This study uses high-frequency water consumption data from 311 smart meters to link consumption with census data. For this purpose a well-established procedure was adopted. Results include the identification of the socio-demographic profiles associated to low, medium, medium-high and high water consumption groups and distinct daily consumption patterns in terms of the period of the day with maximum consumption: (i) morning period, (ii) morning and lunch period, (iii) dinner period. The main socio-demographic drivers to accurately understand water consumption within their different patterns were identified and refer to the characteristics of the population – rented middle size dwellings, middle size families, average educated (high school level) and professionally active population.

INTRODUCTION

Water is used for a variety of activities and sectors such as households, industries, agriculture and the production of electricity. Taking a global perspective, water use has almost tripled over the past 50 years with economic growth and the growing world population (UNESCO 2009). Nevertheless, several European countries appear to have reduced pressures on water resources by reducing or stabilising their abstraction rates per capita between 1989 and 2007 (Eurostat 2010).

The important influential factors determining the sustainability and reliability of urban water supply are socio-economic developments, such as population and economic growth, and the impacts of climate change (WSAA 2010). While structural changes may either increase or decrease domestic consumption, appliance technological changes tend to improve water use efficiency.

Water demand management has been typically approached as an engineering problem, rather than a socio-demographic one. Nevertheless, to reduce the uncertainty associated with consumption, previous studies (Loureiro 2010; Polebitski & Palmer 2010; Willis et al. 2011; Browne et al. 2013; Mamade 2013; Pullinger et al. 2013) identified socio-demographic key-factors (e.g. income, dwelling type, household size), consumption practices, among others (e.g. temperature, precipitation, efficient devices), that contribute to explain its behaviour throughout the time. This understanding is very important for network operation and planning. However, since consumption is context-dependent, the use of general relations to characterise it may introduce bias in the results. Therefore, the availability of expedite procedures to assess the impact of key-factors on consumption is very relevant. Nevertheless, some of these procedures require conducting surveys, for socio-demographic and/or behaviour data collection, involving a large amount of work and human resources (Gilg & Barr 2006; Spinks et al. 2011; Fielding et al. 2013). Census data provide detailed information at the subsection level about the buildings, dwellings, families and population and might be very useful to obtain a first understanding about key-factors that influence household consumption.

The approach proposed by Loureiro et al. (2009) and Loureiro (2010) and later improved by Mamade (2013) and Mamade et al. (2013) allows improving the understanding about spatial demand distribution within the water distribution network, which is fundamental to reduce the uncertainty in network operation and planning and to identify clients with a large potential to improve water's efficient use. Even though the promotion of efficient use in the households may lead to a reduction in the billed consumption, in a first instance, it can be of major interest for a utility. For example, in the case of water scarcity, the utility may need to identify and promote the reduction of water inefficiencies in the households. Therefore, this approach may support utilities in the selection of target clients to run water efficiency campaigns. In the case of networks with high operational costs (e.g. due to energy pumping, system layout), efficiency improvement through the shift of some appliances uses, throughout the day or the year, can contribute to a significant reduction in peak water-energy consumption. In this study, the approach was applied to identify the most important socio-demographic factors that influence household consumption, whereas in the case of Loureiro (2010) it was applied to characterise the consumption in network sectors, where the number of households can vary between 500 and 5,000.

MATERIAL AND METHODS

The procedure proposed by Loureiro et al. (2009) and Loureiro (2010) and later improved by Mamade (2013) and Mamade et al. (2013) to profile household consumption taking into consideration socio-demographic variables includes the following stages: (i) telemetry consumption data collection, processing and analysis; (ii) census data collection, processing and analysis; and (iii) consumption profiling taking into consideration socio-demographic variables.

In the first stage, continuous and real-time domestic consumption data from each client were collected during a time interval long enough to characterise data properly (i.e. at least 3 months). Since data were collected with an irregular time step, data normalisation was carried out to reduce data to 15 minutes regular time step. For each time series, a set of consumption variables was calculated to: (i) evaluate leakage in the household (minimum night consumption value between 0:00 and 6:00); (ii) find the average consumption taking into consideration different weekday scenarios (daily average consumption); and (iii) assess consumption variability (instantaneous and daily peak consumption). Relative to the minimum night consumption period (when customer demand is least and leakage is dominant), Loureiro (2010) verified that it was variable in this case study (among the network sectors and along the year), which agrees with other studies that refer that this period is case-study dependent (Fantozzi & Lambert 2012; Loureiro et al. 2012). However, this period was always included in the window of analysis adopted for leakage estimation (0:00–6:00).

