You are on page 1of 14

Journal of Hydrology 414–415 (2012) 148–161

Contents lists available at SciVerse ScienceDirect

Journal of Hydrology
journal homepage: www.elsevier.com/locate/jhydrol

The hydrological performance of a green roof test bed under UK climatic conditions
Virginia Stovin a,⇑, Gianni Vesuviano a, Hartini Kasmin b
a
Department of Civil and Structural Engineering, The University of Sheffield, Mappin Street, Sheffield S1 3JD, UK
b
Faculty of Civil & Environmental Engineering, Universiti Tun Hussien Onn Malaysia, 86400 Parit Raja, Batu Pahat, Johor, Malaysia

a r t i c l e i n f o s u m m a r y

Article history: This paper presents new rainfall and runoff data from a UK green roof test bed which has been collected
Received 3 November 2010 almost-continuously over a 29-month period from 01/01/2007 to 31/05/2009. Overall, the monitoring
Received in revised form 31 May 2011 period was fairly typical of the location’s long-term climatic averages, although the data set includes
Accepted 18 October 2011
some extreme events in June 2007, which were associated with serious flooding locally. To focus on
Available online 28 October 2011
This manuscript was handled by
the system’s performance under rainfall events likely to be of interest from an urban drainage/storm-
Konstantine P. Georgakakos, Editor-in-Chief, water management perspective, return period analysis has been applied to identify those storm events
with the assistance of Michael Brian Butts, with a rainfall depth in excess of 5 mm and a return period greater than one year. According to these cri-
Associate Editor teria, 22 significant events have been identified, of which 21 have reliable runoff records. Overall the roof
provided 50.2% cumulative annual rainfall retention, with a total volumetric retention equivalent to 30%
Keywords: during the significant events. The annual performance figures are towards the lower end of a range of
Green roof international data, probably reflecting the fact that rainfall depths may be higher and evapotranspiration
Stormwater management rates lower than in some more continental climatic settings. The roof’s finite retention depth means that
Retention the maximum possible retention percentage declines as storm depth increases, and retention varied from
Return period between 0 and 20 mm, or 0% to 100%. Although some attenuation and delay of peak runoff is generally
Detention
observed (mean peak flow reduction of 60% for the 21 significant events), the irregularity of natural rain-
Substrate
fall patterns, combined with the variable influence of detention storage in specific events, makes the
identification of peak-to-peak lag times difficult and arguably meaningless. Regression analyses have
been undertaken to explore the potential to predict the roof’s hydrological performance as a function
of storm characteristics. However, these are shown to have poor predictive capability, even for the system
from which they were derived. Through a detailed examination of data from three contrasting events, it is
argued that the inter-event processes are too complex to be captured by this type of modelling approach.
Instead, an understanding of the hydrological processes affecting the flux of moisture into and out of the
substrate is required to explain the observed runoff response. Locally-derived evapotranspiration rates
and the roof’s observed maximum retention capacity are utilised to provide pragmatic guidance on the
retention performance to be expected in response to selected design events.
Ó 2011 Elsevier B.V. All rights reserved.

1. Introduction ment processes (i.e. infiltration, attenuation, conveyance, storage


and biological treatment) these techniques are seen to constitute
Stormwater runoff from urban roofs makes a significant contri- a ‘more sustainable’ approach to stormwater management com-
bution to sewerage-derived flooding and urban water quality prob- pared with conventional piped drainage. The SUDS philosophy sug-
lems. In most developed cities, roofs may account for approximately gests that modern approaches to stormwater management need to
40–50% of the impermeable urban surface area. Any technique that control the quantity of runoff, but at the same time address urban
reduces the rate and volume of roof runoff has the potential to con- water quality problems and provide amenity value. Green roofs
tribute to improved stormwater management. have the potential to achieve these three aims simultaneously,
In the UK, the Environment Agency and the Scottish Environ- and genuinely represent an opportunity for engineering to work
ment Protection Agency promote the use of Sustainable Drainage in harmony with natural environmental processes to contribute
Systems (SUDS) for the management of surface water runoff. SUDS to sustainable urban environments. In addition to the quantity-
include, amongst others, green roofs, soakaways, swales, infiltra- quality-amenity SUDS ‘triangle’, the SUDS philosophy also em-
tion basins and ponds. Because of their reliance on natural catch- braces two other important concepts. The first is that rainfall is
generally best controlled as close to the source as possible. The sec-
⇑ Corresponding author. Tel.: +44 (0)114 222 5051; fax: +44 (0)114 222 5700. ond is the concept of a treatment train, where multiple SUDS de-
E-mail address: v.stovin@sheffield.ac.uk (V. Stovin). vices may be combined to provide more robust and beneficial

0022-1694/$ - see front matter Ó 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.jhydrol.2011.10.022
V. Stovin et al. / Journal of Hydrology 414–415 (2012) 148–161 149

stormwater management than any single element can achieve on versity drivers (e.g. to provide habitat for the Black Redstart in Lon-
its own. Internationally the terms BMPs (Best Management Prac- don (Gedge, 2003)) rather than for stormwater management
tices), LID (Low Impact Development), WSUD (Water Sensitive Ur- reasons. Assuming sufficient moisture is available within the roof
ban Design) and Green Infrastructure refer to similar concepts. substrate, green roofs may also promote evaporative cooling in
Stormwater controls need to be designed to accommodate both the summer, thereby offsetting some of the energy requirements
extreme events and to deal effectively with more routine events. associated with air conditioning systems and reducing urban heat
The former is required for flood protection, whilst the latter is re- island effects (Sailor, 2008). Green roofs may also help to mitigate
quired to protect river morphology and receiving water quality. climate change via carbon sequestration (Getter and Rowe, 2008).
In the UK, the expectation is that SUDS devices should be designed Green roofs should be seen as providing an ‘all-round’ contribution
to provide protection from downstream flooding for a minimum of to sustainable urban environments. Further information on these
a 1 in 30 year return period event (CIRIA, 2007b). It is therefore broader performance aspects can be found in: CIRIA (2007a), Dun-
critical to understand how green roofs will function both routinely nett and Kingsbury (2004), Banting et al. (2005), www.green-
and in response to specific design return period rainfall events. roofs.org and www.livingroofs.org, amongst others.
Green roofs provide an opportunity to reduce and attenuate Although green roofs are by no means a new idea (see Köhler,
storm runoff at source (Fig. 1). Extensive green roofs typically take 2004; Herman, 2003), their widespread use, particularly for storm-
the form of a ‘carpet’ of plants, supported by lightweight growing water management purposes, has only emerged more recently.
media and overlying a drainage layer (Fig. 2), whereas intensive Over the last 20 years green roofs have become relatively common-
green roofs incorporate more deeply-planted vegetation. place in Germany, Austria and Switzerland; 14% of the flat roofs in
During a storm event the key hydrological mechanisms operat- Germany (13.5 km2) supported green roofs in 2003 (Herman,
ing within the green roof are the interception of rainfall by the 2003). In the UK the latest CIRIA SUDS guidance (CIRIA, 2007b)
plant layer, infiltration and storage in the substrate, and reservoir strongly endorses green roofs as a SUDS technology. CIRIA has also
storage in the drainage layer. Any excess moisture will be directed produced guidance which specifically relates to the use of green
out of the drainage layer into the downstream drainage system. roofs and living walls on buildings (CIRIA, 2007a). However, uptake
Moisture will be lost to the atmosphere as a result of evapotrans- to date has been rather limited in the UK. Possible reasons for this
piration during dry periods. include: the perceived risks associated with a technology that has
In addition to stormwater source control, green roofs may also not been widely used in the UK; the lack of UK validated data and
increase the design life of a roof, improve air quality (Banting modelling tools to enable green roofs to be evaluated against alter-
et al., 2005), add aesthetic value to urban architecture and enhance native stormwater management approaches; and the current lack
biodiversity (English Nature, 2003). Many of the roofs that have of any real incentives for the UK water utilities, local authorities
been constructed to date in the UK have been stimulated by biodi- or developers to implement any form of SUDS at a significant scale.
It is sometimes argued that the land-use requirements associ-
ated with SUDS make them incompatible with high density urban
environments; green roofs differ from many other SUDS ap-
proaches in that they have no additional land-take requirement be-
yond the footprint of the building.

