Professional Documents
Culture Documents
Abstract: Apparent water losses can be problematic to water companies’ revenues. This type of loss is very difficult to detect and quantify
and is often associated with water meter anomalies. This study was motivated by a water company’s challenge that links a decrease in water
consumption to water meters’ malfunction. The aim is to develop a strategy to detect decreasing water usage patterns, contributing to meter
performance assessment. The basis of the approach is a combination of statistical methods. First, the time series of billed water consumption
is decomposed using Seasonal-Trend decomposition based on Loess. Next, breakpoint analysis is performed on the seasonally adjusted time
Downloaded from ascelibrary.org by 23.22.233.225 on 11/21/21. Copyright ASCE. For personal use only; all rights reserved.
series. After that, the Mann–Kendall test and Sen’s slope estimator are used to analyze periods of progressive decrease changes in water
consumption, defined by breakpoints. A quantitative indicator of this change is proposed. The strategy was successfully applied to eight-time
series of water consumption from the Algarve, Portugal. DOI: 10.1061/(ASCE)WR.1943-5452.0001492. This work is made available under
the terms of the Creative Commons Attribution 4.0 International license, https://creativecommons.org/licenses/by/4.0/.
develop mechanisms for predictive water planning based on data company should investigate.
analysis. In particular, a mechanism able to detect abnormal de- The paper is organized as follows: the first section presents a
creasing trends in water consumption, which they believe is asso- brief description of the problem and the dataset analyzed. The next
ciated with a possible water meter malfunction. section explains the methodologies applied, followed by a descrip-
This paper aims to develop a procedure capable of detecting tion of the water consumption analysis’s main results on a case
study of seven households and one hotel, ending with the discus-
decreasing water consumption changes, contributing to meter per-
sion and conclusion and future work.
formance assessment. The decrease in water consumption (pro-
gressive or sudden) is, as explained by Monedero et al. (2016),
associated with water meter malfunction—one of the possible
causes for apparent losses. Thus, it should be investigated by water
Methodologies
utilities in order to mitigate this type of loss. The proposed ap- In this section, the methods that constitute the strategy proposed are
proach intends to contribute and provide a helpful tool for the water described in detail.
utilities within decision analysis on water meter replacement. The
strategy developed uses several statistical methods to detect signifi-
cant decreases (changes) in water consumption after removing the STL Decomposition
seasonal pattern. A time series is a sequence of observations indexed by time t,
In this case study, an empirical analysis of billed water con- t ¼ 1; : : : ; n, usually ordered in equally spaced intervals. Often,
sumption data from seven households and a hotel from the Algarve, time series exhibit seasonal behavior and adequate control for a sea-
Portugal, was performed. Analyzing data on the real conditions on sonal component is essential before using any statistical model. Let
which the water meters operate should provide more precise results consider that a time series is a combination of three components:
than those obtained through laboratory analysis. Time series de- trend-cycle, which represents an upward or downward movement
composition is frequently used in many areas, such as in electricity over the time horizon; seasonality, which is a repetitive pattern over
consumption to evaluate electrical meter data [e.g., Miller and time; and the irregular, which is remaining after the other compo-
Meggers (2017) and Miller (2019)], in hydrological studies to nents have been removed (Hyndman and Athanasopoulos 2018).
understand groundwater behavior [e.g., Lafare et al. (2016)], and Both trend and seasonality components are potential confounding
to predict or detect changes in water consumption [e.g., Ohana- features in analysis, so their identification and removal are impor-
Levi et al. (2020), Quesnel and Ajami (2017), Hester and Larson tant. Additionally, this characteristic impacts the computation of the
(2016), and Gelažanskas and Gamage (2015)]. Seasonal-Trend trend-cycle term T t that bears the time series’s interannual variations
Decomposition of time series based on Loess [(Seasonal-Trend de- but excluding possible variations related to interannual changes in
composition by Loess (STL)] (Cleveland et al. 1990) has been suc- seasonality.
