Exploring the role of pedological and climatic aspects in the medium-term decline of green roofs hydrological performance: an experimental study in a Mediterranean environment Open Access

The combined effects of climate change and increased soil imperviousness led to an increase in urban flooding risk (Eckart et al. 2017). Among the sustainable strategies for mitigating such phenomena, green roofs (GRs) are areas of living vegetation installed on the tops of buildings (Fletcher et al. 2015). With reference to stormwater management, the GRs appear able to make a significant contribution to traditional stormwater management technologies during rainfall events, by implementing two main processes: retention (precipitation storage) and detention (runoff delay) (De-Ville et al. 2018a). The extensive range of experimental and modelling investigations carried out over the years to assess the hydrological effectiveness of these infrastructures unanimously conclude that GRs are a viable solution for stormwater management. However, it should be noted that the retention volumes significantly differ due to the influence of climate conditions and design factors (D'Ambrosio et al. 2021; Mobilia et al. 2021). A typically higher percentage of retention is indeed observed in thicker roofs and climate situations characterized by sporadic rains of moderate size (Berndtsson 2010; Chenot et al. 2017). The need for framing this type of drainage system within the perspective of a long-term cost-benefit evaluation led researchers to start focusing on a phenomenon of particular interest: the aging.

In fact, GRs are considered reactive mediums that can undergo changes in their physical and hydraulic parameters over time (Bouzouidja et al. 2018a). According to several studies (Tafazzoli 2023), aging-related processes may have a great influence on hydrological performance. The detention performance of GRs is indeed greatly influenced by porosity and hydraulic conductivity, which determine how quickly water can flow through the soil matrix, whereas the retention performance is more linked to the distribution of pore sizes, which affects water release and determines the permanent wilting point and maximum water holding capacity (De-Ville et al. 2017). The substrate (growing medium) plays a relevant role in GR hydrological dynamics (Woods Ballard et al. 2015). Due to the presence of the vegetation and its exposure to the atmospheric agents, it is the GR layer most impacted by aging. The physical and hydraulic properties of substrates are influenced by the development of vegetation, particularly by the root's growth (Gadi et al. 2017). According to a study conducted by Gan et al. (2023) on a well-graded sand with clay substrate, it has been observed that an increased root development in the GR substrate can lead to higher hydraulic conductivity. Other literature evidence of the effects that plant-life can have on soil porosity and infiltration rates also comes from the agro-forestry sector. Studies conducted so far found actually contrasting root effects on soil hydraulic properties depending on which are the dominant processes, including root growth (or decay) and the density and diameter of roots (Lu et al. 2020). Dexter (1987) verified the effects of root growth on the reduction of porosity. However, literature evidence also proved that root decay might trigger the opposite effect: enhancement of pore space, formation of preferential flow pathways and hydraulic conductivity increase (Lu et al. 2020). In particular, the effects induced are a function of the diameter of the roots and the typology of the soil. As an example, coarse roots (diameter > 2 mm) cause local compaction and macro-pore development, leading to an increase in saturated water content and hydraulic conductivity (Lu et al. 2020). Moreover, the enhancement of the saturated hydraulic conductivity due to root development can be expected in fine-grained soils, while the opposite effect occurs in coarse-grained soils (Lu et al. 2020).

Already in the early years of the twenty-first century, few authors had begun to investigate GRs aging, identifying very different trends. Mentens et al. (2006) did not detect any effect of aging on GRs' hydrological performances. De-Ville et al. (2017, 2018a, b) specifically pointed out that the medium-term hydrological changes due to aging are minor compared with natural variations due to climate. Getter et al. (2007), Yio et al. (2013) and Yang & Davidson (2021) observed a positive effect of aging on GRs' hydrological performance over the medium-term due to organic matter and pores increase. However, according to Getter et al. (2007), this could be accompanied by drawbacks in detention performances due to the increased occurrence of macro-pore channels. Bouzouidja et al. (2018a) detected a negative effect of aging on GRs' hydrological performances due to pores decrease and saturated hydraulic conductivity increase. The latter was observed also by Alagna et al. (2020) over a 10-month monitoring period.

