Impact of de-icing salt runoff in spring on bioretention efficiency

Urban stormwater is laden with pollutants, such as heavy metals, nutrients, herbicides, polycyclic aromatic hydrocarbons, and various organic compounds (Eriksson et al. 2004; Valtanen et al. 2014). These contaminants threaten the natural environment as urban stormwater is often discharged into waterways without prior treatment. Blue-green infrastructures such as bioretention cells (BR) are increasingly used to address this issue (Fletcher et al. 2015). BR has been shown to effectively manage and treat stormwater runoff through mechanical, chemical and biological processes (Kratky et al. 2017).

Although seasonal differences have been shown to have an impact on the effectiveness of BR (Roseen et al. 2009; Paus et al. 2016; Beral et al. 2023), in winter, plants are dormant, microbial activity is reduced and the physical properties of the water and soil change due to low temperatures, all of which could potentially hinder BR performance (Muthanna et al. 2007; Blecken et al. 2011; Søberg et al. 2014; Géhéniau et al. 2015). Moreover, in many cold regions, large amounts of de-icing salts and/or abrasives may be spread on roads and sidewalks to ensure safety. For instance, Canada annually uses over five million tonnes of de-icer, the most common being sodium chloride (NaCl) (Environment and Climate Change Canada 2022). Spreading de-icing salt generates salty runoffs when the snow melts (Ramakrishna & Viraraghavan 2005), especially during initial spring warming or the first rainy precipitation of the season (Snodgrass et al. 2017; Taka et al. 2017; Burgis et al. 2020; Goor et al. 2021). Salty runoffs could decrease BR effectiveness through a negative effect on the plants and substrate.

In fact, increased osmotic pressure caused by sodium chloride may induce water stress in living organisms (Hintz & Relyea 2019). In plants, osmotic stress results in reduced water absorption by roots and consequently in root development, leaf area, stomatal conductance and plant size (Liem et al. 1985; Munns & Tester 2008; Davis et al. 2014). This reduction in plant performance may consequently diminish their contribution to BR efficiency (Read et al. 2009; Payne et al. 2018; Beral et al. 2023). Intracellular Na accumulation can also interfere with cellular machinery and may cause nutritional deficiencies due to competition for other nutrient transporters (Cekstere et al. 2010). Excess salt disturbs numerous metabolic pathways, leading to chlorosis, necrosis and even plant death (Endreny et al. 2012; Equiza et al. 2017; Shelke et al. 2019; Prodjinoto et al. 2021). Resistance to salt stress varies widely among plant species, which suggests that it should be considered during plant selection for BRs exposed to high salinity (Munns & Tester 2008; Dmuchowski et al. 2022). However, it is unclear how spring runoffs laden with de-icing salt would affect plant contribution to BR efficiency, as these salty runoffs, no matter how severe, are of relatively short duration and occur early in spring when the plants are still dormant.

Excess sodium chloride can also alter the physicochemical properties of the soil. Upon NaCl dissociation, the sodium cation competes with micro- and macro-elements on soil cation exchange sites. Once these elements are dislodged, they leach out of the substrate (Paus et al. 2014; Géhéniau et al. 2015; Søberg et al. 2020, 2017). A higher Na percentage at the expense of calcium favours a reduction in clay flocculation and a dispersion of colloids (Ramakrishna & Viraraghavan 2005; Fay & Shi 2012). This results in a more compacted soil structure, leading to lower permeability and hydraulic conductivity in BR (Denich et al. 2013). Since chloride is considered more mobile and less reactive (Marsalek 2003), BR might only briefly delay its release (Søberg et al. 2014; Burgis et al. 2020). Higher salinity may also lead to the remobilization of pollutants previously adsorbed by the substrate, through the release of contaminated colloids (Tromp et al. 2012; Kakuturu & Clark 2015; Huber et al. 2016), or the creation of dissolved metal-chloride complexes (Marsalek 2003; Reinosdotter & Viklander 2007).

Our study aims to investigate the impact of de-icing salts on BR effectiveness, specifically focusing on (1) the impact of NaCl-laden runoffs on the hydrology and contaminants filtration in BR mesocosms, (2) the persistence of the effect on BR after salt runoffs cease during the vegetation period and (3) the resilience of vegetation exposed to spring de-icing salt runoffs.

