarden from the surrounding area. Clover is a known nitrogen-fixing plant (Bolger et al. 1995) and is thus an undesirable plant species for biofiltration because the potential addition of nitrogen to the system via fixation of atmospheric nitrogen very likely counteracts nitrogen removal from stormwater. The higher TN concentrations in the HW2 effluent compared with HW1 may also be attributed to the presence of clover in this system. Despite the poor reduction in TN concentrations, the ER system still demonstrated reliable removal of other pollutants (Figure 8).

The presented data in Figure 8 showed that, while statistically not significant, TSS, TP, and Zn pollutant concentrations in the influent (after sediment grooves) were somewhat lower, compared with concentrations in the tank (before sediment grooves). This indicates minor accumulation of the suspended particles inside sedimentation grooves. For the grooves to be effective, either flow velocities in the gutter should be kept low to avoid scouring the grooves, or the sediment grooves need more frequent cleaning, which would increase maintenance costs. For instance, the TSS reduced by 30% in one simulated inflow event at ER, when the grooves were clean, but in another inflow event at the same site, the TSS concentration increased from 120 mg/L before the sediment grooves to 130 mg/L after the grooves, indicating that collected sediment was scoured from the grooves and washed into the biofilter.

The median stormwater runoff volume reduction in HW1, HW2, and ER biofilters were 60, 53, and 78%, respectively. Reductions in runoff volumes were highest at ER (83% on average), due to the absence of an impervious liner, and this reduction was enough to counter TN leaching. The volumetric reductions observed in this study were within the range of a previous field study which reported that the monitored bioretention basin retained 25–89% of the inflow volume (Mangangka et al. 2015). Similarly, the effluent volume reduction was 33–84% from five lined bioretention systems (Lucke & Nichols 2015). The high volumetric reduction observed in the present study, combined with pollutant concentration reductions, translated to high pollutant load reductions (Table 2). The performance of these systems thus complied with the local stormwater treatment targets of 80, 45, and 45% reductions in loads of TSS, TN, and TP, respectively (EPA Victoria 2021). Even the ER system, where some leaching of TN and the lowest mean concentration reduction were observed, still delivered a high TN load reduction and reliable performance because the system was able to retain a high quantity of water.

The results indicate that Palmetto Soft Leaf Buffalo lawn was able to prevent the accumulation of metals in biofilters. Similar to the study by Yeh et al. (2009), metal concentrations were higher in the Palmetto Soft Leaf Buffalo roots compared with shoots at all sites. This is not surprising because plant roots are in direct contact with the filter media and the point at which metals enter the plants. However, this correlation was only significant for Zn concentrations at ER (p = 0.02; Figure 9), while (p > 0.12) in the other cases. Plants tissues usually have different growth rates and properties that affect nutrient concentrations, therefore, there was no clear correlation between metal concentrations in the leaves and roots. Visual observation of the Palmetto Soft Leaf Buffalo grass during the study indicated that it was well suited to the site conditions. The grass was green, healthy, and completely covered the filter media. However, having a ZAM biofilter incorporated within a nature strip and exposed to pedestrian traffic could impact system performance and longevity due to compaction of the filter media compaction and vegetation damage. While this issue was not observed during this study, perhaps because pedestrian traffic was light, this should be assessed, particularly for systems installed in locations with higher pedestrian traffic loads. There was no surface clogging observed over the 8-month monitoring period, proving the effectiveness of ZAM biofilter design. However, it is acknowledged that these were quite new systems, only about 3 years old when the study was conducted. Therefore, it is possible that the continuous accumulation of sediments on the surface layer of the systems with time might reduce the infiltration rate.

This study tested the development and implementation of novel concept called ZAM biofilters for stormwater management through a series of laboratory experiments and monitoring of field systems that examined the performance of filter media configurations designed to protect against clogging, plant choice, and field operation of ZAM systems. The results showed that the protective media layer delayed the onset of clogging compared with the design with no protective layer. The extent to which clogging was delayed appeared to be partially determined by particle size distribution favouring designs whose protective layers were comprised of a smaller mean particle size. However, the recommended protective layer configuration also considered cost, availability, and construction complexity.

The choice of easy-to-maintain lawn grasses for ZAM biofilter revealed that biofilters planted with lawn grass species, including Kenda Kikuyu, Empire Zoysia, Santa Ana Couch, Village Green Kikuyu, and Palmetto Soft Leaf Buffalo, could be effective for reducing nitrogen concentrations from stormwater. Some species, such as Kenda Kikuyu and Santa Ana Couch, established faster than others and showed less variable performance during dry conditions. If the preferences of the chosen lawn grass matched the site conditions, lawn grasses can meet local regulatory requirements and best practice standards for stormwater management.

The field study of three ZAM biofilters confirmed their high nutrient and heavy metal removal from stormwater. Designs with a submerged zone had higher TN removal, however, all designs achieved significantly higher pollutant load reductions than required by guidelines for stormwater treatment targets. Surface accumulation of sediments and heavy metals was observed, however no clogging was present after 3 years of system operation. The Palmetto Soft Leaf Buffalo grass used in the system was performed well, both visually and in terms of pollution removal, however, sediment grooves in the kerb upstream of the systems delivered only minor pre-treatment benefits. Overall, the findings of these three studies confirm the potential for well-designed ZAM biofilters to achieve stormwater management requirements with no additional maintenance requirements compared with standard streetscape landscaping. It is recommended that future monitoring of infiltration rates and lawn grass health over the long term be undertaken to provide further field-scale validation of this technology.

This project was funded by the Living Rivers funding grant awarded by Melbourne Water and Manningham City Council, and Cooperative Research Centre (CRC) for Water Sensitive Cities (Project C4). We acknowledge the significant technical contribution to the project made by Simon Brink (Manningham Council) as well as other Council employees who designed, constructed, and maintained ZAM biofilters. Emily Payne is acknowledged for her help during the development of the parts of project proposal and technical input. We thank Feiran Huang and personnel from Monash University's EPHM lab who contributed to the laboratory and field work.

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

The authors declare there is no conflict.