cessfully used in studies of water consumption (Hester and Larson Seasonal-Trend Decomposition of time series based on Loess
2016). Researchers often use the STL default smoothing parameters (STL) is used to decompose the time series fY t ; t ¼ 1; : : : ; ng into
because specifying these parameters can be subjective rather than a trend (T t ), seasonal (St ), and irregular (I t ) components using non-
objective. To overcome this issue, the procedure stl.fit (Cristina et al. parametric regression and is described as
2016) is used to obtain a more accurate decomposition based on the
error measure mean absolute error (MAE). Y t ¼ T t þ St þ I t ; t ¼ 1; : : : ; n ð1Þ
In the context of water consumption analysis, some recent stud- STL was chosen over other decomposition methods in the liter-
ies have adopted the methodology of time series decomposition and ature because it has attractive modeling features. Such as a nonlinear
breakpoint detection, such as Quesnel and Ajami (2017) and Hester trend component, the seasonal component is allowed to change over
and Larson (2016). The latter points out that decomposition meth- time and is robust in the presence of outliers. Note that these
ods have been applied in several hydrological studies, and structural observations are problematic since it influences the estimation of
change models have been used in water quality. Also, the opinion of parameters (e.g., in ARIMA models) that will mislead any statistical
these authors is that the breakpoint analysis appears underutilized in conclusion. STL is an iterative nonparametric procedure that repeat-
water management. Therefore, the procedure developed in this case edly uses a Loess smoother to refine and improve estimates of the
study uses breakpoint analysis (Zeileis et al. 2003) to find points trend and seasonal components [see Cleveland et al. (1990) for more
in time at which statistical patterns change after the seasonal pat- details]. In common with all nonparametric regression methods,
tern was removed. These breakpoints are associated with segments STL requires the subjective selection of smoothing parameters.
where the increases or decreases of the trend are more pronounced. The two main parameters are the seasonal (s.window) and trend
work because it is less sensitive to outliers (Hyndman and Koehler nitude of the trend (or slope) is obtained. If the result points to a
2006) when compared with the root mean square error, and it significant and negative trend, the breakpoint is relevant and is in-
presents several advantages as explained in Willmott and Matsuura cluded in the new set bp , where bp ⊆ bp. This analysis is per-
(2005). In the forthcoming analysis, the seasonally adjusted time formed using the functions mk.test() and sens.slope(), available in
series the R package trend (Pohlert 2020).
Y t ¼ Y t − St ð2Þ
Relative Magnitude of the Change
t ¼ 1; : : : ; n, is considered.
The RMC is proposed to assess water consumption behavior
change. It is calculated through a ratio that compares the consump-
Detect Structural Changes tion pattern before and after a breakpoint (see Fig. 1). The idea of
Some structural disturbances can be observed in the dynamic struc- RMC is in line with the same principle as in Gocic and Trajkovic
ture of the time series. Is important to stay alert for breakpoints, (2013) and is defined by
points in time at which statistical patterns change, as they represent
periods of the time horizon where the increases or decreases of the slpafter − slpbefore
trend are more pronounced. The R package strucchange (Zeileis RMC ¼ ð5Þ
jslpbefore j
et al. 2002) is used to detect structural breaks to test for structural
changes in linear regression models, estimating the number of seg-
ments (m) and the set of the breakpoints bp ¼ ft1 ; t2 ; : : : ; tm−1 g, where slpafter and slpbefore = nonparametric Sen’s Slopes at a
minimizing the Bayesian information criterion (BIC) and the breakpoint tk ∈ bp . This unitless indicator measures the relative
residual sum of squares (RSS) (Zeileis et al. 2003). Additionally, change, or the relative difference, between the slope after (slpafter )
this procedure has a bandwidth parameter (h) that determines the and the slope before (slpbefore ) the breakpoint, in reference to the
minimal segment size (h × n observations). In function breakpoints slope before the breakpoint. A negative value of RMC is an alert for
() (Zeileis et al. 2002) the default bandwidth parameter is h ¼ 0.15 a possible meter anomaly. The more negative is RMC, the higher is
but, it is up to the user to change h to a more suitable value based on the reduction of water consumption after the considered breakpoint.
the knowledge of the data. For example, Verbesselt et al. (2010) use
a minimum of 12% of data (23 observations) for a nine-year time
horizon; in Awty-Carroll et al. (2019), this minimum was two years Strategy Outline
(46 observations); and in Hester and Larson (2016), the minimum The idea behind the approach is the following:
was three years (36 observations). In line with the previous refer- Step 1: Estimation of the seasonal component using stl.fit() by
ences, the present study contributes with two new expressions for minimizing MAE and removing it;
obtaining the minimum length between consecutive breaks Step 2: Breakpoints analysis (bp) on y ;
Step 3: Nonparametric trend analysis for analyzing significant
frequency decreases on the segments and selecting the “relevant” break-
h¼ ð3Þ
n points (bp );
Step 4: RMC is calculated [Eq. (5)] for each breakpoint in bp .
