Given that the common medium in existing green roofs is a single layer composed of organic and inorganic substrates, seven pilot-scale dual-substrate-layer extensive green roofs (G1–G7), which include nutrition and adsorption substrate layers, were constructed in this study. The effectiveness of porous inert substrates (activated charcoal, zeolite, pumice, lava, vermiculite and expanded perlite) used as the adsorption substrate for stormwater retention was investigated. A single-substrate-layer green roof (G8) was built for comparison with G1–G7. Despite the larger total rainfall depth (mm) of six types of simulated rains (43.2, 54.6, 76.2, 87.0, 85.2 and 86.4, respectively), the total percent retention of G1–G7 varied between 14% and 82% with an average of 43%, exhibiting better runoff-retaining capacity than G8 based on the maximum potential rainfall storage depth per unit height of adsorption substrate. Regression analysis showed that there was a logarithmic relationship between cumulative rainfall depth with non-zero runoff and stormwater retention for G1–G4 and a linear relationship for G5–G8. To enhance the water retention capacity and extend the service life of dual-substrate-layer extensive green roofs, the mixture of activated charcoal and/or pumice with expanded perlite and/or vermiculite is more suitable as the adsorption substrate than the mixture containing lava and/or zeolite.

INTRODUCTION

Due to the high proportion of total impervious surfaces in urban areas, the roof area has attracted more attention, and interest in green roofs is increasing in many countries (VanWoert et al. 2005; Carson et al. 2013). Green roofs are considered to have multiple potentially significant benefits, especially in urban areas, including reducing and attenuating rainwater runoff (Nawaz et al. 2015), decreasing the urban heat island effect and building energy consumption (Fang 2008), providing wildlife habitats and biodiversity enhancement (Dunnett et al. 2008), capturing air pollution (Speak et al. 2012), obtaining noise reduction (Van Renterghem & Booteldooren 2009), and possessing considerable aesthetic value. Green roofs are typically divided into two main engineering categories: extensive and semi-intensive/intensive. Extensive green roofs have a thin substrate layer (less than 200 mm, often between 50 and 150 mm) with low level planting, typically sedum or lawn, and they can be very lightweight in structure (FLL 2008; Morgan et al. 2013). Intensive green roofs have a thicker substrate layer (more than 150 mm, often at least 200 mm) to allow deeper rooting plants, such as shrubs and trees, to survive (FLL 2008; Morgan et al. 2013). Semi-intensive roofs have a substrate layer thicker than the extensive but thinner than the intensive, which are vegetated with lawns, ground covering plants and low shrubs (FLL 2008; Berndtsson 2010). Extensive roofs are the preferred option for retrofitting onto existing buildings as well as incorporating into new developments because of the advantages of relatively light weight, limited and low-skill labor requirements, almost free maintenance and a higher survival rate (Carson et al. 2013; Nawaz et al. 2015). Since semi-intensive/intensive roofs require frequent maintenance including cutting, watering, and fertilization (FLL 2008; Berndtsson 2010), these roofs are more often used when a rooftop garden accessible to the building occupants is desired (Morgan et al. 2013). Despite the differences between green roof types, they generally all contain the same principal components, including a waterproof and root-resistant membrane, a drainage layer, a filtering layer, a growing substrate (medium) layer and vegetation (Berndtsson 2010).

The capacity of green roofs for runoff retention depends on many factors, such as the rainfall characteristics (rainfall depth, duration and intensity), green roof design (vegetation, type and depth of substrate layer, drainage, slope of the green roof), the antecedent dry weather period, the age of the green roof and the local climate (Berndtsson 2010; Seidl et al. 2013). Many studies agree that it is the substrate properties (depth, type and moisture conditions before a rain event) that have the major influence on the water retention capacity of green roofs (Monterusso et al. 2004; VanWoert et al. 2005; Dunnett et al. 2008). Green roof substrates should be lightweight, well-draining, primarily inorganic, and capable of supporting good plant growth (Morgan et al. 2013). Both vermiculite and expanded perlite are common substrate/media for green roofs because of the advantages of relatively light weight, good permeability and water holding capacity (DB440300 2009; DB11 2015). Inorganic low-density materials with large water-holding capacity, such as pumice, scoria or lava, have also been used as a substrate in green roofs (Emilsson et al. 2007). The common growing medium in existing green roofs is a single layer of a mixture of organic and inorganic substrates (FLL 2008). Organic fertilizer may also be added appropriately (Emilsson et al. 2007; FLL 2008).