The findings achieved so far are limited and often contradictory (Hanumesh et al. 2021). This can be attributed partially to the scarcity of long-term hydrological records and the differences in geographical locations, growth medium characteristics, and climate conditions. In order to address the knowledge gap surrounding this topic, it is highly recommended to undertake further research. Based on preliminary findings (D'Ambrosio et al. 2022; D'Ambrosio & Longobardi (under review)) and focusing on the comparison of observed data concerning two experimental extensive GRs located in southern Italy, between two specific time windows – i.e., a virgin early stage (2017–2019) and a aged stage (2022–2023) – the presented research aims to pursue the following objectives:

• highlighting the quantitative changes in the hydrological performance of the two experimental GRs over a 7-year monitoring period, through a comparative analysis of average rainfall retention coefficients representative of virgin and aged substrate conditions;

• investigating the potential changes in the hydrological retention behaviour of the two experimental GRs, which would explain the relevant quantitative changes in the retention coefficients mentioned in the previous point;

• investigating the potential causes of variations in the hydrological behaviour of GRs within an interdisciplinary assessment context, where evidence on pedological and climatic dynamics can potentially support and substantiate hydrological observations.

It should be emphasized that the aim of this study is not to model the ageing phenomenon, which is characterized by an inherent complexity in understanding the role temporally played by the various factors involved. Anyway, in the discussion section, an experimental simulation is illustrated that takes into account ageing and the modelling parameterization required to accurately account for this phenomenon, in line with what has been observed experimentally. This approach suggests a simple and intuitive procedure to account for the ageing of GRs, from a hydrological point of view, for more objective and robust urban planning studies.

The GRs substrate layer consists of a mixture of blond peat, Baltic brown peat, zeolites, and simple non-composted vegetable primer (coconut fibres). As added nourishment, mineral fertilizer made of organic nitrogen fertilizer (bio stimulant algae) was also included. Organic soils such as the selected peat-based mix are generally preferable in GRs since they provide lightweight substrates with good aeration and increased plant-available water. These soils have a large total porosity, in turn including the volume fraction of the relatively large, inter-aggregate pores (or macro-pores) that actively transmit water, and the relatively small, intra-aggregate pores (or micro-pores) that allow a substrate to retain water (Getter et al. 2007; Rezanezhad et al. 2009).

Furthermore, their hydraulic conductivity depends on the degree of decomposition of organic material. In particular, those having more decomposed organic matter have lower values of hydraulic conductivity as well as swelling and shrinking more than those with less decomposed organic matter (Kutilek & Nielsen 1998). However, if compared to the inorganic soils, they fail to provide long-term stability (Xue & Farrell 2020). It is undeniable, from a visual inspection, that over a span of 7 years, the two GRs have undergone notable changes. The vegetation, as evident from Figure 1(b), naturally experienced an evolutionary process. Leaf development occurred in the initial months post-planting, quickly attaining a foliage structure closely resembling what is observed today. The most significant factor influencing the evolution of vegetation over time is the growth of the root system within the substrate. The latter, in fact, further influenced by shrinkage and/or compaction phenomena caused by atmospheric agents, now appears completely different compared to its initial set-up phase (Figure 1(c)).

The 58 rainfall–runoff events are classified into two distinct datasets (Tables 3 and 4 in Supplementary material). Data reported in Table 3 (Supplementary material) refer to the 2017–2019 monitoring period, whereas data reported in Table 4 (Supplementary material) refer to the 2022–2023 monitoring period and are respectively representative of a virgin and an aged substrate condition. An equal number of 29 rainfall–runoff events was covered in each group.

