Enhancing Urban Resilience: Stormwater Retention and Evapotranspiration Performance of Green Roofs Under Extreme Rainfall Events

Abstract

Rapid urbanisation and climate change have intensified extreme rainfall events, exacerbating stormwater runoff and overwhelming urban drainage systems. Nature-based solutions, such as green roofs with integrated retention capacity, offer promising strategies to mitigate these challenges. This study investigates the influence of substrate thickness and retention volume on the stormwater retention and evapotranspiration (ET) performance of three green roof variants under extreme rainfall scenarios (natural and 5-, 30- and 100-year events). Using lysimeter-based experimental setups, we show that the overall retention capacity is highly dependent on the filling status of the retention layer. Near full capacity, retention performance decreases significantly, resulting in runoff behaviour similar to that of conventional green roofs, while empty systems store up to 99% of rainfall. In addition, ET rates tend to decrease in systems with higher substrate layers and larger retention spaces due to reduced surface evaporation and greater thermal insulation. However, higher substrate layers store more water, allowing plants to maintain transpiration during dry periods, potentially increasing total cumulative ET over time. Overall, this study highlights the importance of designing intensive retention green roofs with dynamic water management to optimise both rainwater retention and ET, thereby increasing urban resilience to increasing rainfall extremes caused by climate change.

1. Introduction

The rapid urbanisation of the world’s population, exacerbated by the twin crises of climate change and water scarcity, poses critical challenges for sustainable urban water management [1]. Since the 19th century, global temperatures have risen by 1.1 °C, leading to an increase in extreme weather events such as heat waves, heavy rainfall and droughts [2]. Urban areas dominated by impervious surfaces are particularly vulnerable, intensifying surface runoff and increasing the risk of flooding during heavy rainfall [3]. These environmental and economic challenges, together with the impacts of climate change, have exposed the inadequacies of conventional wastewater systems and highlighted the need for innovative and multifunctional solutions in urban water management [4].

In response, nature-based solutions such as blue–green infrastructure (BGI)—including green roofs, permeable pavements and constructed wetlands—offer promising strategies to mitigate flooding and enhance urban resilience [5,6,7]. Integrated urban water management (IUWM) frameworks, along with smart water technologies, further contribute to managing the growing complexity of urban water systems [8]. These solutions require urban planners to integrate water management with broader urban development, pollution control and waste management strategies [9].

An IUWM strategy improves resource efficiency by treating the urban water cycle as an interconnected system. This approach addresses the water needs of the residential, industrial, agricultural and ecological sectors, while serving as a comprehensive framework for planning and managing urban water systems [10]. Successful implementation depends on collaboration across jurisdictions within the urban region, enabling cities and utilities to create resilient water systems that meet current and future demands driven by urban growth [11]. Ultimately, IUWM is reshaping the way cities interact with water, offering innovative perspectives on the design and management of water resources and infrastructure [12].

Urban densification and expansion, coupled with extreme rainfall events, have intensified stormwater runoff, putting significant strain on existing infrastructure, including networks and centralised treatment plants, and prompting cities to consider countermeasures such as the integration and implementation of urban BGI [13]. Urban BGI refers to managed and engineered infrastructures that use vegetation for decentralised urban water management and to provide sustainable solutions for climate resilience [14]. Examples include swales, infiltration trenches, green facades and green roofs that support water sensitive urban design [15].

Green roofs offer multiple benefits, including reducing urban heat island effects through evapotranspiration (ET) [16], enhancing biodiversity [17], and improving air quality [18]. They also play a critical role in stormwater management [19], making them a promising tool for mitigating the impacts of extreme weather events [20].

Numerous studies have investigated the performance of green roofs under conditions of extreme heat [21,22,23] or heavy rainfall [24,25,26,27]. However, there remains a notable gap in understanding the performance of these systems under extreme rainfall conditions, especially in European climates. In addition, many studies predominantly focus on extensive green roofs as these are more common in practice [28]. Extensive green roofs are a low-maintenance form of roof greening characterised by drought-tolerant plants, such as succulents, and a relatively low substrate layer, typically not exceeding 15 cm [29]. In contrast, intensive green roofs provide superior rainwater retention [30]. These systems are characterised by a substrate layer of at least 15 cm and vegetation with higher water requirements, ranging from grasses and perennials to shrubs and small trees [31,32]. Despite their potential benefits, intensive green roofs have been comparatively underexplored in previous research.

Recent studies have investigated retention green roofs, which provide enhanced water retention through raised substrate layers and integrated water storage layers [33]. In addition, water can be transported into the substrate layer by capillary forces [34], acting as an indirect irrigation mechanism. This stored water can also be used to irrigate other green spaces in a process known as BGI cascading [13,35]. Few studies have investigated the effect of substrate thickness [30,36] and retention volume [34,35] on the runoff and ET performance of green roofs. However, experimental studies and the consideration of intensive green roofs often receive insufficient attention in the context of retention green roofs.

A key component of the post-rainfall water cycle is ET, which includes both evaporation from plants and the land surface and transpiration from plants [37,38]. It significantly influences the effectiveness of vegetated BGI systems. As ET plays a major role in the overall water balance, it must be considered in the design of BGI such as green roofs, which aim to reduce the overall volume of runoff [39]. Therefore, it is important to integrate ET into the design of volume reduction strategies and account for vegetated BGI systems.