Studies have revealed green roofs to be both sources of water contaminants, as well as sinks. The most common impact on green roof runoff quality comes from N and P (Speak et al. 2014). High nutrient levels have been frequently found in green roof runoff (Monterusso et al. 2004; Teemusk & Mander 2007; Vijayaraghavan et al. 2012) with the N and P amounts being directly related to organic matter content (Speak et al. 2014). Research efforts have been made to prevent nutrients leaching from green roofs, including amending green roof soil with biochar (Beck et al. 2011), incorporating a brown-seaweed (Turbinaria conoides) in the growth substrate (Vijayaraghavan & Joshi 2015), using controlled release fertilizers (Emilsson et al. 2007). Wang et al. (2013) and Gong et al. (2014) provided a type of dual-substrate-layer extensive green roof, in which the substrate consisted of a nutrition layer (a commercial controlled-release fertilizer or turfy soil) and an adsorption layer (a mixture of perlite and vermiculite). Wang et al. (2013) found the dual-substrate-layer green roofs effective at promoting the green roof runoff quality, and Gong et al. (2014) revealed that the runoff water quality (TN, NH4+-N, NO3-N and TP) would be improved to some extent with the ageing of dual-substrate-layer green roofs. This study is a follow-up of these two previous studies on dual-substrate layer extensive green roofs, and the hydrological performance of the green roofs using porous inert substrates with high sorption capacities (activated charcoal, zeolite, pumice, lava, vermiculite and expanded perlite) as the adsorption substrates was investigated.

The specific objectives of this study were to (1) assess the ability of the dual-substrate-layer extensive green roofs to retain and detain rainfall from individual precipitation events by comparing with a traditional single-substrate-layer extensive green roof, (2) compare the rainfall–runoff response of the dual-substrate-layer extensive green roofs with that of a nearby conventional bare roof for individual precipitation events, and (3) conduct regression analysis to develop a relatively accurate model, which can predict the retention (%) of the dual-substrate-layer extensive green roofs for cumulative rainfall depth. It is expected that the results from this study will offer a reference for the design of extensive green roofs.

MATERIALS AND METHODS

Green roof design

Eight pilot-scale extensive green roof assemblies, namely, G1, G2, G3, G4, G5, G6, G7 and G8, were established in early June on the roof of the Environmental Experimental building on the Tianjin University campus, Tianjin, China. The length and width of each green roof were 0.7 m and 0.5 m, respectively, and the height from the ground was 6.0 m. All assemblies were placed on a 2 ° slope to coincide with the roof drainage slope. As illustrated in Figure 1, each assembly consisted of the following layers: a bituminous base roof, a waterproof and root-resistant membrane, a commercial plastic storage drainage plate (25 mm thick), a geotextile filter layer (6 mm thick) to prevent small particles from entering the drainage layer, a substrate layer (100–200 mm thick) and a vegetation layer on the top. The reference roof is a bituminous membrane roof.
Figure 1

Cross section of the green roof.

Figure 1

Cross section of the green roof.

The substrates in G1–G7 were of dual layer, and the dual substrate layer consisted of two parts: (1) the uppermost was a nutrition substrate layer (50 mm thick) for plant growth; and (2) the lower was an adsorption substrate layer (50–100 mm thick) for water retention and preventing pollutants (especially nitrogen and phosphorus) leaching from the nutrition layer. BAOLVSU and turfy soil (1:1) were used as the nutrition substrate. BAOLVSU is a commercial substrate especially made for green roofs. The fertility of BAOLVSU can be slowly released, which contributes to the prevention of nutrient leaching into the green roof outflow. However, compared to turfy soil, the heat preservation property of BAOLVSU in winter is relatively poor, and the cost is slightly higher. Therefore, the nutrition substrate in this study combined BAOLVSU and turfy soil to complement each other. The adsorption substrates used in the green roofs of G1–G7 were activated charcoal, zeolite, pumice, lava, expanded perlite and vermiculite. In environmental protection, these materials are generally used as filtering media and adsorbents. The substrate in G8 was of a single layer to be compared with the performance of G1–G7, and the mixture of expanded perlite, turfy soil and vermiculite, which is a very common substrate/media for a traditional green roof (DB11 2015), was selected as the substrate for G8. The water-saturated densities (kg/m3) of the substrates in G1-G8 varied from 570 to 970 and the corresponding weight (kg substrate/m2 roof) varied between 85.5 and 145.5, which met the building roof's weight restrictions (250 kg/m2 roof) in this study. The substrate properties in the green roofs are provided in Table 1.

