Effects of ecological roofs in water quality: an experimental study over a humid tropical climate

Although ecological roofs (vegetated or non-vegetated) provide many benefits, it can also leach substances such as nutrients and metals that can affect downstream aquatic ecosystems. Therefore, this work aims to investigate the rainwater quality from ecological roofs in Recife, located in the Northeast Brazil, using local species in a tropical and humid climate. Using four test cells of 1 m2 (one non-vegetated filled with expanded clay aggregate, two vegetated with cactus “Coroa-de-Frade” and grass “Grama Esmeralda”, and one control roof), we analyzed thirteen water quality variables regarding irrigation parameters: pH, electrical conductivity, turbidity, nitrate, ammonia, phosphate, bicarbonate, carbonate, calcium, magnesium, sulfate, potassium, boron, sodium, and sodium adsorption ratio. We simulated rain events controlling its intensity and analyzed a sample of natural rainwater event. All roofs neutralized the pH. Control and clay roofs were source of bicarbonate and calcium, responsible for more alkaline effluents. Carbonate and ammonia were below the recommended limits for irrigation purposes for all roofs. Green roofs were source of nitrate, ammonia, and boron. Neither roofs were source or sink for sulfate and chloride for all analyzed samples. Regarding the natural rainwater experiment, only green roof with Coroa-de-Frade exceeded the recommend irrigation parameters for potassium and phosphate. A post-treatment is required for irrigation purposes. We recommend a first-flush system followed by a filter with sand and activated carbon.


Introduction
Impervious surfaces increase due to populational growth can modify the hydrologic cycle and increase runoff, which can lead to urban floods. To meet UN Sustainable Development Goals (U.N. 2016), low impact development infrastructures, such ecological roofs (Versini et al. 2016), can help to mitigate the impacts of impervious surface area growth (Huang et al. 2018) by retaining and detaining stormwater and subsequently evapotranspire and infiltrate excess water and problems caused by Green House Gases emissions (Besir and Cuse 2018). Among these practices, ecological roofs (Liu et al. 2021) are an excellent alternative considering economic and benefit costs.
A green roof consists of several components, including the water proofing membranes, drainage material, filter layer, substrate, and vegetation (Shafique et al. 2018). Each component has its role for the overall performance. Due to numerous benefits, green and brown roofs, which differentiate only by the presence or absence of vegetation, are being used in many countries. Benefits include the increase in thermal comfort (Santos et al. 2019), mitigate urban heat island effect (Razzaghmanesh et al. 2016), noise reduction (Van Renterghem 2018), air quality improvement (Viecco et al. 2021), contribution to the landscape and recreational and habitat functions (Williams et al. 2014).
Despite ecological roofs benefits, it is not consensus regarding the water quality issue. In fact, ecological roofs could become either a source or sink of runoff pollutants (Wang et al. 2017). The runoff water quality of ecological roofs is a result of substrate and vegetation establishment, microbial activity, roof age, meteorological conditions and seasons, irrigation water quality and atmospheric deposition (Berndtsson 2010;Harper et al. 2015;Hashemi et al. 2015;Karczmarczyk et al. 2018;Akther et al. 2020). Due to several structural designs, substrate, and vegetation combinations, it is important to better understand optimal configuration considering local hydrological variables.
Following a global trend for green infrastructure incentive policies (Liberalesso et al. 2020), some cities in Brazil regulate the construction of green roofs including Recife in the Pernambuco State, Northeast Brazil (Law No. 18112 2015). Even though Brazil gradually advances in green roofs legislation, still lack studies regarding their impacts. Particularly, legislation in Brazil requires the strict separation of sewage and rainwater into two networks. Rainwater runoff is not treated before reaching rivers. In case of ecological roofs acting as source of pollutants, this could lead to water bodies contamination (Karczmarczyk et al. 2018). Some studies have already investigated the influence of green roofs in Brazil (Vieira et al. 2013;Noya et al. 2017;Santos et al. 2019;Castro et al. 2020).
Choosing the best combination of vegetation and substrate is not an easy task. Besides assessing ecological roof benefits, vegetation species must be adaptable to local climate, and soil need to have enough nutrients for vegetation establishment. Local species must be considered to avoid the insertion of exotic species in the environment. Especially in a tropical climate, extensive green roof vegetation species need to endure a hot environment, high solar exposure, and wind. In this matter, it is essential to evaluate the combination of vegetation and substrate for this climate. Therefore, this study aims to investigate the effects of ecological roofs on water quality in a humid tropical climate and assess the use of runoff water for irrigation purposes.