and the maximum number of breaks The scheme of the strategy is shown in Fig. 1. In Step 1, the stl.
fit() was chosen because it showed better performance than the
n 0.5 “standard stl” (Cristina et al. 2016). Nevertheless, it is possible to
breaks ¼ ¼ ð4Þ
2 frequency h use another decomposition method instead, e.g., classical decom-
position. Step 2 uses the function breakpoints() (Zeileis et al. 2002)
where n = length of the time series and frequency is the seasonal with h and breaks estimated by Eqs. (3) and (4), respectively. In the
pattern’s length. case of Step 3, a function that includes mk.test() (Pohlert 2020) was
implemented by the authors in R. It considers bp and y as input
and bp as output. Then, to obtain the RMC values using the TS
Detect Decreasing Trends
slopes, another function was build by the authors. The new code
Research on an increasing or decreasing trend in time series is an was implemented in R version 4.1.0. The data, models, and code
important issue. It is usually based on the parametric test for the are available online (Cordeiro et al. 2020). Moreover, a significance
linear model and in the Mann–Kendall (MK) test [Mann (1945) level of α ¼ 5% is considered in the statistical analysis performed.
Water Consumption Analysis meter replacement in July 2014). There is also a clear seasonal ef-
fect in RH1, RH2, RH3, and NRH3. Concerning the trend of water
Dataset consumption, RH2 and RH3 showed a downward. This pattern is
also observed in smaller periods in the other time series. Moreover,
To illustrate the procedure (Fig. 1), eight-time series of billed water the magnitude of volumes consumed is significantly different be-
consumption from the same region of the Algarve, Portugal, were tween the households and the hotel. In households (RH and NRH
used [available online (Cordeiro et al. 2020)]. The data were pro- type), the water consumption range from 0 to 120 m3 , while at the
vided by the water company Infraquinta and Loulé Municipality hotel, the consumption is higher, varying from 0 to 3,000 m3 .
and are described in the following:
• Two locations: Quinta do Lago Resort, a well-known tourist
destination for holidays; and a small rural village with less than Time Series Decomposition
3,000 inhabitants, mainly retired and older people. Distance be- The empirical analysis starts by deciding between a robust approach
tween locations is approximately 26 km; of STL or not. Outliers were detected by inspecting the weights as-
• Residential Household (RH): four monthly time series of water sociated with the residuals as in Cleveland et al. (1990). Next, for
consumption, time horizon between January 2011 and December each time series, the procedure stl.fit (Cristina et al. 2016; Cordeiro
2016 (nRH ¼ 72), from the village; 2016) was applied and an object of class stl with components T t , St ,
• Nonresidential Household (NRH): three hourly time series of and I t , as in Eq. (1), is returned. Measurements of error is an alter-
water consumption, time horizon between November 2013 and native way of conveying information about the model adjustment,
December 2018, from Quinta do Lago. However, the water and for that, the MAE was chosen. This accuracy measure has al-
companies are only concerned with the monthly levels of water ready been used in the context of water planning and management,
consumption, so the hourly data were aggregated into monthly e.g., Kandiah et al. (2016). The decomposition plots are shown
by taking the average of each month (nNRH ¼ 62); in Fig. 3.
• Hotel: monthly time series of water usage with a time horizon
from July 2006 until December 2016 (nHotel ¼ 120);
• Smart meters collected the data of NRH and RH and Hotel by Decreasing Changes
nonsmart meters. In this subsection, the objective is to detect periods of changes in
Fig. 2 shows water consumption variability in households and water usage patterns. The detection of breakpoints, which are points
hotels. Most of the time series shows a seasonal pattern associated in time k at which statistical patterns change, are estimated on the
with the year’s season, increasing in the summer and decreasing seasonally adjusted water consumption, as in Eq. (2). The procedure
during the winter. tests for structural changes in linear regression models, considering
The hotel presents a strong seasonal pattern and a decreasing the maximum number of breaks and the minimum length between
trend until 2014 that increases after mid-2014 (due to the water breaks estimated using Eqs. (3) and (4), respectively. As a result,
75 15
m3
m3
50 10
25 5
0 0
2012 2014 2016 2012 2014 2016
Downloaded from ascelibrary.org by 23.22.233.225 on 11/21/21. Copyright ASCE. For personal use only; all rights reserved.