Furthermore, an in-depth analysis was conducted to prove whether rainfall properties, in particular rainfall cumulative depth, might have indeed affected the estimates of RC_M and, thus, AI indexes. Rainfall depth thresholds were selected and the two RC datasets (for virgin and aged period) were divided into two groups according to such thresholds. AI indexes for the two experimental GRs were accordingly assessed for rainfall depth thresholds, highlighting more or less marked aging effects on RC properties depending on rainfall characteristics.

In a previous study concerning the investigation of the hydrological behaviour for the experimental systems (Longobardi et al. 2019), a detailed investigation of the dataset in Table 3 of Supplementary material was reported to identify the dominant variables in order to accurately predict RC values. For this purpose soil water content prior to rainfall events and rainfall properties (cumulate, duration, peak intensity) were accounted for in a multi-step regression approach.

In order to detect changes in the GRs' hydrological behaviour, a replication of the mentioned study is proposed here using the data collected during the most recent monitoring period (Table 4 of Supplementary material). Following the same empirical database approach, once again the correlation between RC, soil water content prior to rainfall events and rainfall properties was thoroughly investigated for both the GRs.

To assess potential variations in the GRs peat substrate matrix, investigations were conducted both in situ and in the laboratory to determine the following physical and hydraulic parameters: porosity, hydraulic conductivity, and water repellency. The overall aim was to compare virgin substrate samples, representative of GR conditions during the first operational period (2017–2019), with aged substrate samples, representative of GRs' current conditions (2022–2023).

The measurements of the hydraulic conductivity of the virgin substrate were conducted indoors, reproducing the 2017 GRs layout (GR2), with its original materials and depths, in a plexiglass box with approximate dimensions of 60 × 60 × 60 cm (Figure 4(b)). The experimental campaign carried out at the beginning of July 2023 involved three tests. In situ tests were conducted to determine the hydraulic conductivity of the substrate of the GRs in their current state (Figure 4(c)). The experimental campaign carried out at the end of July 2023 involved tests at three different points for each GR (Figure 4(d)), appropriately chosen to investigate heterogeneous soil samples (proximity of vegetation, presence of roots, visible shrinking). The climatic conditions at the time and location of the tests, characterized by large dry periods and average maximum temperatures of about 32°C, ensured that the substrate of the GRs was initially dry, a condition which was also verified afterwards in the laboratory.

HYDRUS-1D 4.17 is an open-source software used worldwide (Šimůnek et al. 2005; Hilten et al. 2008) to simulate the one-dimensional movement of water in unsaturated porous media based on the partial differential equation proposed by Richards (1931). The mathematical equations of van Genuchten (1980), among others, can be used to describe the VW content and hydraulic conductivity functions. As an example, two events were selected from the databases, one occurring in the virgin period (Table 3 in Supplementary material) and the other in the aged period (Table 4 in Supplementary material). In order to exclude the influence of precipitation characteristics and the previous water content on the retention process, the two selected events present some similarities. The first event (20/02/2018) is characterized by a rainfall of 11.43 mm and, for GR2, by an average water content of 31% and an RC of 29%; whereas the second event (02/06/2023) is characterized by rainfall of 12.95 mm and, for GR2, by an average water content of 14% and an RC of 48%. Using HYDRUS-1D 4.17, a GR model was developed and calibrated for both events (Tables 3³ and 4 in Supplementary material). The GR model accurately reproduces the stratigraphy and functioning of GR2. Therefore, a 10 cm layer of peat soil and a 5 cm drainage layer characterize the soil column (Mobilia & Longobardi 2020). Its parameterization is detailed in Supplementary material, Table 5(a). The hydraulic conductivity attributed to peat, i.e., the standard value given by van Genuchten (1980) to the ‘loam’ soil class, is of the same order of magnitude as the hydraulic conductivity experimentally obtained on dry virgin samples as described in Section 2.3.2. Atmospheric boundary conditions with surface layer and free drainage characterize upper and lower fixed boundary conditions, respectively. Hourly rainfall rate and initial VW contents are variable conditions updated in simulations based on monitoring data. Moving to the second period of analysis, corresponding to the aged state, it was decided to assess if the model was still able to predict the behaviour of the same roof with sufficient accuracy provided that model parameters are calibrated for the virgin state event. Subsequently, it was investigated whether a re-tuning of the substrate modelling parameters could somehow help restore sufficient levels of prediction accuracy (Supplementary material, Table 5(b)).