Materials and methods
The study workflow chart is presented in Fig. 1. The experiment was conducted in Recife in the Pernambuco State, Northeast Brazil. We chose to study three ecological (two green roofs and one brown roof) and one control roof to compare results. We conducted three experiments, a simulated rainfall using an irrigation system, a natural rainfall event, and another experiment which consisted in taking samples at different time stamps within the same rainfall event. For all samples, we monitored 16 water quality variables.

Study area
The city of Recife is located in the Pernambuco State, Northeast Brazil. The city has low thermal amplitude and is under a humid tropical climate (type AS' under the Köppen-Geiger climate classification) with rainy season from April to July.
Four experimental units of ecological roofs were built in the Federal University of Pernambuco (8.0549º S 34.9527º W). All experimental units were built with dimensions of 1.30 × 1.30 m (useful area of 1 m 2 ) at a height of 2.75 m above soil. Roofs were properly waterproofed following Brazilian Technical Standards test for infiltration in buildings  (ABNT 2013). Three ecological roofs were settled: one green roof covered with substrate and a cactus locally called "Coroa-de-Frade" (Melocactus zehntneri), which establishes in poor nutrient, water scarcity and high temperature conditions (Fig. 2a); one green roof covered with substrate and a grass vegetation locally called "Grama Esmeralda" (Zoysia japonica), which has more ramified roots and covers the whole soil (Fig. 2b); one brown roof only containing expanded clay aggregate (Fig. 2c). The remaining of four roofs is a control roof (Fig. 2d) to compare results from the three ecological roofs. All roofs except the control roof have a drainage bed followed by a 10 cm sand layer to favor water percolation.

Experiment setup
To conduct this research, a rainfall simulator system was built as shown in Fig. 3. A well that supplies the engineering center in the University was used as water source for the experiment. The reservoir (01) of 500 L was filled using groundwater, which was pumped to the system using a water pump (03) of 0.55 KW. To control the flow rate in the system, a hydrometer (04) and a manometer (05) were used. To properly simulate rainfall, sprinklers (07) with a 360º water irrigation angle were placed upside down in the center of each roof. Polyethylene containers (08) were used to harvest runoff water at each simulation.
The water flow in the rainfall simulation system was controlled using a water valve (02). A linear correlation between pressure within the system and water discharge was obtained. In this way, it was possible to maintain the same rainfall intensity throughout all simulated experiments. It is important to note that our system has significant hydraulic losses. Then, the same pressure within the system has different water discharge for each roof. For this reason, linear regressions were calculated for each roof individually using three data points each.
Using the intensity-duration-frequency equation (IDF) of Recife (Prefeitura do Recife 2016), we identified rainfall intensities and chose the 5-year return period to simulate the experiment. This return period is consistent for urban micro-drainage devices.

Runoff water quality monitoring
Three experiments were conducted in the study: • During consecutive days, rainfall was simulated with an intensity of 163 mm/h (5-year return period and 5-min duration). Water samples were collected from the reservoir (01) to identify the prior water quality condition and compare with runoff water. A 2-day dry period anteced-ent to experiments was respected to prevent interferences in results. • A non-simulated rainfall (natural) event was collected.
Samples were taken and analyzed right after the rainfall cessation. • Water was sampled in three different times within the same experiment. The same intensity of 5-year return period and 5-min duration was chosen for this analysis. The water was sampled at: the first water percolated through the roof, at the 2-and-a-half-minute mark, at the 5-min mark until the last water was percolated. These timestamps were chosen respecting water detention time of all roofs.
The fieldwork was conducted from June to July of 2019. All water samples were analyzed at the Water Quality Laboratory of the Ecological Roofs Research Group and 16 water variables were obtained according to Table 1. pH and electric conductivity were analyzed in loco using the multi parameter water quality monitor Multi 350I from WTW. Turbidity was measured using the Hanna Instruments HI 93703 turbidimeter. All other water quality measures were obtained using the photocolorimetric method (AT100PBII by Alfakit). The sodium adsorption ratio (SAR) was calculated using Eq. 1 (Oster and Sposito 1980) and values are expressed in (mmol c L −1 ) 1 ∕ 2 . To guarantee precision in estimates, triplicate samples for all variables were obtained and mean of three values considered the true value. In case of discrepancy between data, the outlier was discarded and mean obtained with the remaining two values.
To evaluate the possibility of using the drained water from the green roof to irrigate the roof itself, we compared whether all results follow the recommendations presented in the publication of the Brazilian Agricultural Research Corporation-EMBRAPA on the "Water quality for irrigation" Fig. 3 Schematic design of the experiment (Almeida 2010). The only water variable not included in the recommended thresholds list is turbidity; however, we chose to measure it as it is a good indicator of substrate carrying.
Only relevant plots cited in results and discussion section are presented in this paper.