RH3 RH4
25 15
20
10
15
m3
m3
10
5
5
0 0
2012 2014 2016 2012 2014 2016
NRH1 NRH2
50 50
40 40
30 30
m3
m3
20 20
10 10
0 0
2014 2015 2016 2017 2018 2019 2014 2015 2016 2017 2018 2019
NRH3 Hotel
120 3000
80 2000
m3
m3
40 1000
0 0
2014 2015 2016 2017 2018 2019 2006 2008 2010 2012 2014 2016
breaksRH ¼ 3 and hRH ¼ 0.17, breaksNRH ¼ 3 and hNRH ¼ 0.19, statistically significant (i.e., pvalue < 0.05), the breakpoints are
and breaksHotel ¼ 5 and hHotel ¼ 0.1 were obtained. After ob- considered “relevant” and included in bp , and its decreasing
taining the breakpoints (bp, in Table 1), the MK test for analyzing rate (Sen’s slope) is obtained (see Table 2). Fig. 4 presents the sea-
trends was applied to each segment. If a decreasing segment is sonally adjusted time series, breakpoints (bp), and relevant
data
data
50 10
25
5
0
15.0
30
25 trend 12.5
trend
20 10.0
15 7.5
10 5.0
30
seasonal
seasonal
20 3
10
0 0
−10 −3
−20
60
remainder
remainder
40 5
20
0 0
−20
2011 2012 2013 2014 2015 2016 2017 2011 2012 2013 2014 2015 2016 2017
Downloaded from ascelibrary.org by 23.22.233.225 on 11/21/21. Copyright ASCE. For personal use only; all rights reserved.
RH3 RH4
25 15
20
10
data
data
15
10 5
5 0
10 7
9 6
trend
trend
8 5
7 4
6
6
seasonal
seasonal
4 2
2
0 0
−2
−4 −2
remainder
10 4 remainder
5 0
0 −4
2011 2012 2013 2014 2015 2016 2017 2011 2012 2013 2014 2015 2016 2017
NRH1 NRH2
50
40 40
30
data
data
30
20 20
10 10
0 0
7
6 12
trend
trend
5
4 10
3
2 8
20 20
seasonal
seasonal
10 10
0
0
−10
remainder
remainder
30 20
20 10
10
0 0
−10 −10
2014 2015 2016 2017 2018 2019 2014 2015 2016 2017 2018 2019
NRH3 Hotel
3000
90
2000
data
data
60
30 1000
0 0
16 1600
trend
trend
14 1400
1200
12 1000
seasonal
seasonal
40 400
20 0
0 −400
remainder
remainder
80 1000
500
40 0
0 −500
−40 −1000
2014 2015 2016 2017 2018 2019 2006 2008 2010 2012 2014 2016
breakpoints (bp , blue dashed line). The dashed grey line corre- analysis requires a minimal segment size (h), which in this case
sponds to a breakpoint with nonsignificant adjacent segments is 14 months (corresponding to hRH4 × 72 ¼ 0.2 × 72). So, this
(slpbefore and slpafter ), i.e., pvalue > 0.05, or the water con- breakpoint was out of reach of being detected. However, if data
sumption is not decreasing. Table 2 contains the information about before 2011 were available, this issue would no longer exist.
the date of the water meter replacement, bp , TS slopes, and Concerning the nonresidential households from the Quinta do
RMC [Eq. (5)]. Lago Resort, NRH1 shows one breakpoint in June 2017. According
The residential household RH1 shows one relevant breakpoint to Table 1, the first segment is statistically significant, so the break-
detected in June 2012 (Table 2; Fig. 4). The water meter readings point is considered relevant (Table 2). The relative magnitude of
after the break are smaller than the first segment, and the magnitude change is RMC ¼ 2.71, revealing an increase in consumption after
of the trend (TS slope) is less negative (slpbefore ¼ −0.50 versus the breakpoint, so there is no motive for alarming the water com-
slpafter ¼ −0.19), with RMC ¼ 0.62, as seen in Table 2. Nonethe- pany. Besides, this water meter was replaced in June 2018. How-
less, the trend continues negative after June 2012 with a significant ever, judging by the graph of the water consumption (Figs. 2 and 4),
slope, justifying an alert and further investigation by the water it seems a hasty decision by the company. In NRH2, two break-
utility. Regarding the RH2, the global trend is decreasing (Figs. 2 points were detected (Table 1), but only one is relevant (November
and 3), and the structural break analysis identifies two breakpoints 2017) with an estimate RMC ¼ −3.18 (Table 2). According to
(Fig. 4). The decrease of the third segment is statistically significant this table, this meter was already malfunctioning for half a year
(Table 1) and so the second breakpoint (December 2015) is rel- when the substitution took place. In NRH3, the procedure detected
evant (Table 2). The RMC ¼ −12 reveals that the downward trend two breakpoints, but only November 2017 is relevant (bp ) with
(December 2015–December 2016) is more pronounced in magni- RMC ¼ −1.10. Note that this decline was detected one year before
tude compared to the previous segment (April 2012–December the meter replacement (Table 2). If a surveillance system integrated
2015). Since the consumption is decreasing, the water company this approach, the water company would have received an alert one
should keep RH2 under surveillance. For households RH3 and year before.
