Full-scale characterization of the effects of a bioretention system on water quality and quantity following the replacement of a mixed stormwater and combined sewer system

In urban areas, impervious surfaces create a greater volume of runoff and, combined with anthropogenic activities, increase the environmental vulnerability of communities (Paul & Meyer 2001; Erickson et al. 2013). Combined sewer systems, draining both wastewater and stormwater, have limited capacity. During heavy rainfall or snowmelt, excess water overflows and discharges untreated runoff into receiving waters, causing significant environmental impacts, including harm to aquatic species (Autixier 2012). Studies have shown that even small increases in the urban impervious surface area can lead to substantial ecological changes in watersheds. For example, intolerant bacterial taxa tend to disappear when more than 12% of a watershed is urbanized (Simonin et al. 2019), while an impervious surface threshold as low as 2% has been found to affect aquatic macroinvertebrate populations in streams (King & Baker 2010).

Climate change is multiplying extreme weather events. In northern regions, the intensity of climate change is and will continue to be high. Across Quebec, predictions suggest that the frequency of heavy rain days will increase, alongside modifications in precipitation characteristics such as duration, seasonality, frequency, the magnitude of extremes, and the degree of interannual variability (Leveque et al. 2021). This puts additional pressure on urban sewer systems, making it essential to reduce stormwater volumes and protect aquatic ecosystems.

A growing awareness of urban water issues has led to the gradual introduction of various sustainable stormwater management practices (Field & Tafuri 2006). These practices aim to control problems associated with runoff volumes and quality, control peak flows to limit the erosion of receiving watercourses, and manage groundwater recharge (Ville de Trois-Rivières 2013). Terminologies like blue-green infrastructure (BGI), best management practices, and sustainable urban drainage systems (SUDS) have emerged to facilitate communication among researchers globally (Fletcher et al. 2015).

A bioretention cell is a vertical-flow SUDS that consists of a shallow depression filled with selected substrate and vegetation adapted to local climatic conditions, maximizing the peak runoff control, detention, treatment, and infiltration of surface water runoff (Hunt et al. 2012). In addition to managing surface water runoff, bioretention cells provide aesthetic and safety benefits (calmer traffic, safer intersections, and wider sidewalks). Although few studies have directly measured the impacts of green stormwater infrastructure on active mobility, bioretention cells are one of the few for which outcomes have specifically been measured (Lemieux et al. 2023). The integration of bioretention cells has also been associated with the creation of more aesthetically pleasing environments, with more greenery. A 2019 study conducted in the United States surveyed 497 laypersons and 117 designers, highlighting a distinct preference for landscapes featuring green stormwater infrastructure (Suppakittpaisarn et al. 2019). However, it also identified how the messiness of green stormwater infrastructure directly impacted the preference. Furthermore, this pleasant environment can be directly linked to the reduction of thermic islands (Santamouris 2014), which have also been shown to degrade distributed drinking water quality (Absalan et al. 2024).

Bioretention systems are relatively recent BGI, with the first examples dating back to the 1990s in Prince George County, Maryland (Roy-Poirier et al. 2010). These systems have been the subject of several studies; however, their performance in cold climates remains poorly understood due to the radically different conditions they experience during the winter and summer seasons. Notably, it has been demonstrated that road maintenance salts have both negative effects (reduction of heavy metal uptake by plants) and positive impacts on contaminant removal (e.g. total suspended solid removal) (Géhéniau et al. 2014). However, their effects on vegetation and microbial communities have yet to be fully understood (Kratky et al. 2017). In addition, the impact of large-scale implementation of bioretention cells on sewershed hydrology has mostly been explored via modeling simulations (Meng et al. 2014; Wang et al. 2019; Gougeon et al. 2023). There is a need to investigate the full-scale effects of these BGI on the reduction of stormwater flows and water quality, especially in cold climates.

