Hydrologic impacts of mat-based retention and detention layers within extensive vegetated roof assemblies Open Access

First-generation extensive green roof systems included only vegetation, growing media (GM), and drainage materials. However, green roof designs are increasingly incorporating lightweight GM alternatives to enhance their retention and detention capabilities. This study evaluates the hydrologic impacts of vegetated roof assemblies (VRAs) that incorporate materials, such as fleece, mineral wool, and a combined reservoir–detention system, under natural precipitation conditions in Toronto, Ontario. Over a year, discharge from testbeds was measured and compared to a traditional green roof and an impervious gravel ballast roof. During the growing season, the VRAs provided similar stormwater retention rates. Green roofs also provided significant detention benefits, reducing peak discharge by 94% and delaying, extending, and increasing discharge delay and duration compared to gravel roofs. Winter performance showed reduced effectiveness, increased peak flows, and shorter discharge delays and durations. Overall, an average VRA runoff coefficient of 0.67 was observed in winter, compared to 0.17 during the warm season. This work demonstrates that although adding retention layers improves the hydrologic performance of green roof systems to varying degrees during warmer months, traditional green roof assemblies may still provide superior annual precipitation volume reduction when winter conditions are considered.

Urbanization often limits the space available for green infrastructure at the ground level. As a result, city planners and designers are looking vertically for opportunities to integrate green spaces within the built environment. Building roofs, which are typically unutilized spaces, are ideal spaces for urban greening. Vegetated roof assemblies (VRAs), also known as green roofs (GRs), mimic predevelopment hydrology by retaining and detaining stormwater on the roof and creating opportunities for evapotranspiration. Cities, including Toronto, New York, Seattle, San Francisco, and Washington, D.C., have enacted legislation that mandates or incentivizes GR on new commercial, institutional, and residential development projects (City of Toronto 2017; Green Roof for Healthy Cities 2023).

The traditional extensive GR layering profile, from top to bottom, includes a vegetation cover, aggregate-based growing media (GM), and drainage layers (Credit Valley Conservation Authority (CV) & Toronto and Region Conservation Authority (TRCA) 2010; Carson et al. 2013; Hakimdavar et al. 2014; Hill et al. 2017). Given the harsh growing conditions present on roofs, the selection of vegetation for GRs is primarily driven by plant survival and ease of maintenance. Consequently, sedum plants that can withstand extreme weather conditions and high winds are drought-resistant and have high coverage percentages (Li & Babcock 2014). As a result, they are extremely popular on commercial VRAs.

In a GR, the substrate layer is predominantly responsible for the storage of stormwater. The Green Roof Guidelines, developed by FLL (The Society for Landscape Development and Landscaping Research), recommend the substrate content and properties for GR and vegetation types (FLL – Landscape Development and Landscaping Research Society 2018). These guidelines, developed in Germany, inform and serve as the foundation for most North American GR design guidelines. The composition of bulk material mixes for soil with mineral aggregates – such as sand, expanded shale, and crushed brick – varies based on the desired roof functionality. Substrates with a high percentage of aggregates promote fast drainage, while soils with more organic matter may be more desirable to promote vegetation health. Aggregate content is also important for ballasting the GR system on roofs, especially for high commercial towers with intense wind uplift pressures. Substrates inevitably evolve overtime because of erosion (Liao et al. 2022) and pedogenesis (Bouzouidja et al. 2018). The impact of substrate aging on stormwater retention is difficult to predict as it is influenced by climatic conditions, vegetation growth, and substrate composition (D'Ambrosio et al. 2024).