In most cities, green roofs are designed for smaller, more frequent rainfall events (e.g., 1-year events) [40]. Due to climate change and the occurrence of more frequent extreme rainfall events, BGI will need to be designed for much larger precipitation intensities, such as 30-year and 100-year events.

Therefore, this study investigates the influence of substrate thickness and retention volume on the performance of intensive green roof structures during extreme weather events. Specifically, we analyse three intensive green roof variants and assess their ET and stormwater retention capabilities under varying rainfall return periods (natural heavy rainfall event and 5-, 30- and 100-year events).

2. Materials and Methods

Three experimental green roofs were constructed to investigate rainwater retention and ET performance under extreme rainfall events. The experimental setups were based on lysimeter systems (Umwelt-Geräte-Technik GmbH, Müncheberg, Germany), which accurately measure weight changes and quantify ET by recording water inputs (precipitation and irrigation, if applicable) and outputs (runoff).

2.1. Experimental Setup

The three lysimeter-based green roofs were designed as variations of intensive retention systems with surface areas of about 0.6 m², with a sensitivity of the weight systems of about 0.1 kg, and a maximum weight load of 120 kg for Green Roof 1, 400 kg for Green Roof 2 and 800 kg for Green Roof 3 (Figure 1). The main difference between the green roofs was the thickness of the substrate and the retention volume.

Figure 2 provides a detailed overview of the green roof structures. All three roof segments were vegetated with turf grass and the same type of substrate was used. Green Roof 1, designed as a conventional system, included a 15 cm substrate layer and a drainage layer with a water retention capacity of 8.7 L m⁻² (Figure 2). Green Roof 2 had also a 15 cm substrate layer and an 8.5 cm retention layer instead of a drainage layer, allowing it to store up to 80 L m⁻². Green Roof 3 had a 30 cm substrate layer and a 17 cm retention layer, with a maximum water storage capacity of 161 L m⁻² (Figure 2).

All functional layers were supplied by Optigrün International AG (Krauchenwies-Göggingen, Germany). Images of the drainage layers and retention boxes used in the study are in the Supplementary Materials (Figure S1).

A soil moisture sensor (model GS1 by METER Group GmbH, München, Germany), based on the measurement of the volumetric water content, was placed 10 cm below the turf layer of each experimental plot.

A tilt counter (Umwelt-Geräte-Technik GmbH, Müncheberg, Germany) was connected to each outlet to determine the volume of the overflow. The tilt counters used registered the outflow in 100 mL increments at a volume flow rate of up to 5 L min⁻¹.

All sensors and the weather station were connected to a measurement and control datalogger (CR1000X, Campbell Scientific, INC., Logan, UT, USA).

2.2. Experimental Site Conditions

The three experimental green roofs are located on the campus of the Helmholtz Centre for Environmental Research—UFZ in Leipzig, Germany (51°21′12.8″ N, 12°25′58.4″ E) and were installed at ground level, approximately 30 cm apart and fully exposed to direct sunlight (see Figure 1). A compact digital weather station (ClimaVUE 50, Campbell Scientific, INC., Logan, UT, USA) continuously monitored temperature and precipitation data with a resolution of one minute. The experiments were carried out between 8 July and 18 October 2024.

The weather data and timing of the three simulated rain events are shown in Figure 3. The average daily temperature for the study period was 18.8 °C. The highest daily temperature was recorded on 4 September with 26.4 °C, while the lowest daily temperature was recorded on 15 October with 7.5 °C.

2.3. Experimental Set-Up

2.3.1. Natural Heavy Rainfall Event

On 12 July 2024, a heavy rainfall of 17.7 mm occurred between 12:57 and 13:43 p.m. According to KOSTRA-DWD-2010R (German Weather Service, Offenbach am Main, Germany), this event can be classified as a 2-year heavy rainfall event and was used to investigate the stormwater retention and ET performance of green roofs. Four days before the natural heavy rainfall event, the substrate of three experimental green roofs was saturated and all retention layers were filled to their maximum capacity.

2.3.2. Simulated Heavy Rainfall Events

The regional heavy rainfall data in Germany are available as open data and are provided by KOSTRA-DWD; they have been used to simulate heavy rainfall events in this study. Once a decade, the German Weather Service publishes a breakdown of heavy rainfall intensities for different return periods (1 year–100 years) and duration levels (5 min–72 or 168 h), divided into grid squares. The KOSTRA-DWD-2010R evaluation published in 2017 for the grid square of the south-eastern part of the city of Leipzig (grid square 51055) was used for the simulations.

Heavy rainfall events with different return periods (5, 30 and 100 years) were simulated from 19 September to 18 October 2024. All heavy rainfall simulations were carried out in the morning between 10:00 and 12:00 a.m. To prepare the experimental installations for the 30-year and 100-year rainfall events, a temporary roof was installed prior above the lysimeters to the start of the experiments and during the predicted rainfall events to prevent substrate saturation.

The Euler type II rainfall distribution, which is considered as the standard in Germany [41], was used. Here, the rainfall intensity, typically at 5 min intervals, reaches a maximum at one third of the time of the test duration [42]. The intensities are distributed in descending order, first to the left and then to the right of the maximum (see Supplementary Figure S2). Return periods of 5, 30 and 100 years were chosen for a duration level of 2 h each. The simulated rainfall for a duration of 2 h was 33 mm for the 5-year return period, 50 mm for the 30-year return period and 61 mm for the 100-year return period.