The mesocosm-scale experiment was conducted in the ‘Phytozone’ greenhouse of the Institut de recherche en biologie végétale (IRBV), on the site of the Montreal Botanical Garden, in Canada (45°33′40.5″N 73°34′22.4″W). The temperature in the greenhouse was set to correspond to the outside temperature, except during cold periods, but the minimum was set to 5 °C (±2 °C) to prevent damage to the equipment due to freezing (Figure S1). BR efficiency was determined for 16 mesocosms that were planted with one of three species or left unplanted, then subjected to semi-synthetic runoffs supplemented with four different concentrations of NaCl in a ‘Latin square’ configuration. The assessment was carried out by monitoring hydrology (reduction in water volume and exfiltration rate) and contaminant concentrations and mass in each compartment of the system (water, plants and substrate). In addition, the resilience of plants to NaCl stress was determined by monitoring their growth and health status.

The bottom 10 cm of the mesocosms were filled with granite gravel as a drainage layer (Ø 5–12 mm; ‘Granite gris du nord’ from Agrebec Inc.). The main media consisted of 60 cm of commercial sandy loam BR media (‘Natureausol’ from Savaria Inc.; characteristics summarized in Table S1). On top, 3 cm of mulch (Savaria Inc. fragmented rameal wood) was added (Figure 1). At the time of planting, the mesocosms were treated with the mycorrhizal inoculant (‘MykePro Landscape’ from Premier Tech Ltd) according to the manufacturer's recommendations. The mycorrhizal inoculants are used with the aim of improving plant health and growth. Their use in municipal plantings is a common practice in Quebec, Canada.

Among the 16 mesocosms, 4 were unplanted controls and the 12 others were planted with 1 individual of one of the 3 following plant species: Juncus effusus, Cornus sericea or Iris versicolor (hereafter referred to as Cornus, Juncus and Iris). These species are among those recommended for BR by the Standards Council of Canada (Standards Council of Canada 2019) and include different biological forms and functional traits (Flora of North America Editorial Committee, eds. 1993). They have also previously been the subject of scientific monitoring in BR studies (Fritioff 2005; Zaimoglu 2006; Najeeb et al. 2011, 2009; Nocco et al. 2016; Lu et al. 2018; Dołęgowska et al. 2022; Beral et al. 2023).

Tap water was used to gently remove nursery soil from the roots before planting, on 1 June 2018. The plants were then acclimatized in mesocosms irrigated with rainwater for 12 months (from June 2018 to June 2019), followed by a 9-month period of irrigation with semi-synthetic runoff water (see next section), during which monitoring was carried out (Beral et al. 2023).

The semi-synthetic runoff water consisted of rainwater from the greenhouse roof, supplemented with nutrients, trace elements and a carbon source (Table S2; Beral et al. 2023). No pathogens or hydrocarbons were used for methodological reasons. Also, no sediment was used, since the removal of total suspended solid (TSS) depends mainly on the particle size of the substrate, and since the medium was the same in all our treatments, no significant difference was expected between treatments. Thus, efforts were focused on pollutants for which plants can potentially play a role in removal.

The urban runoff applied (loading frequency and quantity) simulated that reported in regional climatic data from Environment and Climate Change Canada (2018). Thus, based on a BR area equal to 10% of the collection area (as recommended by Coffman et al. 1993; Yang & Chui 2018; Standards Council of Canada 2019), each mesocosm was irrigated with 10 L, three times per week (Monday, Wednesday and Friday) from March to August 2020.

We assume that salty runoffs occur mostly during the first few rainy precipitations in early Spring, which accelerate snowmelt and wash out the salts spread during Winter (Williams et al. 2000; Goor et al. 2021). Thus, to simulate spring road-salt runoff, sodium chloride was added to the semi-synthetic runoff on four occasions, from 12 to 23 March 2020. During the salt runoff period, each mesocosm per species type (three species + one unplanted) received one of four NaCl concentrations. The tested NaCl concentrations were 0 mgCl/L as negative control, 250 mgCl/L – corresponding to Canadian drinking water guidelines (Health Canada 1997), 1,000 mgCl/L – a concentration usually found in runoff from roads spread with a 95/5 sand/salt mixture and previously tested in scientific studies, or 4,000 mgCl/L – a concentration usually found in runoff from roads spread with pure salt (Mayer et al. 1999; Denich et al. 2013; Paus et al. 2014; Géhéniau et al. 2015; Taka et al. 2017). After the salt runoffs, the mesocosms were irrigated with regular semi-synthetic runoff without de-icing salt until the end of the experiment (25 August 2020).