Each of the GR system's layers provides a range of ecosystem services. The vegetation helps reduce the urban heat island effect through evapotranspiration and increases the urban biodiversity by providing a habitat for pollinators and various bird species (Berndtsson 2010; Vacek et al. 2017). Vegetation also improves urban air quality through dry deposition processes and slowing down photochemical reactions that lead to secondary air pollutants (Yang et al. 2008; Rowe 2011; Speak et al. 2012). The substrate layer mitigates urban flooding by retaining the rain and delaying its entry into the sewer systems. Studies have shown wide ranges in retention from 10 to 100% (Berndtsson 2010; Carson et al. 2013; Hakimdavar et al. 2014). Substrate conductivity is widely acknowledged as a driving parameter of a GR's hydraulic behaviour, which in turn controls peak flow discharge rates and overflow (Benoit et al. 2025). Modelling studies have reported that the widespread implementation of GRs can reduce the likelihood of sewer systems being overwhelmed, resulting in sewer overflows or pressurization of the sewer network (Benoit et al. 2025; Orsi et al. 2025). The media also promote stored water uptake and evapotranspiration by the vegetation, thereby reducing the volume of water passing into the grey infrastructure and contributing to urban runoff (Berndtsson 2010; Carson et al. 2013; Hakimdavar et al. 2014).

Recently, GR systems have transitioned away from traditional engineered substrates due to their low water retention efficiency. During rain events, stormwater does not saturate the substrate uniformly; instead, preferential flow paths form, causing leakage before the entire pore volume is fully utilized (Zhang et al. 2018; Sims et al. 2019). To address this issue, manufactured retention materials, such as fleece and mineral wool (MW), are gaining popularity in GR construction. These materials promote horizontal flow, enabling more efficient rainwater absorption and delaying the onset of discharge. Enhanced water-holding capacity supports evapotranspiration, further reducing the volume of stormwater that ultimately enters city sewer systems.

Fleece layers consist of fabric made from recycled synthetic fibres and provide a durable medium for root stabilization (Carson et al. 2013; Abualfaraj et al. 2018). Previous studies have investigated GRs using fleece layers either as a single layer in combination with the substrate (Abualfaraj et al. 2018) or two layers in series (Carson et al. 2013; Hakimdavar et al. 2014). MW, composed of long rock mineral fibres, is compacted into dimensionally stable, hydrophilic felts (Vacek et al. 2017; Kostadinović et al. 2022). These retention layers can be integrated with or without conventional GR growing mediums.

Although retention and detention layers like fleece and MW are occasionally included in the research, the specific impact of this material on overall roof performance is rarely examined directly. Previous studies have reported rainfall retention rates of 10–100% for experimental roofs containing fleece (Carson et al. 2013; Hakimdavar et al. 2014; Abualfaraj et al. 2018) and 32–96% for those using MW (Arkar et al. 2019; Alim et al. 2023). Hakimdavar et al. (2014) also reported peak flow and delay for a roof with fleece, while Abualfaraj et al. (2018) calculated runoff coefficients for fleece VRAs. However, these studies lacked a control roof such as a gravel ballast or traditional extensive GR, making it unclear whether the inclusion of retention/detention layers offers significant hydrological advantages over more traditional GR systems. To our knowledge, Rowe & Getter (2022) are the only researchers to explicitly evaluate the role of a manufactured retention layer in GR hydrology. They found that adding an MW layer improved the retention of standard GRs to a level comparable to GR modules with an underlying water storage layer.

Given the growing demand for lightweight GRs, it is essential to critically evaluate the hydrologic functionality of roofing systems incorporating mat-based manufactured retention and detention layers. This study assesses the impact of these materials by comparing the hydrologic performance of several common VRAs used for commercial roofs exposed to natural rainfall and snowfall. Over a year-long evaluation, runoff reduction and attenuation metrics, including curve number (CN) and runoff coefficients, are analysed. Finally, stormwater management performance is related to plant health observations to provide insights into the trade-offs between hydrologic benefits and plant survivability for different assembly systems.

Retention/detention layers were constructed with materials including fleece (Bio-Fleece, BioRoof Systems Inc., Burlington, ON; Figure Ci), MW (HTC Green Roll D13, Urbanscape via Green Roof Diagnostics, Culpeper, VA; Figure Cii), extensive mix FLL (Gro-Bark, Caledon, ON; Figure Ciii), honeycomb reservoir (Green Roof Diagnostics, Culpeper, VA; Figure Civ), and polyester detention fabric (Green Roof Diagnostics, Culpeper, VA; Figure Cv). Additional material information is included in the Supplementary Information. Shallow VRAs (i.e., fleece and MW) were constructed on top of a wooden frame, so that the vegetation was flush with the testbed edges and not recessed.

