You are on page 1of 11

Urban Water 1 (1999) 323±333

www.elsevier.com/locate/urbwat

Modelling the performance of rainwater collection systems: towards a


generalised approach
A. Fewkes *
Department of Building and Environmental Health, The Nottingham Trent University, Burton Street, Nottingham NG1 4BU, UK
Received 24 November 1999; received in revised form 19 April 2000; accepted 11 May 2000

Abstract
The collection of rainwater from roofs, its storage and subsequent use make a signi®cant saving in the use of potable water. This
paper investigates how spatial and temporal ¯uctuations in rainfall can be incorporated into behavioural models, which simulate the
performance of rainwater collectors. Temporal variations are considered at two di€erent time scales, that is daily and monthly
intervals. The ®rst model uses a daily time interval to model system performance. A set of rainwater collector performance curves for
®ve geographic locations is developed. From the location-speci®c curves, a set of average curves is determined which is shown to be
suciently accurate for estimating rainwater collector performance. The second model uses a larger time interval of one month. The
e€ect of daily ¯uctuations in rainfall is incorporated into this model using a storage operating parameter. Values of the parameter
were determined by matching the output from the daily model with that of the monthly model. Generally the rainwater collector
performance predicted by the monthly model using average values of the storage operating parameter is shown to correlate well with
the corresponding values determined using a daily time interval model. The monthly model provides a simple and versatile method
of modelling the performance of rainwater collectors. Ó 2000 Elsevier Science Ltd. All rights reserved.

Keywords: Water conservation; Rainwater; Storage; Modelling; Simulation

1. Introduction and supply. The analysis is based upon queuing theory.


Behavioural models simulate the operation of the res-
The collection of rainwater from roofs, its storage ervoir with respect to time by routing simulated mass
and subsequent use is a simple method of reducing the ¯ows through an algorithm which describes the opera-
demand on both the public water supplies and waste tion of the reservoir. The operation of the rainwater
treatment facilities. Without extensive treatment the collector will usually be simulated over a period of years.
rainwater is suitable for a range of uses such as WC The input data which is in time series form are used to
¯ushing, garden irrigation and clothes washing. The simulate the mass ¯ow through the model and will be
capacity of the rainwater store is important both eco- based upon a time interval of either a minute, hour, day
nomically and operationally. or month.
McMahon and Mein (1978) identify three general This investigation concentrates upon the use of
types of reservoir sizing models, namely; critical period, behavioural models to simulate the performance of
Moran and behavioural models. Critical period methods rainwater collectors. The main problem associated with
identify and use sequences of ¯ows where demand ex- the modelling process relates to how the spatial and
ceeds supply to determine the storage capacity. The se- temporal ¯uctuations in rainfall can be incorporated
quences of ¯ows or time series used in this method are into the model. Spatial variations in rainfall are incor-
usually derived from historical data. Moran related porated into the model using rainfall time series from
methods are a development of MoranÕs (1959) theory of di€erent UK locations. Temporal ¯uctuations in rainfall
storage. A system of simultaneous equations are used are investigated using two behavioural models each with
with this method to relate reservoir capacity, demand di€erent time intervals.
The ®rst model uses a daily time interval, which ig-
nores ¯uctuations with a time scale less than a day, to
*
Fax: +44-115-9486438. predict system performance for di€erent combinations
E-mail address: alan.fewkes@ntu.ac.uk (A. Fewkes). of roof area, demand, store volume and rainfall level. A

1462-0758/00/$ - see front matter Ó 2000 Elsevier Science Ltd. All rights reserved.
PII: S 1 4 6 2 - 0 7 5 8 ( 0 0 ) 0 0 0 2 6 - 1
324 A. Fewkes / Urban Water 1 (1999) 323±333