One major concern is the potential release of heavy metals into groundwater from the runoff managed in the bioretention cells or other stormwater infiltration solutions. It is especially important to protect groundwater when it is vulnerable to potential contamination coming with stormwater infiltration. In general, stormwater infiltration is generally not allowed in areas with high groundwater levels or extreme soil infiltration rates, as these conditions can either prevent filtration or lead to clogging. In Trois-Rivières, the local aquifer is used for drinking water, but the bioretention systems on Saint-Maurice Street pose no direct risk, as they are outside the city's intake zones. However, with the global push to implement SUDS in response to climate change, it is important to assess their effects on groundwater recharge, sewer overflow reduction, and unintended groundwater contamination. Bioretention systems generally achieve over 95% removal of total metals. However, the removal efficiency for dissolved metals can vary (Kratky et al. 2017). Several studies have focused specifically on the release of metals by bioretention systems (Muthanna et al. 2007; Paus et al. 2014; Søberg et al. 2017). In some cases, such as Trois-Rivières, heavy metal concerns arise not from stormwater runoff but from naturally high concentrations in the subsoil, particularly manganese. Due to the health risks for bottle-fed infants, Quebec will enforce a maximum allowable concentration of 0.12 mgMn/L for manganese in drinking water by June 2024 (Government of Quebec 2023), in addition to the previously established aesthetic objective of 0.02 mgMn/L. Given its relevance to drinking water, there is a need to further study manganese in relation to other stormwater contaminants in bioretention cell systems. Manganese in groundwater is not a problem unique to Quebec, and to our knowledge, no study has specifically examined manganese and the potential interactions between dissolved organic matter (potentially coming from SUDS) and subsoil manganese.

The goal of this project was to select a site for a full-scale study of the impact of bioretention cells as an approach to climate change adaptation, considering their potential impacts on hydrology and water quality. This study sought to demonstrate the extent to which the implementation of bioretention cells reduces sewer flow and improves (or impairs) the quality of runoff and drainage water, with a focus on manganese. The specific objectives were to: (1) install a monitoring system to measure flowrates into and out of the cells, (2) model the bioretention cells to assess their effectiveness in reducing overall runoff volumes and peak flows at a local and sewershed scale, (3) measure the removal of contaminants in runoff through the bioretention cells, (4) assess the impact of bioretention cells on groundwater quality, including iron and manganese, that are relevant for groundwater sources of drinking water, and (5) provide recommendations for the full-scale implementation of bioretention cells in cold climates considering their impact on sewershed-scale hydrology.

The chosen study site was in the City of Trois-Rivières, Québec, Canada. A wide residential street requiring new water and sewer infrastructure was identified for the construction of a series of bioretention cells to assess their performance for the improvement of stormwater quality and the reduction of stormflows. The site was selected because of its width, providing sufficient space, and the combined sewers needed replacement and were separated into storm and sanitary sewers. Given that the city uses both surface water and groundwater as drinking water supply sources, a site was chosen outside the capture zone of their wells to assess their potential impact, should this approach be considered to improve groundwater quantity.

In recent years, the City of Trois-Rivières, located at the mouth of the Saint-Maurice River in Québec, has faced a series of extreme weather events causing power outages, sewage overflows, and costly infrastructure degradation. Recognizing the need for climate change adaptation, the city developed a climate adaptation plan in 2013 (Ville de Trois-Rivières 2013), highlighting risks to its infrastructure. To reduce these vulnerabilities, the municipality aims to implement sustainable urban planning. In 2015, the city launched the Grand Projet de la Rue Saint-Maurice, involving the installation of 54 bioretention cells along one of its predominantly residential streets. Although water quality data are available for a subset of the bioretention cells and a parallel project examined plant growth (Dagenais et al. 2018; Beral et al. 2023a, b), one representative cell was studied in greater detail: the bioretention cell n°4 (BR4) is located at 425 St Maurice Street, Trois-Rivières, QC G8V 1G7, Canada. The aim was to assess its performance at a local scale with sewer monitoring, and the extrapolation of results was made through modeling and simulation.

The SWMM was segmented into 88 sub-basins, with an average surface area of 2.05 ha. This division was carried out by engineers from the City of Trois-Rivières. The model consists of 68 connection points (nodes) and 72 sections (links). The nodes represent the sewer basins in the area, while the links correspond to the pipes. The characteristics of these elements, such as pipe lengths, diameters and maximum depth of nodes, were established by the City of Trois-Rivières. An initial modeling phase was carried out for all these sub-basins of a total area of 111.7 ha. In a second phase, adjustments were made to the model to create a specific local version, encompassing Saint-Maurice Street exclusively. This local system considers only those sub-basins directly connected to Saint-Maurice Street, which represents an area of 16.6 ha.

