The combined performance of green roof and rainwater harvesting: retention capacity and water saving Open Access

Historically, mankind resorted to rainwater harvesting (RWH) to tackle water shortages, either due to limitations of other water sources or to temporal constraints of water availability. Nowadays, this is still the main motivation underlying the implementation of RWH in many developing regions, regardless of the water resources available (e.g., Africa, South America, Southeast Asia), and in developed regions with limited water resources (e.g., California, Australia).

However, another driver emerged in highly urbanized areas in developed regions: the need to minimize the rainwater discharge into the stormwater drainage system. In the 1960s, roughly two-thirds of the world's population (2 billion) lived in rural areas, but since 2007, more than half of the world's population lives in urban areas (7.95 billion in 2022) (Ritchie et al. 2024). One of the consequences of urban growth that has already taken place, more noticeably in the developed regions, is the increase in impervious surfaces. These impervious areas are forecasted to increase further throughout the globe with population and urbanization growth, and this implies reductions in surface retention, infiltration and evapotranspiration. This results in an increase in the volumes and peak discharges into the sewers, which, combined with their limited capacity and difficulty in intervening in the existing drainage infrastructures, increase the risk of combined sewer overflows and urban floods. Adding the frequency and magnitude of extreme rainfall events due to climate changes to this situation, it is understandable that there is increased interest in sustainable rainwater management (Wilderer 2004), also referred to as Best Management Practices (Griffiths et al. 2020), Water Sensitive Design (Mguni et al. 2022), Integrated Water Management (Mitchell 2006), Low Impact Development (Griffiths et al. 2020), Sustainable Urban Drainage Systems (Ferrans et al. 2022) or Sponge City (Hamidi et al. 2021) in the literature.

Sustainable rainwater management relies, to a large extent, on the promotion of decentralized solutions to complement/alleviate the existing stormwater drainage system. At the building level, this has been done through incentives to implement solutions such as green roofs and RWH systems. These incentives range from financial support to stormwater fee reduction (Liberalesso et al. 2020). For instance, the city of Hamburg grants a 50% rebate in the water fee for green roofs with a thickness of more than 5 cm and an impervious surface reduction of 20 m² per cubic metre of storage volume for RWH systems with tanks of at least 2 m³. Some cities/regions/countries (e.g., New Delhi, Indore, Kanpur and other locations in India; New Zealand) promoted RWH, making it mandatory for new buildings (Gabe et al. 2012). In others, it is the green roofs that are mandatory (e.g., Saint Laurent, Toronto, in Canada; Tokyo, in Japan; Los Angeles, Portland, San Francisco, in USA; Recife, Porto Alegre, São Paulo, in Brazil; Córdoba, in Argentina; Zurich, Basel, in Switzerland; Copenhagen, in Denmark; Stuttgart, Munich, Hannover, in Germany; Turin, Bolzano, in Italy) (Wilkinson et al. 2022). In most cases, the main motivation underlying the policies towards promoting green roofs and RWH is reducing stormwater volume and peak discharge, but there are other motivations. In India and Brazil, water shortages were the more relevant drivers, whereas in Tokyo, it was the heat island effect reduction that motivated its green roof policy.

The literature on RWH is extensive (e.g., see Silva et al. 2022 for a review; Younos & Parece 2016; Campisano et al. 2017), but it has mainly focused on water savings. Water saving entails a quantitative and a qualitative dimension, since the rainwater needs to meet some minimal quality standards for it to be used. This is logical from the building owner/investor perspective since the financial viability of a RWH system is driven by the relation between the savings associated with the reduction of public water supply consumption and the investment and operation costs of the system. In addition to the RWH potential in each specific case, the water cost plays a key role in financial viability. In fact, higher RWH system performance in terms of water saving may not imply financial viability because water is, in most regions of the globe, an inexpensive commodity.

Green roofs, on the other hand, have been extensively studied (see Zheng et al. 2021 for a review) in terms of their hydrological performance, both in terms of rainwater detention (peak discharge reduction) and retention (volume reduction). Despite the generalized conclusion that green roofs contribute to reducing the rainwater discharged into the urban drainage networks, their performance is extremely variable. This variability is not only due to the location climate (e.g., see Voyde et al. 2010; Stovin et al. 2012; Viola et al. 2017) and the green roof characteristics but also to the specific temporal pattern of the rainfall (Zheng et al. 2021).

However, despite the various studies on the hydrologic performance of green roofs, physical modelling requires a range experimental data for calibration, which, even in the case of their availability from weather stations, are site-specific and may not be representative of each possible location of green roof (e.g., radiation due to shading; wind velocity due to building roughness; temperature due to heat island effect).