Resilience of the vegetation subjected to spring salt runoffs was determined by monitoring plant growth and health. To characterize the impact of spring NaCl runoffs on the hydrology and filtration of contaminants as well as the impact's persistence in time after the salt runoffs, we monitored water loss, exfiltration rate, influent and effluent water quality, as well as substrate properties. The fate of contaminants and plant contribution to the efficiency of BRs subjected to these salt runoffs were determined by following the evolution of the substrate chemical composition, and contaminant uptake in plant tissues.

Plant health and growth were monitored once a month during the growing period. Average leaf area (Equation S1), total leaf area (TLA; one-sided; Equation S2), leaf area index (LAI; Equation S3) and unit ground surface area covered by the plant (Equation S4) were estimated. Health status was assessed as classes (from 5 healthy to 0 dead). At the time of dismantling, the leaves (plus the reproductive parts), stems in the case of the Cornus and roots were cut, collected separately and oven-dried at 70 °C to constant weight, to determine their dry biomass weight.

Contaminant concentrations were analysed based on leaf dry material collected from (1) three additional nursery individuals per species at planting time, (2) all 12 individuals of the species included in this experiment at the end of the acclimatization period, (3) a composite sample per species at the end of the first experimental period in 2020, as presented in Beral et al. (2023) and (4) all 12 individuals at the end of the experimental growing period following the salt application. The concentrations for N, P, K, Ca, Mg and 15 trace elements were analysed by Environex Inc., an accredited laboratory, using standard methods (see Note S1 for more information). The portion of contaminants taken up by the plants was obtained by multiplying leaf concentration by leaf biomass. Leaf biomass produced was also used as a growth rate indicator.

At each watering event, the basic physicochemical changes (pH, DO (dissolved oxygen), salinity, EC (electrical conductivity), ORP (oxidation–reduction potential), and TDS (total dissolved solid)) were measured with a multiparameter probe (HI98194, Hanna Instruments^®) both at the inflow and at the outflow. The pollutant concentrations were sampled on the second and fourth salt runoff event days, then twice during the following month and then once a month, for a total of eight samples. For the effluent, only the first 300 mL were collected, which is conservative in terms of pollutant removal, because it is generally the most concentrated (Sansalone & Buchberger 1997; Sansalone & Cristina 2004). A wide range of contaminants were analysed by the Interuniversity Research Group in Limnology – Université de Montréal (GRIL-UDEM) laboratory according to standard methods (Baird et al. 2017): NO_x (nitrate (⁠⁠) + nitrite (⁠⁠)), total nitrogen (TN), ammonium (⁠⁠), total phosphorus (TP) and ortho-phosphates (⁠⁠); total metals (Ca, Cd, Cr, Cu, Fe, K, Mg, Mn, Ni, Pb, Zn), total organic carbon (TOC) and dissolved organic carbon (DOC). For DOC, NO_x and ⁠, the analyses were performed after a 0.45 μm filtration using a polyethersulfone (PES) membrane. Multiplying the pollutant concentrations (g/L) analysed by the monthly average water volume (L per event) allowed us to calculate the contaminant mass (grams per mesocosm per event). Trace element concentrations below the detection limit (BDL) were set equal to these limits for data analysis to avoid overestimating trace element removal.

Samples were taken within horizontal soil cores (diameter 3.8 cm and length 15 cm) before salt runoffs (28 February 2020) and at the time of dismantling (25 August 2020), in each mesocosm, at −10 and −30 cm from the soil surface. The holes made in February were refilled with the original substrate previously autoclaved and then sealed again. Substrate chemical composition was analysed for N, P, K, Ca, Mg and 14 trace elements (see Note S1 for more information) by Environex Inc. Sodium adsorption ratio (SAR) was calculated with Equation S5 following the Regional Salinity Laboratory (Regional Salinity Laboratory (U.S.) 1954).