For these simulations, rain events were selected between May and December 2020 based on rainfall data obtained from the nearest meteorological station operated by the City of Trois-Rivières. The selection of events was based on the duration and intensity of rainfall. These events were thunderstorms, representing the longest and most intense occurrences during this 2020 period. Four rainfall events from 2020 were chosen for the present study: the first event (July 19, 2020) was short and intense; the second (July 27, 2020) and the third (August 4, 2020) were, respectively, of medium and long duration, while the last event (August 11, 2020) was a short thunderstorm. Precipitation depths ranged from 13.2 to 91.3 mm, corresponding to return periods of less than 1 in 2 years to 1 in 20 years (Audet et al. 2012).

The bioretention cells were added to the SWMM using the software's ‘Low Impact Development’ (LID) feature, which proposes various types of green infrastructure to be included in the modeling. Four different simulations on separate rainfall events were carried out to calibrate the LID parameters. The parameters tested were determined from the values suggested in the SWMM manual as well as the scientific literature (Autixier et al. 2014a, b; Joshi et al. 2021; Wang et al. 2021). To assess the suitability of this calibration, simulated data were compared with measurements taken using a flow probe installed in 2019 by the City of Trois-Rivières in a sewer pipe downstream of Saint-Maurice Street. To evaluate this calibration, the normalized root mean square error method was used. Further details are provided in the referenced work (Bouattour 2021).

The installation of the monitoring instrumentation to know both inflow and outflow rates was completed on March 24th, 2021. Both the inlet and outlet of the BR4 were equipped with ‘V-Notch’ weir boxes. The boxes were made of stainless steel with an angle of 45° for the inlet, while, in ‘HDPE’ polymer and an angle of 24°, they were used at the outlet. In the sedimentation pit, the water levels were measured every minute by a Solinst Levelogger Edge M10 water pressure sensor adjusted for atmospheric pressure by a Solinst Barologger Edge atmospheric pressure sensor (Solinst 2023a, b). To enable water to reach the bioretention system's area, the weir box must fill up to the level of the pipe conveying the water to bioretention. Furthermore, the inflow basin also has an overflow system connected to the sewer. This instrumentation system was calibrated at the Hydraulic Laboratory of Polytechnique prior to installation (Doucet 2022). In the outflow well, the water level in the ‘V-Notch’ weir was measured every minute using an ultrasonic transducer Siemens XPS-5 connected to an ultrasonic controller Siemens Sistrans LUT400. To mitigate turbulence or water movement that might disrupt level readings, the outlet box was divided into two compartments, connected by an opening at the bottom of the separator. Since the ‘V-Notch’ weir was regularly submerged, the water level of the entire well was also measured using a water pressure probe, which was adjusted for atmospheric pressure with a Solinst Levelogger Edge M10 and a Solinst Barologger Edge sensor (Solinst 2023a, b). The setup was calibrated at the CREDEAU Laboratory of Polytechnique for both situations of free flow and submergence of the weir box (Doucet 2022).

The groundwater level was continuously measured in three piezometers (P-1, P-3, and P-4, Figure 1) using Solinst Levelogger probes. Water table level values were adjusted using the atmospheric pressure value measured in the outlet well. Due to technical issues related to the memory capacity limitations of the equipment used, data were not collected continuously throughout the measurement campaign. The measurements taken are organized into a series of data at different intervals between 2021 and 2022. Since the Piezometer P-1 showed strong agreement with the initial expectations and corroborated other observations made regarding the outflow rate, it was selected as the reference for calibrating the data from Piezometers P-3 and P-4. The calibration process began by calculating the initial differences (deltas) between P-1 and the P-3/P-4 data series from their first measurements. Specifically, the deltas were determined by comparing P-1's measurements to those of P-3 and P-4 at the same time points. These differences were then subtracted from the respective measurements of P-3 and P-4 data, effectively adjusting their initial data to match the reference regarding P-1. This adjustment ensured that any discrepancies between P-1 and the other piezometers were corrected, thereby standardizing the data across all instruments.