2. Literature review

To date, hydrological green roof research has been undertaken


using test facilities at Michigan State University (VanWoert et al.,
2005), the Penn State Center for Green Roof Research (Berghage
et al., 2007), and European facilities at KULeuven (Mentens et al.,
2003) and Neubrandenburg (Köhler et al., 2002). These test facili-
ties complement a number of full-scale instrumented green roof
demonstration projects (e.g. Hutchinson et al., 2003). Typically
vegetation type, planting substrate and slope have been investi-
gated (e.g. VanWoert et al., 2005), with continuous runoff monitor-
ing being provided either by tipping bucket gauges or rainwater
Fig. 1. Rainfall runoff response (schematic). collection tanks incorporating depth sensors. These studies consis-
tently show that high levels of stormwater attenuation can be
achieved with the use of green roofs. For example, Kolb (2004) re-
ported that green roofs reduced annual runoff by 45–70%, Moran
et al. (2004) report 60% total rainfall retention and an 85% reduc-
tion in peak flowrate. However, seasonal effects can be significant.
Johnston et al. (2004) noted that runoff could rise to 100% with
prolonged heavy rainfall. In Portland, where ‘‘ecoroofs’’ must be
irrigated for 3 months in the summer, Liptan (2003) suggests that
a 10–35% volume reduction during the wet season and 65–100%
volume reduction in the dry season can be achieved. Van Seters
et al. (2007) report similar variations for an extensive green roof
in Toronto. Palla et al. (2010) provide a useful review of these
and other similar studies.
Initial losses, or the depth of rainfall that is retained by the roof,
are a function of the antecedent dry weather period (ADWP), but
Fig. 2. Green roof hydrological processes. also of the weather conditions associated with the ADWP. In wet
150 V. Stovin et al. / Journal of Hydrology 414–415 (2012) 148–161

winter conditions a roof will be slower to recover its retention monitoring programmes have generally been quite short (2 years
capacity than in summer, when evaporation rates will be expected or less), and few, if any, of the existing data sets have been evalu-
to be significantly higher. ated in terms of how typical they were of long-term climatic record
Many authors state a nominal maximum retention depth, but or in relation to event return periods. Based on published data it is
few provide sound justification. Some erroneously assume that surprisingly difficult to provide a reasonable estimate for the level
the porosity (or the water holding capacity) of the substrate im- of retention to be expected for a specific roof configuration in re-
plies a certain level of retention. However, this can be misleading, sponse to either an annual time-series of rainfall or a design event
as – in practice – the roof will never achieve the ‘oven-dry’ condi- corresponding to a specific return period.
tions required to achieve such high ‘potential’ retention levels. A The primary objectives of this paper are to present new hydro-
better indication of retention is obtained from field measurements logical performance data relating to a continuously-monitored
of rainfall and runoff obtained from real storm events. green roof test bed in the UK, and to provide an interpretation of
Voyde et al. (2010) provided a comprehensive analysis of data the data in terms of return period analysis and drainage design.
collected from an extensive living roof in Auckland, New Zealand. To enable generalisation of the data set such that performance un-
They identified a cumulative retention efficiency of 66% based on der unseen climatic inputs (for example in other locations, or in re-
12 months of continuous monitoring, with a median of 82% reten- sponse to climate change) can be estimated, hydrological
tion per event for rainfall events greater than 2 mm. Based on six modelling tools are required. Therefore, a further objective is to as-
different roof configurations, no statistically significant differences sess the suitability of regression-based and process-based model-
in runoff response as a result of substrate composition, substrate ling tools for predicting the observed runoff phenomena.
depth or season were identified. The authors acknowledged that
their data record did not include any events greater than or equal
to the 2 year return period 24 h duration event; they also observed 3. Methodology
that retention efficiency decreased as storm depth increased and
when antecedent conditions resulted in high levels of moisture 3.1. Description of the test rig
within the green roof substrate.
Berghage et al. (2007) operated a comprehensive test pro- The test bed (3  1 m) (Fig. 3) uses a standard commercial (Alu-
gramme on behalf of the USEPA. DeNardo et al. (2005) measured masc/Zinco) extensive green roof system, comprising a sedum veg-
retentions in the order of 4–6 mm for a range of autumn storms etation layer growing in 80 mm of substrate. The basic
on test beds with 12 mm drainage +78 mm substrate (90 mm total configuration is as shown in Fig. 2. The base of the rig is laid at a
depth). Their work on evapotranspiration suggests that water loss slope of 1.5°. The substrate is composed of a mixture of crushed
(or depression storage recharge) will initially be around 3 mm/day brick and fines. A fine particle filter membrane separates the sub-
for vegetated (Sedum spurium) and 1.5 mm/day for unvegetated strate from the underlying FloradrainÒ FD 25 ‘egg box’ drainage
roof surfaces (at 27 °C). This rate is a function of temperature, layer. The drainage layer alone has a nominal retention capacity
but also declines exponentially with time as the amount of avail- of 3 l/m2 (i.e. 3 mm rainfall).
able moisture declines. Their data suggests that 48 h after satura- Rainfall is monitored using an Environmental Measures ARG-
tion, approximately 5 mm rainfall might be retained by a 100 tipping bucket rain gauge sited adjacent to the test bed. The
vegetated roof. This figure corresponds to 5% of the substrate rain gauge has a resolution of 0.2 mm. Runoff from the roof is col-
depth. Palla et al. (2008) presented preliminary findings from a lected in a tank below the test bed via a gutter at its downstream
deep (350 mm substrate) full-scale roof operating in a Mediterra- end. The collection tank is tapered to provide greatest sensitivity
nean climate (Genoa). Although complete retention was observed for the first 0.5 mm of runoff. Depth in the collection tank is mon-
for many storm events, those that did result in runoff typically itored using a pressure transducer to provide a continuous record
demonstrated retention depths that corresponded to between 3% of runoff via an appropriate calibration relationship. The data res-
and 11% of the total substrate depth (e.g. for a large event olution is approximately 2  104 mm in the most sensitive zone,
(138.2 mm), the volume retained was 13.13 mm, or 3.75% of the decreasing to 7  103 mm; in both cases considerably higher than
substrate depth). In contrast, Uhl and Schiedt (2008) present data the rain gauge (0.2 mm). The collection tank is automatically emp-
that suggest a maximum retention of 40 mm in a 100 mm sub- tied every time the depth stored approaches the capacity of the
strate (i.e. 40% of substrate depth). tank. Data from the pressure transducer and the rain gauge is
Care must be taken when presenting/interpreting green roof
hydrological performance data. Palla et al. (2010) provided nine-
teen months’ data comprising 30 individual storm events. 11
events exceeded 40 mm depth, and the total monitored rainfall
was 1165.1 mm. On an event-by-event basis the mean retention
was given as 51.5%. The authors suggest that this is evidence of
excellent storm water control performance. However, from a
drainage engineering perspective the mean retention per event
may be less meaningful than an understanding of the system’s
overall annual retention or its response to extreme events. The pre-
sented data shows that, for many of the larger events, retention
was actually below 25%. As these larger events account for the bulk
of the total rainfall, the total volume retained was around 35%. An
annual retention performance of, say, 60% should not be used to in-
fer that the roof will have the capacity to retain 60% – or even 30% –
of a significant rainfall event. Similarly, there are many different
roof configurations (substrate type and depth, slope, size, plant-
ing), which can all be expected to perform differently, and the
same roof configuration will not demonstrate the same perfor-
mance characteristics in two different climatic regimes. Green roof Fig. 3. University of Sheffield green roof performance monitoring test bed.
V. Stovin et al. / Journal of Hydrology 414–415 (2012) 148–161 151