GRITLab 1 houses a weather station that, among other parameters, collects 5-min intervals of air temperature (°C) and rainfall (mm) data via a Campbell Scientific programmed CR1000 Datalogger. Two additional rain gauges were utilized as quality assurance and quality control (QA/QC) techniques to validate the rainfall data (US EPA 2002). These included a weather station located at GRITLab 2 on the roof of the John H. Daniels Faculty of Architecture, Landscape and Design Building in Toronto, roughly 260 m northwest of GRITLab 1 and the Environment and Climate Change Canada (ECCC) ‘Toronto City’ station (Climate ID: 6158355), which is located roughly 950 m north of the lab (Government of Canada 2022). Each testbed is equipped with a tipping bucket rain gauge device (Model TB6, HyQuest Solutions) with one tip representing 6.28 mL of water. Discharge data were recorded at 5-min intervals and stored via Onset HOBOware data loggers.

Data were downloaded weekly. Precipitation parameters (precipitation depth, mm; precipitation volume per testbed area, L; and peak precipitation rate, mm/min and L/min) were computed for precipitation events that triggered discharge from the grey control roof. In the warm season (April to mid-November), precipitation and discharge events were extracted from the data using the following criteria in order to be consistent with parallel research projects (Pelayo Cazares 2022; Saade 2022): rainfall and discharge events were defined by one tip (0.2 mm of rainfall or 6.28 mL of discharge) and a minimum inter-event time of 1 h was applied.

Throughout the winter (mid-November to March) events were classified as:

• Rain when precipitation and grey roof discharge occurred within 1 h of each other.

• Snow was recorded as precipitation at GRITLab 1 and both QA/QC rain gauges without any discharge from the ballast roof, and

• Thaw when discharge was recorded from testbeds more than 1 h after precipitation was recorded.

Conventionally, retention is computed for rainfall-dominated conditions where infiltration and runoff processes occur immediately. In winter, when precipitation is dominated by snowfall and temperatures are persistently below 0°C, frozen precipitation accumulates on surfaces and does not infiltrate or discharge until sufficiently warm conditions occur to melt the snow and thaw frozen substrate. As a result, discharge from a thaw event may include accumulated water from several discrete snowfall events that have occurred over multiple days or weeks. Tipping bucket measurements of snowfall are less accurate than those for rainfall, often leading to an underestimate of precipitation amounts (Savina et al. 2012; Buisán et al. 2017). Losses due to wind effects are more common (Buisán et al. 2017).

Due to the challenges associated with accurate snowfall, snow melt, and substrate thaw measurements, negative winter retention values occasionally occurred, indicating that the GR discharge was greater than the recorded precipitation due to the melting of accumulated snow or ice (Berghage et al. 2009). All winter retention percentages were rounded to the nearest ±5%, given the expected measurement errors. Any minor negative retention values (<0 to −5%) were attributed to measuring uncertainties associated with the weather station and GR tipping buckets, and a retention of 0% was assumed in these cases. Any negative values greater than 5% were excluded from the analysis, since it was assumed that these resulted from unreliable precipitation data. Overall, 19 winter rain events and 8 snowfall events were included in the analysis.

A Tukey Honestly Significant Difference (HSD) test was used to determine the significant differences (at a 95% confidence level, or α = 0.05) of the calculated parameters across treatment types in R (R Core Team 2023).

Finally, plant health was documented using testbed photographs taken after installation (August 2, 2022) and at the end of the study period (July 7, 2023). Visual analysis was conducted, with red colouring plants indicating plant water stress and green colouring indicating good plant health.