Four complete sampling events at the inlet and outlet of the bioretention in 2021 and one sampling in 2022 were conducted. The details of each sampled rainfall event are provided in Table 1. Other attempts were made during the study period, but they proved unsuccessful due to a too-small quantity of water entering the catch basin ahead of the entrance of the bioretention. On occasion, debris at the catch basin's entrance resulted in stormwater bypassing the catch basin's entrance. In these cases, the water did not reach the level of the weir box and did not flow into the bioretention cells.

The method used to sample water quality at the inlet of the bioretention cell was based on the automated collection of sequential grab samples (Ma et al. 2009). At the beginning of the event, discrete samples were taken every 5 min, and this interval gradually increased to 1 h between samplings. This approach estimates an event-averaged concentration without the need for an automatic flow-weighted sampler. The same method was used at the outlet, with the only difference being that they were done over 24 h or more. The sampling interval can be defined as ‘first-flush enhanced’, meaning that it is configured to collect discrete samples every 30 min during the first half of the sampling period and then every hour for the second half of the 24-h duration. This allowed for the collection of rainfall samples at the outlet throughout the entire drainage period of the bioretention system, which had a longer duration than water entering the bioretention cell.

Regarding groundwater, discrete samples were collected from the three piezometers on multiple occasions during the measurement campaign. In total, five sampling campaigns were conducted over the study period, with two sampling events corresponding to that of the inlet and outlet sampling. The groundwater sampling methodology follows the Guide d'échantillonnage à des fins d'analyses environnementales (Sampling Guide for Environmental Analysis) (MDDEP 2011). Prior to each sampling, piezometers were purged to obtain representative groundwater samples. The piezometers were purged slowly until stabilization of pH, dissolved O₂, electrical conductivity, and temperature to avoid disruption of the water in the column that could lead to errors in the data. A peristaltic pump was used to extract water from the piezometers. However, sampling in winter proved to be challenging due to frozen water, resulting in a limited number of winter samples.

The analysis of all the samples collected was divided between the laboratories at Polytechnique Montreal and the Interuniversity Research Group in Limnology and Aquatic Environment (GRIL) at the University of Montreal. The samples were transported and kept on ice. The following analyses were performed: ammonium (⁠⁠), dissolved organic carbon (COD), nitrate (NO₃⁻), total nitrogen (TN), total phosphorus (TP), phosphate (⁠⁠), sodium (Na⁺), chloride (Cl⁻), magnesium (Mg²⁺), calcium (Ca²⁺), potassium (K⁺), chromium (Cr), manganese (Mn), iron (Fe), nickel (Ni), copper (Cu), zinc (Zn), cadmium (Cd), and lead (Pb).

The results of the groundwater samples obtained from the three piezometers were averaged for each sampled event, resulting in a single average value for the groundwater during each event. Since the piezometers are located at distinct points within BR4 (at the center and both ends), the average of the concentrations measured at these three locations was considered an adequate representation of the entire section of the groundwater that interacts with the bioretention cell.

While previous research has largely relied on modeling simulations to assess the impact of bioretention cells on sewershed hydrology (see the Introduction section), our study offers a fresh perspective by directly measuring the accumulation and release of contaminants in bioretention cells. Our objective is to evaluate the real-world effectiveness of bioretention systems in reducing contaminants at the scale of an urban watershed, with a specific focus on their performance in a cold Quebec climate. This approach is particularly timely, given the evolving nature of this field and the scarcity of real-world data on bioretention systems in such climates. By examining the adaptation of bioretention systems in Trois-Rivières, our research has provided valuable insights into their practical application and performance in managing stormwater runoff in cold-weather conditions.

Indeed, at the large scale of the drainage basin, the implementation of bioretention cells does not appear to have any noticeable effectiveness. A comparison of the simulations with and without bioretention cells shows runoff volume reduction percentages of less than 1% for the four simulated events. Similar results were observed by Chen et al. (2019) in the Darst Sewershed area in Peoria, Illinois, where runoff volume reduction ranged from 0.18 to 2.8%. This finding can be explained by the limited area occupied by bioretention cells in relation to the drainage basin's size (1.9 versus 111.7 ha, respectively). In fact, there is a relationship of proportionality between bioretention cell efficiency and the percentage of the surface area covered by the cells (Autixier 2012; Autixier et al. 2014a). In the current study's case, the 54 cells along Saint-Maurice Street are not sufficient to generate enough impact on the entire area examined. Furthermore, a small creek has been channeled into the stormwater drainage network and provides an important constant flow.

