Journal of Environmental Management 180 (2016) 82e93

Contents lists available at ScienceDirect

Journal of Environmental Management

journal homepage: www.elsevier.com/locate/jenvman

Research article

Uncertainty analysis of daily potable water demand on the

performance evaluation of rainwater harvesting systems in residential
buildings
Arthur Santos Silva*, Enedir Ghisi

polis, SC 88040-900, Brazil
Federal University of Santa Catarina, Department of Civil Engineering, Laboratory of Energy Efficiency in Buildings, Floriano

a r t i c l e i n f o a b s t r a c t

Article history: The objective of this paper is to perform a sensitivity analysis of design variables and an uncertainty
Received 30 November 2015 analysis of daily potable water demand to evaluate the performance of rainwater harvesting systems in
Received in revised form residential buildings. Eight cities in Brazil with different rainfall patterns were analysed. A numeric
6 May 2016
experiment was performed by means of computer simulation of rainwater harvesting. A sensitivity
Accepted 11 May 2016
Available online 19 May 2016
analysis was performed using variance-based indices for identifying the most important design pa-
rameters for rainwater harvesting systems when assessing the potential for potable water savings and
underground tank capacity sizing. The uncertainty analysis was performed for different scenarios of
Keywords:
Rainwater harvesting
potable water demand with stochastic variations in a normal distribution with different coefficients of
Performance evaluation variation throughout the simulated period. The results have shown that different design variables, such
Water demand as potable water demand, number of occupants, rainwater demand, and roof area are important for
Sensitivity analysis obtaining the ideal underground tank capacity and estimating the potential for potable water savings.
Uncertainty analysis The stochastic variations on the potable water demand caused amplitudes of up to 4.8% on the potential
for potable water savings and 9.4% on the ideal underground tank capacity. Average amplitudes were
quite low for all cities. However, some combinations of parameters resulted in large amplitude of un-
certainty and difference from uniform distribution for tank capacities and potential for potable water
savings. Stochastic potable water demand generated low uncertainties in the performance evaluation of
rainwater harvesting systems; therefore, uniform distribution could be used in computer simulation.
© 2016 Elsevier Ltd. All rights reserved.

1. Introduction countries is to assure the institutions capacity to manage water

As water is a key resource for human living and development, its
management should be improved properly through the years. A
report by the United Nations World Water Assessment Programme
has shown that the sustainable use of water drives consequences
far beyond social, economic and environmental dimensions, as it
influences human health, availability of food and energy, industrial
development and urbanization growth (WWAP, 2015).
Economy and population growth worldwide, urbanization,
migration and industrialization as well as the growth of production
and consumption lead to more freshwater withdrawal and envi-
ronmental pollution. In general, the major priority for many
* Corresponding author.
E-mail addresses: arthurssilva07@gmail.com (A.S. Silva), enedir.ghisi@ufsc.br
(E. Ghisi).
resources by integrating it to socio-economic development and
reduction of poverty.
The Water National Agency (ANA, 2013) states that Brazil has a
high freshwater availability compared to other countries. The
agency analysed data over the period 1961e2007 and verified that
the average annual rainfall in Brazil is 1761 mm, and it ranges from
500 mm in the northeast region to over 3000 mm in the Amazon
region.
The surface water availability in Brazil is 91,000 m3 in the whole
country, considering 95% of permanence water flow. This avail-
ability was calculated by the Water National Agency (ANA, 2013) by
taking the twelve most representative watersheds to calculate the
indicator of incremental water flow for drought period (that means
the 95% of occurrence of permanence flow). The natural long-term
water flow of the watersheds was not used as an indicator due to
the interference of drought periods and seasonal behaviour.
Although there is high water availability, it is expected that the