Linear mixed-effect models (LMMs) were used to compare hydrological performance (ET and exfiltration rate), water quality changes (contaminant mass removal), plant growth (average leaf area, leaf number, TLA, LAI, plant volume, biomass), NaCl concentrations, interaction and time for repeated measures data (see example of the results for water quality parameters in Table 1). The individual mesocosm effect was added to the model as a random factor. All models were checked for normality and homogeneity of variance by visual inspection of residuals against fitted values (Q–Q plots). Variables that did not meet normality or heterogeneity assumptions were modified using the appropriate transformation (square root) when judged necessary. An ANOVA was performed as a measure of variance analysis on the different LMMs focusing on the effect of NaCl concentration, the interaction of NaCl with time or species and the triple interaction. When interactions were statistically significant (α = 0.05), full LMMs were then partitioned per species or per date (or per month for parameters with more than 10 repetitions). Then, if the difference between NaCl concentrations was significant, a pairwise comparison post hoc Tukey's HSD test with a significance threshold set to p = 0.05 was run. Statistical analyses were performed using R software (ver. 4.1.2).

Overall, regardless of the de-icing salt concentration applied during runoffs, plants showed no growth defects, deficiencies or signs of NaCl-related stress after they emerged from dormancy. Also, no major changes in contaminant concentration in plant tissues due to de-icing salt were observed. The contribution of plants to contaminant removal was mainly limited to macronutrients, which were absorbed and then translocated to aerial parts. For the other trace elements, absorption by plants proved to be of minimal importance, especially since, when absorbed, these elements remain mainly in the root system.

Even though the ANOVA results showed that NaCl concentration significantly influenced the number of leaves per individual, and the interaction of NaCl with dates influenced plant volume, the results of post hoc tests were always above the significance threshold. There was no significant effect of NaCl, or an interaction effect involving NaCl, on LAI, TLA, average leaf area or leaf, stem or root dry biomass (Table S3). Phytosanitary status remained at the maximum (healthier class) during the entire experiment. Na increased in roots and leaves at higher NaCl-concentrated runoffs (Tables S4 and S5). No other major change in contaminant concentrations in plant tissues related to de-icing salt was observed.

We found that plants were able to absorb enough primary (N, P, K) and secondary (Ca, Mg) macronutrients to maintain tissue concentrations close to those measured initially when they left the nursery, regardless of the concentration of salt applied during runoffs (Tables S4 and S5). By absorbing into their tissues at least 37% of the primary macronutrient mass introduced into the system by the semi-synthetic runoff (Table S6), all species were able to contribute to the runoff filtration. Plants contributed to a lesser extent to the filtration of secondary macronutrients, with absorption levels representing 2–30% of the mass provided by the semi-synthetic runoff. The concentration of micronutrients (Mn, Zn, Fe) was lower than initially in the leaves, as well as in the roots for Mn, whereas it increased in the roots for Fe. That said, as for the secondary macronutrients, plants played a secondary role in filtration, with variable efficiency depending on the species. On the other hand, plants played no detectable or only a minor role in metals uptake, since concentrations of most metals (Cr, Cu, Ni, Pb) were low or below detection limits in leaves and roots.

Overall, the BR maintained a high mass removal efficiency for most trace elements and nutrients regardless of salinity during the whole experiment (Tables S7 and S8). Removal of Cr, Mn, Ni, Pb and Zn was very high, with an average efficiency of at least 82%. For Cu and Fe, removal was slightly less effective, with respectively 66 and 43%. Na and the main soil-structuring cations (Ca, Mg) were released, on average, during the experimental period.