Each simulation was run once, with the same initial conditions for all cases: (i) a completely emptied retention layer, and (ii) a saturation deficit in the substrate layer. To create the saturation deficit, the retention chamber was emptied a few days before the experiment and, if rain was expected, a temporary roof was installed as described above. The test periods were chosen to ensure the highest possible outdoor temperatures, and no precipitation was expected to interfere with the experiment.

2.4. Data Analysis

Determining Water Balance Parameters

A water mass balance equation of the experimental green roof systems is expressed as Equation (1):

ΔS + ΔR = P + P_Sim − ET − F

(1)

where the water inflow variables of the three experimental green roofs are precipitation P and/or simulated precipitation P_Sim. Water outflow leaves the experimental system either by runoff F or by ET. Changes in the water content of the substrate ΔS and in the retention/drainage layer ΔR are also considered (Figure 4). All values are expressed in L m⁻². For simplicity, ΔR also includes the water that can be stored in the geotextile (fleece).

Furthermore, weight changes ΔW (of the lysimeters) were expressed as stated in Equation (2):

ΔW = P + P_Sim − ET − F

(2)

Thus, evapotranspiration was determined by a rearrangement of Equation (2).

The performance parameters shown in

Table 1. Calculated performance parameters for three experimental green roof designs.

Performance Parameter	Unit	Definition	Evaluation
Peak evapotranspiration ETp	${\displaystyle \frac{\mathrm{mm}}{\mathrm{h}}}$	Highest hourly ET performance in a certain time frame (absolute value)	${ET}_p=max\left({ET}_i\right)$
Daily evapotranspiration ETd	${\displaystyle \frac{\mathrm{mm}}{\mathrm{d}}}$	Sum of the hourly ET values over 24 h	${ET}_d={\displaystyle \sum _{i=0}^{t=24h}}{ET}_i$
Mean runoff coefficient Cm according to DWA [43]	-	Ratio between the total runoff and total precipitation of a rainfall event	$C_m={\displaystyle \frac{F}{P+P_{Sim}}}$
Peak runoff coefficient CS according to DWA [44]	-	Ratio between maximum runoff and precipitation intensity	$C_S={\displaystyle \frac{F_{max}}{P_{max}}} or C_S={\displaystyle \frac{F_{max}}{P_{Sim, max}}}$
Stormwater retention rate RR	%	Percentage of stormwater retained in the system	$RR={\displaystyle \frac{∆S+∆R}{P+P_{Sim}}}\times 100$
Runoff delay tlag	$\min$	Time difference between the start of the rain event and the first runoff event	$t_{lag}=t_{F,0}-t_{P,0} or$ $t_{lag}=t_{F,0}-t_{PSim,0}$

were calculated to compare the three experimental green roof designs.

A t-test with a significance level of α = 0.05 was used to determine the statistical significance of the differences in the performance parameters having at least three values per green roof design.

3. Results and Discussion

3.1. Natural Heavy Rainfall Event

On 12 July 2024, a heavy rainfall event of 17.7 mm occurred, corresponding to a two-year return period event. The substrates of the three experimental green roof designs were almost saturated and all retention layers were almost filled to their maximum capacity just before the rainfall event started.

Figure 5 presents the ET performance and meteorological data following the event. Among the three green roof designs, Green Roof 1 exhibited the highest peak ET rate (0.8 mm h⁻¹), followed by Green Roof 2 (0.7 mm h⁻¹) and Green Roof 3 (0.6 mm h⁻¹). Daily ET rates showed similar trends across all experimental green roof designs. The average daily ET over the measurement period was 4.2 mm d⁻¹ for Green Roof 1, 4.1 mm d⁻¹ for Green Roof 2 and 4.3 mm d⁻¹ for Green Roof 3, respectively, whereas there were no statistically significant differences among them. On the day of the heavy rainfall event, ET rates were notably lower, with Green Roof 1 recording 2.4 mm d⁻¹, Green Roof 2 showing 2.3 mm d⁻¹ and Green Roof 3 reaching the highest value of 2.9 mm d⁻¹. On 14 July 2024, the rates slightly increased to 3.7 mm d⁻¹, 3.9 mm d⁻¹ and 4.1 mm d⁻¹, respectively. By 15 July 2024, the ET further increased to 5.2 mm d⁻¹ for Green Roof 1, 5.3 mm d⁻¹ for Green Roof 2 and 5.1 mm d⁻¹ for Green Roof 3, respectively.

The ET of green roofs is influenced by several factors, including weather conditions, vegetation type and soil moisture [45]. During heavy rainfall events, the substrate of green roofs becomes saturated, resulting in reduced ET rates. The reason for this is the reduced capacity for further water uptake by plants when the substrate is fully saturated, thereby limiting the ET process. Xu, Chen [45] investigated the ET characteristics of different vegetation types on an intensive green roof under varying weather conditions. The authors found that ET rates and evaporative cooling effects differed significantly among vegetation types, with shrubs exhibiting the highest ET rates, followed by arbors, and grasslands exhibiting relatively lower ET rates. Importantly, the study highlighted that solar radiation and air temperature are the most important meteorological parameters inducing ET on green roofs. Evaporative cooling was observed to be highest in summer and on sunny days, while it was reduced on rainy days, indicating a decrease in ET during these periods. In addition, a study of the rainwater retention of various green roofs found that no ET effects were observed during periods of no rainfall due to the lack of water availability and the limited water retention capacity of the green roof [46]. This suggests that soil moisture availability is a key factor influencing ET rates on green roofs. In conclusion, saturation of the green roof substrate during heavy rainfall events leads to reduced ET rates as the water uptake capacity of the plants is reduced under such conditions.