Each client was also characterised in terms of the dimensionless daily pattern for different weekday scenarios (working days and weekends). k-means clustering (Han & Kamber 2006) was used to group clients in terms of consumption variables and daily consumption patterns. In the water consumption domain, cluster analysis has been used successfully to segment clients into homogeneous groups in terms of consumption and/or socio-demographic characteristics (Gilg & Barr 2006; Fontdecaba et al. 2012). According to Han & Kamber (2006), k-means is a very popular clustering approach that aims at obtaining various data partitions and then evaluate them by minimising the squared Euclidean distance between the objects and mean point of each cluster.

In the second stage, census data regarding buildings, dwellings, families and population were gathered from the 2011 Portuguese geo-referenced database at the statistical subsection level (INE 2011). A statistical subsection is the territorial unit which identifies the smallest homogeneous area, whether built-up or not, existing in the statistical section. It usually represents a block in urban areas and a locality or part of a locality in rural areas. Afterwards, a principal components analysis was run to reduce the large number of demographic variables that were highly correlated (Tabachnick & Fidell 2001). This statistical technique is applicable to a set of variables with the aim of identifying which variables in the set form a coherent subset (e.g. variables that are highly correlated with one another but largely independent of other subsets of variables combined into factors). Thus, this type of statistical analysis allows the reduction of the number of observed variables to a smaller number of principal components which account for most of the variance of the observed variables.

Subsequently, the socio-demographic resulting factors were used to correlate with the consumption profiling, by running a Spearman correlations matrix analysis.

RESULTS AND CONCLUSIONS

Case study

The methodology was tested in a group of 311 households (687 inhabitants), belonging to three small network sectors of a water distribution system, located in the north region of Portugal. In terms of metering equipment, rotating piston meters (nominal diameters between 15–20 mm, R400 metrological class, pulse resolution 1 l/pulse) were used to measure consumption in each dwelling. Telemetry systems, combining radio and GSM communications, were implemented to collect real-time consumption data from each client between January and April 2009. Customers' consumption data were collected with an irregular time step, using a fixed radio network, stored locally and then sent periodically to a central database using GSM communications. The above-mentioned network sectors are predominantly residential and correspond to territorial areas with the following dwelling characteristics: apartment blocks (Sector 1), semi-detached and detached houses (Sector 2) and detached houses with gardens (Sector 3). For this territorial area, 18 statistical subsections were identified from the census and respective data were gathered using a geoprocessing tool (Loureiro 2010; Mamade 2013). This tool calculates a set of socio-demographic variables for each network sector, through the combination of georeferenced data from the census with network data relatively to the service connections, using a geographic information system. Therefore, this tool converted census data, from statistical subsections, into socio-demographic variables associated to the network sectors. Analysed subsections have a majority of buildings with less than two floors (76%), mainly corresponding to detached houses (Figure 1). In terms of dwellings' area, it varies essentially between 50 and 200 m2 (75%). A significant proportion of residents (44%) are active population (i.e. people who spend most of the working days outside home).
Figure 1

Identification of statistical subsections, service connections and characterisation of study areas in terms of the number of floors, residents' age and dwellings' area.

Figure 1

Identification of statistical subsections, service connections and characterisation of study areas in terms of the number of floors, residents' age and dwellings' area.

Figure 2 allows the identification of the clients with the largest water use among the 311 households. In this figure, clients were ordered in terms of daily average consumption for working days. The respective minimum night consumption value is also displayed. In spite of the significant variability of daily consumption represented in Figure 2 (between 0 and 5 times the mean), the average value for all weekdays is 245 L/(client·day). For working days it is similar −240 L/(client·day) and for weekends is slightly higher (257 L/(client·day)). The consumption increase on weekends might be due to the fact that a significant part of the clients is associated to active population that may use more water (e.g. dish washing, clothes washing) during weekends. For the period of analysis, where indoor water uses predominate, only around 30% of clients have an average consumption (for all weekdays) above the national reference value (333 L/(client·day)) (ERSAR 2013). The analysis of minimum night consumption is also important to identify the occurrence of possible household leakage. Average minimum night consumption was also low (0.4 L/(client.h)), which represents only 4% of indoor use and is reduced comparatively with Brown (1984), where leakage rates ranged from 5 to 13% of water use. In this study, only 10% of the clients have average night consumption values above 1.0 L/(client·h). For this set of clients, the respective average value was 3.7 L/(client.h), accounting for a significant proportion of indoor use (20%). Therefore, the expedite analysis of minimum night consumption allowed the identification of important water inefficiencies in some clients due to possible leakage. This analysis contributes to improve the quality of service within a water utility with smart meters installed.
Figure 2

Daily average consumption and minimum night consumption for all clients.