logged using a Campbell Scientific data logger (CR1000) at 1-min intensity (Qp); Peak attenuation (PA); Time to start of runoff (Rs);
intervals. Runoff delay (peak-to-peak) (tp); and Runoff delay (centroids) (tc).
As the surface area of the test rig is only 3 m2, it is anticipated From a stormwater management perspective, there is a require-
that the monitored data will underestimate detention (i.e. the lag ment for modelling tools that enable both the volume and timing
and attenuation of the runoff hydrograph) compared with a full- of runoff to be predicted in response to an arbitrary or design rain-
scale green roof installation. However, retention effects (initial fall input. Regression analysis was undertaken to establish whether
losses) are a function of substrate depth and moisture holding useful equations could be identified. In all cases normality checks
capacity, and are not expected to be affected by the physical scale (Shapiro–Wilk test) were undertaken, and suitable transformations
of the experimental set-up. Whilst acknowledging this limitation were applied to improve normality. In two cases (ADWP and i), the
in the detention performance data, it should be noted that the pri- Shapiro–Wilk significance was below 0.05, indicating that normal-
mary emphasis in this paper is on the roof’s retention performance. ity assumptions were not met. Although it may be argued that
It should also be noted that the aim of the study was to provide non-parametric tests (e.g. Spearman’s rho) would be more suitable
information directly relating green roof runoff to rainfall, rather under such circumstances, non-parametric tests lack the predictive
than to provide a comparison between a conventional rooftop capacity associated with parametric approaches, so they were not
and a green roof. This was also the case in the Voyde et al. utilised. For all regression analyses, the coefficient of determina-
(2010) study, although they suggested that – based on published tion (Pearson’s R2) is used to indicate correlation strength, i.e. the
literature – between 62% and 91% of rainfall becomes runoff from proportion of variability that can be explained by the proposed
conventional asphalt and gravel ballast rooftops. The percentage model.
runoff is likely to be higher for tiled and pitched roofs. Initially scatter plots and single-parameter regression analyses
were used to identify predictive relationships for retention based
3.2. Overview of the data record on the storm property parameters that were expected to influence
retention most strongly, i.e. rainfall depth, ADWP and rainfall
The green roof rainfall–runoff monitoring record contains intensity. Comprehensive multiple regression analyses were then
29 months’ continuous rainfall data (01/01/07 to 31/05/09). How- performed to generate predictive relationships for all eight hydro-
ever, pressure transducer data collected during the periods logical performance characteristics using forwards, backwards and
17/01/08 to 03/03/08 and 17/03/08 to 01/04/08 are unavailable stepwise linear regression, this time including all five of the storm
due to instrumentation problems believed to be linked to a period parameters. The multiple linear regression analyses were per-
of cold weather. formed systematically to ensure that the model generated in each
Section 4 provides a detailed review of the rainfall time series. case had the highest possible R2 and that non-significant terms
In Section 4.2 comparison with the long-term climatic record for were eliminated. It should be noted that the multiple regressions
this location (Met Office, 2010) is provided, whereas in Section 4.3 were performed using sub-sets of the five storm parameters to en-
the storm event characteristics are interpreted with respect to rel- sure that non-independent parameters were not erroneously com-
evant depth-duration-frequency data taken from the Flood Estima- bined. Specifically, rainfall depth was described using either TP or i
tion Handbook (FEH CD-ROM, NERC, 1999). Rainfall events with an and RD (TP = i x RD) and storm intensity was described using either
estimated return period of greater than one year have been termed i or Rp.
‘significant’ – as distinct from ‘routine’ – events for the purposes of ADWP has certain limitations as a predictor of the actual mois-
this paper. ture retention capacity in the green roof substrate at the onset of a
storm event. This is discussed further in Section 6. Additional mul-
3.3. Data analysis tiple linear regression analyses were therefore undertaken using a
new set of parameters describing antecedent weather conditions.
Section 5 provides a detailed analysis of the runoff response These were: the total rainfall in the 24 h, 7 days and 14 days pre-
data. As outlined earlier, stormwater management requirements ceding the storm event (P24, P7 and P14); and the mean temperature
are such that the roof’s response to both routine and extreme over the past 24 h or the long-term mean monthly temperature for
events is of interest. The data has therefore been analysed in terms the relevant month (T24 and T30).
of cumulative annual retention and both mean and median reten- The influence of interactions between the various storm and
tion per storm event. antecedent weather parameters was also evaluated through the
The rainfall record has been classified into 468 individual rain- application of non-linear multiple regression analyses. In this case
fall events. No minimum rainfall depth was specified, so the storm the large number of potential combinations prevents an objective
event data set contains events with as little as 0.2 mm rainfall, and systematic approach; instead subjective judgement and trial-
which is the minimum depth recordable by the rain gauge. Individ- and-error were used to identify and evaluate potentially-meaning-
ual events were defined as being separated by continuous dry peri- ful combinations.
ods of at least six hours. This definition was selected based on The raw data collected from the field logger was processed
VanWoert et al. (2005), Getter et al. (2007) and Voyde et al. using custom-written MATLAB scripts. All the detailed statistical
(2010), amongst others. analyses were undertaken using SPSS 16, whereas Microsoft Excel
The per storm event analysis has been sub-divided into reten- was used for all other analyses.
tion averaged across all storms, storms larger than 2 mm (for com-
parison with Voyde et al. (2010)), and for the greater than 1 year
return period ‘significant’ events. The significant event data was 4. Rainfall event analysis
also examined to determine whether seasonal variations in the
roof’s response were evident. 4.1. Sheffield’s climate
The following five storm parameters were evaluated: Rain dura-
tion (RD); Rainfall depth (TP); Antecedent Dry Weather Period Like the rest of the United Kingdom, the climate in Sheffield is
(ADWP); Mean storm intensity (i); and Peak 5-min storm intensity generally temperate (Fig. 4). Between 1971 and 2000 Sheffield
(Rp). Eight hydrological performance variables were defined and averaged 824.7 mm of rain per year, with December the wettest
derived from the monitored rainfall and runoff data: Runoff depth month (91.9 mm) and July the driest (51.0 mm). The mean yearly
(TR); Retention depth (TS); Retention % (VR); 5-min peak runoff high temperature is about 12 °C, with July being the hottest month.
152 V. Stovin et al. / Journal of Hydrology 414–415 (2012) 148–161

Fig. 4. Average monthly temperature and rainfall data for Sheffield (1971–2000). Source: UK Met Office.

The mean minimum temperature in January and February is 1.6 °C, (2182.9 mm). The rainfall depth was 10% higher than would typi-
and snow is not uncommon during the winter. cally have been expected according to the 1971–2000 monthly
averages (Met Office, 2010). The summer of 2007 experienced
4.2. The data record in context some extreme conditions, with June experiencing more than four
times, and July more than double, the normal monthly totals. Many
Clearly the data record does not constitute a long record in areas of the UK were affected by severe flooding at this time. Con-
hydrological terms. The following section therefore examines versely, totals for April, August, September and October 2007 fell
how representative the specific weather conditions experienced below 50% of the norm. The summer of 2008 was wetter than aver-
are for the location. Fig. 5 presents the monthly rainfall time-series, age, whilst conditions over the following winter (November 08–
alongside the record from Sheffield’s official Met Office recording April 09) were notably drier.
station at Weston Park. The Met Office gauge is located at 4339E
3872N, less than 1 km west of the study gauge. Any differences be- 4.3. Analysis of the rainfall time-series
tween the two records are therefore likely to reflect the effects of
shading from buildings that surround the study location. During The rainfall record has been classified into 468 individual rain-
the study period the total rainfall was 2038.8 mm at the study fall events. For consistency with previous green roof performance
location, 93.4% of the depth recorded at the official Met Office site analysis, individual events were defined as being separated by con-

Fig. 5. Monthly rainfall totals for the study period.