Results and discussion
Water simulation system calibration Figure 4 shows results for the pressure and water discharge regression. Calculated minimum Rainfall Intensity (pressure equals to zero) varies between 68 and 114 mm/h for the grass roof and control roof, respectively. However, empirically, we noticed that minimum pressure to generate water flow in the system considering hydraulic losses varied from 2 to 2.5 m of water gauge (mwg). Under these conditions, we chose a rainfall simulation of 5 min using the IDF of Recife (Prefeitura do Recife 2016) as reported in the methodology section. Naturally, the low possibility of choosing the intensity of precipitation duration is a limitation in the designed rainfall simulation system. The elevation of the reservoir above the level of the simulation system could be a solution to this issue, where the water pump would feed the reservoir and no longer the simulation system directly. As we are testing extreme rainfalls (for 5 min duration), this would represent the worst-case scenario regarding lixiviate nutrients, as extreme rainfall is associated with leaching more nutrients (Berndtson 2010).

pH, carbonate, bicarbonate, calcium, and magnesium
All samples showed an increase in pH for both natural and simulated scenarios (Fig. 5). The control roof showed the highest pH value 7.70 ± 0.12 (mean ± standard deviation). This may have been due to the dissolution of alkaline substances related to the cement and the slab waterproofing material, especially calcium carbonate. Lima (2012) evaluated the water quality in cisterns also built with cement, she observed that the younger the cistern, the higher the alkalinity levels.
The experimental roofs in this study were built in less than 2 years of the conducted experiment; therefore, we can attribute the increase in pH to the roof age. In the studied ecological roofs, the presence of substrate and vegetation minimized this effect. Among the green roofs results were quite similar, a pH of 7.38 ± 0.24 for the grass and 7.33 ± 0.24 for the cactus. As for the brown roof, results of 7.50 ± 0.29 were obtained. Carbonates and bicarbonates results are correlated with pH effect, as they are substances present in cement, especially in the form of calcium carbonate CaCO 3 and sodium bicarbonate NaHCO 3 , derived from limestone rocks. In all analyzed samples, we could not detect presence of carbonate. However, the bicarbonate in control roof (34.2 mg/L ± 4.5 mg/L) and expanded clay (25.5 mg/L ± 4.6 mg/L) showed increase in its concentration. Cactus (13.6 mg/L ± 2.2 mg/L) and Grass green roof (11.2 mg/L ± 1.9 mg/L) showed similar bicarbonate concentrations and overall performed as sink for bicarbonate. In control and expanded clay roofs, the contact of water with the cemented surface of the slab probably promoted the dissolution of these ions. On green roofs, due to the presence Overall, pH neutralization is beneficial for all green roofs. The results agree with studies using different substrates and vegetations in the literature (Razzaghmanesh et al. 2014;Vijayaraghavan and Joshi 2014;Buffam et al. 2016). In studies in the semi-arid region of Pernambuco, neutral-alkaline pH results were also found (Farias 2012;Santos 2022;Silva 2017). Comparing to the irrigation recommendation levels by Almeida (2010), all roofs met the pH (between 6 and 8.5), bicarbonate (maximum of 10 meq/L, i.e., 610 mg/L) and calcium (maximum of 30 meq/L, i.e., 1063.8 mg /L) thresholds.
It is not possible for the three ions (carbonate, bicarbonate, and hydroxides) to coexist in the same sample, as the bicarbonate reacts with the hydroxide. Whenever a sample has a pH between 4.5 and 8.3 (all samples are in this pH range), the alkalinity is only due to the presence of bicarbonate, explaining the absence of carbonate ion in our results. This effect is due to the balance between carbon dioxide, carbonate, and bicarbonate in the sample.
All roofs revealed to be neither source nor sink of magnesium, except for Coroa-de-Frade which was able to absorb part of magnesium (10,12 mg/L ± 2,86 mg/L to 4,70 mg/L ± 1,14 mg/L). Overall, all roofs recorded magnesium concentration values below the recommended irrigation limit of 486 mg/L by Almeida (2010).