http://dx.doi.org/10.1016/j.jenvman.2016.05.028
0301-4797/© 2016 Elsevier Ltd. All rights reserved.
 A.S. Silva, E. Ghisi / Journal of Environmental Management 180 (2016) 82e93
water demand in Brazil will increase from 570 m3/s in 2015 to
630 m3/s in 2025 (ANA, 2010).
According to the Water National Agency (ANA, 2010) the major
problems of water supply in Brazil are due to the limited productive
capacity of the existing surface or underground systems. For the
whole country, 45% of the cities have satisfactory water availability
to guarantee urban water supply, while 46% require expansion and
improvement of existing systems, and 9% require investments in
new water sources (both underground and surface).
The use of rainwater as a water source is an alternative to
mitigate freshwater availability. Rainwater harvesting can reduce
freshwater direct withdrawal and reduce efforts for providing
infrastructure and maintenance of existing and new water supply
systems. It can be used for non-potable end-uses without the need
of great amount of chemical or biological treatment (Fewkes,
2000). As a matter of fact, rainwater harvesting is important to
reduce the dependence on centralized water supply, as well as
reducing maintenance, operation and infrastructure costs of public
water supply (Gurung and Sharma, 2014; Silva et al., 2015).
There are many numerical and simulation methods for deter-
mining rainwater system performance, such as sizing tank capacity,
calculating spillage and efficiency, determining the potential for
potable water savings and other performance indicators.
A Brazilian standard (NBR 15527) addresses some methods for
sizing rainwater tank capacity. From the literature, simulation
methods are good tools for evaluating rainwater system perfor-
mance as they consider stochastic variations in rainwater avail-
ability and consumption through predefined time steps (ABNT,
2007). The problem of simplified or “practical” sizing methods is
that they consider average values for some important variables, but
not the heterogeneity of each location and operation of each
building.
Fewkes (2000) studied spatial and temporal variations in rain-
fall and how they can be incorporated in behavioural models that
simulate rainwater harvesting systems. Two well-known models
were used, i.e., “yield after spillage” (YAS) and “yield before
spillage” (YBS) (Jenkins et al. (1978) apud Fewkes (2000)). The
authors concluded that the simplified monthly model could be
used as an alternative for modelling the performance of rainwater
collectors. Ghisi et al. (2012) have also analysed temporal variations
of short- and long-term rainfall data for evaluating rainwater sys-
tem performance. The difference from longer periods to short pe-
riods was never greater than 5% on the potential for potable water
savings indicator, which confirms that short-term time series can
be used as they provide reliable results. Campisano and Modica
(2014) showed that the timestep (daily and sub-daily data) in-
fluences the performance of rainwater harvesting systems, espe-
cially in small tank sizes and high water demand.
Rupp et al. (2011) performed a comparative analysis of different
methods presented in the Brazilian standard NBR 15527 (ABNT,
2007) for sizing rainwater storage tanks. A computer programme
called Netuno was also used. Netuno estimates the potential for
potable water savings for different tank capacities based on daily
rainfall. All of the normative methods showed to be inadequate for
sizing rainwater tanks compared to Netuno. Also, Cordova and Ghisi
(2011) compared Netuno with the YAS and YBS models and
concluded that the three methods show results that are statistically
equivalent.
There are other studies that have assessed rainwater harvesting
system performance. Appan (2000) studied rainwater harvesting
from rooftop of high-rise buildings in Singapore by using a simu-
lation model and hourly water demand for supplying non-potable
water. Zhang et al. (2009) analysed rainwater harvesting in high-
rise buildings in Australia and concluded that bringing together
rainwater usage and efficient water fixtures (by reducing water
83
demand) could contribute to reduce potable water consumption
and improve water distribution efficiency in a long-term period.
Imteaz et al. (2011) analysed an existing rainwater harvesting sys-
tem on Swinburne University Hawthorn campus, in Australia, by
improving the tank capacities. Sample and Liu (2014) analysed
near-optimum solutions considering runoff capture reliability and
water supply, considering different input variables and locations.
Some studies have also analysed the benefits of rainwater har-
vesting to reduce urban rainwater runoff, such as Burns et al. (2014)
and Petrucci et al. (2012).
By analysing different combinations of roof area, rainwater de-
mand and potable water demand for rainwater harvesting in Brazil,
Ghisi (2009) concluded that the sizing and evaluation of rainwater
system should be performed for each specific situation, as input
variables for simulation approaches influence all performance in-
dicators. System sizing methods based on local or general recom-
mendation could lead to low efficiency and high-cost systems.
Different aspects for sizing rainwater tanks or calculating per-
formance indicators were found in the literature. However, pa-
rameters related to potable water demand variability (which is a
key factor for all possible analyses) were not properly addressed in
previous studies.
Thus, the objective of this paper is to perform a sensitivity
analysis of common design variables of a rainwater harvesting
system, and an uncertainty analysis of daily potable water demand
for system performance assessment in residential buildings.
2. Method
Method was divided in five phases: (1) definition of the simu-
lation model and general demonstration of the experiment; (2)
definition of rainfall data and the independent variables of the
experiment; (3) detailing potable water demand parameters; (4)
definition of the dependent variables; and (5) statistical treatment
of the results.
2.1. Simulation model and experiment
The Netuno computer programme was used to perform the
simulations. Netuno (Ghisi et al., 2014a,b) was validated by Rocha
(2009) and has been used with success in other researches (Ghisi,
2009; Ghisi et al., 2012, 2014a,b; Ghisi and Schondermark, 2013).
The aim of the Netuno programme is to estimate the potential
for potable water savings for different tank capacities for rainwater
storage. For the calculation procedures, Netuno uses input data such
as daily rainfall, potable water demand per capita, rainwater de-
mand (as a percentage of potable water demand), number of oc-
cupants, roof area, tank capacity, initial rainfall to be discarded and
runoff coefficient.
The algorithm begins calculating the captured water volume QðtÞ
according to Eq. (1) for each day t. The next step is to determine the
water volume consumed in day t using Eq. (2), and also to calculate
the total available rainwater volume in the tank at the end of each
day using Eq. (3). From the values obtained using Eqs. (1) and (3),
for each time step t that represents the number of days in the
rainfall time series, the potential for potable water savings is esti-
mated using Eq. (4).
QðtÞ ¼ PðtÞ  A  cp (1)


dp  DðtÞ  n
CpðtÞ ¼ min (2)
Sðt1Þ þ QðtÞ
84 A.S. Silva, E. Ghisi / Journal of Environmental Management 180 (2016) 82e93