set of curves is produced which enable the performance 


Dt ;
of rainwater collection systems to be predicted in dif- Yt ˆ min …1†
Vtÿ1 ;
ferent UK locations. The main limitation of this ap-
proach is the requirement for daily rainfall time series, 
Vtÿ1 ‡ Qt ÿ Yt ;
which can be both costly and dicult to manipulate. Vt ˆ min …2†
S ÿ Yt ;
The second method of modelling uses a larger time
interval of one month resulting in a more compact where Rt is the rainfall (m) during time interval, t, Qt the
model and economic data set. However, the coarser rainwater runo€ (m3 ) during time interval, t, Vt the
monthly time interval does not take into account rainfall volume in store (m3 ) during time interval, t, Yt the yield
¯uctuations with a time scale less than one month, from store (m3 ) during time interval, t, Dt the demand
which may result in an inaccurate prediction of system (m3 ) during time interval, t, S the store capacity (m3 ),
performance. The poor resolution of the monthly in- and A is the roof area (m2 ).
terval model compared to the daily model is countered The YAS operating rule assigns the yield as either the
by the introduction of a parameter, referred to as the volume of rainwater in storage from the preceding time
storage operating parameter. The short time scale ¯uc- interval or the demand in the current time interval
tuations of the daily model are in e€ect replicated in the whichever is the smaller. The rainwater runo€ in the
monthly model by the storage operating parameter. current time interval is then added to the volume of
Values of the parameter are selected so that the monthly rainwater in storage from the preceding time interval
model mimics the system performance predicted by the with any excess spilling via the over¯ow and then sub-
corresponding model using a daily time interval. This tracts the yield.
approach provides a simple and versatile method of The YBS operating rule is
modelling the performance of rainwater collectors which 
Dt ;
takes into account temporal ¯uctuations in rainfall. Yt ˆ min …3†
Vtÿ1 ‡ Qt ;

Vtÿ1 ‡ Qt ÿ Yt ;
2. Background and previous research Vt ˆ min …4†
S:
The generic con®guration of a rainwater collection The YBS operating rule assigns the yield as either the
system is illustrated in Fig. 1. Behavioural models have volume of rainwater in storage from the preceding time
been used by other researchers (Jenkins, Pearson, interval plus the runo€ in the current interval or the
Moore, Sun, & Valentine, 1978; Latham, 1983) to in- present demand whichever is the smaller. The rainwater
vestigate the performance of rainwater stores. Jenkins runo€ in the current time interval is then added to the
et al. (1978) developed a behavioural model and iden- volume of the rainwater in storage from the preceding
ti®ed two fundamental algorithms to describe the time interval before subtracting the yield and allowing
operation of the rainwater store. The yield after spillage any excess to spill via the over¯ow.
(YAS) operating rule is Jenkins et al. (1978) used the YAS algorithm and a
monthly time interval. The reliability or performance of
the rainwater store can be expressed using either a time
or volume basis. In either case, a reliability or perfor-
mance of 100% indicates complete security in provision
of service water. Latham (1983) used a behavioural
model but de®ned the reservoir operating algorithm in a
more general form:

Dt ;
Yt ˆ min …5†
Vtÿ1 ‡ hQt ;

…Vtÿ1 ‡ Qt ÿ hYt † ÿ …1 ÿ h†Yt ;
Vt ˆ min …6†
S ÿ …1 ÿ h†Yt ;
where h is a parameter between 0 and 1. If h ˆ 0, then
Fig. 1. Rainwater collection system con®guration. Rt ± rainfall (m) the algorithm is YAS and if h ˆ 1, the algorithm is YBS.
during time interval, t, Qt ± rainwater runo€ (m3 ) during time interval, Fewkes and Butler (2000) report the results of a
t, Mt ± mains supply make-up (m3 ) during time interval, t, Vt ± volume
preliminary mapping exercise which evaluated the ac-
in store (m3 ) during time interval, t, Yt ± yield from store (m3 ) during
time interval, t, Dt ± demand (m3 ) during time interval, t, S ± store curacy of behavioural models for the sizing of rainwater
capacity (m3 ), A ± roof area (m2 ). (Note: Duration of time interval, t collection systems using both di€erent time intervals and
can be hourly, daily or monthly.) reservoir operating algorithms applied to a compre-
A. Fewkes / Urban Water 1 (1999) 323±333 325