Table 1

Substrate properties in the green roofs of G1–G8

MaterialParticle size (mm)Porosity (%)Packing density (kg/m3)
BAOLVSU 45 ± 5 360 ± 10 
Turfy soil – 60 ± 5 750 ± 15 
Pumice 25 ± 5 76 ± 4 350 ± 10 
Activated charcoal 10 ± 2 90 ± 2 560 ± 14 
Zeolite 10 ± 1 43 ± 5 1,470 ± 25 
Lava 25 ± 5 40 ± 2 820 ± 20 
Expanded perlite 8 ± 3 50 ± 2 150 ± 5 
Vermiculite 10 ± 2 85 ± 5 300 ± 4 
No. Nutrition substrate Adsorption substrate Total void fraction (%) 
Material Depth (mm) Void fraction (%) Material Depth (mm) Void fraction (%) 
G1 BAOLVSU and turfy soil (1:1) 50 45 ± 5 Pumice, activated charcoal and zeolite (4:4:2) 50 50 ± 5 48 ± 5 
G2 BAOLVSU and turfy soil (1:1) 50 45 ± 5 Pumice, activated charcoal and zeolite (4:4:2) 100 50 ± 5 48 ± 5 
G3 BAOLVSU and turfy soil (1:1) 50 45 ± 5 Lava, activated charcoal and zeolite (4:4:2) 100 50 ± 5 48 ± 5 
G4 BAOLVSU and turfy soil (1:1) 50 45 ± 5 Lava, pumice and zeolite (4:4:2) 100 55 ± 4 52 ± 3 
G5 BAOLVSU and turfy soil (1:1) 50 45 ± 5 Expanded perlite, activated charcoal and vermiculite (2.5:5:2.5) 100 45 ± 3 45 ± 3 
G6 BAOLVSU and turfy soil (1:1) 50 45 ± 5 Expanded perlite, activated charcoal and vermiculite (1.5:7:1.5) 100 45 ± 3 45 ± 3 
G7 BAOLVSU and turfy soil (1:1) 50 45 ± 5 Pumice and activated charcoal (1:1) 100 50 ± 5 48 ± 5 
G8 Expanded perlite, turfy soil and vermiculite (2.5:5:2.5) Depth = 200 mm Void fraction = 45 ± 3 45 ± 2 
MaterialParticle size (mm)Porosity (%)Packing density (kg/m3)
BAOLVSU 45 ± 5 360 ± 10 
Turfy soil – 60 ± 5 750 ± 15 
Pumice 25 ± 5 76 ± 4 350 ± 10 
Activated charcoal 10 ± 2 90 ± 2 560 ± 14 
Zeolite 10 ± 1 43 ± 5 1,470 ± 25 
Lava 25 ± 5 40 ± 2 820 ± 20 
Expanded perlite 8 ± 3 50 ± 2 150 ± 5 
Vermiculite 10 ± 2 85 ± 5 300 ± 4 
No. Nutrition substrate Adsorption substrate Total void fraction (%) 
Material Depth (mm) Void fraction (%) Material Depth (mm) Void fraction (%) 
G1 BAOLVSU and turfy soil (1:1) 50 45 ± 5 Pumice, activated charcoal and zeolite (4:4:2) 50 50 ± 5 48 ± 5 
G2 BAOLVSU and turfy soil (1:1) 50 45 ± 5 Pumice, activated charcoal and zeolite (4:4:2) 100 50 ± 5 48 ± 5 
G3 BAOLVSU and turfy soil (1:1) 50 45 ± 5 Lava, activated charcoal and zeolite (4:4:2) 100 50 ± 5 48 ± 5 
G4 BAOLVSU and turfy soil (1:1) 50 45 ± 5 Lava, pumice and zeolite (4:4:2) 100 55 ± 4 52 ± 3 
G5 BAOLVSU and turfy soil (1:1) 50 45 ± 5 Expanded perlite, activated charcoal and vermiculite (2.5:5:2.5) 100 45 ± 3 45 ± 3 
G6 BAOLVSU and turfy soil (1:1) 50 45 ± 5 Expanded perlite, activated charcoal and vermiculite (1.5:7:1.5) 100 45 ± 3 45 ± 3 
G7 BAOLVSU and turfy soil (1:1) 50 45 ± 5 Pumice and activated charcoal (1:1) 100 50 ± 5 48 ± 5 
G8 Expanded perlite, turfy soil and vermiculite (2.5:5:2.5) Depth = 200 mm Void fraction = 45 ± 3 45 ± 2 

Sedum lineare, which is a widely used plant species in green roofs, was selected in this study because of its ability to survive in drought conditions, extreme temperature and low nutrient conditions (Cook-Patton & Bauerle 2012; Wang et al. 2013). Once planted, the vegetation kept growing naturally without either fertilization or watering. After one month, the vegetation was in a good growing condition with coverage reaching over 90% in G1–G7 and 80% in G8, and the vegetation conditions were similar in G1–G7 roofs. Moreover, the plants have also survived the dry and cold seasons. The stems and leaves of Sedum turned from green to dark brown in the cold season (from 0 °C to −14 °C), but began to bud once the substrates thawed and covered the green roofs in early spring.