V. Stovin et al. / Journal of Hydrology 414–415 (2012) 148–161 153

tinuous dry periods of at least six hours. However, the use of 6 h 12 h rainfall intensities have therefore also been compared with
Antecedent Dry Weather Period (ADWP) is not universal, and many FEH rainfall intensities. Overall there were 22 events with an ADWP
drainage designs are based on an assumed ADWP of 24 h. It is clear greater than or equal to 6 h which experienced event mean or max-
that the way in which a storm event is defined will significantly imum 1, 6 or 12 h intensities greater than that expected for a
influence any conclusions reached about the overall retention 1 year return period. Of the 22 significant rainfall events, one
and detention performance of the green roof. For example, if more was found to occur on 20/01/08 and is excluded from any further
events are included with short ADWPs this may result in lower analysis due to the lack of reliable runoff data.
mean retention per event than if all analysed events are preceded The WaPUG Code of Practice (2002) is widely-used by UK drain-
by longer ADWPs. Conversely, if the smallest events are completely age engineers as a basis for the calibration and verification of sewer
excluded from analysis, many events with 100% retention will be flow models. It provides the following definition of a storm event
eliminated, and the mean retention percentage per storm event suitable for sewer flow model verification purposes: ADWP > 24 h;
may be reduced as a consequence. Rainfall depth >5 mm; and rainfall intensity >6 mm/h for more
As highlighted above, few – if any – previous green roof perfor- than 4 min. As explained above, a minimum ADWP of 6 h has been
mance studies have provided return period analysis for monitored adopted here for consistency with previous green roof research.
events. The analysis that follows will therefore attempt to consider However, the events all exceed the WaPUG Code’s minimum depth
the roof’s performance for ‘significant’ (here defined as >1 year Re- and intensity criteria. For the 21 significant events highlighted
turn Period) events, in addition to its overall performance. Where above, the minimum rainfall depth is 8.8 mm and all events expe-
relevant, a distinction will also be made between events with short rience rainfall intensities of 7.2 mm/h or more for at least 5 min
(6–24 h) ADWPs and longer (>24 h) ADWPs, as well as events with (Table 1).
TP > 2 mm. In the following section the green roof’s performance will be
The full rainfall time-series has been analysed against Sheffield evaluated, both in terms of annual retention and the retention
return period data taken from the FEH CD-ROM (NERC, 1999). The and detention associated with significant events.
FEH rainfall data supersedes the FSR Rainfall (NERC, 1975), and it
should be noted that the later analysis incorporates more data
and better spatial resolution than the FSR data. In many parts of 5. Green roof performance characteristics
the UK FEH rainfall depths for the 100 year return period storm
are up to 1.4 times greater than the equivalent FSR event. It should 5.1. Annual
be appreciated, therefore, that events of the same depth may have
a lower FEH return period compared with previously-determined Considering the approximately 27 months for which both a
FSR values. rainfall and runoff record are available and accurate, a total of
Fig. 6 compares all the monitored storm depths and durations 1892.2 mm of rain fell. 949.4 mm was retained by the green roof,
against the relevant FEH data set for Sheffield. Fig. 6 clearly shows which equates to a cumulative annual retention of 50.2%. This
– as expected – that most events fall below the 1 year return period may be compared to previously-published annual retention data.
threshold. Fourteen events have a return period of greater than one From a range of studies – including both test beds and full-scale
year, with three greater than two years and one greater than installations – annual volumetric retention figures appear to range
10 years. The largest event (starting on 13/06/2007) had 99.6 mm between 50% and 80% (Hutchinson et al., 2003; Moran et al., 2005;
rainfall over 42:29 h, and has a return period of approximately 1 Liu and Minor, 2005; Carter and Rasmussen, 2006; VanWoert et al.,
in 16 years. The rainfall record contains a reasonable distribution 2005; Getter et al., 2007; Jarret and Berghage, 2008; Uhl and
of short and long-duration events. Schiedt, 2008; Mentens et al., 2006; and Voyde et al., 2010). The
The depths and durations presented in Fig. 6 are event totals. It performance of the Sheffield test bed therefore falls at the lower
is possible for an event represented in this way to appear to have a end of previously-reported data. This probably reflects the fact that
return period less than 1 year, even if rainfall intensities exceed the the climate in Sheffield is relatively cool and wet, with moderate
1 in 1 year threshold for part of the event. The peak 1 h, 6 h and evapotranspiration rates even in the height of summer. The test

Fig. 6. Rainfall characteristics for the 29-month data series compared with FEH return period estimates for Sheffield.
154 V. Stovin et al. / Journal of Hydrology 414–415 (2012) 148–161

Table 1
Rainfall characteristics for the significant rainfall events (Return Period > 1 year).

Event start Rain duration Rainfall ADWP Mean intensity Peak 5-min Return period
(hh:mm) depth (mm) (hh:mm) (mm/h) Intensity (mm/h)
Event 1h 6h 12 h
RD TP ADWP i Rp
18/01/2007 01:11 24:17 27.0 10:26 1.11 21.6 – – >1 >1
20/01/2007 19:47 24:18 38.6 9:02 1.59 14.4 1.54 >1 >1 >1
13/05/2007 12:34 21:30 29.8 16:04 1.39 12.0 1.20 – >1 >1
12/06/2007 05:38 2:03 12.8 199:14 6.24 28.8 1.56 >1 – –
13/06/2007 15:39 42:29 99.6 31:58 2.34 21.6 15.91 >2 >2 >10
15/06/2007 17:54 9:19 16.2 7:46 1.74 12.0 – >1 >1 –
24/06/2007 22:12 22:41 58.0 6:00 2.56 28.8 4.73 >1 >2 >5
26/07/2007 06:56 13:29 12.6 13:25 0.93 33.6 – >1 – –
15/01/2008 02:51 10:23 21.0 7:19 2.02 9.6 1.22 – >1 >1
28/05/2008 14:33 8:11 11.0 6:24 1.34 19.2 – >1 – –
02/06/2008 20:40 23:27 21.6 34:04 0.92 9.6 – >1 >1 –
26/06/2008 13:22 9:05 16.6 21:57 1.83 16.8 – – >1 –
19/07/2008 00:13 5:00 8.8 12:49 1.76 26.4 – >1 – –
31/07/2008 10:12 7:15 19.8 27:22 2.73 48.0 1.39 >2 >1 >1
01/08/2008 01:44 2:24 18.8 8:17 7.83 50.4 2.71 >5 >1 –
12/08/2008 05:13 7:02 18.0 13:28 2.56 24.0 1.25 >1 >1 –
16/08/2008 19:05 9:26 18.8 49:56 1.99 14.4 1.13 >1 >1 –
03/09/2008 13:43 7:29 11.4 14:31 1.52 28.8 – >1 – –
05/09/2008 04:36 22:29 46.8 14:38 2.08 26.4 2.24 >1 >1 >1
04/10/2008 17:19 15:09 27.8 30:23 1.83 7.2 1.38 – >1 >1
28/04/2009 16:37 6:39 14.6 9:39 2.20 24.0 1.02 >1 >1 –
MEAN 14:00 26.17 25:56 2.31 22.74
MEDIAN 9:26 18.8 13:28 1.83 21.6
MINIMUM 2:03 8.8 6:00 0.92 7.2
MAXIMUM 42:29 99.6 199:14 7.83 50.4

bed’s substrate depth is also shallower than some, although it is depth of runoff recorded for the storm event. The system’s reten-
typical of current commercially-available extensive systems. The tion depth (i.e. initial losses) is calculated as TP–TR. This is denoted
relatively poor performance is probably an underestimate of as TS (for Total Storage). The Volumetric Retention (VR) is the ratio
long-term performance, given that the rainfall total during the TS/TP, and is expressed as a percentage. It may be seen that the
monitored period was 10% higher than the long-term mean. mean retention for significant storms is approximately 43%. How-
The cumulative retention (on average 50.2% of 824.7 mm mean ever, this is a very variable figure; the retention on a single storm
annual rainfall, or just over 400 mm per annum) demonstrates that basis varies from zero to 100%. The total depth retained over the 21
the roof can make a significant contribution to the total volume of events was only 29.3% of total rainfall, as retention is naturally
stormwater that might otherwise impact upon watercourses and/ lower for the larger storms. Similarly, the median retention was
or require treatment (approximately 4000 m3 per impermeable 30.2%. Nonetheless, evidence that green roofs can retain even one
ha per year). However, it is also critical to have a fuller understand- third of the runoff from significant events provides strong support
ing of the roof’s ability to retain and detain flows from larger ex- for their role in urban surface water management strategies.
treme events which are more likely to contribute to significant The maximum observed retention depth was 19.8 mm
catchment flooding or CSO problems. (13/05/07). Subsequent analysis of the data therefore assumes that
the roof’s practical maximum storage capacity is approximately
5.2. Storm event 20 mm.
Fig. 7 shows that for return periods of between 1 and 2 years,
In two of the storms (02/06/08 and 01/08/08) the runoff mea- retention of up to 100% is possible. This is to be expected, given
surement system failed to record the latter part of the event, as a that, for a one-year return period, the rainfall depth associated
result of debris becoming trapped in the collection barrel’s sole- with short duration events (<12 h) falls below the roof’s maximum
noid valve. Because the start of runoff had been captured, it was retention capacity of 20 mm (see Fig. 6). Two year return period
possible to fit a conceptual model (Kasmin et al., 2010, also dis- events of less than 3 h duration are also smaller than the roof’s
cussed later in this paper) to the data and thereby identify the maximum retention capacity.
retention depth (initial losses) with a high degree of confidence. There is very limited data for return periods of greater than two
However, confidence in the temporal response characteristics of years. The maximum retention observed for the four monitored
the model is not sufficient to allow it to be used it to identify events in this category was 29.3%, whilst the 16 year return period
any of the temporal runoff parameters, which have not therefore event provided less than 15% retention. For higher return period
been calculated for these two events. This provides 21 storms for events, the expected rainfall depth (Fig. 6) significantly exceeds
volumetric analysis and 19 for temporal profiles. the roof’s maximum retention capacity. For example, the 1 in
The 21 storms vary widely (Table 1), ranging in duration (RD for 10 year 24 h duration event depth is 72 mm; compared with the
Rainfall Duration) from just over 2 h to almost 2 days and in depth roof’s maximum retention capacity of 20 mm, this implies that
(TP for Total Precipitation) from 8.8 to 99.6 mm. Two alternative the maximum retention possible will be 28%. For a 1 in 10 year
rainfall intensities are presented in Table 1. i (mm/h) denotes the 6 h event (42 mm) the maximum retention possible would be 48%.
mean storm intensity, i.e. TP/RD, whereas Rp denoted the peak Voyde et al. (2010) considered all storm events with rainfall
intensity measured over 5 min, which on average was ten times depths greater than 2 mm, on the assumption that rainfall events
higher than the mean storm intensity. <2 mm are unlikely to produce runoff from a conventional roof.
Table 2 summarises the corresponding green roof performance They found a mean per-event retention efficiency of 78%, with a
characteristics. The runoff depth (TR for Total Runoff) (mm) is the median value of 82%. The comparable analysis for the Sheffield
V. Stovin et al. / Journal of Hydrology 414–415 (2012) 148–161 155

Table 2
Green roof performance characteristics.