The winter season was wetter than normal. The Toronto City Weather Station recorded 242 mm of rainfall and 145 cm of snowfall. The water equivalent of snowfall was calculated by dividing the measured amount of snowfall (in cm) by 10 and converting it to unit (mm), resulting in a total precipitation of 387 mm, well above the historical average of 286 mm (Government of Canada 2024). At GRITLab 1, 19 rain events totalling 350 mm and eight snow events totalling 38.4 cm were observed (total precipitation 388 mm). Despite this discrepancy in the rain-to-snow frequency, the total precipitation is equivalent. Precipitation events at the lab were classified based on the timeline of precipitation to testbed discharge, so we suspect that some snowfall, which melted rapidly, may have been misclassified as rainfall. Regardless, the total recorded winter precipitation at the lab was equivalent to the local climate station. The winter precipitation events ranged from 2.6 to 59.6 mm (average 25.7 mm), and one winter rain event of note surpassed the 2-year IDF curve at 50.2 mm and an intensity of 2.3 mm/h. The breakdown of storm event classification by size is provided in Table S1. The combination of the individual winter rain and snow events prior to thaw events resulted in 15 winter precipitation–discharge events (Table S3).

The GRs completely captured 12 of 27 (44%) warm season rainfall events; of these, five were small rain events (<5 mm) and seven were medium-sized (5–20 mm) (Table S2). Large rain events caused discharge from all treatment bed types, except for one large storm that occurred after 11 dry days. Of the medium-sized events, only one generated discharge from all roof assemblies. This was not the largest storm in the medium category but did have an antecedent dry period of less than 6 days, resulting in discharge from previously saturated roofs. Overall, the fleece VRA produced discharge most often (48% of the time), followed by the traditional GR (44%), MW (39%), MW&GM (37%), and CRD (35%) beds. VRAs with retention/detention materials released discharge from all winter events, while the traditional GR occasionally did not produce discharge (Table S3).

The VRAs retained the majority (88–93%) of the received precipitation (Figure S1a). The traditional GR retained 90% of the rainfall, while the three VRAs, which included MW MW&GM, and CRD, retained, on average, 92% of the rainfall. The Tukey test found essentially no difference (p = 1.00) between the MW&GM and CRD retention rates, indicating that the additional storage and detention layer did not influence the roof retention performance. Our observed retention rates were notably higher than those in other recent field studies, including Rowe & Getter (2022), who observed an 81% retention rate in their pilot-scale rock wool GR. Located in central-west Michigan, adjacent to Lake Michigan, the Nursery building, where Rowe & Getter (2022) completed their project, experienced 43 rain events in the span of 8 months, of which 12 were of high intensity (>20.0 mm) with the largest being 119.4 mm. Our study, in contrast, only had eight events larger than 20.0 mm, with the largest being 71 mm. Arkar et al. (2019) evaluated a full-scale MW GR in Wien, Austria, which has a similar Koppen-Geiger climate classification as Toronto (Dfb). Rainfall intensity in Arkar et al. (2019) was three times larger than ours and, consequently, authors reported much lower retention (35–40%) for their VRAs with an MW retention layer.

The fleece VRA had a total study retention of 88%. Field studies with a fleece retention layer have reported a wide range of results. Over 4 years of data collection during warm weather months in New York City (NY), Abualfaraj et al. (2018) constructed a full-scale GR using a singular fleece layer combined with 25.4 mm of GM, yielding an overall retention of 77%. The New York field study was dryer overall than this project, with an average annual precipitation of 292.4 mm, but had a greater frequency of rain events. Antecedent moisture may have contributed to the lower observed retention. The other full-scale studies that used two fleece layers in series resulted in 36 and 10–100% retention (Carson et al. 2013; Hakimdavar et al. 2014).

The presence of a vegetated system produced significantly greater retention than the ‘Grey’ gravel ballast control roof for medium and large events. The configuration of retention and detention layers only produced small differences for large rain events (Figure 4(b)). VRAs with MW provided 10% more retention than VRAs with fleece or GM only. Regardless of the smaller retention percentages observed for large storms, the VRAs could retain 70–80% of the rainfall.

The gravel ballast roof was a surprisingly effective system for stormwater control, retaining 213.3 mm, or 45%, of the rainfall received. The role of ballast in building hydrology is often overlooked since these roofs do not offer co-benefits present in vegetated roofs. However, previous studies have shown that coarse aggregates like gravel can retain a significant amount of rainfall. Pilot-scale outdoor studies have reported that gravel beds can retain 10–50% of rainfall (De Cuyper et al. 2005; VanWoert et al. 2005; Rowe & Getter 2022). Carson et al. (2017), who evaluated a full-scale gravel ballast roof in New York, observed rainfall retention of 27% over two and a half years.