N-ammoniacal, N-nitrate and phosphate
Regarding nutrients, most of the analyzed samples N-ammoniacal was kept below the detection level (0.1 mg/L). We detected N-ammoniacal in 6 samples out of 30, with a maximum value of 0.6, below the Almeida (2010) recommended threshold. Inflow water for all tests was below detection level. In these 6 samples, roofs acted as sources in a non-clear pattern. In fact, N-NH 4 + is a water-soluble cation that can easily be fixed by negatively charged organic matter and clay through adsorption or ion exchange (Todorov et al. 2018), explaining the low nitrate levels in percolated water. The green roof containing the Coroa-de-Frade   . 5a) and phosphate (Fig. 5b). C o n s i d e r i n g N -n i t r a t e , C o r o a -d e -F r a d e green roof obtained the highest concentrations (6.43 mg/L ± 2.95 mg/L) and in one of the tests (10.30 mg/L), it exceeded the established limit for nitrate by Almeida (2010) of 10 mg/L (Fig. 6). The grass green roof and expanded clay roof obtained the same means of 1.16 mg/L and deviations of 0.45 mg/L and 0.79 mg/L, respectively. On the control roof, there were no major changes in the concentration of the outflow (0.78 mg/L ± 0.69 mg /L) to the inflow (0.73 mg/L ± 0.85 mg/L). The ecological roofs did not show denitrification conditions, meaning that the removal of nitrate mainly depended on plant absorption condition, being consistent with conclusions of Gong et al. (2021). There is a more significant increase in nitrate concentrations for the simulated events compared to natural rainfall, even though the concentrations of N-nitrate in the inflows are similar. This may have occurred due to the intensity of simulated rain compared to natural rain, allowing for higher nitrate leach. This hypothesis is in agreement with Berndtson (2010), which indicated that intense rainfall is responsible for higher concentrations of nitrogen and phosphorus species than less intense rainfall, although other factors may influence these results.
For the analysis of natural rain, results from grass roof were also found to be above the recommended threshold. On 26th June and 27th June, the concentration on the inflow water was 3.00 mg/L and 2.20 mg/L, respectively, both above the recommended limit. In studies of the semi-arid region, the green roofs studied were also sources of phosphorus (Farias 2012;Santos 2022;Silva 2017), with all their effluents above the limit established by Almeida (2010). The reason for the green roof with Coroa-de-Frade leached the highest concentrations of N-ammoniacal, N-nitrate and phosphate was probably because the low substrate interception and plant nutrient fixation. The cactus species was not using all the available nutrient resulting in leaching at the observed levels. The use of a substrate in lower quantity of nutrients would be an alternative to reduce the concentrations of nitrogen and phosphorus compounds in the effluents. Future analyses are needed to verify whether the amount of leached nutrients has stabilized, since nutrient concentrations are higher for newly constructed roofs and decrease over time (Todorov et al. 2018).

Turbidity and electric conductivity
For the turbidity, according to Fig. 7a, we notice that the cactus green roof obtained the highest values for the simulated events (277.7 NTU ± 39.2 NTU), followed by the expanded clay roof (17.5 NTU ± 13.5 NTU), by the grass roof (12.7 NTU ± 5.3 NTU) and the control roof (4.2 NTU ± 4.9 NTU). For better visualization of the results, the y axis of Fig. 7a is in logarithmic scale. Turbidity is associated with the type of substrate (Morgan et al. 2011) and the roof age, as young roofs may carry more fine particles. All roofs have the same substrate. Thus, the greater turbidity from the cactus roof can be explained by the higher soil exposure, and by the shape of the vegetation's roots, since they are not as ramified as that of the grass ones. The turbidity for the event with natural rain was lower than the average of the simulated events (less expressive in the control roof), probably because the intensity of the simulated rain was higher than the natural rain.
Regarding electric conductivity, in the experiments for the simulated rain, only the green roofs decreased the electric conductivity (EC), and the roof with Coroa-de-Frade had the highest removal efficiency. The inverse effect was noticed for the natural rainfall analyzed. All roofs increased EC, with the greatest increase in the Grass roof (Fig. 7b). Only the EC of the green roofs, when using the average values of the simulated experiments, are within the parameters of Almeida (2010) which has a maximum value of 300 mS/cm. For the experiment using natural