hensive range of operational conditions. The prelimi- sites ranged from coastal to inland areas and were lo-
nary analysis of their study indicated the hourly YAS cated within the Midlands and Wales. The sites and
model could be used as a standard of comparison their average rainfall levels are summarised in Table 1.
against which other models could be compared and Temporal ¯uctuations are considered at two di€erent
calibrated. The YAS reservoir operating algorithm was time scales, namely daily and monthly intervals. Initially
found to give a conservative estimate of system perfor- a daily time interval model using the YAS reservoir
mance irrespective of the model time interval and operating algorithm was developed and secondly a more
therefore preferred for design purposes compared to the computationally ecient monthly time interval model.
YBS operating algorithm. The short time scale ¯uctuations, that is daily variations,
Di€erent combinations of roof area, store capacity are incorporated into this model using the general op-
and demand were expressed in terms of two dimen- erating parameter h. Values of the parameter h were
sionless ratios, namely the demand fraction and storage determined by matching the output from the daily
fraction. The demand fraction is given by, D/AR, where model with the monthly model. The matching of the
D is the annual demand (m3 ), A the roof area (m2 ) and R output from the models using the storage operating
is the annual rainfall (m). The storage fraction is given parameter in e€ect enables the short time scale ¯uctua-
by, S/AR, where S is the store capacity (m3 ). tions to be included in the model and provides a po-
The detailed analysis undertaken by Fewkes and tentially versatile method of modelling the performance
Butler (2000) enabled constraints to be proposed for the of rainwater collectors.
application of hourly, daily and monthly models ex- The objectives of this paper are summarised below:
pressed in terms of storage fraction. It was recom- 1. Determine a generic set of curves using a YAS daily
mended that hourly models should be used for sizing time interval model, for a range of storage and de-
small stores with a storage fraction below or equal to mand fractions which can be used to predict the per-
0.01. Daily models can be applied to systems with formance of rainwater collectors within the
storage fractions within the range 0.01±0.125. Monthly geographical area de®ned by this study.
models were only recommended for use with storage 2. Evaluate the accuracy of the generic daily time inter-
fractions in excess of 0.125. Generally, daily models can val rainwater collection system performance curves.
be used to predict the performance of all stores except 3. Incorporate the e€ect of daily ¯uctuations into a
small stores with a storage fraction less than or equal to monthly time interval model using the storage operat-
0.01. ing parameter h.
Values of the parameter h were also determined for a 4. Determine a generic set of values of the parameter h,
monthly model operating over a comprehensive range of for a range of store capacities and storage and de-
demand and storage fractions. The system performance mand fractions which can be used, in conjunction
predicted by the monthly (h) model was found to cor- with a monthly time interval model, to predict the
relate well with corresponding values determined using a performance of rainwater collectors within the geo-
daily YAS model over a range of demand fractions graphical area de®ned by this study.
likely to be encountered in practice. However, the results 5. Evaluate the accuracy of a monthly time interval
of the study by Fewkes and Butler (2000) were based on model using generic values of the parameter h.
one geographic location and it was recognised that
further research was required to con®rm the results of
their study are generally applicable to other areas of the 4. Development of the behavioural model
UK.
The con®guration of the rainwater collection system
is illustrated in Fig. 1. The rainwater collection sizing
3. The proposed behavioural model investigation (rcs) model consists of two parts:
(i) Provision of rainfall supply and usage demand
Research undertaken by Jenkins et al. (1978) and patterns or time series.
Latham (1983) related to a limited range of volume
stores aimed at conserving an excess of 95% of house- Table 1
hold water. Previous work by Fewkes and Butler (1999) Average annual rainfall levels of sites
considered a wide range of store volumes but was based Site Average annual rainfall (mm/yr)
upon rainfall data collected from one geographic loca-
tion within the UK. The proposed research addresses Bredenbury 750
how both spatial and temporal ¯uctuations in rainfall Newport 940
Nottingham 620
can be incorporated into the modelling process. The
Rhymney 1600
spatial ¯uctuations in rainfall are incorporated, by using Yns-y-Fro 1100
rainfall data from ®ve di€erent sites within the UK. The
326 A. Fewkes / Urban Water 1 (1999) 323±333