Event start Runoff Retention Retention 5-min peak runoff Peak Time to Start of Runoff delay Runoff delay
depth (mm) depth (mm) (%) intensity (mm/h) attenuation (%) Runoff (min) (peak-to-peak) (min) (centroids) (min)
TR TS VR Qp PA Rs tp tc
18/01/2007 01:11 25.33 1.67 6.18 11.28 47.79 15 7 42.74
20/01/2007 19:47 36.75 1.85 4.79 11.55 19.81 18 5 70.22
13/05/2007 12:34 9.98 19.82 66.50 4.17 65.22 262 90 198.57
12/06/2007 05:38 0.01 12.79 99.95 0.02 99.93 483 103 83.38
13/06/2007 15:39 86.48 13.12 13.17 15.24 29.43 350 9 261.32
15/06/2007 17:54 16.19 0.01 0.04 8.97 25.26 9 7 39.04
24/06/2007 22:12 57.96 0.04 0.07 14.92 48.20 12 4 51.69
26/07/2007 06:56 10.06 2.54 20.18 13.89 58.67 14 6 66.64
15/01/2008 02:51 19.75 1.25 5.97 6.63 30.96 9 9 41.30
28/05/2008 14:33 1.67 9.33 84.79 2.52 86.89 78 44 31.12
02/06/2008 20:40* 15.10 6.50 30.20 451
26/06/2008 13:22 0.27 16.33 98.36 0.59 96.48 359 283 181.41
19/07/2008 00:13 4.25 4.55 51.66 3.39 87.16 13 35 78.77
31/07/2008 10:12 3.15 16.65 84.10 13.66 71.54 41 54 4.50
01/08/2008 01:44a 13.30 5.50 29.30 4
12/08/2008 05:13 5.24 12.76 70.89 2.41 89.97 73 77 78.54
16/08/2008 19:05 9.46 9.34 49.68 10.12 29.70 396 18 85.03
03/09/2008 13:43 1.45 9.95 87.30 1.32 95.43 218 43 121.65
05/09/2008 04:36 44.53 2.27 4.84 16.72 36.66 35 49 54.63
04/10/2008 17:19 24.35 3.45 12.40 4.80 33.40 56 12 75.36
28/04/2009 16:37 3.32 11.28 77.23 6.57 72.62 106 272 100.39
MEAN 18.51 7.67 42.74 7.83 59.22 142.95 54.16 87.70
MEDIAN 10.06 6.50 30.20 6.63 58.67 56 18 75.36
MINIMUM 0.01 0.01 0.04 0.02 19.81 4 49 4.50
MAXIMUM 86.48 19.82 99.95 16.72 99.93 483 283 261.32
a
Runoff record missing from the end of the event – recreated by model fit.

storm water. The mean retention was 71% for storms with an Rs
over one hour, but only 24% for storms with an Rs below one hour.
Various parameters have been defined that attempt to link the
timing and magnitude of this peak runoff to the peak in the rainfall
hyetograph. The simplest approach – which literally evaluates the
time difference between the two peaks (tp) – ranges from instanta-
neous to nearly five hours. One value (05/09/08) is actually nega-
tive. This definition is highly sensitive to the timing of tips in the
rain gauge. An alternative approach is to evaluate the difference
between the centroids of the rainfall and runoff distributions (tc).
Both approaches suggest that the lag is of the order of one hour.
However, careful inspection of the temporal rainfall and runoff
profiles (see Section 6) suggests that runoff will follow rainfall al-
most instantaneously, unless there is spare retention capacity. In
Fig. 7. Retention as a function of return period. most instances the apparent lag effect is a function of retention
(initial losses) and not a result of detention (temporary storage).
It should be noted, however, that more significant temporal runoff
dataset suggests mean and median per-event retention values for delays are likely to arise in a full scale green roof.
>2 mm rainfall events of 61% and 62% respectively. It should be
noted, however, that whilst the Sheffield dataset includes four 5.3. Seasonal analysis
events with return periods in excess of two years, the Auckland
data contained no events larger than the 2-year 24 h return period Several previous studies (Hutchinson et al., 2003; Johnston
event. Furthermore, Voyde et al. (2010) acknowledge that the per- et al., 2004; Liptan, 2003; Liu and Minor, 2005; Mentens et al.,
formance of the roof would be expected to be reduced for large 2006; Uhl and Schiedt, 2008; Van Seters et al., 2007) have pre-
storm events or when the roof is already at field capacity. sented retention results on a seasonal basis, typically showing that
The peak runoff intensity (Qp) is evaluated over a 5 min time the roof’s retention capabilities are reduced in winter and in-
interval. It ranges from zero (the 12/06/2007 event barely gener- creased in summer, as a result of the different seasonal rainfall to-
ated any runoff) up to 16.7 mm/hr. The magnitude of peak attenu- tals and evapotranspiration rates. Interestingly, however, Voyde
ation (PA) has been calculated for the aggregated 5-min data, with et al. (2010) were unable to detect any statistically significant sea-
the mean attenuation found to be around 60%. Again, this is a sig- sonal variations in their New Zealand data set.
nificant finding for urban drainage applications. The set of storms in the present analysis is highly biased to-
The time to start of runoff, Rs was defined as the time between wards the summer months. From the 21 selected storms, 12 oc-
the start of the rainfall and the time at which the total runoff ex- curred in June, July or August and three each in spring, autumn
ceeded 0.01 mm. Ten storms had an Rs value above one hour, while and winter (the one rejected storm also occurred in winter). Fur-
eleven had a value below one hour. Empirical analysis confirmed thermore, the three storms with the largest return periods all oc-
that a high Rs was associated with a high level of retention; every curred in the summer. Consequently, this very uneven spread of
storm with an Rs over one hour retained an above average depth of events causes difficulty in performing a meaningful seasonal anal-
156 V. Stovin et al. / Journal of Hydrology 414–415 (2012) 148–161

ysis of the roof’s performance. In addition, due to the large climatic event performance. Initially scatter plots and single-parameter
variations that occur naturally within seasons, the value of such an regression analyses were used to identify potential predictive rela-
analysis is questionable. For a very basic overview, however, the tionships for retention based on Total Rainfall, ADWP and Rainfall
retention of the green roof on a seasonal basis is given in Table 3. Intensity.
Table 3 suggests that the greatest retention occurs during spring Fig. 8a shows a strong relationship between rainfall depth and
(1st March to 31st May) and not summer. However, the storm runoff depth, which is not surprising. The plot essentially shows
events that occur during the summer are on average 45% larger. – as was evident in Table 2 – that the total runoff tends to be
Furthermore, during the study period the mean cumulative sea- 0–20 mm below the total rainfall. The amount of retention (be-
sonal rainfall was 303.8 mm in summer but only 167.3 mm in tween 0 and 20 mm) is assumed to depend on the storage available
spring. Additionally, the mean peak intensity of the summer events in the system at the onset of the storm event. Fig. 8b shows the
was 42% higher than the spring events, 26.2 mm/h compared to same phenomenon in terms of the percentage retention. The max-
18.4 mm/h. The mean retention depth is almost equal in spring imum possible percentage retention decreases with any increase in
and summer, in both cases near to the limit of the green roof’s rainfall depth over 20 mm. The observed event data all fall on or
retention capacity. It may be concluded that the perceived lack of below this optimal performance curve. Table 4 provides informa-
retention during the summer months is primarily due to the larger tion on the significance of the regression relationships. The t-statis-
rainfall depths of the summer storms. tic indicates that TP is a determinant of both TR and VR, with a
significance of p = 0.01. However, R2 values of 0.72 and 0.33 respec-
5.4. Regression analysis tively provide limited confidence in the predictive capability of the
derived relationships.
From an engineering, stormwater management, perspective, Based on the assumption that a longer ADWP provides more
there is a requirement for modelling tools that enable runoff per- time for the green roof substrate to recharge its storage capacity
formance to be predicted in response to an arbitrary or design rain- by drying out, attempts were made to correlate both retention
fall input. The following analysis therefore explores whether depth and percentage retention with ADWP (Fig. 8c and d). How-
regression analysis can provide a robust predictive tool for storm ever, the plots show large amounts of scatter, and the R2 values

Table 3
Seasonal retention of the green roof.