Due to a malfunction, the tipping bucket of Green Roof 1 failed to record runoff, resulting in a lack of data for evaluation and visualization. In contrast, Green Roof 2 and Green Roof 3 showed significant delays in runoff (Figure 6). The first runoff from Green Roof 3 was recorded 19 min after the onset of rainfall, while for Green Roof 2, it was recorded after 22 min. The total runoff volume was 8.7 mm for Green Roof 2 and 9.6 mm for Green Roof 3, corresponding to mean runoff coefficients C_m of 0.49 and 0.54, respectively. Peak runoff coefficients C_s were 0.49 for Green Roof 2 and 0.31 for Green Roof 3.

The performances of the three experimental green roof designs differed in their runoff characteristics. Although the total runoff delay was similar, Green Roof 2 showed a lower total runoff over a shorter period (64 min) when compared to Green Roof 3 (498 min). Green Roof 3 provided a greater runoff delay, while Green Roof 2 demonstrated better retention, as reflected in the runoff parameters. The lower peak runoff coefficient of Green Roof 3 suggested more effective peak flow attenuation. However, the difference in mean runoff coefficients between the two roof designs remains unclear. Possible reasons can be the variability in substrate composition or measurement uncertainties.

A direct comparison with previous studies is challenging due to differences in experimental conditions. Richter and Dickhaut [33] reported a mean runoff coefficient of 0.2 for a green roof similar to Green Roof 2. However, the authors analysed 453 rainfall events, 75% of which had less than 8 mm of rainfall, making direct comparisons difficult. Palla, Gnecco [47] observed mean runoff coefficients between 0.3 and 0.5 for high-intensity rainfall events (108–194 mm h⁻¹ over 15–21 min) on an extensive green roof with a 12 cm substrate layer. The mean runoff coefficients of Green Roofs 2 and 3 were slightly higher, differing by approximately 20–50%, which may be attributed to variations in rainfall intensity, roof design or substrate properties.

Figure 7 shows the water balance of two experimental green roof designs, due to a malfunction, the tipping bucket. The greatest runoff was recorded for Green Roof 3 (53%). Only 2.9 mm was retained in the retention layer. Considering a free volume of 13.2 mm, the green roof design could theoretically have retained more than four times as much water in the retention layer. Even considering the maximum error of the soil moisture sensors of ±2 Vol%, the free volume in the retention layer was at least 7.2 mm. However, we suppose that inhomogeneous moisture distribution in the substrate may be the decisive factor. Green Roof 2 also retained less water in the retention layer than expected. With a free storage volume of 14.1 mm, only about one half, i.e., 7.4 mm, was retained. Even considering the error of the soil moisture sensors, the free storage volume of 12.6 mm was significantly higher than the actual volume retained. The high intensity of the rain event, during which the soil was unable to absorb the water, may explain why significantly more water was drained. For Green Roof 3, the higher volume retained in the substrate layer could also play a role (15%). However, considering the water stored in both retention and substrate layers, the rainwater retention was 54% for Green Roof 1, 41% for Green Roof 2 and 31% for Green Roof 3. This result seems to be reasonable, because the highest water deficit in the system (ΔS + ΔR) was recorded for Green Roof 1 during the drying period, meaning that the system was able to absorb more water. However, the rainwater retention in Green Roof is only an estimate, in contrast to that in Green Roofs 2 and 3. In the case of the two retention green roof designs, Green Roof 2 achieved a higher stormwater retention rate (RR).

Several publications have investigated the limitations of green roofs during heavy rainfall events, focusing on the water retention capacity of the substrate [46,48]. These studies show that while green roofs are effective in managing stormwater, their retention capacity can be exceeded during intense storms, leading to increased runoff. For example, Wang, Garg [49] studied the effect of different green roof configurations on rainwater retention capacity. The results suggest that while green roofs can retain a significant amount of rainfall, their effectiveness is reduced during heavy rainfall events, resulting in increased runoff. Similar, a study of the performance of green roofs for stormwater control by Raimondi and Becciu [50] found that while green roofs can be effective in reducing and delaying runoff, their retention capacity is limited during heavy rainfall events, resulting in increased overflow.

To sum up, the results indicate that while green roofs are beneficial for stormwater management, their retention capacity is finite and can be exceeded during heavy rainfall events, resulting in increased runoff.

3.2. Simulated Heavy Rainfall Events

3.2.1. Simulation of the 5-Year Event

Figure 8 shows the ET alongside the weather data, including simulated rainfall. During the two-hour simulation period, the ET at the start was set to zero due to significant weight fluctuations, which caused excessive measurement error. However, during heavy rainfall events, the ET is generally negligible [48,51], underlining the validation of this approach.

The highest ET peak was recorded for Green Roof 1 at 0.7 mm h⁻¹, followed by Green Roof 2 (0.4 mm h⁻¹) and Green Roof 3 (0.4 mm h⁻¹). The daily ET was similar for each test site with 2.8 mm d⁻¹, 2.6 mm d⁻¹ and 2.4 mm d⁻¹ for Green Roof 1, Green Roof 2 and Green Roof 3, respectively. No statistically significant differences were found among all the test sites. The overall lower daily ET rates compared to the natural rain event can be attributed to significantly lower daily temperatures during the time period of the experiments that were carried out in September.