( 1961e2013. The median can show the rainfall pattern of each

0
SðtÞ ¼ max Sðt1Þ þ QðtÞ  CpðtÞ
min
V  CpðtÞ
Pd
i¼1 CpðtÞ
E ¼ 100 
DðtÞ  n  T
where QðtÞ is the rainwater volume captured in time step t [L/day];
PðtÞ is the daily rainfall in time step t [mm/day ¼ L/m2.day]; A is the
roof surface area for capturing rainwater [m2]; cp is the runoff co-
efficient of the roof surface [non-dimensional]; CpðtÞ is the rain-
water volume consumed in time step t [L/day]; dp is the rainwater
demand [% of potable water demand]; DðtÞ is the potable water
demand in time step t [L/capita.day]; Sðt1Þ is the rainwater volume
available in the tank in time step t  1 [L]; SðtÞ is the rainwater
volume available in the tank in time step t [L]; E is the estimated
potential for potable water savings [%]; t is the time step [day]; T is
the total time period (total number of days) [day]; V is the rain-
water tank capacity [L]; n is the number of occupants in the
building [non-dimensional].
The rainwater system was modelled considering underground
tank capacities ranging from 0 to 50,000 L and a header tank with
capacity equal to the average daily rainwater demand, in each
simulated case. The initial rainfall discarded was fixed as 2 mm per
rainfall event according to the Brazilian Standard NBR 15527 (ABNT,
2007). This action is needed to discard sediments and dirt that can
affect the quality of the stored rainwater. The runoff coefficient was
fixed as 0.8, representing the actual quantity of rainwater captured
and conducted to the underground tank.
When there is no sufficient rainwater in the header tank, a
pump has to conduct rainwater from the underground tank. When
the header tank capacity is less than or equal to 10%, the pump is
turned on to pump rainwater from the underground tank to the
header tank. In order to avoid contamination there is no connection
between the potable water tank and the rainwater tanks.
When there is no rainwater available either in the rainwater
header tank or in the underground tank, potable water from the
public system supplies both potable and non-potable end-uses.
2.2. Rainfall data
In order to verify the effects of each input variable on the per-
formance evaluation of the rainwater system, eight Brazilian cities
with very distinct rainfall patterns were chosen for the simulation
experiment.
Daily rainfall data was obtained from the Brazilian National
Meteorological Institute (INMET, 2014) for the period 1961e2013.
Average rainfall and geographical location for each city are shown
in Table 1.
Fig. 1(a) shows the individual monthly rainfall data for the eight
cities in terms of median, quartiles and outlier data over the period
specific city and the differences that can occur in each year over the
(3)
period. For example, the average value for January in Floriano
polis
is 196 mm, but rainfall reached 625 mm in 1997.
Fig. 1(b) shows the average monthly rainfall over 1961e2013 for
each city. On an annual basis, Manaus has the highest rainfall and
(4)
has rainfall availability almost uniform
Petrolina, the lowest. Bage
over the whole year, while the other cities have heterogeneous
rainfall behaviour along the year.
2.3. Independent variables
Input variables for Netuno (independent variables for the
experiment) were considered as design variables for sizing rain-
water systems and for analysing the potential for potable water
savings and ideal underground tank capacity.
Table 2 shows the list of independent variables. For this study, a
residential building was considered as having variable roof area in
three levels (100, 200 and 300 m2). The number of occupants was
also considered as a variable in three levels (2, 4 and 6 occupants).
Rainwater demand is a variable that varies intensively among
buildings due to socioeconomic factors as income, age of occupants,
degree of education (Inman and Jeffrey, 2006) and also because of
the heterogeneity of the sample or by non-detected factors
(Marinoski et al., 2014). Thus, three levels of variation were adopted
(30, 50 and 70% of potable water demand value).
The independent variable called “potable water demand” is
properly detailed in Section 2.4.
2.4. Potable water demand
The “potable water demand” is the main independent variable
analysed throughout this study. Three key parameters of potable
water demand routines in the simulation experiment were
considered: (1) daily average per capita demand; (2) interval of
routine repetition; and (3) sampling coefficient of variation.
The initial parameter for defining the routines is the daily
average potable water demand, which was considered in two levels
of variation (100 and 200 L/capita.day). For each average demand, a
random sampling procedure through Latin Hypercube (Helton
et al., 2006) was performed to define a representative set of data.
Normal distribution was considered for probability density func-
tion of parameters in four coefficients of variation (10, 20, 30 and
40%). Weekly, monthly and annual variations were considered. Eq.
(5) shows the normal distribution as a function of the mean and
coefficient of variation.
For the weekly interval, a 7-day sample size was generated and
repeated throughout the rainfall time series. For the monthly in-
terval, a 30-day sample size was used and repeated, and for the
annual interval, a 365-day sample size was used and repeated
through the rainfall time series. Three different samples were
considered for each coefficient of variation. A uniform potable

Table 1
Rainfall and geographical location for the eight Brazilian cities considered in this study.

City e state Latitude/Longitude Average rainfall over 1961e2013 [mm/year]

 0 00  0 00
Manaus e AM 03 06 00 /60 01 00 2249
Natal e RN 05 470 4200 /35 120 3200 1589
Petrolina e PE 09 230 3400 /40 300 2800 452
Cuiaba
e MT 15 350 4500 /56 050 4900 1364
Brasília e DF 15 470 3800 /47 520 5800 1495
Corumb
a e MS 19 000 3200 /57 390 1000 898
Floriano
polis e SC 27 350 4800 /48 320 5700 1559
Bage
e RS 31 190 5100 /54 060 2500 1403
 A.S. Silva, E. Ghisi / Journal of Environmental Management 180 (2016) 82e93 85

Fig. 1. Rainfall data for the eight cities over the period 1961e2013.

Table 2
Independent variables of the simulation experiment using the Netuno programme.

Independent variables Unit Quantity of levels Levels of variation

Roof area m2 3 100; 200; 300

Potable water demand L/capita.day 26 104 patterns as explained in Section 2.4
Number of occupants e 3 2; 4; 6
Rainwater demand % of potable water demand 3 30; 50; 70
Underground rainwater tank capacity L 200 050,000 L (at increments of 250 L)

water demand distribution was considered as a baseline for as well as the peak demand that could occur and its random
comparison. behaviour are shown.

" #
1 ðx  mÞ2
f ðxÞ ¼ pffiffiffiffiffiffi$exp$ 
s 2p 2s2
where: f ðxÞ is the variable probability of being less than or equal to
x [non-dimensional]; e is the number of Euler [non-dimensional]; x
is the sampling data (random variable) [L/capita.day]; m is the mean
of the random variables [L/capita.day]; s is the standard deviation
of the random variables [L/capita.day].
Fig. 2 shows of the random sampling, generated in the case of
potable water demand over the three periods (annual, monthly and
weekly) and average demand equal to 100 and 200 L/capita.day for
the four different coefficients of variation and one uniform distri-
bution, on three sampling repetition procedures. The clear differ-
ences among the potable water demand for each day of the interval
2.5. Dependent variables
(5)
The potential for potable water savings estimated from Eq. (4)
was calculated for each simulation based on each underground
tank capacity. Thus, the potential savings were calculated for a great
quantity of tank capacities (as shown in Table 2), enabling the
determination of an ideal tank capacity.
For each simulation, the fraction of rainwater use was calculated
using Eq. (6) for each variation of 1000 L in the underground tank
capacity. This usage represents the real percentage of rainwater
that is used. For example, if rainwater demand is 30% for a specific
case and the potential for potable water savings reaches 24%, the
value of “percentage of rainwater usage” is 80% (equal to 24 divided
by 30).
86 A.S. Silva, E. Ghisi / Journal of Environmental Management 180 (2016) 82e93

Fig. 2. Sampling for potable water demand (100 and 200 L/capita.day), for each interval of variation (weekly, monthly and annual), coefficient of variation (uniform, 10, 20, 30 and
40%) and sampling repetition (1st, 2nd and 3rd).