(ii) Simulation of system operation. tors at each of the ®ve locations identi®ed in Table 1. At
For each of the ®ve sites (Table 1) investigated, each location the water-saving eciencies were modelled
computer coded system simulation models were devel- for a range of store capacities and roof areas. The dif-
oped based upon both monthly and daily time intervals. ferent combinations of roof area and store capacity are
The general form of the reservoir operating algorithm expressed in terms of the demand fraction and storage
was used in the system simulation models as de®ned in fraction (Section 2). The performance of rainwater col-
Section 2 (Eqs. (5) and (6)). Twenty years of either daily lection systems at each location was modelled with de-
or monthly historic rainfall time series (Walsh, 1999) mand fractions of 0.25, 0.5, 0.75, 1.00, 1.25, 1.5, 1.75
were used as input to the respective system simulation and 2.00 each with a storage fraction range of 0.005 to
models. 0.4. The modelled system performances at demand frac-
The demand component of the models was limited to tions of 0.25, 1.00 and 2.00 are illustrated in Figs. 2±4.
WC usage which accounts for approximately 30% of At a demand fraction of 0.25 (Fig. 2) the performance
potable household water usage in the UK (Department curves predicted for each site are close together, almost
of the Environment and Welsh Oce, 1992) and was coalescing into a single curve. The modelled perfor-
assumed to occur at a constant daily or monthly rate. mance of rainwater collectors at this demand fraction
This assumption is reasonable because the demand time appears therefore to be relatively insensitive to ¯uctua-
series generated by WC usage does not exhibit excessive tions in daily rainfall patterns experienced at each lo-
daily or monthly variance (Fewkes, 1999). However, if cation. When the demand fraction is increased to 1.00
the demand from other domestic appliances such as the (Fig. 3) the performance curves are not so tightly
washing machine was considered the demand patterns grouped indicating increased sensitivity to daily ¯uctu-
would not be constant and the demand time series re- ations in rainfall patterns at the di€erent sites. The
quired. performance curves at a demand fraction of 2.00 (Fig. 4)
The performance of the rainwater collection system is begin to coalesce together again indicating an insensi-
described by its water-saving eciency (ET ) (Dixon, tivity to site speci®c rainfall patterns.
Butler, & Fewkes, 1999). Water-saving eciency is a The overall close proximity of the water-saving e-
measure of how much mains water has been conserved ciency curves at each demand fraction ratio for each of
in comparison to the overall demand of the WC and is the ®ve sites suggested system performance could be
given by Eq. (7) adequately represented by a set of average or generic
PT curves. The average water-saving eciency of a rain-
Yt
ET ˆ PTtˆ1  100; …7† water collector at demand fractions of 0.25, 0.50, 0.75,
tˆ1 Dt 1.00, 1.25, 1.50, 1.75 and 2.00 each with a storage
where T is the total time under consideration. fraction range of 0.005±0.4 is illustrated in Fig. 5.

5. The daily time interval model 6. The accuracy of the rainwater collection system
performance curves (daily time interval model)
In this section, a daily time interval model using a
YAS (h ˆ 0) reservoir operating algorithm is used to The water-saving eciencies predicted by the average
model the water-saving eciencies of rainwater collec- curves (Fig. 5) for demand fractions of 0.25, 0.50, 1.00,

Fig. 2. The variation of water-saving eciency with location and storage fraction (D/AR ˆ 0.25).
A. Fewkes / Urban Water 1 (1999) 323±333 327

Fig. 3. The variation of water-saving eciency with location and storage fraction (D/AR ˆ 1.00).

Fig. 4. The variation of water-saving eciency with location and storage fraction (D/AR ˆ 2.00).

Fig. 5. Average variation of water-saving eciency with storage fraction and demand fraction.

1.50 and 2.00 each with a storage fraction range of the intercept of the straight line was set to zero before
0.005±0.4 were plotted in turn against the correspond- determining the gradient (m) of the straight line and the
ing values predicted for each of the ®ve di€erent loca- coecient of determination (R2 ). The values of m and
tions. For each of the 25 plots a straight line was ®tted R2 for each site and demand fraction are given in
to the data points using linear regression. In each case Table 2.
328 A. Fewkes / Urban Water 1 (1999) 323±333

Table 2
The accuracy of average rainwater collection system performance curves (daily time interval model)

D/AR Bredenbury Newport Nottingham Rhymney Yns-y-Fro

m R2 m R2 m R2 m R2 m R2

0.25 0.998 0.997 1.006 0.975 0.996 0.986 0.998 0.99 1.002 0.994
0.5 0.993 0.996 1.012 0.99 0.991 0.996 0.998 0.995 1.006 0.997
1 0.986 0.996 1.014 0.998 0.987 0.997 1.008 0.992 1.005 0.999
1.5 1.006 0.998 1.016 0.998 0.988 0.999 0.99 0.995 1.001 1
2 1.008 0.998 1.004 0.998 1 1 1 0.971 0.999 1