Figure 2

Daily average consumption and minimum night consumption for all clients.

Figure 3(a) shows a time series for the client ID 60992, where several periods occur with significant minimum consumption values, which may be due to leakage or an semi-open tap. Therefore, the availability of new services that use telemetry data to alert clients about possible household leakage can remarkably contribute to detect and reduce these inefficiencies. Relatively to the distribution of minimum night consumption during the period of analysis, Figure 3(b) shows the histogram of this variable for the 32 clients with a respective average value above 1.01 L/(client·h). Although minimum night consumption is zero most of the time, it is visible that some other categories have significant consumption: between 0 and 1 at 9% and between 3 and 100 at 29%. Consumption values less than 1 L/(client.h) may correspond to a leaky tap that drips at the rate of one drip per second (EPA 2014), whereas values between 3 and 5 may be associated with toilet leaks, which constitute one of the most common type of indoor leaks (Vickers 2001) and values above 100 L/(client.h) to open taps.
Figure 3

(a) Consumption time series for the client with the ID 60992, with average night consumption equal to 13 (L/client.h); (b) minimum night consumption histogram for clients with an average value above 1 L/h.

Figure 3

(a) Consumption time series for the client with the ID 60992, with average night consumption equal to 13 (L/client.h); (b) minimum night consumption histogram for clients with an average value above 1 L/h.

Consumption clustering

At the statistical subsection level, only five statistical subsections have total daily average consumption for working days higher than the national reference value – 333 L/(client·day) (ERSAR 2013) (Figure 4). This group of statistical subsections is located in Sector 3 (Z3_303, Z3_704, Z3_705, Z3_706) and Sector 1 (Z1_419) where detached houses with large areas predominate. Figure 4 also represents the proportions of dwellings with areas between 50 and 100 m2, which indicates that subsections with lower average consumption correspond to units with a higher proportion of dwellings with small areas.
Figure 4

Variation of average consumption for working days in each statistical subsection with dwelling area 50–100 m2.

Figure 4

Variation of average consumption for working days in each statistical subsection with dwelling area 50–100 m2.

In order to group subsections and clients in terms of consumption, k-means clustering was used. At the statistical subsection level, four groups were obtained, indicating that only in the medium-high and high groups is the daily average consumption higher than the national reference value, 333 L/(client·day) (ERSAR 2013). At the client level, four groups were obtained, where more than 50% were classified in the groups of low or medium consumption, having an average value less than 251 L/(client·day) (Table 1).

Table 1

Cluster analysis results for the daily average consumption (L/client/day) at statistical subsection level and at client level (working days)

Analysis levelClusterNAverageStd. deviationMinimumMaximum
Subsection Low 181 28 137 217 
Medium 242 20 221 281 
Medium-high 388 39 355 434 
High 560 44 529 591 
Total 18 287 126 137 591 
Client Low 128 56 49 0.0 153 
Medium 93 251 50 158 330 
Medium-high 71 411 55 332 541 
High 19 787 227 602 1,517 
Total 311 240 211 0.0 1,517 
Analysis levelClusterNAverageStd. deviationMinimumMaximum
Subsection Low 181 28 137 217 
Medium 242 20 221 281 
Medium-high 388 39 355 434 
High 560 44 529 591 
Total 18 287 126 137 591 
Client Low 128 56 49 0.0 153 
Medium 93 251 50 158 330 
Medium-high 71 411 55 332 541 
High 19 787 227 602 1,517 
Total 311 240 211 0.0 1,517 

k-means was also applied to group dimensionless daily average patterns for working days and weekends, for each client. A daily pattern is a curve uniting some representative value of consumption at each successive point throughout the 24 hours of the day. Different days of the week may correspond to different patterns. There are significant differences between working days and weekends daily consumption patterns (Figure 5). Figure 5 represents the typical daily behaviour for each cluster, given by the median value for each instant throughout the 24 hours. During working days, the differences between the three groups of patterns were more noticeable and they were classified according to the maximum consumption period: (i) morning (n = 54), (ii) morning–lunch (n = 103), (iii) dinner (n = 133). Results indicate that the morning peak pattern is less frequent among the clients than the remaining patterns. On the other hand, daily consumption patterns for weekends are much less variable, indicating that the consumption behaviours during the weekend are very similar among the clients.
Figure 5