Also, the studies on the combined performance of RWH systems and green roofs are limited (e.g., Almeida et al. 2023). In this research effort, the performance of combining a green roof with an RWH system in the city of Turin, Italy, is explored, resorting to (i) an empirical model to estimate the green roof runoff and (ii) a simulation model to mimic the RWH system operation. The goal is to provide a planning-level tool that is easy to implement and use by both academia and industry, and requires only information that is either a design parameter (e.g., rainwater collection area; water consumption) or generally accessible data (e.g., daily rainfall series). The comparison with the performance achieved with a traditional roof is also presented, considering that a traditional roof encompasses a variety of possible impervious roofing solutions (e.g., clay tiles, metal sheets, polymeric membranes).

The present research uses the city of Turin as a case study to demonstrate the tool, providing continuity to past research on the topic (e.g., Carollo et al. 2022). The city of Turin is the administrative centre of the Piemonte Region and is located in the north of Italy, about 240 m above sea level. The climate is Cfa (humid subtropical), according to the Köppen-Geiger classification (Beck et al. 2023). The city extends over an area of 130 km², having 871,850 inhabitants distributed over 36,158 residential buildings (Istat 2011).

Most of the stations are located at elevations identical to Turin (239–368 m), but two are located at higher elevations (Bric della Croce, 710 m, and Pino Torinese, 608 m). These latter stations recorded a lower average annual rainfall (731 mm) than the stations at lower elevations (870 mm), so they were not considered as data sources. Among the remainder, data from the Caselle station was chosen to conduct the study, because it has the longest record series (54 years between 1973 and 2022) and the Pearson correlation with the remaining stations is always above 0.842 (Table 1), meaning that there are no important differences in the rainfall regime over the city of Turin. The length of the rainfall series affects the results of the RWH simulation. In the present case study, the length of the chosen rainfall series, 54 years, is even longer than what is generally suggested for RWH performance analysis (e.g., 30 years, according to Palla et al. 2011). Nevertheless, other authors used rainfall series shorter than 30 years for RWH simulation (e.g., 25 years – Agudelo-Vera et al. 2013; 10 years – Mitchell 2007 and Geraldi & Ghisi 2017).

According to the Turin public water utility, the annual potable water consumption in 2022 was 7.5 × 10⁷ m³ (SMAT Torino 2023). Considering the residential population, the average per capita water consumption is 241 litres per day per person (l/(day.person)). However, this amount includes all water consumption within the city of Turin, both for residential and non-residential uses and for the residents and visitors (i.e., commuters and tourists). Studies specifically on the domestic water demand at the building level in Piemonte Region (e.g., Busca 2018; Salvatico 2020) estimate a mean water consumption of 150 l/(day.person), varying between 144 and 158 l/(day.person) on weekdays and weekends, respectively. The non-potable water fraction was estimated to be one-third of the total consumption (Carollo et al. 2022).

Based on previous research on RWH carried out in Turin (Carollo et al. 2022), the following characteristic buildings were selected to capture the RWH system's performance differences:

• Type 1 – roof area = 240 m²; number of individuals = 11 (non-potable water consumption = 0.55 m³/day).

• Type 2 – roof area = 460 m²; number of individuals = 22 (non-potable water consumption = 1.10 m³/day).

• Type 3 – roof area = 600 m²; number of individuals = 29 (non-potable water consumption = 1.45 m³/day).

Complementary to assuming a constant per capita water consumption, two scenarios were considered to model seasonality:

• Scenario 1 aims to capture households in which non-potable water consumption increases during the summer months (June, July and August), namely motivated by irrigation needs and the occupants spending more time at home (e.g., children on vacation while parents work).

• Scenario 2 pretends to capture the alternative situations, in which the occupants are outside the household during the summer (e.g., vacations outside Turin).

The proposed empirical model for green roof, which transforms rainfall into runoff, has to be calibrated with observations of green roofs with structure (e.g., thickness of the soil layer) and geographic location that are coherent with the case study, located in Turin, Italy. Therefore, a literature review was conducted to obtain data on rainfall and runoff from green roofs in Italy using the following combination of keywords in Scopus: ‘green roof’ and ‘Italy’ and ‘runoff’, resulting in 47 publications (13 in conference proceedings/book series and 34 in international journals). Some studies, despite the keywords being present, were not focused on the hydrological performance of green roofs, but rather on other aspects (e.g., greenhouse gas emissions – Lugo-Arroyo et al. 2024; thermal performance – Cirrincione et al. 2021). Several of the references focused on modelling runoff from green roofs with physical models (e.g., Palla & Gnecco 2019; Pugliese et al. 2022; Mobilia et al. 2023; Pampaloni et al. 2024). Still, various studies report experimental research on green roofs, but the vast majority do not disclose the data or just present it in a graphical or aggregated form (e.g., Mobilia & Longobardi 2020; Raimondi et al. 2023). From the few studies disclosing the data at the rainfall event level (e.g., Longobardi et al. 2019; D'Ambrosio et al. 2022; Pumo et al. 2023), the study by Palermo et al. (2019) was selected since it has the largest (in terms of the number of rainfall events) and most complete (in terms of the number of variables) dataset. In the present work, linear and non-linear multiple regression analysis were used to develop empirical models to estimate the runoff from green roofs using SPSS, version 29. The models were developed to estimate directly the runoff or indirectly the runoff coefficient (runoff/rainfall). The best subsets and Akaike Information Criterion were used to select the variables to include in the models.