In order to evaluate the accuracy of the water-saving Models for each of the ®ve sites using monthly time
eciencies predicted using the average performance intervals and appropriate values of the parameter h
curves arbitrary limits were imposed upon the values of (Section 2) are used to predict system performance. The
m and R2 . It was assumed the accuracy of the water- water-saving eciencies were modelled using monthly
saving eciencies predicted by the average curves would (h) models for demand fractions of 0.25, 0.50, 0.75, 1.00,
be preserved within acceptable tolerances if the follow- 1.25, 1.50, 1.75 and 2.00 each with a storage fraction
ing limits were observed: range of 0.005±0.4. In each case, the value of the
1:11 P m P 0:89; parameter h was chosen whereever possible to predict
…8† the water-saving eciency within 0.01% of the value
R2 P 0:9: determined by the corresponding daily YAS model
The limits of m result in average water-saving eciencies (Section 5).
within the boundaries of ‹10% compared to the per- The variation of the parameter h, against storage
formances predicted by the location-speci®c curves. fraction for each site and demand fraction is illustrated
Generally, the values of m and R2 are within the ranges in Figs. 6±13.
of 0.986±1.016 and 0.986±1, respectively. The analysis With demand fractions of 0.25 and 0.5 (Figs. 6 and 7)
suggests the average of generic performance curves can the curves for each site are closely grouped together and
be used to predict the water-saving eciencies of rain- demonstrate a characteristic skewed distribution with
water collection systems within the boundaries of the the value of the parameter h, tending to zero as the
geographical area investigated. storage fraction increases. When the demand fraction is
increased to 0.75 (Fig. 8) the curves are again skewed
but are more loosely grouped for values of storage
7. The monthly model using the storage algorithm fraction exceeding 0.2. At demand fractions of 1.00,
parameter h 1.25, 1.50, 1.75 and 2.00 (Figs. 9±13, respectively) the
curve for Nottingham exhibits a skewed distribution
In this section, an alternative method of modelling with the value of the parameter h tending to zero as the
rainwater collection system performance is investigated. storage fraction increases. However, for each of the

Fig. 6. The variation of parameter h with location and storage and storage fraction (D/AR ˆ 0.25).
A. Fewkes / Urban Water 1 (1999) 323±333 329

Fig. 7. The variation of parameter h with location and storage and storage fraction (D/AR ˆ 0.50).

Fig. 8. The variation of parameter h with location and storage and storage fraction (D/AR ˆ 0.75).

Fig. 9. The variation of parameter h with location and storage and storage fraction (D/AR ˆ 1.00).

other four sites modelled the curves exhibit a di€erent closely grouped together within the storage fraction
characteristic shape with the value of the parameter h range 0.005±0.15.
tending to unity as the storage fraction increases. Gen- The curves exhibit two general regions of be-
erally, the curves for all demand fractions and sites are haviour, one where the curves for each geographic
330 A. Fewkes / Urban Water 1 (1999) 323±333

Fig. 10. The variation of parameter h with location and storage and storage fraction (D/AR ˆ 1.25).

Fig. 11. The variation of parameter h with location and storage and storage fraction (D/AR ˆ 1.50).

Fig. 12. The variation of parameter h with location and storage and storage fraction (D/AR ˆ 1.75).

location are closely grouped together and another grouped together for storage fractions up to approxi-
where the grouping is less pronounced. At low demand mately 0.04 (Figs. 6 and 7). At larger demand frac-
fractions of 0.25 and 0.5, the curves are closely tions from 0.75 to 2.00 (Figs. 8±13) the curves remain
A. Fewkes / Urban Water 1 (1999) 323±333 331

Fig. 13. The variation of parameter h with location and storage and storage fraction (D/AR ˆ 2.00).