If the criterion indicated in Eq. (7) complies with the respective Thus, three dependent variables were analysed for each simu-
tank capacity, this capacity is considered the ideal one. A value of lated case, representing the rainwater harvesting system
1.00%/m3 was adopted for this study purposes. performance:

E (1) The potential for potable water savings (E);

pusage ¼ (6) (2) The ideal underground tank capacity (Videal );
dp
(3) The fraction of rainwater usage (pusage ).

Dpusage .
 1:00 % m3 (7)
DV
2.6. Data treatment
where: dp is the rainwater demand [% of potable water demand]; E
is the potential for potable water savings [%]; pusage is the fraction of The first analysis was an assessment of the results for the
rainwater usage [non-dimensional]; Dpusage is the variation in the dependent variables. Histograms of frequency of occurrence for the
percentage of rainwater usage [%]; DV is the variation in the tank potential for potable water savings, ideal underground tank ca-
capacity (equal to 1000 L) [L]. pacity and fraction of rainwater usage, caused by the simulation
 A.S. Silva, E. Ghisi / Journal of Environmental Management 180 (2016) 82e93
experiment, are shown.
The second step was a sensitivity analysis of the independent
variables shown in Table 2 against the dependent variables. The
influence of variables was determined up to the second order ef-
fects using analysis of variance criteria. As the experiment was
created considering different levels of variation of independent
variables, a complete factorial design was performed (Montgomery
and Runger, 2012). The chosen sensitivity indicator is the F-value,
calculated according to Eq. (8) for first and second order effects.
MQ ðAÞ
FðAÞ ¼
MQ ðErrorÞ
where: MQ ðAÞ is the mean square of A variable (analogously for B
variable); FðAÞ is the F-value of A variable (analogously for B vari-
able); MQ ðErrorÞ is the mean square of the non-computed vari-
ances, i.e., the variances above third order effects.
The third step was an uncertainty analysis. The dependent
variables “potential for potable water savings” and “ideal under-
ground tank capacity” were organised based on their correspon-
dence with the respective influent variables. Thus, the amplitude of
uncertainty caused by the stochastic variations in potable water
demand (i.e., its parameters) was determined for each subgroup of
data, as shown in Eq. (9).
Amplitudei ¼ Maxi  Mini
where: i is each subgroup divided by daily potable water demand,
rainwater demand and roof area (45 subgroups); Amplitudei is the
amplitude of uncertainty caused only by the potable water demand
parameters in each subgroup i; Maxi is the maximum value of each
dependent variable for each subgroup i; Min i is the minimum value
of each dependent variable for each subgroup i.
For understanding some results from the uncertainty analysis, a
Pearson’s correlation approach was used to assess the relationship
between the dependent variables and some key parameters on
rainfall data. The total number of days without rain (dry periods)
was determined for each city, as well as the mean value of the
duration of the dry periods and standard deviations. Days with
rainfall events lower than 2 mm were considered as with no
rainfall.
The fourth analysis was a comparison of results obtained for
each dependent variable with the uniform potable water demand
distribution. Subgroups were divided into daily potable water de-
mand, rainwater demand, roof area, coefficient of variation of the
stochastic sample (from 10% to 40%) and interval of variation (7, 30
or 365 days), totalling 540 subgroups. The measure used was the
difference from the results in each subgroup in comparison to the
uniform distribution, shown in Eq. (10), (which was not affected by
either the coefficient of variation or the interval of variation). This
comparison can show the effect of using a deterministic value
instead of a detailed distribution for potable water demand.
 