The objectives of climate change adaptation measures generally involve investigating ways to mitigate surge events. In this context, several studies have shown that bioretention cells are capable of reducing the number, frequency, and volume of sewer overflows and thus contribute to reducing their impacts on receiving environments (Autixier et al. 2014a). The design of resilient systems for urban hydrology under climate change with an anticipated increase of intensity and duration of precipitation events will require the characterization of the distribution of future flowrates and an investigation of peak flow and volume reductions as a function of bioretention implementation scenarios.

In our study, the sewer network in this area was a separate sewer system prior to the construction of the project (apart from a 300 m section that was combined). The City of Trois-Rivières opted to overhaul the entire sewer system in the area into a separate system, meaning these flowrates no longer contribute to combined sewer overflows. However, modeling results suggest that the addition of bioretention cells, without separating the sewers and removing the flow of the sewered creek, would not be sufficient to reduce overflows without a large increase in the surface area of bioretention cells throughout the drainage basin. This insight highlights the necessity of comprehensive multicriteria decision analyses to guide urban planning and policy decisions. By comparing the costs, benefits and co-benefits of bioretention cells with those of the other combined sewer overflow reduction strategies, stakeholders can gain valuable insights into the most effective and economically viable solutions.

A detailed economic analysis can reveal the relative cost-effectiveness of implementing a fully separated sewer system versus expanding the bioretention cell network. Such comparisons are essential to consider as many combined sewers have not reached the end of their useful life, and green stormwater infrastructure potentially offers a solution for preventing urban flooding at a local level. Furthermore, as cities integrate life-cycle analyses into their infrastructure decision-making to mitigate overall emissions, data are needed on how much peak flow and contaminant release can be achieved with the addition of green stormwater infrastructure. Detailed modeling of the evolution of bioretention cell performance over time, considering factors such as clogging and vegetation growth, could also provide valuable data concerning the limits of bioretention cells in cold climates and their long-term effectiveness.

Depth data identified two intervals during which there were no interactions between the groundwater and the filtered water: June 23 to June 26, 2021 and August 04 to October 09, 2021. The observed continuous flow in the bioretention system's outlet well, even when the water table is assumed to be below the drains, suggests that the water table may have been above the drains during certain intervals that were previously identified as having no interactions between the groundwater and the filtered water. Such intervals were identified twice: the first extended from June 23 to June 26, 2021, and the second covered a few hours on August 04, 2021.

Regarding the first interval, the water table depth varied between 1.39 and 1.43 m. Since the drains have a depth of 1.40 m, the water table was 3 cm below the drains at its lowest level. However, given the challenges encountered during data processing, this difference was considered negligible compared to errors related to data calibration. As for the second interval, the groundwater depth curves indicated that during August 2021, the groundwater level was decreasing. However, this second interval corresponded to the period around which the water table would have passed below the drains. Uncertainty remains about the exact moment when the shift of the water table on either side of the drains could have occurred. For these two intervals, conclusions drawn from flow data were, therefore, retained like those derived from groundwater depth data.

The four other events exhibited sufficient runoff to supply the bioretention cell after filling the sedimentation pit. For the events of March 26th, 2021, June 8th, 2021, and June 26th, 2021, we note that the level inside the inlet sump before the onset of rain and after the event is high. The water level in the inlet sump after the end of the event decreases very slowly. In comparison, during the September 24th, 2021 event, the level in the sump was initially low and quickly returned to this level after the sampled event. This observation indicates the inlet sump is not watertight and allows runoff water to infiltrate into the ground and groundwater to infiltrate into the sump. During the thaw period (March–April), the catchment of runoff water by the sump, combined with low temperatures and a high water table, seems to maintain a high level in the inlet sump. During this period and until the end of July, the bioretention system can be called upon more frequently, as the margin between the base level in the sump and the discharge threshold to the bioretention system decreases slightly. This compares with the dry period (August–October) when the water table is lower. During this period, the bioretention system is hardly used at all, as the level in the sump is kept too low. Therefore, only major rainfall events can bring sufficient runoff to the inlet to allow the level to reach the weir threshold.