Similar results were reported for studies conducted in September in previous studies. Gößner, Mohri [52] recorded an ET of approximately 1.5 mm d⁻¹ for both a green roof with a design similar to Green Roof 1 and a retention green roof similar to Green Roof 2. Feng, Burian [23] reported an average daily ET of approximately 3 mm d⁻¹ for an intensive green roof.

Following the 5-year heavy rainfall event, the ET increased on the second day (21 September) at all experimental sites. For Green Roofs 2 and 3, this increase was already apparent on the first day (20 September). This trend can be attributed to greater overall water availability, particularly in the substrate layer, which dried more slowly due to lower temperatures and retained a greater volume of water.

Runoff was only observed for Green Roof 1, while for Green Roofs 2 and 3, all precipitation was either retained or evaporated within the system. Figure S3 shows a correlation between runoff and precipitation, consistent with the findings of Wang, Garg [53]. The highest mean runoff rate recorded 40 min after the start of the rain simulation was 0.7 mm min⁻¹, which coincides with the peak precipitation rate of 1.1 mm min⁻¹. This gives a peak runoff coefficient C_s of 0.6. Zhang et al. (2021) reported a peak runoff coefficient of 0.6–0.7 for extreme rainfall events (>20 mm) on green roofs with drainage, which is close to the value observed for Green Roof 1 [48]. The total runoff from Green Roof 1 was 26.5 L m⁻², giving an average runoff coefficient C_m of 0.80. The first runoff from Green Roof 1 was recorded 18 min after starting the irrigation.

The water balance after 20 h illustrates the distribution of heavy rainfall within the systems (Figure 9). For Green Roof 1, 80% of the water input exited as runoff, while approximately 9% evaporated. The rainwater RR was 11% in this case, with most of the retained water (7%) stored in the drainage and fleece layers. When comparing Green Roofs 2 and 3, a similar pattern was observed, where about 91% of the retained water was held in the retention layer. A smaller proportion remained in the substrate layer—2% in Green Roof 2 and 4% in Green Roof 3, respectively. The RR, calculated as the sum of ΔR and ΔS, was highest in Green Roof 3 (95%), which was closely followed by Green Roof 2 (93%).

These results are consistent with those of previous studies. Busker, de Moel [34] reported a RR of 12% during heavy rainfall events (>20 mm h⁻¹) for green roofs without a dedicated retention layer. Similarly, Gong, Yin [26] found that a 50% increase in substrate depth resulted in a 5% increase in retention. In contrast, our study showed that doubling the substrate depth resulted in only a 1.5% retention increase, suggesting that other factors such as substrate composition or drainage dynamics may play a more significant role. Wang, Garg [53] showed that substrate depth had a relatively small effect compared to substrate composition and the presence of a retention layer. However, other studies suggest that increasing the total retention volume is strongly correlated with a higher RR [54]. This pattern is evident in the present study, where Green Roofs 2 and 3, which had larger retention volumes, showed more than an eightfold increase in retention compared to Green Roof 1.

3.2.2. Simulation of the 30-Year Event

The ET in Figure 8 shows a significant decrease in case of a 30-year event when compared to a 5-year event, possibly due to the significantly lower temperatures during the 30-year experiment. Among the green roofs, Green Roof 1 had the highest ET peak of 0.5 mm h⁻¹, while Green Roofs 2 and 3 showed much lower peaks of 0.2 mm h⁻¹ each. The pronounced peak in Green Roof 1 occurred immediately after the 30-year heavy rainfall event, suggesting a possible correlation with the saturated substrate. The daily ET was much lower than in the 5-year simulation, with the highest ET_d at 1.6 mm by Green Roof 1, followed by 1.2 mm and 1.0 mm for Green Roof 2 and Green Roof 3, respectively. No significant differences were observed in the daily ET_d of the green roofs. Similar results were reported by Jahanfar, Drake [55], who documented a mean daily ET of approximately 1.8 mm d⁻¹ for October.

Currently, no studies have directly examined the behaviour of green roofs with saturated substrates, and such conditions are often excluded from models [25]. A commonly observed phenomenon of green roofs is surface runoff, typically between 5 and 15 mm [56]. However, if surface runoff occurred in Green Roof 1, it was not detectable but could have affected the ET data.

The first runoff of Green Roof 1 occurred 12 min after the irrigation started, i.e., 6 min earlier than in the case of the 5-year rainfall event. Wang, Garg [53] observed a similar reduction in runoff delay (5 min) with an increase in rainfall from 30 mm to 50 mm, but at higher flow rates. They also found that runoff always started at the same cumulative rainfall volume. In the present study, in the case of Green Roof 1, this volume was 2.9 L (5-year event) and 2.7 L (30-year event). The total runoff was 38.4 L m⁻², and the mean runoff coefficient C_m of 0.77 was close to the value of the 5-year simulation. The highest mean runoff was reached with 0.7 mm min⁻¹ after 40 min (Figure S4). This value was also reported by Wang, Garg [53]. The peak runoff coefficient C_s was, at 0.46, somewhat lower when compared to the previous simulation.