As expected, the presence of plants significantly influenced the removal of nutrients and, to a lesser extent, that of metals. This contribution varied with time (see the significant elements for the species*dates interaction in Table 1). Removal of nutrients (TP, PO₄, TN, NO₃, NH₄, K) was consistently higher in planted mesocosms than in unplanted ones (Tables S7 and S8). Indeed, we found a TP, PO₄, TN, NO₃, NH₄ and K removal efficiency on average per species of at least 88, 95, 72, 74, 94, and 72%, respectively, while unplanted mesocosms removed only 52, 71, 12, −142, 86, and 0%, respectively. This plant effect was not affected by spring salt runoffs at the tested concentrations since in the statistical analysis, the interactions of plants (species) with salt (NaCl) and time (Date) were never statistically significant (see NaCl*species or NaCl*date*species in Table 1).

If we exclude the control curve (no salt), removal efficiency for several metals presented a trend consistent with the salt gradient applied, i.e., the higher the salt concentration, the greater the removal. The effect was short-term or even limited to the period of salt runoffs'. The mass removal of Cr and Ni was significantly lower at 250 mgCl/L than at the other NaCl concentrations during the runoffs, while the effect persisted until 2 weeks after the runoffs for Pb and Zn (Figure 5). Cu showed a similar trend, although often slightly above the significance threshold. The curves of the control group (no salt) for these metals (Figure 5, Cu not shown) show a reduced removal rate between mid-April and the beginning of July. Sampling or measurement errors in May (see outliers in Figure 5) and the fact that no sampling was done in June could account for these results. Cd, Fe, K and Mn mass removal showed no significant effect of salinity, with or without interactions.

For the removal of macronutrients TP, TN, NO₃ and Mn, the full LMM ANOVA shows a significant effect of NaCl with date (Table 1). However, testing the effect of salt for each date using an ANOVA or a post hoc, no significant effect was identified.

As expected, effluent EC, salinity and TDS increased significantly and ORP decreased with spring salt runoffs of increasing salinity (Figure 5). This effect persisted until the end of July, i.e., 4.5 months after the end of the salt runoffs (1.5 months for ORP, i.e., end of April). Temperature, pH and DO were not impacted by NaCl runoffs.

Overall, there were no major changes in the concentrations of substrate contaminants, except for SAR, and nitrate in planted mesocosms (Table S1).

Between ‘before the salt runoffs’ and ‘dismantling’, the spring salt runoffs increased the concentration of Na in the substrate, while that of Ca and Mg remained stable, resulting in an increase in SAR. For mesocosms treated with the most concentrated NaCl runoffs, the SAR averaged 2, compared to unsalted ones, with a SAR of only 1. Regardless of the species, planted mesocosms lowered the average NO_x concentration below the detection limit (BDL), whereas unplanted mesocosms did not. Compared to values before the runoffs, cation exchange capacity increased slightly, regardless of the NaCl concentration of runoffs, and without significant changes in pH.

All plant species tested were unaffected by the salt runoffs applied during their dormancy, regardless of the salt concentration. Also, spring salt runoffs had no effect on the hydrology of the BRs, or therefore on water volume or peak flow reduction. Conversely, spring salt runoffs impacted the filtration of many contaminants, increasing metals retention and releasing soil-structuring cations for a relatively short period of time after the salt runoffs ceased. The more concentrated in NaCl the runoffs were, the more intense and persistent these effects became.

Very high salinity in BR has been shown to have a negative impact on plant performance and survival during the growing season and consequently on their contribution to BR efficiency (Szota et al. 2015). In our experiment, we observed no phytotoxic effect of NaCl, such as growth defects, deficiencies or NaCl-related stress on plants, even at the highest NaCl concentration (4,000 mgCl/L) tested. Plant resistance to a given salinity level generally depends on the duration of its exposure and its stage of growth (Hessini et al. 2015). Mesocosms were exposed to salt runoffs for only relatively short periods of time during plants' dormant period, in order to mimic spring de-icing salt runoff under real conditions (Williams et al. 2000). It is thus likely that the potential deleterious effects of spring NaCl runoffs were limited, due to negligible plant water uptake during dormancy.

A spring salty runoff of short duration could have had a negative impact if it had resulted in a sustained increase in soil water salinity. However, in our experiment, the level of salinity in the effluent dropped rapidly after the salt runoffs ceased, suggesting that salt concentration in the substrate interstitial water returned close to normal soon after, in the growing season. Baraza & Hasenmueller (2021) also observed a rapid drop in salinity following spring salt runoff when monitoring the salinity in the pore water of roadside soil. Similarly, high salinity was found to be of short duration in surface water or groundwater downstream of a watershed with or without modern stormwater management practices (Snodgrass et al. 2017). The duration of high salinity levels may depend on soil granulometry, composition and drainage (Ramakrishna & Viraraghavan 2005). In BR, the moderate to high exfiltration rate maintained to avoid stagnant water likely facilitates the evacuation of salt after exposure.