The four other events exhibited sufficient runoff to supply the bioretention cell after filling the sedimentation pit. The events of March 26th, 2021, June 8th, 2021, and June 26th, 2021 presented similar results, and the plot for March 26th, 2021 is observed (Figure 5(a)). For these three events, the water level inside the sedimentation pit was high prior to the start of rainfall and remained high after the sampled event ended. The water level in the pit decreases slowly following events. In comparison, during the event on September 24th, 2021 (Figure 5(b)), the water level in the pit was low and returned to this level after the sampled event. This observation confirms that the pit is not watertight and that this design has a major impact on the performance of the bioretention cell, as it allows runoff water to infiltrate into the ground but also allows groundwater to infiltrate into the pit.

Considering the observed levels of groundwater on site, we can conclude that the conditions during the thaw period (March–April), along with the collection of runoff water by the pit, combined with low temperatures and a high groundwater level, maintained a relatively high water level in the pit. Thus, during this high water table period until the end of July, the bioretention cell received water more frequently in contrast to the drier period (August–October) when the groundwater level is lower, and the pit must fill before water enters the bioretention cell. During the drier period when the groundwater table decreases, the bioretention cell is only minimally used. Only major rainfall events can provide sufficient runoff at the inlet to allow the water level to reach the weir's threshold.

During the four events that generated enough stormflow to fill the sedimentation pit up to the weir, a variation in the water level between the two threshold lines of the bioretention cell and the weir was observed. For instance, in the event of March 26th, 2021, the water level rose until it reached the weir's level and then stabilized above this line. This indicates that the water filled the box and entered the bioretention cell through the dedicated pipe. Then, as the rainfall event continues, precipitation gradually decreases, as does runoff, and we observe that the level in the pit decreases rapidly once it falls below the weir's threshold, stabilizing at the threshold of the pipe leading to the bioretention cell. This observation indicates that the weir box was not completely watertight, despite the sealant used for water loss prevention, and that the water in the pit can exit through other points. However, this loss from the measurement system is negligible compared to the flow entering through the ‘V-Notch’ weir.

It is interesting to note that no accumulation of water was observed in the bioretention cell area over the study. For all events, the bioretention system was able to accept and infiltrate the water input, including an extreme event of 392 mm of rain that flooded parts of the City of Trois-Rivières on June 08, 2021. The outflow during this event was such that it exceeded the measurement capacity noted by the plateau on the curve during the V-notch flooding, even though the equipment was designed for submergence situations. The second event during which the outflow exceeded the measurement capacity occurred during the thaw and was due to heavy rain in addition to the snowmelt (26-03-2021). Additionally, while rainfall occurred on June 26th, the peak flow at the outlet was observed on June 27th. This delay is due to the time required for the water to travel through the bioretention cell and reach the outlet, resulting in peak flow measurements being recorded on the day following the rainfall event.

For the event on September 24th, 2021, we were unable to obtain a flowrate measurement due to excessively low water levels in the outlet well. It can be noted that with the water input from the event, the level gradually increased until September 26; however, according to the recorded data, this level never reached the weir. This issue is only visible in the event sampled during the dry season when the water table is lower, and the soil's infiltration potential is higher. Therefore, to obtain a flowrate measurement at the outlet of the bioretention cell, the presence of a base flow is necessary for the water level to reach the ‘V-notch’ threshold.

To summarize, similarly to the inlet of the bioretention system, a base flow was required to allow the drainage flow reading. The research team also faced data collection issues with the ultrasonic transducer. However, it is important to note that no water accumulation was observed on the bioretention cell over the entire study period. The bioretention cell was effective in draining the collected runoff water.

Over the five sampled events between 2021 and 2022, two gave complete results (inflow and outflow) to analyze water quality: March 26th, 2021 and June 26th, 2021. Over these two events, we observed a release of nitrogen, particularly in the form of nitrates. It is also notable that the NO₃⁻ release was more important during the thaw period. Indeed, during the rainfall on June 26th, the release observed for nitrates is lower, and for TN and ⁠, there is removal. In a bioretention system, the release of nitrogen would be attributed to the decomposition of nitrogen-rich organic matter by microorganisms in the soil (Li & Davis 2014). This observation is typical of new bioretention cells, as there was compost in the installed substrate, but the NO₃⁻ in the substrate tends to decrease with time (Dagenais et al. 2022). Also, as the plants are in dormancy in the winter, they do not absorb as much NO₃⁻.