Richter and Dickhaut [33] investigated the runoff behaviour of retention green roofs during a 30-year heavy rainfall event (35 mm in one hour, August 2019). They reported a mean runoff coefficient of 0.48 for a roof design comparable to Green Roof 1, and 0.15 for a roof design comparable to Green Roof 2. The highest runoff contribution of the drainage roof was 1.4 mm min⁻¹. Green Roof 1 had a higher mean runoff coefficient but a lower maximum runoff rate, indicating a greater lag. Green Roof 2 outperformed the comparable retention green roof in Richter and Dickhaut [33] by exhibiting no runoff. This could be due to its larger retention volume (80 mm vs. 30 mm) and its substrate layer being twice as high (15 cm vs. 7 cm). The lack of information on water saturation and pre-filling of the retention layer makes direct comparison difficult. Additionally, Richter and Dickhaut [33] investigated extensive green roofs, which generally store less water than intensive roofs. Nevertheless, both studies confirmed the superior runoff control of retention green roofs.

The water balance after 20 h of the 30-year rainfall experiment is shown in Figure 9. The distribution of runoff water was similar to that of the 5-year event. For Green Roof 1, the percentage of runoff was slightly lower (76%), while the percentage of water stored in the drainage system or in the fleece was slightly higher (14%). This could be due to the drainage system being largely empty following a two-week irrigation break prior to the simulation. However, as the storage volume of the drainage system could not be measured, no definitive conclusions can be drawn in this regard. The amount of evaporated water was slightly lower in Green Roof 1 (5%), while the amount of water retained in the substrate was identical to the first simulation (4%). For Green Roofs 2 and 3, the water distribution was the same as it was in case of the 5-year rainfall event. In both designs, slightly more water was stored in the retention layer than it was in the first experiment (93% for Green Roof 2 and 92% for Green Roof 3). A slight increase in water retention was also observed in the substrate layer (5% for Green Roof 2 and 6% for Green Roof 3). The ET was significantly lower in both green roof designs (2% for each). The RR increased for each test site, with the highest RR of 99% for Green Roof 3, followed by 98% for Green Roof 2 and 19% for Green Roof 1, respectively.

3.2.3. Simulation of the 100-Year Event

The results of the simulation of the 100-year event are shown in Figure 8 and Figure 9. Regarding ET, the results were similar to those for the 30-year simulation (Figure 8). The highest peak ET was recorded for Green Roof 1 at 0.5 mm h⁻¹ (vs. 0.3 mm h⁻¹ for Green Roof 2 and 0.2 mm h⁻¹ for Green Roof 3, respectively). Daily ET was slightly higher than in the previous experiment. No statistically significant differences were found among the experimental sites.

Similar to the 30-year simulation, the higher ET observed for Green Roof 1 may be attributed to greater substrate saturation, which likely reduced runoff and enhanced ET. A slight increase in daily ET was also observed for Green Roof 2 (1.7 mm d⁻¹, compared to 1.2 mm d⁻¹ for the 30-year event), especially on the day of the simulation. No significant change was observed for Green Roof 3 (1.2 mm d⁻¹, compared to 1.0 mm d⁻¹ for the 30-year event). These results suggest that temporary substrate saturation resulted in a transient increase in ET. An increase in daily ET was observed at all sites on the second day after the simulation, which was consistent with the findings from the 5-year heavy rainfall event.

As shown in Figure S5, which depicts runoff delay patterns for the 100-year event, runoff was recorded only for Green Roof 1, beginning 11 min after rainfall onset. At this time, the cumulative rainfall volume was 3.2 L, which was as high as other cumulative rainfall volumes at runoff initiation. Thus, the 100-year heavy rainfall event also demonstrated that runoff begins at approximately the same cumulative rainfall volume, regardless of the rainfall intensity. The total runoff of Green Roof 1 was 49 L, yielding an average runoff coefficient C_m of 0.82. As in previous simulations, the maximum runoff was recorded 40 min after rainfall initiation, at 0.9 mm min⁻¹. The peak runoff coefficient C_s of 0.42 was comparable to that of the 30-year heavy rainfall event (0.46). In addition to the peak runoff coefficient C_s, the runoff delay t_lag (12 min for the 30-year event and 11 min for the 100-year event), as well as the mean runoff coefficient C_m (0.77 for the 30-year event and 0.82 for the 100-year event) exhibited similarities. The mean runoff coefficient for the 5-year event was also in the same range (0.80). A possible explanation is that the percentage increase in precipitation from the 30-year to the 100-year event (23%) was significantly lower than the increase from the 5-year to the 30-year event (50%).

Silva, K. Najjar [57] simulated a heavy rainfall event of 116 mm h⁻¹ over 30 min to investigate the runoff behaviour of extensive green roofs (14 cm substrate, 5 cm drainage). The total rainfall (57.9 mm) was comparable to that in the present study but occurred over a shorter duration. The mean and peak runoff coefficients reported were 0.3 and 0.4, respectively, with the latter showing the same value as in the case of Green Roof 1, while the mean runoff coefficient was higher. This suggests that the green roofs in Silva, K. Najjar [57] absorbed more water, possibly due to differing initial moisture or climatic conditions as the simulation was conducted in Brazil.