Simulated rains

Tianjin, China, located on the eastern coast of mid-latitude Eurasia where the East Asian monsoon prevails, enjoys a warm, temperate, semi-humid, monsoon climate. The average annual rainfall in Tianjin varies between 520 mm and 660 mm, and the precipitation in June, July and August accounts for approximately 75% of the total. To study the hydrological performance of the extensive green roofs under different rainfall events, six types of simulated rains were performed on roofs G1–G8 using a man-made, movable, auto-controlled artificial rain generator. Each type of simulated rain was repeated four times, and thus the green roofs experienced a total of 24 simulated rain events. The simulated rains (Table 2) were decided by the latest rainstorm intensity formula of Tianjin, China, as follows:
formula
1
where q is the rainstorm intensity (L/s·ha), P is the design recurrence period (year), and t is the rainfall duration (min).
Table 2

The pattern of simulated rains of R1–R6

No.Design recurrence period (year)Duration (min)Intensity
Total rainfall depth (mm)
(L/(s ha))(mm/min)
R1 60 119.95 0.72 43.2 
R2 60 152.26 0.91 54.6 
R3 10 60 212.43 1.27 76.2 
R4 20 60 241.81 1.45 87.0 
R5 50 40 355.15 2.13 85.2 
R6 100 30 480.81 2.88 86.4 
No.Design recurrence period (year)Duration (min)Intensity
Total rainfall depth (mm)
(L/(s ha))(mm/min)
R1 60 119.95 0.72 43.2 
R2 60 152.26 0.91 54.6 
R3 10 60 212.43 1.27 76.2 
R4 20 60 241.81 1.45 87.0 
R5 50 40 355.15 2.13 85.2 
R6 100 30 480.81 2.88 86.4 

Sampling and analysis

The entire experimental period was from early June 2015 to late October 2015. During the investigation period, the recorded average minimum and maximum temperatures were 18 °C and 32 °C, respectively, and the relative humidity of the air had an average of 59%. There were no large fluctuations in temperature among the different sampling time periods. Before each experiment, all green roofs were kept dry for approximately 4–10 days to ensure that the substrate moisture was less than 20%. The substrate moisture was measured by a soil-moisture content analyzer (provided by Shandong Hengmei Electronic Technology Co., Ltd, Weifang City, Shandong Province, China). When the runoff started to occur during a simulated rain event, runoff samples were collected from the green roofs through sampling orifices (Figure 1) using pre-cleaned plastic cans every 5 or 10 min, and runoff volume was measured until runoff finished (state of drip leakage).

The quantity retention rate (Q) and runoff coefficient (RC) were used to evaluate the runoff-retaining capacity of a green roof and were calculated as follows:
formula
2
formula
3
where Hi is the total rainfall depth (mm), and He is the total runoff depth (mm).
Given that the water retention capacity of the nutrition substrate layer was weak and the vegetation conditions were similar in G1–G7 roofs, the maximum potential rainfall storage depth (MPRSD) per unit height of adsorption substrate layer was used to evaluate and compare the maximum water holding capacity of the dual-substrate-layer green roofs, which was calculated as follows:
formula
4
where Hc is the recorded cumulative rainfall depth (mm) and Qc is the corresponding quantity retention rate; Had is the adsorption substrate layer depth (cm) in a green roof.
The MPRSD per unit height of the single-substrate-layer green roof (G8) was calculated as follows:
formula
5
where H is the substrate layer depth (cm) in a green roof.

To investigate the statistical difference of the runoff quantity performance from the pilot-scale green roofs, tests for significant differences among the roofs were performed at the significance level of 0.05 using one-way analysis of variance (ANOVA) and were followed by a Duncan post hoc test (p < 0.05). Regression analysis was undertaken to develop predictive relationships between cumulative rainfall depth (mm) and green roof retention (%), and the strength of correlation was indicated by the coefficient of determination (R2). The hydrological model's accuracy was evaluated by comparing the modeled stormwater retention values with the observed performance data under one natural rain event with larger total rainfall depth during the investigation period. The duration, intensity and total rainfall depth of the natural rain were approximately 10 hours, 0.49–7.15 mm/h (the average was 4.21 mm/h) and 42.1 mm, respectively.

All of the statistical analyses were performed with SPSS and Origin software.

RESULTS AND DISCUSSION

Overall hydrological performance of the green roofs

As listed in Table 3, during the investigation period, rainfall runoff attenuation of individual precipitation events ranged from 14% to 37% for G1, 27% to 62% for G2, 21% to 49% for G3, 26% to 57% for G4, 48% to 82% for G5, 36% to 77% for G6, 35% to 70% for G7, and 50% to 89% for G8, showing large fluctuations in stormwater retention capacity among the green roofs with different substrates under different rainfall events. Based on the average retention of the total rainfall for R1–R6, G8 (65%) demonstrated the best stormwater retention performance in this study, followed by G5 (60%), G6 (49%), G7 (45%), G2 (38%), G4 (36%), G3 (30%) and G1 (22%). However, based on the MPRSD per unit height, the sequence of the rainfall attenuation performance turned out to be G5 > G6 > G7 ≈ G1 > G2 > G4 ≈ G8 > G3 (see Tables 4 and 5).