Chloride, sodium, and SAR
All roofs were neither sources nor retained chloride, maintaining all outflow concentrations inside standard deviation of inflow samples. Chloride inflow concentration mean was 60.70, while outflow means varied from 57.40 to 59.44. Sodium and RAS values were found to be much lower the limits established by Almeida (2010) of 919.6 mg/L and 15 mmol c ∕L 1 ∕ 2 , respectively, for all roofs studied.
Silva (2017) found similar results for green roofs, from 19.1 to 20.9 mg/L of sodium and 1.4 mmol c ∕L 1 ∕ 2 to 2.3 mmol c ∕L 1 ∕ 2 . This is indicative of the absence of sodicity in the soil, a factor that could decrease hydraulic conductivity and affect vegetation growth and water retention.
Regarding potassium (Fig. 8a), the effluents from the brown roof with expanded clay (1.30 mg/L ± 0.75 mg/L), the control effluent (1.22 mg/L ± 0.53 mg/L) and the grass roof (1.37 mg/L ± 0.88 mg/L) were below the limit recommended by Almeida (2010) of 2 mg/L. The effluents from the green roof with Coroa-de-Frade obtained an average of potassium samples approximately 5 times higher than the recommended limit for irrigation (10.94 ± 1.49). Silva (2017) found values between 18.0 and 18.8 mg/L for effluents from the green roofs of both cactus species, Babosa and Coroa-de-Frade, in the semi-arid region of Pernambuco State (Fig. 8).
The well water used for the simulated rainfall analysis contained high levels of sulfate compared to natural rainwater. The presence of these sulfate concentrations in well water may be due to the decomposition of soils and rocks present in the aquifer such as gypsum (CaSO 4 ) and magnesium sulfate (MgSO 4 ). Sulfate analysis of natural rainwater resulted in values below the detection level (0.1 mg/L). The green roof with Coroa-de-Frade was a source of sulfate ( Fig. 8b) both for the simulated experiments and for the one with natural rain, the others did not show such significant differences in relation to the initial concentrations. The green roof with Coroa-de-Frade was a source of sulfate for all experiments analyzed. According to Almeida (2010), Water quality results for potassium, sulfate, and boron the maximum recommended limit for sulfate is 960.6 mg/L. Therefore, all samples were within the recommended values for irrigation. Sulfate (which also obtained values below the detection level of 0.1 mg/L) and nitrate are substances that contribute to acid rain, as they can produce sulfuric (H 2 SO 4 ) and nitric acids in an aqueous medium (HNO 3 ). The low concentration of these ions contributed to a more neutral pH of the rain of 6.4. According to the studies analyzed by Hashemi et al. (2015) normally the rain pH is between 5 and 6. Studies in the dry period are necessary to evaluate the effect of these variables.
No boron levels ( Fig. 8c) above the lower limit of quantification (0.1 mg/L) were detected in natural rainfall, brown roof with expanded clay and control roof. In agreement with Ferrans et al. (2018), the green roofs were sources of boron, being more pronounced in the one that contains Coroa-de-Frade as vegetation. The presence of Boron is associated with the composition of the substrate and in Coroa-de-Frade to the low degree of assimilation by the vegetation. The higher concentration of boron in the simulated experiments was possibly due to the presence of boron-containing aquifer rocks, as groundwater was used for simulated. Silva (2017) found values far above sulfates in his analyses (2685 mg/L to 3050 mg/L) and values below Boron (0.6 mg/L to 0.9 mg/L) than those found in the present study. Figure 9 shows the main results of the temporal discretization sampling in a single event experiment. The cactus green roof was the one that showed the greatest changes with the collection of samples at different times, especially there was a more expressive decrease in the variables: N-ammoniacal, calcium and potassium. Variables that exceeded the recommended parameters for irrigation (Almeida 2010) using natural rainfall were phosphate and potassium. This experiment by sampling at different times of the same precipitation Water quality results over time N-ammoniacal, calcium, potassium, and phosphate event indicates that a device for the disposal of first waters would be an alternative to mitigate the leaching effects of potassium but not phosphate. The first water disposal device is a sanitary barrier that diverts the first waters of each rain to discard the waters that wash the atmosphere and the catchment surface. This type of device has proven to be very efficient (Carvalho et al. 2018). Silva (2017) developed a filter containing sand and activated carbon that was able to decrease the phosphate levels of effluents from green roofs containing the two types of cactuses: Coroa-de-Frade and Babosa. Thus, a treatment system containing a first water diverter and a sand and activated carbon filter could keep all parameters in the recommendation levels of Almeida (2010) and should be further investigated.

Conclusions
Considering only the natural rainfall experiment, all the analyzed water quality variables from the grass green roof, the brown roof with expanded clay aggregate and the control roof, fit into the recommended limits. In this way, the system is water self-sufficient in the context of water quality. As for the cactus green roof, phosphate and potassium were outside the limits established for irrigation, requiring a subsequent treatment for use for irrigation.
Considering the possible forms of treatment and the results of the analysis by sampling at different times of the same precipitation event, a series system with a first water diverter and a sand filter with activated carbon are good alternatives to mitigate the phosphate concentration. and potassium and makes them suitable for use in irrigation. The efficiency of this system should be further investigated and test application feasibility.