 
Differencej ¼ limitj  unifj 
where: j is each subgroup divided by daily potable water demand,
rainwater demand, roof area, coefficient of variation, and interval of
variation (540 subgroups); limitj is the upper or lower limit of each
dependent variable in each subgroup j, caused only by the repeti-
tion of the sample; unifj is the result for the uniform distribution for
each dependent variable in each subgroup j; Differencej is the dif-
ference from the limits in each subgroup j from the result of the
uniform distribution.
87
3. Results
Results are divided in assessment of dependent variables,
sensitivity analysis of input variables, uncertainty analysis of
potable water demand parameters, and comparison with the uni-
form potable water demand.
3.1. Assessment of dependent variables
Fig. 3 shows the assessment of the dependent variables by
showing the amplitude of variation. For the potential for potable
(8)
water savings (Fig. 3(a)), the histograms show that there is one
input variable (or more than one) that caused three distinct clusters
of histograms, in all cities. Petrolina had the greatest amplitude of
variation for this dependent variable. This variable depends on the
rainwater demand.
For the ideal underground tank capacity (Fig. 3(b)), Petrolina
had the greatest amplitude of variation and greatest average;
Manaus had the lowest average. The capacities varied from 2000L
to 31,500 L in Petrolina. The fraction of rainwater usage (Fig. 3(c))
indicates that all cities had high probability of reaching the
respective rainwater demand; however, the amplitude of variation
is also high and there are many cases with low fraction of rainwater
usage in all cities.
(9) 3.2. Sensitivity analysis of input variables
Fig. 4 (a) shows the influence of each variable on the potential
for potable water savings, and Fig. 4(b) shows the ideal under-
ground tank capacity (only the ten most influent variables are
shown). The F-value was normalized to enable the comparison
between sensitivity indices in the different simulated rainfall data.
In Fig. 4(a) one can notice that rainwater demand (as a per-
centage of potable water demand) was the most important variable
affecting the potential for potable water savings, representing the
major part of the data variance. This variable was the cause of the
distinct behaviour that can be seen in the histograms shown in
Fig. 3(a).
However, for Petrolina the most important variables were
number of occupants and potable water demand. Rainwater de-
mand was not influent in this city, i.e., the variation from 30 to 70%
of the potable water demand was not significant, as there is a lack of
rainfall availability in Petrolina.
There are many influent variables affecting the ideal under-
ground tank capacity (Fig. 4(b)), such as potable water demand,
number of occupants, and rainwater demand (and their second
order effects). However, each city has different influent variables. In
Floriano
polis and Bage
the number of occupants was the most
important variable and had the major part of the data variance. In
Natal and Corumba
, the second order effects between number of
occupants and potable water demand, and number of occupants
and rainwater demand were more important than each separate
variable. In Petrolina, roof area was more important than potable
water demand or number of occupants. In the other cities (with
(10) higher rainfall) daily water demand was the most important vari-
able for underground tank sizing.
The results of the sensitivity analysis can lead to more effective
strategies for improving the potential for potable water savings and
reducing the rainwater tank capacity. By considering both depen-
dent variables (tank capacity and potable water savings), roof area,
potable water demand, number of occupants and rainwater de-
mand are key variables for the simulation. They should be deter-
mined with as most accuracy as possible for guaranteeing that the
simulation can represent the reality.
88 A.S. Silva, E. Ghisi / Journal of Environmental Management 180 (2016) 82e93
Fig. 3. Histograms of frequency of occurrence for the dependent variables for the eight cities.
3.3. Uncertainty analysis
The amplitude of uncertainty is shown in Fig. 5 for potential for
potable water savings and ideal underground tank capacity,
respectively (Amplitudei measure). Histograms show the distribu-
tion of the uncertainty values, and enable the visualization of
values more likely to occur and the distance of the extreme values
from the current distribution. These are the outlier values, i.e.,
values with minimum probability of occurrence and far from the
actual data distribution.
Table 3 shows the quantitative values of uncertainty, i.e., the
amplitude of variation for the two dependent variables. In this
calculation the outlier values were properly removed. For Flo-

polis, the uncertainty in the potential for potable water sav-
riano
ings varied from 0 to 4.2%, with mean value of 1.1%; while the
uncertainty in the ideal underground tank capacity varied from 0 to
7.4%, with mean value of 3.8%. The uncertainty in the ideal under-
ground tank capacity was calculated dividing the amplitude of
uncertainty in each group (in L) by the mean tank capacity (in L),
obtaining percentage figures. Such percentages in each group were
used for performing the statistics shown in Table 3.
The uncertainty in the potential for potable water savings is
quite low in comparison with the magnitude of the values. In Flo-

polis, the highest uncertainty was obtained for the case with
riano
potable water demand equal to 600 L/day, roof area equal to 100 m2
and rainwater demand equal to 70%. The potential for potable water
savings was 52.4% in average.
As for the ideal underground tank capacity, the uncertainty can
be high in some cases. The highest uncertainty was obtained for
Petrolina (9.4%), which was related to the interval of 19,000 L to
27,750 L (8,750 L of amplitude) for the case with potable water
demand equal to 800 L/day, rainwater demand equal to 30% and
roof area equal to 300 m2.
As for the outlier values some discussion is necessary. Brasília
has shown a single value of 10.8% of uncertainty in the potential for
potable water savings in the case of 400 L/capita.day, rainwater
demand equal to 70% and roof area equal to 100 m2. Brasília and
Cuiab
a showed uncertainty in the ideal underground tank capacity
 A.S. Silva, E. Ghisi / Journal of Environmental Management 180 (2016) 82e93 89

Fig. 4. Sensitivity analysis of independent variables.

of 13,000 L and 12,500 L, respectively, for the same case. An attempt
to understand these high but outlier values was performed using
Pearson’s correlation.
The number of days without rainfall (period between two
rainfall events) was calculated for each city. Table 4 shows the re-
sults, in which the mean number of days of each dry period is
shown, along with the coefficient of variation between periods.
The Pearson’s correlation was performed for the potential for
potable water savings and the coefficient of variation of the dura-
tion of dry periods, reaching correlation of 0.798 and p-value of
0.018. The same was performed for the ideal underground tank
capacity and the coefficient of variation of the duration of dry pe-
riods, reaching correlation of 0.833 and p-value of 0.01. These
values indicate that the correlation hypothesis cannot be dis-
regarded, with 98% confidence, i.e., the outlier values in the po-
tential for potable water savings and ideal underground tank
capacities were caused by the coefficient of variation of duration of
dry periods for each city.
Nevertheless, a combination of the stochastic potable water
demand with the heterogeneity of the rainfall events has caused
90 A.S. Silva, E. Ghisi / Journal of Environmental Management 180 (2016) 82e93

Fig. 5. Frequency of occurrence of the amplitude of uncertainty.

Table 3
Statistics of amplitude of uncertainty in the potential for potable water savings and underground tank capacity caused by potable water demand parameters in each city.