closely grouped for storage fractions up to approxi- above 0.75 the monthly time interval model is relatively
mately 0.15. insensitive to values of the parameter h. This general
Generally, the values of the parameter h at low ®nding is supported by previous research, Fewkes and
storage volumes range from 0.25 to 0.85 and the curves Butler (1999), relating to the Nottingham area which
are closely grouped, indicating as would be expected, indicated monthly models were suciently accurate for
that the store is sensitive to temporal ¯uctuations in predicting the performance of rainwater collectors with
rainfall. The values of the parameter h at higher storage storage fractions in excess of 0.125.
volumes would be expected to be less sensitive to tem- The average variation of the parameter h against
poral and spatial ¯uctuations in rainfall because of the storage fraction associated with each of the demand
damping e€ect by the store. The loose grouping of the fractions and the overall average curve for the complete
curves at higher storage volumes suggested that above a demand fraction range of 0.25±2.00 are illustrated in
limiting storage fraction the store becomes insensitive to Fig. 14.
rainfall ¯uctuations with a time scale less than one The characteristic shape of the overall average curve
month and a larger band of values of the parameter h is interesting because for storage fractions exceeding
will adequately describe the operation of the store. The 0.025 the variation in the parameter h is small. The
loose grouping of the curves is typi®ed at demand maximum value of h is 0.675 (S/AR ˆ 0.025) and the
fractions of 1.25 and above. The value of the parameter minimum value 0.525 (S/AR ˆ 0.4). A possible approx-
h for Nottingham tends to zero whilst for the other lo- imate method of sizing rainwater collection systems
cations h tends to unity. using the monthly (h) model would be to use an overall
Further investigation supported this argument. For average value of h equal to 0.6.
example, Nottingham (S/AR ˆ 0.4, D/AR ˆ 1.25, h ˆ 0),
water-saving eciency (ET ) ˆ 81.28% and when h ˆ 1,
ET ˆ 81.95%. Whilst for Bredenbury (S/AR ˆ 0.4, D/ 8. The accuracy of a monthly time interval model using
AR ˆ 1.25, h ˆ 0), ET ˆ 79.44% and when h ˆ 1, average values of the parameter h
ET ˆ 79.51%. With a daily time interval model and h ˆ 1
the values of water-saving eciency (ET ) for Notting- The accuracy of the water-saving eciencies pre-
ham and Bredenbury were 80.91% and 80.47%, respec- dicted by the monthly (h) model using average values of
tively. Therefore in this particular case for both h (Fig. 14) was investigated using linear regression
Nottingham and Bredenbury the choice of a YAS or analysis. The water-saving eciencies predicted by the
YBS monthly time interval model would result in a monthly (h) model were plotted against the corre-
maximum di€erence in water-saving eciency (ET ) of sponding values predicted for each site using the ap-
‹1% compared to the YAS daily time interval model. propriate daily YAS model. For each plot, a straight
The monthly time interval models at higher storage line was ®tted to the data points using linear regression
volumes are therefore, relatively insensitive to the pa- and the values of m and R2 determined (Table 3). In
rameter h and consequently, as would be expected to each case the intercept of the straight line was set to
temporal variations in rainfall. Generally, at storage zero.
fractions above 0.05 for demand fractions up to 0.5 and Overall the agreement between the water-saving e-
at storage fractions above 0.15 for demand fractions ciencies predicted by the monthly (h) model and the
332 A. Fewkes / Urban Water 1 (1999) 323±333

Fig. 14. Average variation of parameter h with storage fraction and demand fraction.

Table 3
The accuracy of a monthly time interval model using average values of the parameter h

D/AR Bredenbury Newport Nottingham Rhymney Yns-y-Fro

m R2 m R2 m R2 m R2 m R2

0.25 1.006 0.744 1.008 0.854 1.006 0.829 1 0.986 1.006 0.911
0.5 1.012 0.9 1.02 0.957 1.015 0.965 1.003 0.998 1.016 0.975
1 0.998 0.994 1.002 0.993 1.002 0.998 0.983 0.997 0.996 0.997
1.5 0.974 0.995 0.977 0.996 0.983 0.988 0.961 0.987 0.969 0.993
2 0.959 0.981 0.959 0.983 0.968 0.97 0.94 0.965 0.975 0.972

daily YAS model is very good as indicated by the large tems, one is based on a daily time interval and the other
proportion of correlation coecients exceeding 0.9. a monthly time scale. The central theme of the inves-
There are three exceptions, which occur at a demand tigation relates to the spatial and temporal ¯uctuations
fraction of 0.25 for the Bredenbury, Newport and in rainfall and their incorporation into the behavioural
Nottingham sites. models. Both models were used to predict system per-
The critical range for values of the parameter h occurs formance, in terms of their water-saving eciency, for
within a storage fraction range of 0.005±0.15. The av- di€erent combinations of roof area, demand, storage
erage variation of the parameter h against storage volume and rainfall level. Initially, the general reservoir
fraction curves (Fig. 14) are closely grouped together up operating algorithm used in behavioural models to
to a storage fraction value of 0.15 with exceptions oc- describe the operation of rainwater stores was re-
curring at demand fractions of 0.25 and 0.5. The di- viewed. The model parameter h was de®ned and two
vergence of the average h curves at these demand extreme reservoir operating conditions identi®ed,
fractions probably accounts for the poor correlation namely yield after storage YAS when h ˆ 0 and YBS
obtained at these values (Table 3). when h ˆ 1.
The monthly (h) model using average values of h The ®rst model used a daily time interval in con-
provides a ¯exible and accurate method of predicting junction with a YAS (h ˆ 0) reservoir operating algo-
water-saving eciency for the range of demand frac- rithm to model the water-saving eciencies of rainwater
tions likely to be encountered in practice. When average collectors at ®ve di€erent UK locations. The perfor-
values of h are applied to low demand fractions inac- mance curves for each location were closely grouped
curacies in predicting water-saving eciency occur but a together suggesting system performance was relatively
good approximation of system performance is modelled. insensitive to daily ¯uctuations in rainfall at each site.
The close grouping of the water-saving eciency curves
at each site also suggested system performance could be
9. Conclusions adequately represented by a set of average curves, one
for each demand fraction investigated.
This study presents two behavioural models, which The correlation between the water-saving eciencies
simulate the performance of rainwater collection sys- predicted by the average curves was good compared to
A. Fewkes / Urban Water 1 (1999) 323±333 333