Table 3

Overall stormwater retention for the green roofs (G1–G8) under the six types of simulated rains (R1–R6)

Total rainfall (mm)G1 Q (%)G2 Q (%)G3 Q (%)G4 Q (%)G5 Q (%)G6 Q (%)G7 Q (%)G8 Q (%)
43.2 (R1) 37 62 49 57 82 77 70 89 
54.6 (R2) 29 51 39 46 73 63 55 80 
76.2 (R3) 20 35 27 31 61 46 39 66 
87.0 (R4) 15 28 23 27 48 37 36 54 
85.2 (R5) 15 28 23 27 49 37 36 52 
86.4 (R6) 14 27 21 26 48 36 35 50 
Total rainfall (mm)G1 Q (%)G2 Q (%)G3 Q (%)G4 Q (%)G5 Q (%)G6 Q (%)G7 Q (%)G8 Q (%)
43.2 (R1) 37 62 49 57 82 77 70 89 
54.6 (R2) 29 51 39 46 73 63 55 80 
76.2 (R3) 20 35 27 31 61 46 39 66 
87.0 (R4) 15 28 23 27 48 37 36 54 
85.2 (R5) 15 28 23 27 49 37 36 52 
86.4 (R6) 14 27 21 26 48 36 35 50 
Table 4

MPRSD per unit height of (adsorption) substrate layer of G1–G8 roofs

G1G2G3G4G5G6G7G8
3.2 2.9 2.1 2.5 4.6 3.6 3.2 2.5 
G1G2G3G4G5G6G7G8
3.2 2.9 2.1 2.5 4.6 3.6 3.2 2.5 
Table 5

ANOVA of the MPRSD per unit height for the green roofs of G1–G8 (α level is 0.05)

SourceSum of squaresDegree of freedomMean squareF-ratioSignificance
Treatment 25.24  3.61  1,730.85 0.000 
Error 0.083 40  0.002    
Total 24.33 47      
Duncan Post Hoc Tests (p < 0.05) 
Treatment No. Subset
 
1 2 3 4 5 6 
G3 2.1      
G8  2.5     
G4  2.5     
G2   2.9    
G7    3.2   
G1    3.2   
G6     3.6  
G5      4.6 
Sig.  1.000 0.213 1.000 0.531 1.000 1.000 
SourceSum of squaresDegree of freedomMean squareF-ratioSignificance
Treatment 25.24  3.61  1,730.85 0.000 
Error 0.083 40  0.002    
Total 24.33 47      
Duncan Post Hoc Tests (p < 0.05) 
Treatment No. Subset
 
1 2 3 4 5 6 
G3 2.1      
G8  2.5     
G4  2.5     
G2   2.9    
G7    3.2   
G1    3.2   
G6     3.6  
G5      4.6 
Sig.  1.000 0.213 1.000 0.531 1.000 1.000 

Compared to the retention value (15%–83% with an average of 57%) observed from previous extensive green roof studies (Nawaz et al. 2015), the total percent retention of the dual-substrate-layer extensive green roofs in this study (14%–82% with an average of 43%) was relatively good, especially considering the larger rainfall events of R1–R6. Generally, the majority of the rainwater retained and detained by the green roof is stored in the substrate layer (Bianchini & Hewage 2012). Therefore, substrate depth is thought to be the primary factor influencing water retention for horizontal sedum roofs (Carson et al. 2013). This explains why the 50 mm adsorption substrate depth (G1) retained less rainwater than the deeper depths (14%–37% versus 21%–89%). Additionally, the data in Table 3 agreed with the general expectation that the total percent retention of rainfall by a green roof decreased as event precipitation increased.

Given that the substrate depth of G8 was deeper than that of G1–G7 (see Table 1), it was more reasonable to evaluate and compare the runoff-retaining capacity of the green roofs by MPRSD per unit height. From the ANOVA of MPRSD per unit height of G1–G8 (Table 5) it can be inferred that substrate type also has major influence on the water retention capacity of green roofs. A similar conclusion can be found in previous studies (VanWoert et al. 2005; Dunnett et al. 2008). Vermiculite and expanded perlite have strong powers of water absorption (DB440300 2009; DB11 2015), which was further demonstrated by the higher water retention capacity of G5 and G6 compared to G2, G3 and G4 (see Table 3 and Figure 2). In comparison to G3, the higher stormwater retention of G2 and G7 (see Table 3 and Figure 2) demonstrated that activated charcoal and pumice also possess relatively high water absorption capacity. Comparing the runoff-retaining capacity of G2 to that of G3, G4 and G7 (see Table 3 and Figure 2), it can be concluded that the water absorption capability of zeolite and lava was poorer than that of activated charcoal and pumice. The total rainfall depth of R4, R5 and R6 was similar, but the intensity kept increasing (see Table 2). However, almost no changes were found in the total percent retention of G1–G7 with the increased rainfall intensity, while the retention value of G8 was reduced by approximately 4% (see Table 3). This result demonstrated the stronger buffering capacity and faster water absorption speed of activated charcoal and pumice.
Figure 2