The RC_M of GR1 and GR2 in the first (virgin) and latest (aged) operational periods and the AI assessment are presented in Table 1. Considering the statistics on the whole sample of selected rainfall–runoff events, the two GRs appear to have very similar hydrological performance. They are characterized by RC_M of approximately 66% during their first operational period. At the end of the monitored period, this percentage decreased to 58%, resulting in a reduction of their hydrological performance, expressed in terms of AI, by approximately 12%.

Behind the average RC variations, it is important to highlight a significant variability in RC at the event scale (Tables 3 and 4 of Supplementary material). This variability also exhibited similar values in the two reference operational periods, ranging from 97 to 4% in the first operational period and from 90 to 5% in the latest observation period. In both periods, it is possible to detect a dependence of RC on rainfall cumulative depth.

With this premise, as mentioned earlier, in order to highlight more or less marked aging effects on average RC properties depending on rainfall characteristics, AI indexes for the two experimental GRs were computed according to rainfall depth thresholds. Results are illustrated in Table 1 for one single threshold value for which significant differences in AI indexes were assessed. Precipitation events characterized by cumulative depths below the threshold (11.2 mm) exhibited notable RC values, averaging between 71 and 78%. The group characterized by cumulative depths exceeding 11.2 mm records, registered on average RC values ranging from 34 to 51%. Events with cumulative depths less than 11.2 mm there was a reduction of their hydrological performance equal on average to 9%, as indicated by AI. This percentage notably increased to 32%, on average, when considering precipitations with rainfall depths surpassing the identified threshold. The AI indexes dependence on event rainfall depth potentially represents evidence of a further change in hydrological behaviour caused by the aging phenomenon at the experimental site.

To detect the effect of GRs aging on the relevant hydrological behaviour, a similar investigation was considered for the sub-set of rainfall–runoff events that occurred in 2022–2023 (Table 4 in Supplementary material). As a result (Figure 6(c) and 6(d)), contrarily to what was observed in the first operational period, RC values are no longer dependent on VW% values but, instead, rather strongly dependent on rainfall depth (correlation coefficients larger than 60%).

The results of tests aimed at measuring the total and macro-porosity of both virgin and aged peat substrate samples, using either a standard or a non-standard method (i.e., without a vacuum pump), are shown respectively in Table 2(a) and 2(b). After analysing Table 2(a), it can be observed that – on average – there are no substantial changes in total porosity values between the virgin and aged peat substrate samples. On the other hand, Table 2(b) – while indicating a low variability in results concerning the virgin peat substrate samples on which multiple tests were conducted – shows an increase in macro-porosity over time, averaging around 10%. Therefore, since the value of the total porosity between the virgin and the aged substrate samples remained virtually unchanged, it can be deduced that the observed increase in macro-porosity must be matched by a reduction in micro-porosity over time.

The hydraulic conductivities of both virgin and aged peat substrate samples were tested and the results are collected in Table 3(a) and 3(b). The hydraulic conductivity values for virgin samples in dry conditions (Table 3(a)) exhibited a narrow range, from 0.0001 to 0.0003 cm/s. On the other hand, the tests performed on aged substrates (Table 3(b)) showed a higher variability, with conductivity values ranging from 0.001 to 0.005 cm/s. This is because over time, in addition to the initial conditions mentioned above, various phenomena may occur in situ that modify the distribution of voids (in terms of macro- and micro-porosities) and increase the heterogeneity of the soil matrix (e.g., root distribution, presence of external and internal cracks, compaction, and shrinkage). It is important to note that the results of the tests conducted on both virgin and aged peat substrates differ by two orders of magnitude. This means that in the presence of aged substrate under unsaturated conditions, the infiltration rate into the substrate is much higher. Tests conducted on moist substrates sometimes negate these differences, but there are cases where aged substrate exhibits infiltration rates twice as fast as those of virgin substrates.