Daily consumption patterns for (a) clusters for working days and (b) clusters for weekends.

Figure 5

Daily consumption patterns for (a) clusters for working days and (b) clusters for weekends.

After the cluster analysis regarding the daily average consumption (four clusters) for working days and daily consumption patterns (three clusters), new variables were created. These variables correspond to the clients' distribution by each of the seven obtained clusters.

Socio-demographic analysis

Prior to the correlations analysis between water consumption and demographic information, several principal components factorial analyses were conducted to aggregate single variables within each census domain (i.e. buildings, dwellings, families and population). Table 3 shows the results of these analyses, only regarding to the factors that obtained a significant correlation with the consumption profiling.

Regarding domain ‘dwellings’, the factor that correlates with household consumption is the one formed by the variables related with the dwellings size and type of property. This factor explains 45.72% of the total variance of the domain, presenting a high reliability (α = 0.93) and was referred to as ‘rented middle size dwellings’. The census domain ‘families’ is represented by a factor composed by the variables regarding the families size, age and professional occupation/status. This factor explains 46.28% of the total variance of the domain, presenting a high reliability (α = 0.93) and was designated by ‘working middle size families’. Finally, the domain ‘population’ is represented by a factor composed of the variables regarding the population studying/working and living in the same municipality, residents aged up to 19 years old, residents with a professional occupation and with mandatory graduation. This factor explains 43.52% of the total variance, showing a high reliability (α = 0.91) and was designated by ‘working adults and school age population’ (Table 2).

Table 2

Results from socio-demographic data factorial analysis (Census 2011)

 Census information domain
DwellingsFamiliesPopulation
Rented middle size dwellingsWorking middle size familiesWorking adults and school age population
Cronbach α 0.93 0.93 0.91 
Eigenvalues 3.65 3.70 6.52 
Variance explained 45.72 46.28 43.52 
 Factorial scores 
Dwellings with 50–100 m2 0.93   
Dwellings with 50 m2 0.80   
Rented dwellings 0.91   
2–3 bedrooms dwellings 0.82   
Families with 1–2 persons  0.89  
Families with 3–4 persons  0.82  
Families without unemployed  0.85  
Families without youngsters (<15 yrs)  0.77  
Population studying/living same district   0.92 
Population working/living same district   0.79 
Residents with <4 yrs   0.77 
Residents with 5–9 yrs   0.91 
Residents with 10–13 yrs   0.63 
Residents with 14–19 yrs   0.87 
Residents with 3rd grade   0.84 
Residents with high school completed   0.65 
Residents with a job   0.78 
 Census information domain
DwellingsFamiliesPopulation
Rented middle size dwellingsWorking middle size familiesWorking adults and school age population
Cronbach α 0.93 0.93 0.91 
Eigenvalues 3.65 3.70 6.52 
Variance explained 45.72 46.28 43.52 
 Factorial scores 
Dwellings with 50–100 m2 0.93   
Dwellings with 50 m2 0.80   
Rented dwellings 0.91   
2–3 bedrooms dwellings 0.82   
Families with 1–2 persons  0.89  
Families with 3–4 persons  0.82  
Families without unemployed  0.85  
Families without youngsters (<15 yrs)  0.77  
Population studying/living same district   0.92 
Population working/living same district   0.79 
Residents with <4 yrs   0.77 
Residents with 5–9 yrs   0.91 
Residents with 10–13 yrs   0.63 
Residents with 14–19 yrs   0.87 
Residents with 3rd grade   0.84 
Residents with high school completed   0.65 
Residents with a job   0.78 

Socio-demographic profiling of consumption

Table 3 shows the correlations between the above referred consumption variables and socio-demographic factors. Regarding the clusters of daily patterns of working days, the strongest correlations are between the ‘peak in the dinner period’ and the rented middle size dwellings, the working middle size families and with the working and school age population. The cluster ‘peak in the morning-lunch period’ is more strongly related with working middle size families whereas the cluster ‘peak in the morning’ is most related to active working population and young people at school age. These findings were less differentiating than those expected and may be due to the fact that we are dealing with a more aggregate level of information, such as the statistical subsection, where the slight differences between patterns are more probably blurred.