Rainfall–runoff response from the bare roof and G1–G4 roofs for R1 (A), from G5–G8 roofs for R1 (a), from the bare roof and G1–G4 roofs for R2 (B), from G5–G8 roofs for R2 (b), from the bare roof and G1–G4 roofs for R3 (C), from G5–G8 roofs for R3 (c), from the bare roof and G1–G4 roofs for R4 (D), from G5–G8 roofs for R4 (d), from the bare roof and G1–G4 roofs for R5 (E), from G5–G8 roofs for R5 (e), from the bare roof and G1–G4 roofs for R6 (F), from G5–G8 roofs for R6 (f). R1–R6 are the six types of simulated rains. The runoff depth (mm) from the roofs is the mean data from four repetitions of each simulated rain event. Standard deviations were between 0.96 and 5.59 (n = 4).

Figure 2

Rainfall–runoff response from the bare roof and G1–G4 roofs for R1 (A), from G5–G8 roofs for R1 (a), from the bare roof and G1–G4 roofs for R2 (B), from G5–G8 roofs for R2 (b), from the bare roof and G1–G4 roofs for R3 (C), from G5–G8 roofs for R3 (c), from the bare roof and G1–G4 roofs for R4 (D), from G5–G8 roofs for R4 (d), from the bare roof and G1–G4 roofs for R5 (E), from G5–G8 roofs for R5 (e), from the bare roof and G1–G4 roofs for R6 (F), from G5–G8 roofs for R6 (f). R1–R6 are the six types of simulated rains. The runoff depth (mm) from the roofs is the mean data from four repetitions of each simulated rain event. Standard deviations were between 0.96 and 5.59 (n = 4).

Due to the poor wear resistance, the physical state of vermiculite and expanded perlite could change over time, and some of the expanded perlite may crumble into powder even during transport, which would reduce the breathability and drainage of the materials and thus the water retention capacity of green roofs. Activated charcoal, zeolite, pumice and lava possess high wear resistance, which could help improve the service life of green roofs. Moreover, both activated charcoal and pumice have very well-developed pore structures, which endow the materials with high water absorption capability and fast absorption speed. Therefore, it can be concluded that a mixture of activated charcoal and/or pumice with expanded perlite and/or vermiculite was more suitable as the adsorption substrate to enhance the water retention capacity and extend the service life of the dual-substrate-layer extensive green roofs. The better runoff-retaining capacity of G5, G6 and G7 based on MPRSD per unit height further proved the above conclusion. Additionally, the plastic storage drainage plate employed in this study is expected to store some rainwater and contribute to the relatively good stormwater retention performance of the green roofs.

Comparison of rainfall–runoff responses

As illustrated in Figure 2, total rainfall depth of the six types of simulated rains (R1–R6) ranged from 43.2 to 87 mm, while the corresponding runoff depth varied between 4.73 and 55.2 mm for the investigated green roofs. Runoff quantity from the green roofs was obviously much lower than that from the conventional bare roof (39.3–81.5 mm). Under R1–R6, the green roofs of G1–G8 delayed the initiation of runoff by an average of 3–21 minutes (Table 6), while the bare roof produced runoff almost as soon as it began to rain. Compared to the average RC of G1–G8 roofs (0.35–0.78, Table 6), the average RC of the bare roof under R1–R6 (0.92) was much higher. Calculated from the product of rainfall intensity and time of runoff initiation, the average maximum cumulative rainfall depth with 100% retention varied from 4.8 mm for G1 to 27.9 mm for G8 (Table 6), showing large fluctuations in stormwater retention capacity among the different green roofs.

Table 6

Characteristic hydrological parameters of the green roofs (G1–G8) under the six types of simulated rains (R1–R6)

No.Average time for delaying the initiation of runoff (min)Average time for discharging runoff longer than the rainfall duration (min)Average maximum cumulative rainfall depth with 100% retention (mm)Average RC
G1 4.8 0.78 
G2 12.6 0.61 
G3 6.5 0.74 
G4 8.9 0.70 
G5 14 18.7 0.40 
G6 12 17.3 0.51 
G7 10 15.2 0.55 
G8 21 27.9 0.35 
No.Average time for delaying the initiation of runoff (min)Average time for discharging runoff longer than the rainfall duration (min)Average maximum cumulative rainfall depth with 100% retention (mm)Average RC
G1 4.8 0.78 
G2 12.6 0.61 
G3 6.5 0.74 
G4 8.9 0.70 
G5 14 18.7 0.40 
G6 12 17.3 0.51 
G7 10 15.2 0.55 
G8 21 27.9 0.35 