Season No. of events Average rainfall depth (mm) Average retention (%) Average retention (mm)
Winter (01 December to 28/29 February) 3 28.9 6.06 1.7
Spring (01 March to 31 May) 3 18.5 76.18 14.1
Summer (01 June to 31 August) 12 26.8 52.12 14.0
Autumn (01 September to 30 November) 3 28.7 35.26 10.1

Fig. 8. Regression plots for selected key rainfall/runoff parameters. Solid black lines correspond to the regression equations given in Table 4.
V. Stovin et al. / Journal of Hydrology 414–415 (2012) 148–161 157

Table 4 metric tests of this type. This further reduces confidence in their
Retention as a function of rainfall depth (TP), ADWP, i and Rp. usefulness and applicability.
Equation R2 t-statistics Multiple linear regressions based on the antecedent weather
1/3
TR = 2.410 + 1.514 ln TP 0.7239 3.60**, 7.06** parameters (P24, P7, P14, T24 and T30) provided improved models
VR0.43 = 11.05  2.173 ln TP 0.3290 4.97**, 3.05** for all hydrological performance parameters except TR and Qp.
TS0.55 = 3.130 + 0.882 ln ADWP 0.2481 9.88**, 2.50* However, given that the predictive value of the TR model was sig-
VR0.43 = 4.832 + 1.136 ln ADWP 0.1746 9.48**, 2.01NS nificantly reduced (R2 = 0.323) and that for percentage retention
VR0.43 = 3.916 + ln i 0.0253 4.68**, 0.70NS
VR0.43 = 1.858 + 3.421Rp1/5 0.0881 0.40NS, 1.35NS
(VR) R2 was only improved from 0.499 to 0.552, this approach
was not judged to provide a significant improvement in overall
NS – not significant. predictive modelling capability. Similarly, the non-linear multiple
*
Significant at p = 0.05.
** regression analysis also failed to provide any predictive relation-
Significant at p = 0.01.
ships that were clearly superior to those generated using linear
regression.
are poor (Table 4). The data set includes a spring storm (28/05/08) The main conclusion from the regression analysis is that it is not
of 11 mm with an ADWP of 6.4 h retaining 84.8% of runoff possible to predict the likely retention depth associated with a par-
(9.33 mm), but also a similarly sized summer storm (26/07/07) of ticular rainfall depth or return period, even when ADWP and rain-
12.6 mm with an AWDP of 13.42 h retaining only 20.2% of runoff fall intensity or other antecedent weather parameters are taken
(2.54 mm). Although, there is an observable link between low into consideration. The next section will explore why some of the
ADWP and low retention, the inverse does not hold true; a long expected relationships were not evident in the data set; at the
ADWP does not guarantee high retention, as the roof’s retention same time it will show that a continuous modelling framework,
capacity is finite. based on substrate moisture flux, can help to both understand
Considering that more intense storms may arguably lead to de- and predict a green roof’s runoff performance.
creased retention, as there is less time for the substrate to absorb
moisture, an analysis was also performed to link percentage reten- 6. Green roof moisture flux model
tion to storm intensity. Of the storms in this analysis, the range of
mean storm intensities varies widely, from 0.92 mm/h to The statistical analysis above failed to provide either a robust
7.83 mm/h. They are somewhat negatively skewed as 18 of the predictive tool or much additional insight into the key processes
21 storms are in the bottom quarter of this range. The overall operating within the green roof system. Of particular interest is
and peak intensity of the storms are not strongly correlated the fact that ADWP was not found to be a good predictor of reten-
(R2 = 0.3469). The total rainfall depth is not correlated to either tion. However, this may be explained through a more careful con-
the overall or peak storm intensity (R2 = 0.0019 and 0.0069 respec- sideration of the roof’s key hydrological processes, specifically the
tively). Regression analysis of percentage retention against both flux of moisture into and out of the substrate.
peak and overall storm intensities showed that these were not sig- The substrate’s maximum water-holding capacity defines the
nificantly related (see Table 4). condition when the substrate can hold no more moisture under
The results of the comprehensive forwards and backwards step- gravity (i.e. field capacity). The moisture content at any given time
wise multiple linear regression analysis are summarised in Table 5, will lie somewhere between field capacity and a minimum practi-
where the t-statistics are presented in the same order as the terms cal moisture content, which may vary in response to ambient con-
in the relevant equation. ditions and to substrate physical characteristics. Standard
The multiple regression analysis for TR improved (R2 = 0.83, laboratory material tests exist for the determination of field capac-
rather than 0.72), with RD, ADWP and i determined as being signif- ity (FLL, 2002). The minimum moisture content may be considered
icant predictors and, in contrast to Table 4, TP eliminated as being to define the depth of permanently retained moisture. In soil sci-
non-significant. Four out of the eight relationships presented in Ta- ence, tension plate tests are utilised to quantify the soil-bound
ble 5 identify ADWP as being a significant determinant of hydrolog- moisture fraction. However, the large mineral fragments character-
ical performance. Yet Fig. 8c and d clearly indicate that the links istic of green roof substrates require the standard test procedures
between ADWP and green roof hydrological performance are com- to be refined before they can be applied with confidence. Nonethe-
plex, and unlikely to be revealed by regression-based analysis. less, the moisture retention capacity of the green roof system is
Apart from the relationship for TR, the regressions do not have clearly finite; in this case the empirical data suggests that it may
strong predictive value (R2 < 0.60). It should also be recalled that reach a maximum of 20 mm. For this reason it is unreasonable to
two of the key predictive variables identified above (ADWP and i) expect retention to vary in response to ADWP beyond this practical
did not comply with the normality assumptions required for para- upper limit.

Table 5
Multiple regression analysis to predict various measures of green roof performance from storm event characteristics.

Regression equation Model R2 t-statistics


0.3 1/5
TR = 4.811 + 7.082RD  0.336 ln ADWP + 0.893 ln i Backward 0.823 5.673**, 8.196**, 2.958**, 4.044**
TS0.58 = 3.329 + 0.990 ln ADWP Forward & Backward 0.247 9.336**, 2.498*
ln Rs = 4.542 + 1.304 ln ADWP Forward & Backward 0.489 16.510**, 4.264**
VR0.45 = 12.656  2.438 ln TP + 1.229 ln ADWP Forward 0.499 5.662**, 3.408**, 2.395*
Q 0:6 1=5 Forward 0.478 2.358*, 3.942*
p ¼ 4:829 þ 9:217RD
t0:38
c – no equation with either method – – –
ln tp = 8.998 – 6.688RD1/5 Forward & Backward 0.354 4.579**, 2.959**
PA50:3 ¼ 2:160  0:432 ln TP þ 1:366R1=5 Backward 0.596 2.162*, 3.289**, 2.949**
p

NS – not significant.
*
Significant at p = 0.05.
**
Significant at p = 0.01.
158 V. Stovin et al. / Journal of Hydrology 414–415 (2012) 148–161