Table 3

Correlation matrix between water consumption clusters and socio-demographic factors

 Spearman's rho coefficient correlationRented middle size dwellingsWorking middle size familiesWorking adults and school age population
Working days daily patterns at client level Peak in the dinner period 0.71** 0.81** 0.82** 
Peak in the morning- lunch period 0.67** 0.68** 0.65** 
Peak in the morning period 0.69** 0.63** 0.74** 
Daily average consumption at client level High 0.20 ns 0.40 ns 0.33 ns 
Medium-high 0.64** 0.72** 0.55* 
Medium 0.80** 0.79** 0.84** 
Low 0.79** 0.61** 0.89** 
Daily average consumption at subsection level (L/client.day)  0.55* 0.23 ns 0.23 ns 
Daily average consumption at subsection level (clustered groups)  0.53* 0.29 ns 0.10 ns 
 Spearman's rho coefficient correlationRented middle size dwellingsWorking middle size familiesWorking adults and school age population
Working days daily patterns at client level Peak in the dinner period 0.71** 0.81** 0.82** 
Peak in the morning- lunch period 0.67** 0.68** 0.65** 
Peak in the morning period 0.69** 0.63** 0.74** 
Daily average consumption at client level High 0.20 ns 0.40 ns 0.33 ns 
Medium-high 0.64** 0.72** 0.55* 
Medium 0.80** 0.79** 0.84** 
Low 0.79** 0.61** 0.89** 
Daily average consumption at subsection level (L/client.day)  0.55* 0.23 ns 0.23 ns 
Daily average consumption at subsection level (clustered groups)  0.53* 0.29 ns 0.10 ns 

** P < 0.001.

* P < 0.01.

Following the analysis of Table 3, ‘low’ and ‘medium’ consumption clusters correlate primarily with rented middle size dwellings and with the working population and young population at school age. In addition, the ‘medium’ consumption cluster is also strongly associated with working middle size families that, in turn, is the only related socio-demographic variable with the "above the average consumption cluster". For the ‘high’ consumption cluster no reliable correlation was found with socio-demographic variables since the number of clients in this cluster was reduced (N = 19).

The correlation matrix at the subsection level only produced reliable results between one demographic factor from the dwellings domain and average daily consumption water, which means that the subsections where rented middle size dwellings are higher are associated with higher average consumption.

CONCLUSIONS

The results of this study on urban water consumption clearly suggest that it is worthwhile distinguishing two different aspects: the way people use water and to what extent they use water. In this case study, clients with lower or medium consumption predominate and during working days the daily consumption pattern group predominates where the instantaneous peak consumption occurs during the dinner period. This study also demonstrated the potential of smart meters to detect inefficiencies, namely household leakage. Regarding the nexus of water consumption–socio-demographic characteristics of the population, working middle size families, mainly living in rented houses, average educated (high school level) and professionally active population are the socio-demographic drivers to accurately understand water consumption within their different patterns. Census data allow obtaining a first understanding about domestic consumption. Furthermore, when the level of analysis is more aggregated, some relations are blurred compared to the pattern of results obtained at the individual level. This clearly suggests that the individual information seems to be preferable when compared to the georeference information, allowing a more comprehensive, diverse and accurate understanding of the mixed social and water consumption profiling. This conclusion is reinforced by the need to obtain individual demographic data, as well as detailed water consumption practices, in order to obtain more reliable findings at the individual level.

ACKNOWLEDGEMENTS

The research leading to these results was funded by the European Union Seventh Framework Programme (FP7/2007-2013) under grant agreement no. 318272 iWIDGET project. The authors would like to thank to AGS and Águas de Barcelos for ensuring the data collection programmes and the data provided.