Thus, our results suggest that, under a cold climate, spring salty runoffs caused by road salt may not represent a high threat for plants in BR with a substrate that has a high draining quality. In real conditions, the 3 years of monitoring of these same plant species, selected for their resistance to the urban environment, planted within bioretention in the city of Trois-Rivieres (Canada), did not show any damage related to salt runoffs (Dagenais et al. 2022). The long-term effect under real-life conditions remains to be evaluated, but Géhéniau et al. (2015) documented no negative effect on vegetation in a BR system operating for 5 years in a parking lot in Montreal (Canada) that was subjected to repeated winter NaCl spraying and snowmelt events. However, in BR, as in the rest of the urban environment, plant tolerance to road salt varies between species, as shown in a recent study in Norway by Laukli et al. (2022). Salt tolerance should remain an important criterion for plant selection for use in BR systems in cold environments (Standards Council of Canada 2019). However, long-term monitoring of BR would also help avoid the potential long-term effects of road salt (Willmert et al. 2018).

It must be noted that our experiment focused only on the effect of salt in runoff, without considering the direct effect of saline spray on exposed plant parts in winter and spring. Equiza et al. (2017) found that species with parts rising above the snow cover, notably evergreens, presented damage likely attributable to salinity. Plant selection for BR in locations that could be subjected to salty spray should thus consider their vulnerability to direct exposure (Laukli et al. 2022; Caplan et al. 2023). Also, in our experiment, plants were watered regularly after the spring salt runoffs. Watering in spring is considered a strategy to alleviate the negative effects of road salt (Łuczak et al. 2021), which can be aggravated by other environmental factors such as drought or high soil pH (Calvo-Polanco et al. 2014; Equiza et al. 2017).

Overall, all BRs successfully reduced water volume and flow. Mimicking spring salt runoffs did not have a significant impact on these hydrological parameters. With an ET rate of 0.4–2.8 mm/day, our results were close to those of experiments without de-icing salt in runoff conducted by Beral et al. (2023) and Wadzuk et al. (2015), who reported 0.1–1.6 or 1.0–6.1 mm/day, respectively. Plants are the main drivers of ET, and since salt had no effect on them, it is not surprising that we found no salt effect on ET rate either.

Complete exfiltration of effluent in less than 3 h was very rapid compared to that reported in other studies (Khan et al. 2012; Yuan et al. 2017) and far shorter than the maximum of 48 h recommended by the Standard Council of Canada (Standards Council of Canada 2019). Denich et al. (2013) showed lower permeability caused by the spring salt runoffs. It is possible that our serial NaCl runoffs mimicking a single spring season were insufficient to generate a change in the properties of the substrate, which contained little clay. While the effluents released more soil-structuring cations with higher salt concentrations, the final concentrations of these cations in the substrate remained more or less similar, irrespective of the NaCl concentration applied. It is possible that the amount of Ca leached was negligible compared to total Ca in the substrate, so that no effect on soil structure, or therefore on water flow, was noticeable within the time frame of the experiment. This result is similar to those of Burgis et al. (2020) who did not observe a cumulative effect of Ca release due to salt over the years. On the other hand, Denich et al. (2013) observed a change in the exfiltration rate in their experiment combining salt treatment with cold temperatures (freeze-thaw). Spraakman and Drake (Ding et al. 2019; Spraakman & Drake 2021) observed an infiltration capacity remaining above the recommended minimum of 25 mm/h even after several years. Therefore, our results support their hypothesis that the increase they found in infiltration rate was more related to the freeze-thaw cycle than to de-icing salt.

Overall, despite a relatively rapid runoff infiltration rate, all of our systems removed a high level of harmful contaminants (measured from the first flushes). However, we observed an effect of salinity on the removal of soil-structuring cations and some metals. In general, the higher the salt concentration, the greater and longer the effect.