The process of ammonium removal or nitrification is slowed down by cold temperatures (Ding et al. 2019). NO₃⁻ removal relies on vegetation presence, which is more active during the summer and may not absorb as much ammonium as observed on June 26th, 2021.

For the two sampled events, we noted an emission of COD. This phenomenon is attributed to the breakdown of organic matter from animals or plants. Various factors may contribute to this increased release, such as the presence of de-icing salts during thaw periods, decreased temperatures, and diminished vegetation coverage (LeFevre et al. 2015). However, the concentrations observed for the two events are relatively low in comparison to those found in the natural environment, hence precluding their role in manganese release.

Regarding Cl⁻ and Na⁺ concentrations, removal is observed in March, while release is observed in June. The first event occurs during the thaw period when de-icing salts are still being used. These observations may be due to the time it takes for the salt to be absorbed and then released later (Lawson & Jackson 2021). Analyzing more events would be necessary to conclude these contaminants. However, it can be noted that for the two events in question, the observed concentrations are not a concern and remain below aesthetic quality objectives (Health Canada 2022).

The results also revealed that iron was removed by 96.86% (Table 3), consistent with the literature that indicates iron is typically retained in bioretention systems. This retention could be influenced by several factors. Iron and other metals often absorb strongly organic matter and clay particles in the soil media, a process facilitated by the high surface area and cation exchange capacity of the bioretention substrate (Zhang et al. 2023). Additionally, at neutral to alkaline pH levels, many heavy metals precipitate as insoluble hydroxides or carbonates, further enhancing their retention.

Table 3

EMCs and removal percentage for each monitored event (red = release; green = removal)

View Large

For manganese, Health Canada provides two recommendations: a maximum acceptable concentration (MAC) of 0.12 mg/L and an aesthetic objective of 0.02 mg/L (Health Canada 2022). The technical document from Health Canada related to manganese (Health Canada 2019) reports that some human studies have suggested a potential link between manganese in drinking water and certain neurological effects in children. The effects observed in children align with the neurological effects seen in key animal studies used to establish the MACs. These effects include behavioral disorders, changes in intellectual function, reduced academic performance, short-term memory, and motor dexterity. However, this complex association depends on factors such as age, gender, and nutritional status. The uncertainties related to the methodology used in these epidemiological studies, including the assessment of manganese intake and confounding factors, as well as the variability in results, prevent us from concluding a causal link between these effects and the presence of manganese in drinking water. Nevertheless, numerous animal studies appear to support the biological plausibility of these effects. Notably, the drinking water quality criteria are based on effects observed in rodents exposed to manganese during the first few months of their lives (World Health Organization 2011). Concerning aesthetic considerations, the presence of manganese in drinking water often raises consumer complaints about water color, which is why Health Canada recommends an aesthetic objective.

In Québec, many community water systems are supplied by groundwater with high concentrations of manganese that could have neurotoxic effects associated with intellectual impairment of children (Bouchard et al. 2011). Manganese in groundwater originates from the soil and sediments. Stormwater can also contain manganese from sources including the wear from vehicles and soil dust (McKenzie et al. 2009). Organic matter entering groundwater has been shown to increase the dissolution and mobility of metals such as manganese (Neidhardt et al. 2014). Over the study period, the groundwater data appear to show a gradual increase in manganese concentrations in the groundwater, exceeding criterion values. During the study period, the consistent outflow observed during dry periods, coupled with groundwater depth data (piezometers) compared to the bioretention's perforated drain depth, indicated continuous groundwater infiltration into the BR-4 drains. This implies that the groundwater table is positioned above the drain level.

Six comparative samplings identified periods of groundwater–bioretention cell interaction (communication from the Trois-Rivières City). Of these six, four revealed interactions coinciding with observed increased manganese concentrations on June 8, 2021, June 26, 2021, and March 19, 2022 events.

While the hypothesis was based on a limited number of event comparisons, the Mann–Whitney U Test and principal component analysis (PCA) were employed on pooled data to compare medians across independent sampling locations and assess potential correlations with other water quality parameters, both with and without groundwater interaction (see Supplementary material, Figures S3 and S4). The Mann–Whitney U Test p-value of 0.000821, significantly less than the alpha level of 0.05, confirmed a significant difference between groundwater and bioretention inlet manganese medians. Additionally, the PCA supported the clustering of outlet and groundwater samples, particularly in relation to manganese and other heavy metals.