In a rainfall event, the substrate will absorb moisture until the for the data to reveal a short ADWP, but for the substrate still to
point when field capacity is reached. If further moisture is added have a high moisture retention capacity. This will arise if the pre-
to the system, runoff will occur. It should be noted that, as green ceding event was small, and did not cause the roof to reach its field
roof substrates typically have very high permeabilities, surface capacity or to initiate runoff.
runoff is not expected to occur. The excess runoff will pass verti- This conceptual model of green roof hydrological processes will
cally down through the substrate and leave the roof via the under- now be employed to explain the runoff behaviour observed in
lying drainage layer, where it will be temporarily detained before three storm events selected from the data series, and – subse-
becoming runoff. quently – to further explain the limitations of the regression-based
Between rainfall events the roof’s storage capacity will be re- modelling approach.
stored via evapotranspiration. Evapotranspiration rates vary sea- Stovin (2010) and Kasmin et al. (2010) have already provided
sonally and daily depending upon meteorological conditions, some discussion on events from this data set; hence the analysis
plant configuration and health. Therefore, a simple relationship be- will focus now on three new, contrasting events. These are
tween ADWP and the restoration of retention capacity should not 26/07/07, 26/06/08 and 04/10/08. The roof responses for these
be expected. For example, Kasmin et al. (2010) suggested that ET events are presented in Fig. 9, both as 5-min discretised temporal
rates on the Sheffield test rig were likely to range from a winter profiles and as cumulative profiles.
average of <0.5 mm/day up to around 3 mm/day in summer. The 26/07/07 event provides an example of a response to high
Furthermore, the ADWP can provide misleading information intensity summer rainfall. The peak intensity – measured over
about the actual level of moisture in the substrate. It is feasible 5 min – was 2.8 mm/5 min, i.e. 33.6 mm/h. Of the 12.6 mm rainfall,

Fig. 9. Rainfall and runoff profiles for three selected events.


V. Stovin et al. / Journal of Hydrology 414–415 (2012) 148–161 159

only 20.2% was retained by the roof. The initial small rainfall was capacity, i.e. approximately 20 mm, and was therefore able to re-
completely retained, as was the initial part of the period of high tain the entire rainfall. In this case the ADWP parameter fails to
intensity rainfall that commenced around 12:00. After that time, provide a useful indication of the roof’s real moisture retention
it is clear that the roof had reached field capacity, and runoff essen- capacity.
tially mirrored the rainfall input thereafter. There is evidence of The 04/10/08 event has been selected to illustrate a typical au-
peak attenuation (60%), but minimal delay in the timing of the tumn response to longer, low intensity frontal-type rainfall. In total
peaks. This almost instantaneous response is not unexpected, as 27.8 mm fell in 15 h. In contrast to the 26/07/07 event, the maxi-
the test bed has a small surface area (3 m2) and a shallow depth mum intensity observed was only 7.2 mm/h. However, the roof’s
(100 mm). Some detention is evident, with the runoff decreasing hydrological response appears unaffected by rainfall intensity. As
exponentially after the rainfall ceased. The limited retention ob- with the 26/07/07 event, some initial losses due to retention are
served in this event reflects the relatively short ADWP of only observed, after which no further retention and only limited deten-
13:25 h. The roof generated runoff from the previous event, sug- tion effects are observed. On this occasion care must again be taken
gesting that it was at, or close to, field capacity at the start of the in any assessment of the link between ADWP and retention capac-
ADWP (although this will not always be the case), and even with ity. Technically the ADWP is 30:23 h. However the rainfall event on
high summer rates of evapotranspiration (3 mm/day, see Kasmin 03/10/08 was only 0.2 mm and did not generate runoff from the
et al., 2010), it is likely that only 2 mm of the roof’s moisture deficit roof. The last rainfall event to generate runoff occurred the day be-
(i.e. retention capacity) would have been restored. The observed fore, with runoff ceasing approximately 48 h before the start of the
retention of 2.54 mm provides some confidence in the validity of current event. ET rates in autumn tend to be lower, so estimating
the process-based conceptual model. 1.5 mm/day would suggest around 3 mm retention. The observed
The 26/06/08 event demonstrated the roof at close to its most retention of 3.45 mm is again consistent with this process-based
effective as a stormwater retention device. 98.4% of the 16.6 mm model.
rainfall was retained. The roof was able to retain this much as a re- Voyde et al. (2010) present a detailed analysis of their roof’s
sult of a prolonged antecedent dry period. Although the ADWP was hydrological response to a set of consecutive rainfall events. Their
calculated to be short, 21:57 h, only 11 mm rainfall fell in the pre- interpretation of the processes leading to observed variations in
ceding 20 days. With assumed summer ET rates of 3 mm/day, the hydrological performance is entirely consistent with that reported
roof is likely to have reached close to its maximum retention here, with high retention being observed following a prolonged dry

Fig. 10. Runoff depths and percentage retention for selected design rainfall events for Sheffield.
160 V. Stovin et al. / Journal of Hydrology 414–415 (2012) 148–161

period and limited retention associated with substrate at close to for significant events; 21 events had good runoff data. The moni-
field capacity. tored period was reasonably typical of Sheffield’s temperate cli-
Jarret and Berghage (2008) and Kasmin et al. (2010) have both mate, though rainfall totals were exceptionally high in summer
demonstrated that a substrate moisture flux model, when com- 2007. The largest event had an estimated return period of approx-
bined with storage routing to represent detention, has the capabil- imately 16 years.
ity to accurately simulate the observed temporal runoff profiles Cumulative retention by the roof was 50.2%. However, mean (or
associated with specific green roof test rig installations. Input data median) retention per event is sensitive to the events that are in-
requirements are the rainfall time series, suitable evapotranspira- cluded in the analysis. Retention performance was greater when
tion rates, an estimate of the roof’s maximum retention depth all storm events were considered, with mean and median retention
and relevant storage routing parameters. As an alternative to the of 70% and 91% per event (all storms) and 61% and 62% (all storms
use of semi-empirical routing methods to model the detention ef- with rainfall >2 mm). This implies that green roofs can make a sig-
fects of green roof systems, She and Pang (2010) have demon- nificant contribution to the mitigation of storm runoff associated
strated that a more mechanistic, physically-based, hydrological with high frequency rainfall events. Retention reduced to 43%
model effectively replicates green roof runoff profiles monitored (mean) and 30.2% (median) for storms with a return period of
at full scale. Unlike regression-based analyses, these process-based greater than one year. The mean peak attenuation for these signif-
approaches have the potential to provide detailed predictions of icant storms was 60%. These figures provide a clear indication that
temporal runoff responses for any roof configuration subjected to green roofs could play a significant part in providing source con-
arbitrary climatic or design storm inputs. However, further re- trols as part of SUDS treatment trains. However, retention varied
search is required to link the controlling parameter values to spe- widely between storm events, ranging between 0 and 100%, and
cific green roof configuration variables, including plant type, the roof provided only 13.2% retention for the largest event
substrate type and depth, amongst others. (16 year return period).
In the meantime, the substrate moisture flux concept may be Regression analyses confirmed that it is not possible to provide
used to provide simple performance estimates corresponding to a useful predictive relationship for the depth or percentage reten-
specific design rainfall depths. If the roof is dry, then retention will tion, even when a range of storm and antecedent weather indica-
approach its maximum value (in this case 20 mm). Twenty milli- tors were considered. Although the response of the green roof
metres is 91% of a 10 year return period 1 h duration rainfall event can be explained by considering moisture fluxes into and out of
for Sheffield. However, if a conservative assumption is made and the substrate, the ADWP parameter fails to characterise the com-
calculations are based on a 48 h period since the roof was satu- plex processes that account for the roof’s antecedent moisture con-
rated, then ET will only have removed 1–6 mm, depending on sea- tent. Compared with regression-based analysis, a continuous
son. 6 mm only represents 27% of the 10 yr rainfall depth. In winter simulation approach based on substrate moisture flux appears
the ET and retention values will obviously be reduced. Fig. 10 pro- likely to provide a more robust framework for predicting the runoff
vides contour plots of the expected runoff from the roof for se- response.
lected design events for Sheffield (NERC, 1999). Data for both 1 h Estimates of the roof’s response to specific design events may be
duration and 24 h duration events is presented. Fig. 10a and c show derived from careful consideration of the likely maximum reten-
runoff depths, whilst Fig. 10b and d show the retention as a per- tion depth and the choice of a suitable ADWP. For example, when
centage of the rainfall. The plots show that high levels of runoff dry, the roof is capable of retaining approximately 20 mm, which
and low percentage retention are to be expected from long return corresponds to 91% of a 10 year return period 1 h duration rainfall
period events. For an extreme event, such as a 100 year return per- event for Sheffield, or 45% of a 100 year 1 h event. For more realis-
iod event, the percentage retention is always small, even when the tic estimates based on 48 h ADWP, the retention values should be
roof’s retention capacity is at its maximum. The 100 year return revised downwards to just 27% and 13% respectively.
period 1 h duration rainfall depth for Sheffield is 44.8 mm; at an
assumed maximum retention capacity of 20 mm, the roof is capa- References
ble of retaining 44.6% of the rainfall. For the 24 h duration event,
120.9 mm, the maximum possible retention falls to 16.5%. Perfor- Banting, D., Doshi, H., Li, J., Missios, P., 2005. Report on the Benefits and Costs of
Green Roof Technology for the City of Toronto, Ryerson University. <http://
mance is proportionately better for less extreme design events.
www.toronto.ca/greenroofs/findings.htm>.
The roof is capable of retaining 100% of the 1 yr return period 1 h Berghage, R., Jarrett, A., et al., 2007. Quantifying Evaporation and Transpirational
event, even at below 50% of maximum retention capacity, whilst Water Losses from Green Roofs and Green Roof Media Capacity for Neutralising
72.2% of a 1 yr return period 24 h event can be retained when Acid Rain. National Decentralised Water Resources Capacity Development
Project. Pennsylvania State University.
the roof’s retention capacity is at its maximum. Carter, T.L., Rasmussen, T.C., 2006. Hydrologic behaviour of vegetated roofs. J. Am.
Clearly it is feasible to replace the Sheffield-specific return per- Water Resour. Assoc. 42 (5), 1261–1274.
iod data with locally-relevant intensity–duration–frequency values CIRIA, 2007a. C644, Building Greener.
CIRIA, 2007b. C697, The SUDS Manual.
to provide a generic approach for estimating the retention perfor- DeNardo, J.C., Jarrett, A.R., Manbeck, H.B., Beattie, D.J., Berghage, R.D., 2005.
mance expected for a green roof installation in response to specific Stormwater mitigation and surface temperature reduction by green roofs.
design rainfall characteristics. It is evident that the prediction of Trans. ASAE 48 (4), 1491–1496.
Dunnett, N., Kingsbury, N., 2004. Planting Green Roofs and Living Walls. Timber
design storm performance is critically dependent on the assump- Press, Cambridge, UK.
tions made regarding an appropriate ADWP; it remains to be seen English Nature, 2003. Green Roofs: Their Existing Status and Potential for
whether the 48 h suggested here becomes adopted as a standard Conserving Biodiversity in Urban Areas. English Nature Report No. 498,
Peterborough.
basis for design amongst SUDS and green roof practitioners. FLL (Forschungsgesellschaft Landschaftsentwicklung Landschaftsbau), 2002.
Guidelines for the Planning, Execution and Upkeep of Green-Roof Sites.
Gedge, D., 2003. From Rubble to Redstarts, Greening Rooftops for Sustainable
Communities, Chicago, 29–30 May, pp. 233–241.
7. Conclusions
Getter, K.L., Rowe, D.B., Andersen, J.A., 2007. Quantifying the effect of slope on
extensive green roof stormwater retention. J. Ecol. Eng. 31, 225–231.
Data on rainfall and runoff has been continuously monitored on Getter, K., Rowe, B., 2008. Carbon Sequestration Potential of Extensive Green Roofs.
a green roof test rig in Sheffield, UK, which has a typical extensive World Green Roof Congress – London, 17–18th September 2008.
Herman, R., 2003. Green roofs in Germany: Yesterday, Today and Tomorrow,
green roof build-up. From a 29 month record of green roof rainfall/ Greening Rooftops for Sustainable Communities, Chicago, 29–30 May, pp.
runoff performance, 22 events were judged to meet defined criteria 41–45.
V. Stovin et al. / Journal of Hydrology 414–415 (2012) 148–161 161