As expected, higher EC, salinity and Na concentration in the effluent following spring salt runoffs were correlated with the NaCl level applied. The NaCl effluent concentration quickly returned to the basal level. These results do not suggest any NaCl accumulation in the substrate. This is consistent with findings by Shetty et al. (2020), who, after around 300 days, found Cl concentration at the basal level. Also as expected, soil-structuring cations were released in higher quantity, since they are known to be dislodged from the soil by NaCl in order to maintain electrical neutrality (Bäckström et al. 2004).

Søberg et al. (2017) did not observe salt-related release of metals in their BR experiment, but several authors have documented a metals remobilization in roadside soils attributable to salts (Amrhein et al. 1992; Bäckström et al. 2004). In contrast, in our experiment, salt runoffs significantly increased the removal of some metals (Cr, Ni, Pb, Zn and possibly Cu). It may be that dislodging Ca and Mg by Na leads to renewed competition among cations for exchange sites, allowing greater metal fixation. Organic matter could also bind metals through complexation mechanisms (Amrhein et al. 1992), but analysis of the substrate showed no increase in organic matter during the experiment. If we compare ‘before salt runoffs ‘ with the control group (0 mgCl/L of NaCl applied) from the substrate, we observe a decrease in SAR. It is possible that the increase in SAR during the periods of NaCl application was compensated by the supply of Ca and Mg over the remainder of the year, preventing the accumulation of Na. This is consistent with the results of an experiment by Denich et al. (2013), in which, after mimicking 15 years of salt spreading, they observed no soil destructuration either.

Improved removal of metals at higher salt concentrations contradicts the results of BR experiments by Paus et al. (2014) and Costello et al. (2020), who found metal leaching at a salt concentration of 1,500 mg Cl/L NaCl, in a 7-year-old BR. Since our mesocosms were relatively young (9 months of runoff application prior to the salt experiment), it is possible that the metal concentrations in the substrate were still low enough to allow the binding of metal (Hunt et al. 2012). A mulch layer on top of the BR can increase adsorption and accumulation capacity (Kratky et al. 2017).

Treating stormwater by plant uptake can represent 5–10% of total metal removal (Muthanna et al. 2007; Read et al. 2008), 2–37% of the total P removal and 35–79% of the total N removal (Beral et al. 2023). In our experiment, the contribution of plants to the removal of contaminants was mainly limited to macronutrients. In addition, after being absorbed, trace elements mainly remained in the root system and would therefore be difficult to harvest for effective removal, while macronutrients were translocated to aerial parts. This pattern of absorption did not change with salt runoffs. As for volume reduction, it is likely that the potential deleterious effects of NaCl were limited due to negligible uptake by plants during dormancy.

The effectiveness of BR subject to road salt has been questioned. Our study suggests that, at the concentrations of NaCl usually found in runoff water, negative effects are limited and should not be an obstacle to implementing a BR system under cold climatic conditions. However, the possible impact of the sprays on exposed plant parts in winter and early spring was not considered in our experiment. The rapid leaching of NaCl from our BR system appears to have prevented its negative effects on plants from manifesting. Longer-term studies should be conducted to confirm that there is no harmful level of Na accumulation over time in the BR substrate. Although rapid leaching of salinity may be beneficial to plants in BR, this also means that such systems are relatively ineffective in preventing de-icing salt outflow to storm sewers or infiltration in the ground, which could eventually contaminate groundwater or water courses.

The authors would like to thank Patrick Boivin for technical assistance. Thanks also to Rolando Trejo-Perez, Camille Giguere and Gwladys Jourdan for field work assistance, and to Karen Grislis for helpful comments on a previous draft of the manuscript.

Conceptualization and methodology: H. B., D. D., J. B., M. K. -V.; Data collection and analyses: H. B.; Original draft: H. B.; Review and editing: H. B., D. D., J. B., M. K. -V.; Supervision: D. D., J. B., M. K. -V.

All authors have read and agreed to the published version of the manuscript.

This research was funded by the NSERC [grant no. CRDPJ-513260-17] and the Estonian Research Council [grant no. PUT1125].

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

The authors declare there is no conflict.