Compared to conventional roofs, green roofs can delay the initiation of runoff, reduce total runoff volumes, reduce peak runoff rates and discharge runoff over a longer period of time (Berndtsson 2010; Nawaz et al. 2015). It can be found from Figure 2 in conjunction with Table 3 that under the same precipitation event, the higher the retention capacity of the green roof, the longer the time at which runoff began to appear and the deeper the maximum cumulative rainfall depth with 100% retention. These findings were in line with expectations. Moreover, Figure 2 demonstrated that the greater the rainfall intensity, the shorter the time at which runoff began to appear from a green roof. However, there seemed to be a weak relationship between the green roof retention capacity and the time consumption for draining runoff after the precipitation, as the average time for discharging runoff longer than the rainfall duration was similar among the G1–G8 roofs (7–9 min, Table 6). Additionally, it can be more clearly observed from Figure 2 that with the increase of total rainfall depth, the retention values of a green roof would decrease and the time for draining runoff would increase. The longer time of runoff initiation, deeper maximum cumulative rainfall depth with 100% retention and lower RC of G8 roof (200-mm adsorption substrate depth) further proved that substrate depth is the primary factor influencing water retention for horizontal sedum roofs.

Regression analysis between cumulative rainfall depth and green roof retention

From the above analysis, it can be concluded that the total percent retention of G1–G8 was determined mainly by the cumulative rainfall depth and runoff-retaining capacity of the green roof, and it had little to do with the rainfall intensity when the total rainfall depth was constant or similar. Since a large number of factors influence the green roof's hydrological performance, there is large variation in a green roof's hydrological performance within studies, and even the same green roof would present different hydrological performances in different regions due to the site-specific factors. Therefore, the hydrological models were developed mainly to quantitatively analyze the relationship between cumulative rainfall depth (mm) and stormwater retention (%) of the dual-substrate-layer extensive green roofs in this study, and they were not a representation of general hydrological models of green roofs.

To prevent lower-bound overestimation of the retention values, especially for small rain events with zero runoff, a step-shaped equation was adopted to express the green roof's hydrological performance with rainfall depth. As illustrated in Figure 3, the resulting regression equations were two-step-shaped. Before runoff began to appear from the green roof, the retention value was 100%. Due to the different water retaining capacities, the maximum cumulative rainfall depth with zero runoff varied among green roofs (Table 6 and Figure 3). When runoff began to appear, the observed retention values displayed a logarithmic relationship (R2 > 0.98) between cumulative rainfall depth and stormwater retention for G1–G4 and a linear relationship (R2 > 0.98) for G5–G8. Rainfall depths of the simulated rains (43.2–87 mm) in this study were below 90 mm, and predictions beyond this value might not accurately represent the green roof's retention performance. Therefore, the equation of each green roof is applicable for rainfall depths up to 90 mm. When the rainfall depth exceeds 90 mm, a hypothesis can be made that the green roofs have reached their maximum water holding capacity, and the retention value can be predicted by a formula, as follows:
formula
6
Figure 3

Regression analysis between cumulative rainfall depth (mm) and retention rate (%) of G1–G8 green roofs. Standard deviations were between 0.0142 and 0.0743.

Figure 3

Regression analysis between cumulative rainfall depth (mm) and retention rate (%) of G1–G8 green roofs. Standard deviations were between 0.0142 and 0.0743.

As shown in Figure 4, the observed total percent retention values under the natural rain were 84% for G5 and 78% for G6, while the modeled stormwater retention values were 83% for G5 and 76% for G6, which were smaller than those observed by an average of 2%. This finding may be because of the longer duration (10 hours) and weaker intensity (0.49–7.15 mm/h with an average of 4.21 mm/h) of the natural rain compared to the simulated rains, which could increase the contact time between the rainwater and the substrate in green roofs and thus the retention time. Moreover, it can be clearly observed from Figure 3 that once runoff appeared from the green roofs, the retention values decreased at a relatively rapid speed with the increase of cumulative rainfall depth, especially in G5–G8 roofs. This result demonstrated the good drainage of the porous inert substrates used in this study.
Figure 4

Model verification for G5 and G6 green roofs under one natural rain event (the duration was 10 hours, intensity was 0.49–7.15 mm/h with an average of 4.21 mm/h, and total rainfall depth was 42.1 mm.). Standard deviations were between 0.0161 and 0.0682.

Figure 4

Model verification for G5 and G6 green roofs under one natural rain event (the duration was 10 hours, intensity was 0.49–7.15 mm/h with an average of 4.21 mm/h, and total rainfall depth was 42.1 mm.). Standard deviations were between 0.0161 and 0.0682.