the corresponding values de®ned by the curves for each di€erent sites located within the Midlands and Wales,
of the individual sites. A set of generic performance including both coastal and inland areas. Further re-
curves has therefore been produced which predict the search is required to con®rm the results are generally
performance of rainwater collection systems within the applicable to other areas of the UK, which may expe-
geographic boundaries de®ned by the sites investigated rience similar levels of rainfall but di€erent daily pat-
within this study. terns. When this further work has been undertaken two
The second method of predicting rainwater collection ¯exible methods of assessing the performance of rain-
system performance used a more computationally e- water collection systems within the UK using either
cient monthly time interval model. In the monthly daily or monthly data will have been developed and
model the e€ect of daily ¯uctuations in rainfall was validated. The monthly model in particular represents a
approximated by the parameter h. The parameter h ef- computationally ecient method of determining the
fectively absorbs the daily variations in rainfall and performance of rainwater collectors. The parameter h
enables the monthly model to replicate the performance has been shown to be an e€ective means of enabling a
of the daily time interval model. Appropriate values of monthly model to replicate the behaviour of a daily time
the parameter h were determined by matching the per- interval model.
formance of the daily model with the monthly model.
The water-saving eciencies predicted by the monthly
(h) model using average values of the parameter h cor-
References
related well with the corresponding values determined
using a daily YAS model. The only exception occurred Department of the Environment and Welsh Oce (1992). Using water
at the lowest demand fraction investigated of 0.25. wisely. London: HMSO.
Generally, a wide variation in values of the parameter Dixon, A., Butler, D., & Fewkes, A. (1999). Computer simulation of
h was observed from each of the sites when the storage domestic water reuse systems: investigating greywater and rainwa-
ter in combination. Water Science Technology, 38(4), 25±32.
fraction exceeded 0.15. However, the water-saving e-
Fewkes, A. (1999). The use of rainwater for WC ¯ushing: the ®eld
ciencies predicted by the monthly model are not sensi- testing of a collection system. Building and Environment, 34(6), 765±
tive to variations of the parameter h, when the storage 772.
fraction exceeds 0.15. This is because the variations in Fewkes, A., & Butler, D. (2000). Simulating the performance of
rainfall pattern are bu€ered by the increasing storage rainwater collection systems using behavioural models. Building
volumes which are associated with storage fractions in Services Engineering Research and Technology, 21(2), 99±106.
Jenkins, D., Pearson, F., Moore, E., Sun, J. K., & Valentine, R. (1978).
excess of 0.15. Consequently, the use of an average value Feasibility of rainwater collection systems in California. Contribu-
of h does not impair the accurate prediction of system tion No. 173. Californian Water Resources Centre, University of
performance. An approximate method of sizing rain- California.
water collection systems using an overall average value Latham, B. G. (1983). The design of single-purpose reservoirs. MASc
thesis, University of Canada.
of h equal to 0.6 is also proposed although the accuracy
McMahon, T. A., & Mein, R. G. (1978). Reservoir capacity and yield.
of this approach has not been investigated in this study. Developments in water science. Amsterdam: Elsevier.
The analysis described in this study has been based Moran, P. A. P. (1959). The theory of storage. London: Methuen.
upon a comprehensive set of rainfall data from ®ve Walsh, H. (1999). Private communication. Environment Agency.

You might also like