The water balance, illustrated in Figure 9, remained consistent with the first two simulations. No significant differences were observed for Green Roof 1 compared to previous simulations. Runoff was 82%, close to previous values (80% for the 5-year event and 76% for the 30-year simulation). ET accounted for 5%, similar to the 30-year simulation. The proportion of water in drainage decreased from 14% to 10% compared to the 30-year event, while the relative volume ∆S stored in the substrate remained nearly constant at 4%. No significant changes were detected in the water balance for Green Roofs 2 and 3. Rainwater retention was 14% for Green Roof 1, 98% for Green Roof 2 and 99% for Green Roof 3, respectively.

Silva, K. Najjar [57] reported an increase in RR from 0.68 to 0.82 with an increase in rainfall intensity from 115.8 mm h⁻¹ to 145.4 mm h⁻¹. Green Roof 1 exhibited a similar trend, with an increased RR from the 5-year to the 100-year heavy rainfall event. However, a slight decrease was observed between the 30-year and 100-year events. Due to differences in initial conditions and climatic factors, as well as the lack of replication, no definitive correlation between RR and rainfall intensity could be established, consistent with the findings of Silva, K. Najjar [57]. Future studies should incorporate repeated trials and standardized initial conditions to better quantify these relationships.

3.2.4. Summary and Comparison of Heavy Rainfall Events

The heavy rainfall simulations for Green Roof 1 showed similar runoff behaviour and retention capacity. The mean runoff coefficient C_m was consistent across all simulations at approx. 0.80. According to DIN 1986-100 [58], this corresponds to that of an unvegetated flat roof (0.8–0.9, depending on the material). No reference value is given for intensive green roofs with less than 30 cm of substrate. Based solely on the mean runoff coefficient, the intensive green roof without additional retention capacity performed similarly to a conventional flat roof. However, as this study focused only on extreme rainfall events, the long-term mean runoff coefficient is expected to be lower. The peak runoff coefficient C_s of 0.50 on average is in line with the FLL [59] guideline, which specifies a value of 0.4 for substrates of 10–15 cm. Higher values observed in practice, as in this study, can be attributed to structural differences such as the installed drainage system. Compared to an unvegetated flat roof (0.8–1.0, DIN 1986-100 [58)], a significant reduction was achieved. The average RR across all simulations was 15%, excluding the natural extreme event, which was estimated to be higher. This is lower than the 24% reported by Wang, Garg [53] for a green roof with no retention capacity. Runoff retardation decreased with increasing rainfall intensity, which is consistent with the findings of Lee, Lee [27]. Runoff initiation occurred at a similar rainfall volume in all simulations, averaging 2.9 L (2.9, 2.7 and 3.2 L, respectively, for T = 5, 30 and 100 years), consistent with Wang, Garg [53].

For Green Roof 2, an average RR of 96% was achieved with an empty retention layer. As the storage capacity was never exceeded, this result was expected. The high RR during the 100-year heavy rainfall event highlights the efficiency of the green roof. Despite the temporary oversaturation of the substrate, no runoff was recorded. However, as the measurement setup did not capture surface runoff, potential bias cannot be ruled out. This result is consistent with the model-based study by Busker, de Moel [34] of retention green roofs with intelligent runoff control, where retention volumes were actively managed based on weather forecasts. The reported RR ranged from 68% to 99% depending on the accuracy of the forecasts, similar to the present results. In contrast, a natural heavy rainfall event with an 82% filled retention layer resulted in a significantly lower RR of 41%. Busker, de Moel [34] reported an average RR of 59% for green roofs without intelligent runoff control. Raimondi, Marchioni [60] highlighted the strong influence of infiltration on runoff performance and retention capacity. As many studies do not consider infiltration, direct comparisons remain challenging. The peak runoff coefficient C_s of 0.49 was similar to that of Green Roof 1, suggesting that Green Roof 2 had a runoff delay effect comparable to that of a conventional green roof under high prefilling conditions. However, its retention performance was superior, with a mean runoff coefficient of 0.49—significantly lower than the 0.80 for Green Roof 1.

Green Roof 3 had an average rainwater RR of 98%, which was only slightly higher than that of Green Roof 2. A high RR was expected due to the large retention volume. However, during the natural heavy rainfall event—when the retention layer was already 92% filled—only 31% of the water was retained, which was significantly less than in the other experimental setups. Nevertheless, Green Roof 3 showed the best peak runoff coefficient C_s (0.31). According to FLL [59], a peak runoff coefficient of about 0.2 is expected for a substrate height of 25–50 cm, which was almost achieved in this case. The mean runoff coefficient C_m was comparable across all green roof designs, indicating similar runoff behaviour at high initial saturation. In contrast, an empty retention space significantly improved water retention. No clear differences in delay effects were observed between the green roof variants, particularly as no runoff delay was recorded for Green Roof 1 during the natural heavy rainfall event. A t-test revealed significant differences in RR between Green Roof 1 and both Green Roofs 2 and 3, while no significant difference was found between Green Roofs 2 and 3.

Daily ET was statistically analysed with respect to return periods and showed significant differences between simulated and natural heavy rainfall events, as well as between 5-year and 100-year events. However, the correlation between ET and temperature suggests that this significance is primarily due to temperature variations rather than rainfall intensity or return period (Figure 10). Further repeated experiments are needed to refine this relationship.

In addition, the trend shows that ET rates decrease in substrates with higher layers and in large retention layers (

Table 2. Summary of the calculated performance parameters for the observed rainfall events. ET_P: peak evapotranspiration, ET_d: daily evapotranspiration, t_lag: runoff delay, C_m: mean runoff coefficient, C_S: peak runoff coefficient and RR: stormwater retention rate.