Although the models developed could accurately predict to a certain extent the dual-substrate-layer green roof's hydrological performance in response to a certain precipitation event, it is far from enough to generalize the green roof's performance based on just a limited number of simulated rain events. The multi-year hydrological performance from the dual-substrate-layer extensive green roofs in urban environments needs to be further investigated in the future.

CONCLUSIONS

This study investigated the hydrological performance of dual-substrate-layer extensive green roofs using porous inert substrates with high sorption capacities. The results revealed that compared to the single-substrate-layer extensive green roof (G8), the dual-substrate-layer extensive green roofs (G1–G7) possessed better runoff-retaining capacity based on the MPRSD per unit height of adsorption substrate layer, and they were more capable of supporting good plant growth. Regression analysis showed that there was a logarithmic relationship between cumulative rainfall depth (mm) with non-zero runoff and stormwater retention (%) for G1–G4 roofs and a linear relationship for G5–G8 roofs. To enhance the water retention capacity and extend the service life of dual-substrate-layer extensive green roofs, the mixture of activated charcoal and/or pumice with expanded perlite and/or vermiculite is more suitable as the adsorption substrate than the mixture containing lava and/or zeolite.

ACKNOWLEDGEMENTS

The authors gratefully acknowledge the financial support from The Key Special Program on the S&T for the Pollution Control and Treatment of Water Bodies of China (2014ZX07203-009).

REFERENCES

Beck
D. A.
Johnson
G. R.
Spolek
G. A.
2011
Amending greenroof soil with biochar to affect runoff water quantity and quality
.
Environmental Pollution
159
,
2111
2118
.
Bianchini
F.
Hewage
K.
2012
How green are the green roofs? Lifecycle analysis of green roof materials
.
Building and Environment
48
(
1
),
57
65
.
Carson
T. B.
Marasco
D. E.
Culligan
P. J.
McGillis
W. R.
2013
Hydrological performance of extensive green roofs in New York City: observations and multi–year modelling of three full–scale systems
.
Environmental Research Letters
8
(
2
),
024036
(13 pp)
.
Cook-Patton
S. C.
Bauerle
T. L.
2012
Potential benefits of plant diversity on vegetated roofs: a literature review
.
Journal of Environmental Management
106
,
85
92
.
DB440300
2009
Code for the design of roof greening DB440300/T 37-2009 (in Chinese)
.
DB11
2015
Specification for roof greening DB11/T 281-2015 (in Chinese)
.
FLL
2008
Guidelines for the Planning, Construction and Maintenance of Green Roofing
(
English edn
).
Forschungsgesellschaft Landschaftsentwicklung Landschaftsbau E.V
.
Gong
K. N.
Wu
Q.
Peng
S.
Zhao
X. H.
Wang
X. C.
2014
Research on the characteristics of the water quality of rainwater runoff from green roofs
.
Water Science and Technology
70
(
7
),
1205
1210
.
Monterusso
M. A.
Rowe
D. B.
Rugh
C. L.
Russell
D. K.
2004
Runoff water quantity and quality from green roof systems
.
Acta Horticulturae
639
,
369
376
.
Morgan
S.
Celik
S.
Retzlaff
W.
2013
Green roof storm–water runoff quantity and quality
.
Journal of Environmental Engineering
139
(
4
),
471
478
.
Seidl
M.
Gromaire
M.
Saad
M.
De Gouvello
B.
2013
Effect of substrate depth and rain–event history on the pollutant abatement of green roofs
.
Environmental Pollution
183
,
195
203
.
Speak
A. F.
Rothwell
J. J.
Lindley
S. J.
Smith
C. L.
2012
Urban particulate pollution reduction by four species of green roof vegetation in a UK city
.
Atmospheric Environment
61
(
10
),
733
746
.
Speak
A. F.
Rothwell
J. J.
Lindley
S. J.
Smith
C. L.
2014
Metal and nutrient dynamics on an aged intensive green roof
.
Environmental Pollution
184
,
33
43
.
Van Renterghem
T.
Booteldooren
D.
2009
Reducing the acoustical facade load from traffic with green roofs
.
Building and Environment
44
(
5
),
1081
1087
.
VanWoert
N. D.
Rowe
D. B.
Andresen
J. A.
Rugh
C. L.
Fernandez
R. T.
Xiao
L.
2005
Green roof stormwater retention: effects of roof surface, slope, and media depth
.
Journal of Environmental Quality
34
(
3
),
1036
1044
.
Vijayaraghavan
K.
Joshi
U. M.
Balasubramanian
R.
2012
A field study to evaluate runoff quality from green roofs
.
Water Research
46
(
4
),
1337
1345
.
Wang
X. C.
Zhao
X. H.
Peng
C. R.
Zhang
X. B.
Wang
J. H.
2013
A field study to evaluate the impact of different factors on the nutrient pollutant concentrations in green roof runoff
.
Water Science and Technology
68
(
12
),
2691
2697
.