The soil water repellency, measured by the repellency index R, varies greatly between virgin and aged peat substrates (Table 4). On the dry substrate, the repellency index for virgin soil was found to be 216.45, indicating an ‘extremely water repellent soil – class 5’ according to the classification in Iovino et al. (2018). However, the repellency index obtained for aged substrates was much lower, indicating a ‘strongly water repellent soil – class 3’. These differences are likely due to the heterogeneous nature of the in-situ soil matrix. The repellency indexes for wet substrates, both virgin and aged, are significantly lower. The virgin substrate is classified as a ‘strongly water repellent soil – class 3’, while the aged substrate is classified as a ‘wettable or non-water repellent soil – class 1’ for both tests.

Based on the experimental evidence, the two GRs experienced a reduction in their RC over the 7-year monitoring period, expressed in terms of AI, by approximately 12%. The results obtained align with the conclusions of a study conducted by Bouzouidja et al. (2018a), who, despite examining a different GR infrastructure and climatic setting, similarly noted a decline in RC over time. Moreover, based on the hydrological observations at the study site, it appears that AI is positively correlated with rainfall depth. This is confirmed by Speak et al. (2013) who found the major role of this parameter in the medium-term hydrological performances of GRs.

For what concerns the infrastructures investigated in the current study, as during the monitoring period, the cumulative rainfall and the frequency of rainy days align with the standard averages of the study area, there might not be an impact of climate on the observed reduced performance. Consequently, the observed changes are most likely linked to the aging of the infrastructure and should be sought in the pedological dynamics influencing this living system. This assumption actually contrasts with the conclusions drawn by De-Ville et al. (2017, 2018a, b), who highlighted that the hydrological changes resulting from aging over the medium term are relatively minor in comparison to natural variations driven by climate. The discrepancies in such results can be attributed to the climatic conditions and typical features of the experimental site. The presented research also highlights that over time, possibly due to changes in the GRs substrate matrix, the RC values are no longer dependent on VW, but instead, are rather strongly dependent on rainfall depth.

Preliminary analyses aimed at investigating the evolution of specific physical and hydraulic characteristics of the peat substrate led to some interesting results for the reported case study. Experimental investigations revealed an increase in the soil macro-porosity during the 7-year observation period, averaging around 10%, with negligible changes in total porosity values. This result matches with that obtained by Bouzouidja et al. (2018b), who pointed out a slight increase in macro-porosity over time. However, some other authors, investigating aging effects in substrates other than those used in this work, report contrasting results (Getter et al. 2007; De-Ville et al. 2017; Yang & Davidson 2021). Furthermore, the role played by root decay in increasing the macro-porosity due to the formation of preferential flow pathways cannot be disregarded (Lu et al. 2020). On the other hand, Bouzouidja et al. (2018a) observed also a minimal reduction in the total porosity in GRs substrates – partially composed of peat dust – during a 30-month observation period. The analyses conducted also revealed a significant increase in hydraulic conductivity, with differences between virgin and aged substrates reaching two orders of magnitude in the case of unsaturated conditions of the substrate. This result, supported by other studies (Bouzouidja et al. 2018a; Alagna et al. 2020), indicates an enhancement in infiltration rates over the medium period, most likely affecting the retention and detention dynamics of GRs infrastructures. The soil water repellency also exhibited a substantial variation between virgin and aged peat substrates, shifting between two distinct classes, as defined in Iovino et al. (2018). The virgin substrate is characterized by a water repellency higher than that of aged substrates. These findings align with the earlier assessments of hydraulic conductivity, indicating that over time, the substrate of GRs allows water to penetrate more easily. This holds true in both dry conditions, where a minimal level of repellency is maintained, and near-saturation conditions, where their repellent nature is completely neglected.