City Uncertainty in the potential for potable water savings [%] Uncertainty in the ideal underground tank capacity [%]

Mean Median Lower Upper Mean Median Lower Upper

Manaus 0.9 0.5 0.1 3.4 2.5 2.5 1.4 4.5

Natal 1.2 1.0 0.1 3.8 3.0 2.8 1.4 6.3
Petrolina 1.8 1.6 0.1 5.4 2.8 2.5 1.3 9.4
Cuiab
a 1.2 1.3 0.1 3.1 2.9 2.6 1.1 8.3
Brasília 1.0 0.9 0.1 3.1 2.6 2.4 1.2 6.3
Corumba
1.5 1.5 0.0 4.8 2.7 2.5 1.0 7.1
Floriano
polis 1.1 0.6 0.0 4.2 4.0 3.8 0.0 7.4
Bage
1.2 0.4 0.0 4.6 2.4 2.3 1.1 3.6

Notes: 1. Outliers were removed for calculating statistics. 2. The uncertainty in the ideal underground tank capacity was calculated dividing the amplitude of uncertainty due to
potable water demand parameters by the ideal underground tank capacity mean value.

Table 4
Statistics for number of days without rainfall events (dry periods).

City Rainfall time Quantity of events Mean duration of events Standard deviation of events Coefficient of variation of Maximum duration of events
series [days] without rainfall without rainfall [days] without rainfall [days] events without rainfall [%] without rainfall [days]

Manaus 19,109 3584 3.3 4.1 123 58

Natal 14,308 1957 5.3 7.8 149 77
Petrolina 14,259 808 16.2 28.6 177 259
Cuiab
a 16,715 2489 4.5 8.7 193 96
Brasília 19,021 2371 5.9 15.0 253 164
Corumba
12,734 1496 7.0 10.1 145 88
Floriano
polis 15,901 2545 4.4 4.5 102 47
Bage
17,085 2238 6.0 5.5 91 49

extreme values of uncertainty for some cities (e.g. Natal, Cuiaba
,
Brasília), specially due to the interval of the potable water demand
variable (annual, monthly, weekly). As the standard deviation is
high for the dry periods there could be some days with high de-
mand that occurred in such dry periods causing uncommon values.
Two specific cases with outlier uncertainty values were taken
from Natal, Cuiaba
and Brasília, as shown in Fig. 6. These cases
belong to the group of 400 L/day, rainwater demand equal to 70%
and roof area equal to 100 m2. The cases differ only by the “sam-
pling” and “coefficient of variation” parameters with weekly
interval parameter. Great variation in ideal tank capacity was ob-
tained in these cases, such as 10,000 L of uncertainty for Natal, and
11,000 L and 13,000 L of uncertainty for Cuiaba
and Brasília,
respectively.
This indicates that the issue of low number of days in the in-
terval of variation (weekly e 7 days) and different random samples
generated a great difference for calculating the potential for potable
water savings and, consequently, the ideal underground tank ca-
pacity. Considering only one sample, this behaviour would not be
detected in this study. The uncertainties of potable water demand
 A.S. Silva, E. Ghisi / Journal of Environmental Management 180 (2016) 82e93 91

Fig. 6. Simulated cases with major uncertainty for Natal, Cuiaba
and Brasília, by showing the potential for potable water savings for different tank capacities. The only difference
between the two cases is that index ‘3’ represents a distribution with 10% (2nd sampling) and index ‘11’ represents a distribution with coefficient of variation of 40% (1st sampling).
The bigger circles refer to the ideal tank capacity and its respective potential for potable water savings in each simulation.

parameters, i.e., their variation in time can lead to some un-
certainties in the final calculation of the potential for potable water
savings and ideal underground tank capacity for some cities.
3.4. Comparison of uniform with non-uniform potable water
demand
The results from the previous section showed that the average
potable water demand can be used rather than a detailed distri-
bution of demand, as the major number of simulated cases showed
low amplitude of uncertainty. In this section, the differences from
each one of the 540 groups of daily potable water demand, rain-
water demand, roof area, coefficient of variation and interval of
variation, were compared with the results of the uniform distri-
bution of potable water demand. Table 5 shows the relative fre-
quency in each interval for difference in the potential for potable
water savings and ideal underground tank capacity in comparison
to the uniform potable water demand (Differencej measure).
From Table 5 it can be seen that there is no significant difference
in the potential for potable water savings when considering the
uniform potable water demand in comparison to the non-uniform
potable water demand. The greatest value was obtained for Pet-
rolina, i.e., a difference greater than 1.00% had 4.1% probability of
occurrence. It can be noticed that most differences in the potential
for potable water savings when considering the uniform potable
water demand in comparison to the non-uniform potable water
demand lie between 0.01% and 0.50% for all cities (which is
insignificant).
As for the ideal underground tank capacity, the differences were
greater. The greatest figure was obtained for Floriano
polis, in which
there is 6.9% probability of having differences in the ideal under-
ground tank capacity when considering the uniform potable water
demand in comparison to the non-uniform potable water demand
greater than 10.0%; and 19.7% probability of having differences
between 5.0% and 10.0%. Most differences in the ideal underground
tank capacity when considering the uniform potable water demand
in comparison to the non-uniform potable water demand were
lower than 2.5%.
A sensitivity analysis was also performed considering as
dependent variable the difference both in the potential for potable
water savings and ideal underground tank capacities for the 540
groups when considering the uniform potable water demand in
comparison to the non-uniform potable water demand. The same
full factorial design was performed. The results showed that the
rainwater demand was the most important variable in the potential
for potable water savings, except for Petrolina, where the most

Table 5
Relative cumulative frequency for difference in the potential for potable water savings and ideal underground tank capacity when considering the uniform potable water
demand in comparison to the non-uniform potable water demand.