The typical mass balance model (e.g., Silva et al. 2015) was used to simulate the RWH system performance, only changing the way in which the first flush is accounted for. First flush diversion is an important part of the RWH process regarding the quality of the collected water. Typically, the first millimetres of rainwater that are collected from the roof during a rainfall event are the most polluted and should be discarded before entering the RWH tank. From a computational viewpoint, however, the first flush is just a volume of water that is not conveyed to the rainwater tank; thus it can possibly be considered as an ‘initial loss’ and aggregated to the other real initial losses (e.g., evaporation/evapotranspiration and retention due to roof type – traditional or green roof, materials and slope). Most research efforts, in fact, either account for the first flush in an overall constant runoff coefficient (e.g., Bashar et al. 2018; Geraldi & Ghisi 2018, 2019; Khan et al. 2021) or discount a constant first flush amount every day (e.g., Carollo et al. 2022). Herein, the first flush was modelled at the rainfall event scale, because (i) assuming an overall constant runoff coefficient is not compatible with the variable runoff from the green roof and (ii) it is reasonable to assume that during rainfall events lasting for several days, there is no need to discharge the first flush every day. In this context, rainfall events were considered independent if there was an antecedent dry weather period (ADWP) of at least 24 h. Other authors (e.g., Stovin et al. 2012) use a 6 h interval to identify independent rainfall events when studying green roofs, but the rainfall data is only available at a daily resolution in the gauge stations that are selected. This is a very common restriction of the rainfall data available from most official sources of information, particularly when looking at long record series. Nevertheless, the use of daily rainfall data is not a limitation, given that computations refer to a large-scale analysis. A wide range of suggested values for the first flush diversion can be found in the scientific literature, from 0.4 to 8.5 mm (e.g., Texas Water Development Board 2005; Thomas & Martinson 2007), depending on the initial and desired turbidity of water. For instance, Kus et al. (2010) found that a 5 mm first flush diversion leads to a quality of water that meets the standards for drinking water. In the present work, the collected rainwater does not have to comply with drinking water standards, and a first flush that corresponds to the first 2 mm of runoff from the roof, in each rainfall event, has been adopted according to Silva & Ghisi (2016).

In the case of the RWH system combined with a green roof, the runoff was estimated at the rainfall event scale, and it was assumed that the retention takes place at the start of the event and is not uniformly distributed over the event duration (i.e., the net-rainfall approach was used). For instance, if the runoff of a rainfall event is estimated to be 90% of the rainfall amount, the rainfall retained in the green roof is the first 10% of the rainfall amount. This implies that, for rainfall events that last for several days, after the retention amount is discounted, the runoff will be equal to the rainfall.

Concerning the replication of this study under different climates, two design parameters need to be defined: RWH catchment area and water demand of the building. Then, the regression equation for the green roof runoff estimation should be calibrated based on data for the specific location, which are climate and technology dependent, since green roof solutions are distinct in different regions of the globe (e.g., substrate thickness tends to be bigger as the climate becomes drier). As a last step, the well-known mass balance simulation for RWH has to be applied.

From the information available in the original data, the maximum intensity was excluded since it is not available from the rainfall data series, remaining as potential predictors the (i) total rainfall amount during the event (P in mm); (ii) rainfall event duration (D in days); and (iii) ADWP in days. Additionally, the mean rainfall intensity (in mm/day) and the natural logarithms of P, D and ADWP were calculated. Table 2 presents empirical models developed for predicting the green roof runoff directly or the runoff coefficient. In the order of appearance of the models in the table, the Adjusted R² is 0.974, 0.873, 0.975 and 0.880, respectively, indicating a better fit of the models using the runoff as a dependent variable.

The sign of the regression coefficients is consistent with what was physically expected: (i) a positive sign for the precipitation implies proportionally more runoff as the total rainfall of the event increases; (ii) a negative sign for the duration, since the durations increase for the same total rainfall amount implies a lower rainfall intensity; and (iii) a negative sign for ADWP, since the moisture content of the soil decreases with the length of the dry period prior to a rainfall event.

For the simulations present below, the model of the linear regression of the runoff coefficient was used since (i) the correlation between the observed and forecasted runoff is identical to the models of the runoff; (ii) the model is simpler (parsimony principle); and (iii) it identifies more accurately the precipitation events with no runoff. Also, the runoff coefficient is physically bounded between 0 and 1, whereas the upper limit of the runoff is the rainfall, which is not a fixed value.