REFERENCES

REFERENCES
Brown
C.
1984
Residential Water Conservation Projects
.
Summary Report
.
The Division
,
Washington, DC
.
EPA
2014
Water Sense. Available from: www.epa.gov/WaterSense/pubs/fixleak.html
.
ERSAR
2013
Annual Report of Water and Waste Services in Portugal (2012)
,
Volume 1
.
ERSAR
,
Lisbon
.
Eurostat
2010
Sustainable Development in the European Union. Available from: http://epp.eurostat.ec.europa.eu/cache/ITY_OFFPUB/KS-02-13-237/EN/KS-02-13-237-EN.PDF
.
Fantozzi
M.
Lambert
A.
2012
Residential night consumption – assessment, choice of scaling units and calculation of variability
. In:
Water Loss 2012
,
26–29 February
,
Manila
,
Philippines
.
Fielding
K. S.
Spinks
A.
Russell
S.
McCrea
R.
Stewart
R.
Gardner
J.
2013
An experimental test of voluntary strategies to promote urban water demand management
.
J. Environ. Manage.
114
,
343
351
.
Fontdecaba
S.
Grima
P.
Marco
L.
Rodero
L.
Sanchez-Espigares
J. A.
Sole
I.
Tort-Martorell
X.
Demessence
D.
De Pablo
V. M.
Zubelzu
J.
2012
A methodology to model water demand based on the identification of homogenous client segments application to the city of Barcelona
.
Water Resour. Manage.
26
(
2
),
499
516
.
Han
J.
Kamber
M.
2006
Data Mining: Concepts and Techniques
.
Morgan Kaufmann Publishers
,
São Francisco
.
INE
2011
CENSOS 2011 – XV recenseamento geral da população e V recenseamento geral da habitação
.
Instituto Nacional de Estatística (INE)
,
Lisbon
.
Loureiro
D.
2010
Consumption analysis methodologies for the efficient management of water distribution systems
.
PhD thesis
,
Universidade Técnica de Lisboa
,
Lisbon
,
Portugal
.
Loureiro
D.
Coelho
S. T.
Ramalho
P.
Alegre
H.
Covas
D. I. C.
Pina
A.
2009
Correlating Water Consumption Behaviour with Billing, Infra-structure and Socio-demographic Factors
.
CCWI 2009
,
Sheffield
.
Loureiro
D.
Alegre
H.
Coelho
S. T.
Borba
R.
2012
A new approach to estimating household night consumption at DMA level
. In:
Water Loss 2012
,
26–29 February
,
Manila
,
Filipinas
.
Mamade
A.
2013
Profiling Consumption Patterns using Extensive Measurements
.
MSc thesis Universidade Técnica de Lisboa
,
Lisbon
,
Portugal
.
Mamade
A.
Loureiro
D.
Covas
D.
Coelho
S. T.
Amado
C.
2013
Spatial and Temporal Forecasting of Water Consumption at the DMA Level using Extensive Measurements
. In:
2nd International Conference on Computing and Control for the Water Industry, CCWI2013
,
2–5 September
,
Perugia
,
Italy
.
Polebitski
A. S.
Palmer
R. N.
2010
Seasonal residential water demand forecasting for census tracts
.
J. Water Resour. Plann. Manage.
136
(
1
),
27
36
.
Pullinger
M.
Anderson
B.
Browne
A.
Medd
W.
2013
New directions in understanding household water demand: a practices perspective
.
J. Water Supply Res. Technol.
62
(
8
),
496
506
.
Spinks
A.
Fielding
K.
Russell
S.
Mankad
A.
Price
J.
2011
Water demand management study: Baseline survey of household water use (Part A)
.
Urban Water Security Research Alliance Technical Report(40)
.
CSIRO
,
City East
,
Australia
.
Tabachnick
B. G.
Fidell
L. S.
2001
Using multivariate statistics
.
UNESCO
(
2009
)
Water in a Changing World – The United Nations World Water Development Report 3. Available from: www.unesco.org/new/en/natural-sciences/environment/water/wwap/wwdr/wwdr3-2009/
.
Vickers
A.
2001
Handbook of Water Use and Conservation: Homes, Landscapes, Businesses, Industries, Farms
.
WaterPlow Press
,
USA
.
Willis
R. M.
Stewart
R. A.
Giurco
D. P.
Talebpour
M. R.
Mousavinejad
A.
2011
End use water consumption in households: impact of socio-demographic factors and efficient devices
.
J. Clean. Product.
60
,
107
115
.
WSAA
2010
.