The permeability of the inlet basin or inadequately positioned or designed inlet grates could impact the heavy metals' (Mn and Fe) concentrations in the groundwater. It can result in either inlet clogging, diverting stormwater away from the bioretention cells, or the creation of a hydraulic gradient that drives water toward the groundwater. Similar results or potential effects on groundwater quality associated with infiltrating stormwater have been observed in several studies (Zhang & Chui 2017; Edwards et al. 2022; Nazarpour et al. 2023).

For the first event (for which we concluded removal), the difference in median values was not significant. Whereas for the second event (for which we concluded release), the central values of the represented data demonstrate a distinctly more pronounced difference.

It is important to consider the distribution of pollutant concentrations based on runoff volume and flow. In the sampled events, the release of manganese during the second event (EV2) appears to be linked to higher runoff volumes, suggesting that increased flowrates may have remobilized contaminants, including manganese. This pattern suggests that pollutant concentrations, especially during peak flow periods, are influenced by the volume of runoff passing through the system. To better understand the mechanism behind manganese release and retention, a detailed analysis of the relationship between runoff volume and manganese concentrations is necessary. Furthermore, a comprehensive, long-term evaluation of manganese behavior is essential to confirm trends and gain insights into the mechanisms. Additionally, exploring potential modifications to the bioretention cell design or operation could enhance its effectiveness in managing manganese, particularly under varying groundwater conditions and high runoff volumes.

A full-scale study of bioretention cell effects on local and sewershed hydrology and water quality was conducted. Novel data were generated in support of the protection of cold climate groundwater sources of drinking water, particularly in areas vulnerable to manganese contamination. While considering the challenges and limitations in data collection from the monitoring system, the following conclusions can be drawn:

• Measurements of inflows, outflows and groundwater demonstrated that groundwater was being drained from the system and that not all events had sufficient precipitation to reach the bioretention cells from the inlet basin.

• Most events were captured by the inlet basins that were not sealed and could be a route of groundwater contamination from contaminated runoff.

• Inlet basins should not be oversized to ensure water flow to bioretention cells; they should be designed to avoid the accumulation of debris at their entrance.

• Water table variability and depth must be considered when identifying the most suitable locations for the implementation of bioretention cells. However, even with groundwater drainage, bioretention cells were effective in reducing runoff volumes and peak flows generated from the sector where they were implemented. The reduction of peak flows and volumes is an important consideration for low-lying regions with high water table depth vulnerable to flooding or sewer overflows.

• At a sewershed scale, a sewered creek largely masked the effects of bioretention cells. Efforts to daylight creeks and separate them from combined sewers are options to consider given the large areas of green stormwater infrastructure that would be needed to significantly reduce peak flows at a sewershed level where creeks have been sewered.

• Bioretention cells were effective for reducing most of the measured metals including iron and TP even in cold water. Nitrogen compounds and COD were variably released into the outlet water.

• Manganese is a concern for drinking water and the potential to increase its mobility was observed. Further study is required to observe the long-term trends of manganese in groundwater and in the outlet water from bioretention cells, which is important for communities that rely on groundwater for drinking water.

• Decisions related to contaminants and contaminant removal by green stormwater infrastructure must consider the balance among the various competing needs for aquifer recharge for water supply, urban flood mitigation, reduction of downstream combined sewer overflows and the various uses of the water resource.

• Although cities typically select sites for green stormwater infrastructure implementation based on required upgrades to sewer infrastructure or correction of specific problems, there is a need to improve overall planning to implement the infrastructure where it will provide the greatest benefits relative to costs. Multicriteria decision analysis including life-cycle analysis could provide additional insight.

• Future research should investigate the reduction of peak stormwater flows and volumes in relation to bioretention implementation scenarios for a variety of return periods considering climate change. Critical scenarios should be evaluated to identify potential additional mitigation measures to ensure the resilience of urban stormwater systems and community safety.

The authors acknowledge the three funding sources: the City of Trois-Rivières, Ouranos, NSERC PURE CREATE, and the Natural Sciences and Engineering Research Council. Special thanks to Marwan Abdelrahman for assistance with preliminary sampling and analysis.

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

The authors declare there is no conflict.