Hutchinson, D., Abrams, P., Retzlaff, R., Liptan, T., 2003. Stormwater Monitoring Two Moran, A., Hunt, B., Smith, J., 2005, Hydrologic and Water Quality Performance from
Ecoroofs in Portland, Oregon, USA. Greening Rooftops for Sustainable Greenroofs in Goldsboro and Raleigh, North Carolina, Greening Rooftops for
Communities Conference, Chicago, 29–30 May, pp. 372–389. Sustainable Communities Conference, Washington DC, 4–6 May.
Jarret, A.R., Berghage, R.D., 2008. Annual and Individual Green Roofs Stormwater NERC (Natural Environment Research Council), 1975. The Flood Studies Report.
Response Models, 6th Greening Rooftops for Sustainable Communities NERC (Natural Environment Research Council), 1999. Flood Estimation Handbook
Conference, Baltimore, MD. (FEH) CD.
Johnston, C., McCreary, K., Nelms, C., 2004. Vancouver Public Library Green Roof Palla, A., Lanza, L.G., La Barbera, P., 2008. A green roof experimental site in the
Monitoring Project, 2nd Greening Rooftops for Sustainable Communities Conf., Mediterranean climate. In: 11th International Conference on Urban Drainage,
Portland, 2–4 June, pp. 391–403. Edinburgh, Scotland, UK, 31/8-5/9.
Kasmin, H., Stovin, V.R., Hathway, E.A., 2010. Towards a generic rainfall-runoff Palla, A., Gnecco, I., Lanza, L.G., 2010. Hydrologic restoration in the urban
model for green roofs. Water Sci. Technol. 62 (4), 898–905. environment using green roofs. Water 2, 140–154.
Köhler, M., Schmidt, M., Grimme, F.W., Laar, M., Paiva, V.L.A., Taveres, S., 2002. Sailor, D.J., 2008. A green roof model for building energy simulation programs.
Green roofs in temperate climates and in the hot-humid tropics – far beyond Energy Build. 40, 1466–1478.
the aesthetics. Environ. Manage. Health 13 (4), 382–391. She, N., Pang, J., 2010. Physically based green roof model. J. Hydrol. Eng. (ASCE) 15
Köhler, M., 2004. Green Roof Technology – From a Fire-Protection System to a (6), 458–464.
Central Instrument in Sustainable Urban Design, 2nd Green Roof Conference, Stovin, V., 2010. The potential of green roofs to manage urban stormwater. Water
Portland, Oregon. Environ. J. 24 (3), 192–199.
Kolb, W., 2004. Good reasons for roof planting – Green roofs and rainwater. Int. Conf Uhl, M., Schiedt, L., 2008. Green roof storm water retention –monitoring results. In:
on Urban Horticulture, Acta Horticulturae, No.643, 295-300. 11th International Conference on Urban Drainage, Edinburgh, Scotland, UK, 31/
Liptan, T., 2003. Planning, Zoning and Financial Incentives for Ecoroofs in Portland, 8-5/9.
Oregon. Greening Rooftops for Sustainable Communities Conference, Chicago, Van Seters, T., Rocha, L., MacMillan, G., 2007. Evaluation of the runoff quantity and
29–30 May, pp. 113–120. quality performance of an extensive green roof in Toronto, Ontario. In: 5th
Liu, K., Minor, J., 2005. Performance Evaluation of an Extensive Green Roof. Greening Greening Rooftops for Sustainable Communities Conf., Minneapolis.
Rooftops for Sustainable Communities Conference, Washington DC, 4–6 May. VanWoert, N.D., Rowe, D.B., Andresen, J.A., Rugh, C.L., Fernandez, R.T., Xiao, L., 2005.
Mentens, J., Raes, D., Hermy, M., 2003. Effect of Orientation on the Water Balance of Green roof stormwater retention: effects of roof surface, slope, and media
Greenroofs. Greening Rooftops for Sustainable Communities, Chicago, 29–30 depth. J. Environ. Qual. 34, 1036–1044.
May, pp. 363–371. Voyde, E., Fassman, E., Simcock, R., 2010. Hydrology of an extensive living roof
Mentens, J., Raes, D., Hermy, M., 2006. Green roofs as a tool for solving the rainwater under sub-tropical climate conditions in Auckland, New Zealand. J. Hydrol. 394,
runoff problem in the urbanized 21st century? Landscape Urban Plan. 77, 217– 384–395.
226. WaPUG (Wastewater Planning Users Group), 2002. Code of Practice for the
Met Office, 2010. <http://www.metoffice.gov.uk/climate/uk/averages/19712000/>. Hydraulic Modelling of Sewer Systems. <http://www.ciwem.org/knowledge-
Moran, A., Hunt, B., Jennings, G., 2004. A North Carolina Field Study to Evaluate networks/groups/wapug/publications/code-of-practice.aspx>.
Greenroof Quantity, Runoff Quality, and Plant Growth, 2nd Greening Rooftops
for Sustainable Communities Conf., Portland, 2–4 June, pp. 446–460.

You might also like