Throughout the winter, the VRAs produced overall seasonal retention of 44–71%, with the traditional GR showing the highest precipitation reduction (Figure S1b). These values exceed the reported range of 0–36% retention for GRs during the winter (Berghage et al. 2009; Johannessen et al. 2017; Braskerud & Paus 2022). The testbeds used in this study were raised and, therefore, were not in direct contact with the building. Consequently, the thermal exchange processes exhibited in the testbeds may not be characteristic of a full-sized GR in direct contact with the roof. The presence of cold air under the testbeds may affect the freezing and thawing processes experienced during the winter. Over the winter, there was only one small event (2.6 mm), and the VRAs captured more than 95% of it (Figure 4(b), Table S2). The inclusion of certain retention and detention materials did not have an impact on overall winter retention. In fact, it was the traditional GR that had the greatest retention in the medium and large size categories at 85 and 52%. The grey roof had significantly lower retention than the VRAs, with a winter seasonal average of 31%. This is notably higher than the winter retention of 14% reported by Mentens et al. (2006) for gravel roofs.

Over the warm season, peak rainfall intensities ranged from 0.14 to 5.95 L/min. All roofs will attenuate flows to some degree due to the time required for runoff to travel from the most distant point to the outlet. This attenuation effect was evident in the ballast roof, which reduced the peak discharge from an average input rainfall intensity of 1.71 L/min to a peak discharge rate of 0.30 L/min, resulting in an 82% reduction in peak flow rate. The VRAs further attenuated peak discharge flows by an additional 10–14%, averaging between 0.07 and 0.14 L/min (or 0.02–0.04 mm/min). This performance slightly outperformed that of previously studied VRAs by Squier-Babcock & Davidson (2020), which had peak flow rates of 0.06 mm/min.

For small-sized rainfall events, all VRAs completely attenuated rainfall peak flows (Figure 5(b)). During the medium events, the VRAs significantly mitigated peak flows more effectively than the gravel ballast roof, achieving an average PFR of 98% compared to the ballast roof's 70%. For large rainfall events, differences in the configuration of retention and detention layers had a minor impact on performance. VRAs incorporating MW and GM (MW&GM and CRD) attenuated 5% more peak flows than the MW-only VRA and 10% more than the VRAs with fleece or GM alone. Despite slightly lower PFR for large storms, the fleece and traditional GRs were capable of reducing rainfall peak flows by 80–90%. These findings align with Hakimdavar et al. (2014), whose full-scale fleece GR completely attenuated peak flows for small-sized events but showed decreased PFRs of 70 and 60% for medium and large events, respectively.

In the traditional GR, the preferential flow paths formed as rainwater infiltrates through the GM, causing discharge to begin before the substrate is saturated, were observed, resulting in an average discharge delay of only 4.5 h (Figure S2a, Table S3). In contrast, the synthetic retention layers in the fleece, MW&GM, and CRD VRAs absorbed rainwater more efficiently, delaying the discharge by an average of 6.5, 7.5, and 7.5, respectively. The MW-only VRA demonstrated the longest delay, with an average discharge delay of 9.7 h. Discharge delay alone provides an incomplete understanding of the retention and detention processes occurring within a GR. For example, during a large rain event in mid-May, the MW&GM and CRD beds completely retained the rainfall, while the MW testbeds began discharging after 11.6 h, skewing the overall average delay time for the MW testbeds. Additionally, discharge from the VRAs often lasted 3–60 h longer than the grey control, with the CRD VRA consistently exhibiting the most prolonged overall discharge times (Figure S2b). This prolonged discharge suggests that the CRD detention layer effectively slows the discharge rate as intended.