Green Roof	ETP (mm h−1)	ETd (mm d−1)	tlag (min)	Cm	CS	RR (%)
Natural heavy rain event
1	0.8	4.2	n.a.	n.a.	n.a.	54
2	0.7	4.1	22	0.49	0.49	41
3	0.6	4.3	19	0.54	0.31	31
Simulated 5-year heavy rain event
1	0.7	2.8	18	0.80	0.61	11
2	0.4	2.6	-	-	-	93
3	0.4	2.4	-	-	-	95
Simulated 30-year heavy rain event
1	0.5	1.6	12	0.77	0.46	19
2	0.2	1.2	-	-	-	98
3	0.2	1.0	-	-	-	99
Simulated 100-year heavy rain event
1	0.5	2.1	11	0.82	0.42	14
2	0.3	1.7	-	-	-	98
3	0.2	1.2	-	-	-	99

). Intuitively, higher substrates should hold more water, allowing plants to sustain transpiration longer, potentially increasing total ET over a dry period. In addition, deeper soils can support more extensive root systems. However, per unit time, ET rates may decrease in higher substrates due to lower evaporative fluxes from deeper, cooler and more insulated soil layers. Higher soil layers also hold moisture deeper and are less exposed to atmospheric demand (i.e., radiation, wind), thus reducing direct evaporation.

Several studies have shown that substrate depth is a critical factor influencing the dynamics of ET in green roofs. Voyde, Fassman [61] observed that shallow green roofs with substrate heights of less than 100 mm tended to dry out rapidly, resulting in high initial ET rates but limited overall water retention and reduced ET over time. In contrast, deeper substrates had lower peak ET rates but were able to maintain ET for longer periods. Similarly, Li and Babcock [62] reported that while green roofs with deeper substrates exhibited reduced daily ET rates, total cumulative ET over longer periods could be comparable to or even higher than that of shallower systems due to prolonged plant transpiration. Furthermore, Silva, Paço [63] emphasised that water availability, rather than substrate depth alone, is a key driver of ET. Their results suggest that although higher substrate layers may buffer drought stress more effectively, the magnitude of ET also depends on the adaptations of plant species to water-limited conditions, meaning that additional depth does not necessarily lead to proportionally higher ET rates.

Several mechanisms may explain the observed decrease in ET rates with increasing substrate thickness. Higher substrate layers tend to retain moisture deeper in the profile, thereby reducing exposure to surface evaporation. In addition, greater substrate depths provide better thermal insulation, reducing surface temperatures and, hence, evaporative losses. Furthermore, unless root systems can fully exploit deeper substrate layers, a significant proportion of stored water may remain unavailable for transpiration, limiting plant water use from these zones.

In conclusion, current evidence suggests that while daily ET rates may decrease with increasing substrate depth—particularly in the early stages due to reduced surface evaporation and thermal insulation effects—total cumulative ET may remain stable or even increase over longer periods. This is primarily due to the increased water storage capacity of deeper substrates, which can sustain plant transpiration during prolonged dry periods and, thus, contribute to long-term ET processes.

4. Conclusions

The results show that near-maximum filling of the retention layer leads to a significant reduction in retention efficiency, causing the runoff behaviour of the retention green roof designs (Green Roofs 2 and 3) to resemble that of a conventional green roof design (Green Roof 1) with respect to key runoff parameters. Both Green Roofs 2 and 3 retained significantly less water than theoretically expected, particularly within the substrate layer, in line with the simulation results. Notably, although Green Roof 3 achieved the most effective peak flow reduction and delay—reflected in lower peak runoff coefficient and longer time to last runoff event—it had the lowest overall retention rate of all the variants. When the retention chamber was almost empty, it successfully retained water during all simulated rainfall events (5-, 30- and 100-year return periods). In addition, spreading the inflow over a two-hour period resulted in a slower saturation of the substrate and a more even distribution of water.

These findings underscore a key challenge: Although Green Roofs 2 and 3 theoretically offer sufficient storage capacity in both the substrate and retention layers, the substrate cannot absorb water anymore if saturated. This causes excess water to bypass the substrate and flow directly into the retention layer. If this layer is already full, immediate runoff occurs. An active control mechanism is, therefore, essential to regulate and drain the retention layer, ensuring its capacity is available for subsequent rainfall events.

Naturally, some runoff exists in any green roof system. The goal is not to eliminate runoff entirely—as even the densest natural forests produce some runoff—but to reduce the volume and rate of flow into urban drainage networks during extreme rainfall events. Green roofs are, thus, intended to mitigate peak flows and delay runoff, rather than act as total retention systems.

ET rates tend to decrease in green roofs with higher substrates and larger retention layers. Higher substrates naturally hold more water, allowing plants to sustain transpiration for longer periods and potentially increasing total ET during dry periods. As deeper layers of soil are cooler and more insulated, they experience lower evaporation rates. In addition, moisture stored deeper in the substrate is less exposed to atmospheric influences such as sunlight and wind, which helps to reduce direct evaporation.

This study evaluated runoff behaviour using small-scale lysimeters over a short observation period. While the findings provide valuable insights, several limitations should be acknowledged: The use of small-scale setups may introduce oasis effects, the results may vary under different weather conditions due to the lack of repeated simulations and measurement accuracy poses potential constraints. Future research on larger scales and over longer durations is recommended to validate and expand on these findings.