From a financial perspective, the viability of the RWH system decreases when the slope of the relation between tank capacity and the non-potable water savings or rainwater retention efficiency decreases, because it shows that the benefit from investing in a larger tank delivers a smaller increase in benefit. Since a detailed cost-benefit analysis is outside of the scope of the present study, it is possible, simply by visual observation, to conclude that the slope of the curves starts to decrease for a tank of roughly 5 m³ in volume, for both the traditional and the green roof. This size of storage capacity corresponds to the non-potable water supply of 3 days (building type 3) to 10 days (building type 1), which is within the storage length recommended to prevent water quality degradation in the tank. For this tank capacity, the combination of green roof and RWH system would allow savings up to 30% of the non-potable water consumed and retain at least 65% of the rainwater precipitation on the roof of the building. On the other hand, a traditional roof would provide water savings above 35%, but would only retain up to 50% of the rainwater. It is worth noting that in Italy, there is a recommendation to select a tank capacity corresponding to 80% of the maximum water savings (Gnecco et al. 2013), which leads to a tank capacity over 10 m³ in this case study. However, this would mean that the rainwater could be stored for up to 20 days, which might result in water quality issues.

As expected, the wetter months allow for higher non-potable water savings, but lower rainfall retention. The performance differences between traditional and green roofs are according to the annual performance results. However, it is noticeable that there is a greater variability in the rainwater retention efficiency when the RWH system is associated with a green roof. This is explained by the fact that the retention of the green roofs decreases, in percentage, with the rainfall. When the retention capacity of the green roof is exhausted, all remaining rainfall will be converted into runoff, which is more likely to occur when the rainfall event is intense. Additionally, in wetter months, the initial retention capacity tends to be reduced, since the green roof may not have time to dry between rainfall events.

Overall, the most relevant variability results from the building type and the consumption pattern. The monthly variation is large, but expected, and does not affect the decision regarding investing in the RWH system from the perspective of the building owner. It does, however, influence the assessment of the potential contribution of these solutions towards flood risk reduction since it was observed that their combined water retention efficiency decreases non-linearly with the magnitude of the rainfall event.

The growth of population and urban spaces is increasing the pressure on the public water infrastructure, both for water supply and stormwater drainage. This led decision makers to adopt more holistic approaches, integrating both centralized and decentralized solutions. Two common decentralized solutions at the building level that are being promoted by municipalities throughout the globe are RWH and green roofs. These solutions have been extensively studied from multiple perspectives (e.g., financial performance, environmental performance, technical performance – hydrologic and thermal), but in most cases, these perspectives have been considered in isolation.

Since the water supply and stormwater management are the most common drivers for the implementation of RWH, the present research aims at assessing the combined performance of RWH and green roofs in these two dimensions, providing a planning-level tool useful for policymakers. The city of Turin was used as a case study, but the approach proposed is applicable in other contexts. Still, it has a limitation which is the need for data from monitored green roofs to develop an appropriate empirical model to estimate the runoff from the green roof. However, the large majority of the green roofs tend to be of the extensive type, due to their lower implementation and operation cost, but mostly to their lower weight. The additional weight of the green roof plays a decisive role in the decision-making process, considering the impact that it may have on the structure of the building, particularly in seismically active regions. For this type of green roof, there are already numerous experimental studies in many climate contexts (e.g., see Zheng et al. 2021 for an extensive review) that allow for, at least, a crude approximation in most contexts.

In Turin, the analysis of a group of typical buildings shows that the advantages of the combination of RWH and green roofs decrease with the increase of the RWH system storage capacity. This is a positive result for the combination of these two solutions, considering that RWH tanks are the most expensive component of the RWH system, and large rainwater retention periods may induce water quality issues.

The sensitivity analysis shows that both the seasonality of water consumption and the use of rainfall data from different weather stations have a limited impact on the RWH system performance, regardless of the combination with the green roof or not. Furthermore, the monthly variation of performance is large, but expected, and does not affect the decision regarding investing in the RWH system from the perspective of the building owner.

Using the non-potable water consumption to calculate two ratios with the other two governing parameters of a RWH system, generalized relations were developed to enable a simple estimation of a RWH system in Turin with either a traditional or a green roof.

V.S. and I.M. acknowledge the funding from the EEA grants through the project GREENER (FBR_OC2_80).

Conceptualization: VS; Data curation: VS, MC, IB, IM; Methodology: VS, IM; Models and simulations: VS, IM; Supervision: IB; Validation: MC; Visualization: IM; Writing – original draft: VS, IM; Writing – review and editing: MC, IB; Funding: VS, IB. The final version was approved by all co-authors.

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

The authors declare there is no conflict.