RH4, only a relevant breakpoint is detected, May 2014 and In the case of the hotel, from the five segments (four break-
November 2015, respectively. Moreover, comparing the TS slopes points), only two are statistically significant, corresponding to
before and after the breakpoint, the magnitude of change is consid- bp ¼ fFebruary 2008; July 2013; July 2014g, as seen in Table 2
erably higher in water meter RH4 (RMCRH4 ¼ −20 vs RMCRH3 ¼ and Fig. 4. The estimated RMC are 0.58, −31.65, and 1.06, respec-
−5). RH4 exhibits an abrupt change at the beginning of 2012 tively (Table 2). The negative value of RMC for the second break-
(Figs. 2 and 4), and based on the records from the water company, point and the decrease in consumption shows evidence that the
the meter was replaced in January 2012 (13th month). As already meter should have been replaced sooner, perhaps between July
explained in subsection “Detect structural changes,” the breakpoint 2013 and July 2014.
75 15
m3
m3
50 10
25 5
0 0
2012 2014 2016 2012 2014 2016
Downloaded from ascelibrary.org by 23.22.233.225 on 11/21/21. Copyright ASCE. For personal use only; all rights reserved.
RH3 RH4
25 15
20
10
15
m3
m3
10 5
5
0
0
2012 2014 2016 2012 2014 2016
NRH1 NRH2
50 50
40 40
30
30
m3
m3
20
20
10
10
0
−10 0
2014 2015 2016 2017 2018 2019 2014 2015 2016 2017 2018 2019
NRH3 Hotel
3000
100
2000
m3
m3
50
1000
0
0
2014 2015 2016 2017 2018 2019 2006 2008 2010 2012 2014 2016
As observed in Table 2, the results indicate that the company the approach in monitoring water consumption by the consumers,
should continue to monitor all households’ meters, except RH1, such as sudden changes and continuous decrease. Besides, the ap-
NRH1, and the hotel since the corresponding RMC values are pos- proach could detect these changes sooner than the actual replace-
itive. To sum up, the previous results illustrate the effectiveness of ment time and in instances in time not foreseen by the water utility.
occur due to inaccuracies in metering, illegal use, meter reading meter performance assessment. It was developed and implemented
errors, data handling, and billing errors. These type of losses has using R language and is supported by reliable statistical methods.
a significant impact on the revenue of the water companies. The Besides, a new quantitative measure, designated as RMC, is pro-
cost for the utility of a cubic meter of real losses is much less than posed. The objective is to obtain a magnitude of the water usage
the price of a cubic meter of apparent losses. This issue is of great change, which can help decide on the meter’s replacement. The
importance because while real losses are easily recognized, appar- water company must also investigate the decline of the consump-
ent losses are not perceived well likely because apparent losses tion (RMC < 0) since it might be related to a possible anomaly in
cannot be directly seen and are mostly related to economic losses. the meter. If so, it is the ideal time to replace it. The inclusion of this
Arregui et al. (2006) present extensive work regarding this issue, new approach in a system that automatically analyzes the data and
and one key observation is that the water company should study updates the information could be the solution sought by water com-
the client’s water consumption patterns (private domestic or indus- panies to mitigate the impact of apparent water losses.