Relative frequency in the interval [%]

Manaus Natal Petrolina

Cuiaba Brasília

Corumba
polis
Floriano

Bage

Difference in the potential for potable water savings [%]

1.00  x 1.0 3.4 4.1 2.7 2.4 3.5 3.0 3.0
0.50  x < 1.00 4.7 7.1 8.5 7.7 5.4 5.5 6.1 3.5
0.25  x < 0.50 10.5 10.9 11.3 13.6 12.3 12.1 14.5 8.5
0.10  x < 0.25 23.8 29.2 25.7 25.2 29.0 28.2 24.6 19.8
0.01  x < 0.10 50.3 44.1 45.4 44.6 44.9 43.7 44.2 52.2
0.00  x < 0.01 9.7 5.3 5.1 6.2 6.0 7.0 7.6 13.0
Difference in the ideal underground tank capacity** [%]
10.0  x 1.3 3.1 1.5 1.7 1.9 1.8 6.9 1.1
5.0  x < 10.0 13.1 11.9 5.7 9.5 10.5 8.1 19.7 9.8
2.5  x < 5.0 28.4 29.1 20.5 31.1 28.6 22.5 25.6 22.5
1.0  x < 2.5 0.0 14.0 20.9 18.6 17.8 22.0 5.8 12.7
0.0  x < 1.0 57.1 41.9 51.4 39.1 41.3 45.7 41.9 53.9

*The percentage difference for ideal underground tank capacity was calculated by dividing the difference e in litres  from the non-uniform distribution with the uniform
potable water demand by the value obtained for the uniform potable water demand.
92 A.S. Silva, E. Ghisi / Journal of Environmental Management 180 (2016) 82e93
important variable was the interval of variation. The greatest dif-
ference for Petrolina (4.3%) as well as the major part of the figures
greater than 0.5% were obtained for a 7-day interval.
As for the ideal underground tank capacity, the most important
variable was the coefficient of variation; except for Natal, where the
most important variable was the second order effect between
rainwater demand and roof area; and for Brasília, where the most
important variable was rainwater demand. The coefficient of vari-
ation played an important role in the ideal underground tank ca-
pacity, i.e., the amplitude of variation when considering the
uniform potable water demand in comparison to the non-uniform
potable water demand was significant. In all cities (except Natal and

polis), the greatest amplitudes of variation when consid-
Floriano
ering the uniform potable water demand in comparison to the non-
uniform potable water demand were obtained for the 7-day sample
size.
4. Conclusions
This study has shown a sensitivity analysis of design variables
and uncertainty analysis of parameters related to potable water
demand in the performance of rainwater harvesting systems. The
dependent variables analysed were potential for potable water
savings and ideal underground tank capacity. The simulation
experiment was performed using the Netuno computer programme
for eight cities in Brazil, considering a rainwater system for a resi-
dential building.
Parameters such as average potable water demand per capita,
coefficient of variation of sampling, interval of data repetition and
sampling number were considered with the variation on design
variables, such as roof area, number of occupants and rainwater
demand.
In general, design variables such as daily potable water demand,
rainwater demand and roof area were the most influent variables in
the assessment of ideal underground tank capacity and potential
for potable water savings. These variables should be estimated with
major accuracy for assessing rainwater harvesting system
performance.
The uncertainty of considering stochastic variation (non-uni-
form distribution) in potable water demand was relatively low in
the potential for potable water savings, but showed some impor-
tance in the ideal underground tank sizing. The upper uncertainty
values varied from 3.1 to 4.8% in the potential for potable water
savings, and from 3.6 to 9.4% in the ideal underground tank ca-
pacity amongst all cities. The uncertainty in ideal underground tank
capacity could be as great as 13,000 L. Some combinations of
samples of potable water demand with rainfall data generated
extreme uncertainties for some cities (e.g. Natal, Cuiab
a, and Bra-
sília) in both dependent variables.
Using an average value of potable water demand for purposes of
performance evaluation of rainwater harvesting systems still gen-
erates uncertainties. However, it could be overlooked taking into
account that most of the considerable uncertainties had low
probability of occurrence as shown in the comparison of uniform
with non-uniform potable water demand.
Acknowledgements
The authors thank the National Counsel of Technological and
Scientific Development (CNPq) for the scholarship that enabled the
accomplishment of this research.
References
~o Brasileira de Normas Te
ABNT, 2007. Associaça
cnicas [Brazilian Association of