As highlighted in the Introduction, the research work presented does not aim to provide an aging model for soil hydraulic properties, for which additional observations would be necessary beyond those actually available and used, namely those related to a virgin period (2017–2019) and an aged one (2022–2023). Nevertheless, the aging phenomenon can be introduced in the hydrological modelling of the observed events, focusing particularly on the parameterization of the model and demonstrating how the latter is consistent and meaningful with respect to experimental observations. Looking at the new parameters of the model, reported in Table 5(b) in Supplementary material, it is clear that: (I) the water content at saturation, a property representative of porosity, registered a slight increase equal to 9% and (II) the hydraulic conductivity increased by an order of magnitude, going from 10.4 to 200 mm/h, coherently with the results of the experimental analysis on aged sample hydraulic properties. While emphasizing that the aim of this work is not to provide a model capable of predicting the behaviour of an aged GR, the proposed modelling exercise nevertheless supports the experimental evidence discussed so far and is a first step toward the definition of future research activities.

The proposed research investigated potential changes in the hydrological performance and in the hydrological behaviour of two extensive GRs experimental test beds, situated in a typical Mediterranean area over 7 years. The focus was on understanding the differences in hydrological characteristics between early-stage and medium to long-term aged systems within a multidisciplinary framework. Rather than modelling the aging process, the study aimed to develop a straightforward procedure to account for GRs aging, from the hydrological perspective, for more objective and robust urban planning studies. Based on experimental analyses and laboratory evidence, the results of this interdisciplinary assessment led to the following results:

• The two investigated GRs experienced a reduction of their hydrological performance measured by an average RC, expressed in terms of the AI, by approximately 12% in 7 years. The aging phenomenon, quantified by the AI index, is impacted by rainfall properties, with a more evident effect in consideration of events characterized by large rainfall depth, reaching up to 32%.

• Behind the hydrological performance, also the hydrological behaviour of the GRs appears changed due to the aging process. In the early-stage period, the retention process was strongly dominated by the VW content prior to the rainfall event, with lower retention associated with large VW. With reference to the last operational period, the retention process appeared instead strongly dominated by rainfall properties, as probably the soil storage is no longer able to act as a sponge modulating the ability to capture rainfall volume.

• Substantial changes in climate dynamics that could have affected the GRs’ hydrological behaviour in the first and latest operational periods were excluded. Hence, the identified variations in the hydrological performance of the GRs, most likely attributed to infrastructure aging, were thoroughly investigated within the realm of pedological dynamics shaping this living system.

• The preliminary pedological experimental campaign successfully unveiled a significant crucial connection between variations in the physical and hydraulic properties of the substrate and changes in the hydrological performance of the GRs. The observed increase in both macro-porosity and hydraulic conductivity, together with the reduction in water repellency, simultaneously reducing both precipitation residence times on the surface and within the soil matrix, provided a compelling rationale for the observed changes in the retention and detention capabilities of the GRs.

GRs are essential strategies in addressing contemporary challenges related to urban and climate changes. Understanding the factors influencing their aging process is undoubtedly a topic of significant scientific interest, with numerous aspects yet to be explored to bridge existing research gaps. As an example, additional aspects that should be taken into account in the aging dynamics of GR substrates include phenomena such as erosion from wind, the presence of microorganisms and small animals, radiation, as well as cycles of warming and cooling of the roofs (Hanumesh et al. 2021). The analyses conducted revealed inherent issues in experimental hydrological analyses. Implementing hydrological models, as well as performing controlled experiments on GR prototypes, is unquestionably necessary to address the complexities associated with the study of GR evolution. Future insights would assist local authorities and stakeholders in making informed decisions and establishing a robust maintenance framework, thereby ensuring the desired GRs' performance throughout their lifespan.

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

The authors declare there is no conflict.