The winter precipitation peak intensities ranged from 0.14 to 1.08 L/min. The ballast roof continued to significantly attenuate flows, decreasing the average rainfall intensity in half from 0.48 L/min to a discharge rate of 0.24 L/min. The VRAs also continued to decrease the peak flow rates by an additional 12–27% on average (0.11–0.18 L/min or 0.03–0.05 mm/min). Surprisingly, the traditional GR provided the largest PFR during the winter season. Discharge rates observed in this study were significantly larger than other published results. Braskerud & Paus (2022) observed runoff rates of 0.002 and 0.03 mm/min during below and above-freezing air temperatures. Teemusk & Mander (2007) further observed the impact of air temperature on thawing processes. They found that surface snow melts first and slowest. As temperatures rise increasingly above freezing, the water that was previously frozen in the GR is released, contributing to greater flows. So, future analysis of this data set should consider determining if there is a difference in testbed peak flow under freezing versus non-freezing ambient air and soil temperatures. The one small thaw event was completely captured by all VRAs (Figure 5(b)). Peak discharge from medium and large precipitation events averaged 0.41 and 0.52 L/min, respectively. The VRAs attenuated more than 50% of the peak flow, and no significant difference was observed between GRs.

For the winter months, the discharge delay of the VRAs was directly compared to the grey roof discharge start time. Contrary to the warm season, the traditional GR had the longest average discharge delay at 13 h, followed by the CRD at 9 h (Figure S2a). The other VRAs (fleece, MW, and MW&GM) delayed the release of discharge by 2 h, on average. Discharge from the grey roof lasted an average of 2.6 h, and the VRAs lasted 2.5–3.4 h, with the traditional GR being the shortest and no other trends associated with roof assembly type (Figure S2b).

CNs and volumetric runoff coefficients (C_vol) were calculated for each discharge-producing rain event observed during the warm season. CN for the ballast roof was 96 or higher. Overall, CNs did not vary significantly between VRAs and ranged between 79 and 87 depending on the event. CNs reported in this study were frequently lower than values reported by others for traditional GR, which have often ranged between 92 and 94 (Fassman-Beck et al. 2016; Carson et al. 2017; Hill et al. 2017). However, lower CNs have been observed by some. Getter et al. (2007) reported CNs ranging from 80 to 94 for a roof in Detroit, MI. Fassman-Beck et al. (2016) noted that low CNs have been observed in studies in Chicago (USA) and Kingston (New Zealand). Fassman-Beck et al. (2016) proposed that CN of 84 may be justified for planning and design calculations. The results of this study support and align well with this recommendation.

Over the years, various GR designs have been adopted without the guidance of performance codes developed by cities to determine the most suitable system for specific building and climate conditions. By evaluating the hydrological performance of multiple VRAs, it was observed that lightweight, soilless retention layer systems (fleece and MW) alleviated stormwater volumes similarly to systems with both manufactured layers and GM (MW&GM and CRD). However, adding a reservoir and detention layer (CRD) enhanced PFRs, discharge delays, and discharge durations during warm seasons.

In contrast, the variance in hydrological performance based on VRA materials was less pronounced in winter. This season exhibited significantly lower retention than in the warm months, increased peak flows due to complex processes associated with snow melting and ground thaw, and shorter discharge delay and discharge durations relative to a ballast roof. Additionally, the hydrological benefits observed in the warm season for VRAs with lightweight retention layers did not translate to the winter season. Instead, the traditional GR contributed to improved annual precipitation volume reduction due to its superior retention performance in winter.

This study highlights the seasonal impact of incorporating manufactured retention layers on the hydrologic performance of GR systems. While this research focused on GR hydrologic performance, other critical factors must be considered when assessing GR systems. Irrigation, an operational practice not included in this study, may be necessary, especially for systems without soil-based retention layers, such as the fleece-based VRAs, to prevent plant stress and ensure that vegetation is healthy. Finally, the hydraulic characteristics of retention layers like fleece and MW will undoubtedly evolve as these materials age. It remains unknown whether stormwater retention changes significantly as materials age and degrade over time.

GR materials were provided by Sedum Master Inc., Green Roof Diagnostics, Next Level Stormwater Management, Sempergreen, BioRoof Systems Inc. and Gro-Bark. Funding for this work was provided to J. Drake through an NSERC CREATE grant and an Ontario Early Career Researcher Award. Use of the GRITLab facilities was provided by the UofT Daniels Faculty of Architecture, Landscape, and Design. Research support provided by J. Saade, V. Sidu, W. Liao, and H. Momin.

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

The authors declare there is no conflict.