Technical Standards]. NBR 15527-Agua de chuva - Aproveitamento de cober-
turas em a
reas urbanas para fins n~ ao pota
veis e Requisitos [Rainwater - Har-
nessing roofs in urban areas for non-potable purposes e Requirements] (Rio de
Janeiro, RJ).
ANA, 2013. Age ^ncia Nacional de Aguas
e Brasil [National Water Agency - Brazil].
Conjuntura dos recursos hídricos no Brasil [Outlook on water resources in
Brazil]: ANA (Brasília, DF).
ANA, 2010. Age ^ncia Nacional de Aguas
- Brasil [National Water Agency e Brazil].
Atlas Brasil. Volume 1-Abastecimento urbano de a
gua: panorama nacional
[Brazil Atlas - Vol. 1-Urban water supply: national outlook]. Engecorps/Cobrape,
Brasilia, DF.
Appan, A., 2000. A dual-mode system for harnessing roofwater for non-potable
uses. Urban Water 1, 317e321. http://dx.doi.org/10.1016/S1462-0758(00)
00025-X.
Burns, M.J., Fletcher, T.D., Duncan, H.P., Hatt, B.E., Ladson, A.R., Walsh, C.J., 2014. The
performance of rainwater tanks for stormwater retention and water supply at
the household scale: an empirical study. Hydrol. Process. 160, 152e160. http://
dx.doi.org/10.1002/hyp.10142.
Campisano, A., Modica, C., 2014. Selecting time scale resolution to evaluate water
saving and retention potential of rainwater harvesting tanks. Procedia Eng. 70,
218e227. http://dx.doi.org/10.1016/j.proeng.2014.02.025.
Cordova, M.M., Ghisi, E., 2011. Analysis of potable water savings using behavioural
models. In: Jha, M.K. (Ed.), Water Conservation. InTech Open, pp. 89e104.
Fewkes, A., 2000. Modelling the performance of rainwater collection systems: to-
wards a generalised approach. Urban Water 1, 323e333. http://dx.doi.org/
10.1016/S1462-0758(00)00026-1.
Ghisi, E., 2009. Parameters influencing the sizing of rainwater tanks for use in
houses. Water Resour. Manag. 24, 2381e2403. http://dx.doi.org/10.1007/
s11269-009-9557-4.
Ghisi, E., Cardoso, K.A., Rupp, R.F., 2012. Short-term versus long-term rainfall time
series in the assessment of potable water savings by using rainwater in houses.
J. Environ. Manag. 100, 109e119. http://dx.doi.org/10.1016/
j.jenvman.2011.12.031.
Ghisi, E., Cordova, M.M., Rocha, V.L., 2014a. Netuno 4.0: Aproveitamento de
agua
pluvial. Programa computacional [Rainwater harvesting e Computer Pro-
gramme]. Avaliable in. www.labeee.ufsc.br.
Ghisi, E., Rupp, R.F., Triska, Y., 2014b. Comparing indicators to rank strategies to save
potable water in buildings. Resources. Conserv. Recycl. 87, 137e144. http://
dx.doi.org/10.1016/j.resconrec.2014.04.001.
Ghisi, E., Schondermark, P.N., 2013. Investment feasibility analysis of rainwater use
in residences. Water Resour. Manag. 27, 2555e2576. http://dx.doi.org/10.1007/
s11269-013-0303-6.
Gurung, T.R., Sharma, A., 2014. Communal rainwater tank systems design and
economies of scale. J. Clean. Prod. 67, 26e36. http://dx.doi.org/10.1016/
j.jclepro.2013.12.020.
Helton, J.C., Johnson, J.D., Sallaberry, C.J., Storlie, C.B., 2006. Survey of sampling-
based methods for uncertainty and sensitivity analysis. Reliab. Eng. Syst. Saf.
91, 1175e1209. http://dx.doi.org/10.1016/j.ress.2005.11.017.
Imteaz, M.A., Shanableh, A., Rahman, A., Ahsan, A., 2011. Optimisation of rainwater
tank design from large roofs: a case study in Melbourne, Australia. Resour.
Conserv. Recycl. 55, 1022e1029. http://dx.doi.org/10.1016/
j.resconrec.2011.05.013.
Inman, D., Jeffrey, P., 2006. A review of residential water conservation tool perfor-
mance and influences on implementation effectiveness. Urban Water J. 3,
127e143. http://dx.doi.org/10.1080/15730620600961288.
INMET, 2014. Instituto Nacional de Meteorologia - Brazil [National Meteorology
Institute - Brazil], 2014. Ministe
rio da Agricultura, Pecua
ria e Abastecimento.
Available in. http://www.inmet.gov.br/portal/.
Jenkins, D., Pearson, F., Moore, E., Sun, J.K., Valentine, R., 1978. Feasibility of Rain-
water Collection Systems in California. Contribution n. 173. Californian Water
Resources Centre, University of California, California.
Marinoski, A., Vieira, A., Silva, A., Ghisi, E., 2014. Water end-uses in low-income
houses in Southern Brazil. Water 6, 1985e1999. http://dx.doi.org/10.3390/
w6071985.
Montgomery, D.C., Runger, C., 2012. Estatística aplicada e probabilidade para
engenheiros [Applied statistics and probability for engineers]. John Wiley &
Sons, Inc. Ediça ~o LTC, Rio de Janeiro.
Petrucci, G., Deroubaix, J.-F., de Gouvello, B., Deutsch, J.-C., Bompard, P., Tassin, B.,
2012. Rainwater harvesting to control stormwater runoff in suburban areas. An
experimental case-study. Urban Water J. 9, 45e55. http://dx.doi.org/10.1080/
1573062X.2011.633610.
Rocha, V.L., 2009. Validaç~ ao do Algoritmo do Programa Netuno para Avaliaç~ ao do
Potencial de Economia de Agua
Pota
vel e Dimensionamento de Reservato
rios


de Sistemas de Aproveitamento de Agua Pluvial em Edificaço ~es [Validation of
the Netuno Programme for assessing the potential for potable water savings
and sizing rainwater harvesting system tanks]. Universidade Federal de Santa
Catarina.
Rupp, R.F., Munarim, U., Ghisi, E., 2011. Comparaç~ ao de me
todos para dimensio-
namento de reservato
rios de
agua pluvial [Comparison of methods for sizing
rainwater tanks]. Ambiente Construído 11, 47e64. http://dx.doi.org/10.1590/
S1678-86212011000400005.
Sample, D.J., Liu, J., 2014. Optimizing rainwater harvesting systems for the dual
purposes of water supply and runoff capture. J. Clean. Prod. 75, 174e194. http://
dx.doi.org/10.1016/j.jclepro.2014.03.075.
Silva, C.M., Sousa, V., Carvalho, N.V., 2015. Evaluation of rainwater harvesting in
 A.S. Silva, E. Ghisi / Journal of Environmental Management 180 (2016) 82e93 93

Portugal: application to single-family residences. Resour. Conserv. Recycl. 94, United Nations Educational, Scientific and Cultural Organization, Paris, France.
21e34. http://dx.doi.org/10.1016/j.resconrec.2014.11.004. Zhang, Y., Chen, D., Chen, L., Ashbolt, S., 2009. Potential for rainwater use in high-
WWAP, 2015. (United Nations World Water Assessment Programme. The United rise buildings in Australian cities. J. Environ. Manag. 91, 222e226. http://
Nations World Water Development Report 2015: Water for a Sustainable World. dx.doi.org/10.1016/j.jenvman.2009.08.008.