trial). Additionally, Oviedo-Ocaña et al. (2020) also suggest that The proposed approach was successfully applied to eight-
customer meter renovation is an alternative to reduce apparent time series of billed water consumption from two locations in the
losses, mitigating inaccuracies associated with the aging of these Algarve, Portugal, representing three different consumption pat-
devices. terns. The data were collected by meters (Residential Households
So far, the usual mechanisms of analyzing the water consump- and the hotel) and smart meters (nonresidential households). A
tion are performed on a bigger scale, such as considering all the complement is the visual inspection of the results and their inter-
consumers/water meters of cities, as in Monedero et al. (2016) and pretation. This is also a valuable tool worth exploring to understand
Hester and Larson (2016). To the best of our knowledge, no similar and derive more insights into the consumption pattern. Moreover,
approach to the one proposed has been developed to examine water the procedure was able to identify points in time that coincide with
consumption trends for a single water meter. Its purpose is to con- the date of replacement of the meters. In some cases, they were
tribute to a better understanding of water consumption by providing detected earlier than the company’s decision on the water meter
useful guidance on the replacement of the meters. It will also pro- replacement. As so, an alert would have been useful for the com-
vide better analysis and a new perspective on water usage within pany to minimize revenue loss.
water management. In future work, it is planned a simulation study on water con-
The case study involved houses with different characteristics sumption to evaluate the approach’s performance based on quanti-
(residential and nonresidential, mainly for holidays), and a hotel, tative metrics. Furthermore, an association of the decreasing trend
from the Algarve. The Algarve is a popular summer holiday desti- with socio-economic aspects could explain the change in consum-
nation with a Mediterranean climate, meaning hot summers and ers’ behavioral patterns. The latter was not performed because it
mild winters. This seasonal effect is estimated by stl.fit() and re- was out of the paper scope. Future research on this topic should
moved. The strategy proceeds as follows: breakpoint analysis, non- be considered to refine and explain the changes in water usage. This
parametric trend analysis, and its magnitude, and a change of water strategy’s extension to identify increasing patterns is straightfor-
consumption indicator are calculated. The combination of the meth- ward, translating into a viable approach to manage NRW, alert con-
odologies was able to detect statistically significant decreasing serving water resources, and encourage responsible water use.
trends associated with points in time (i.e., breakpoints) where these
changes occur and in some cases even earlier than the company’s
decision on the water meter replacement. This is an added value of Data Availability Statement
the strategy. If the water meter is replaced too late, a significant rev-
enue loss (caused by meter under-registration) may occur. On the The data, models, and the code used that support this study’s find-
other hand, if detected after the actual replacement means that it ings are available in a repository online: https://github.com/Clara
was replaced too soon. Therefore, this approach could become a Cordeiro/StrategyWaterMeterPerformance.
useful tool to use by the water companies because it provides a bet-
ter understanding of water consumption patterns and makes a con-
scious decision regarding the water meter maintenance (to replace Reproducible Results
or not). The proposal for a unitless indicator of the magnitude of
change allows companies to analyze water meters from different Soraia Pereira (Universidade de Lisboa) downloaded all the ma-
infrastructures, with different water volumes such as a household terials, installed, ran the models using the data and functions in
or a hotel. The decision to replace the water meter is of the water “Workspace_data_functions.RData”, and reproduced the results in
company’s responsibility, which must weigh the cost-benefit of Tables 1 and 2.
Borges work has been supported by national funds through els.” Accessed June 1, 2021. http://pkg.robjhyndman.com/forecast.
FCT—Fundação para a Ciência e Tecnologia through project Hyndman, R. J., and G. Athanasopoulos. 2018. Forecasting: Principles
UIDB/04728/2020. M. Rosário Ramos was partially supported and practice. 2nd ed. Melbourne, Australia: OTexts.
Hyndman, R. J., and A. B. Koehler. 2006. “Another look at measures of
by National Funding from FCT—Fundação para a Ciência e Tec-
forecast accuracy.” Int. J. Forecasting 22 (4): 679–688. https://doi.org
nologia under the project UIDB/04561/2020. /10.1016/j.ijforecast.2006.03.001.
Kadenge, M. J., V. G. Masanja, and M. J. Mkandawile. 2020. “Optimisa-
tion of water loss management strategies: Multi-criteria decision analy-
References sis approaches.” J. Math. Inf. 18 (Feb): 105–119. https://doi.org/10
.22457/jmi.v18a9166.
Arregui, F., J. Soriano, J. Garcia-Serra, and R. Cobacho. 2013. “Proposal of Kandiah, V. K., E. Z. Berglund, and A. R. Binder. 2016. “Cellular automata
a systematic methodology to estimate apparent losses due to water meter modeling framework for urban water reuse planning and management.”
inaccuracies.” Water Sci. Technol. Water Supply 13 (5): 1324–1330. J. Water Resour. Plann. Manage. 142 (12): 04016054. https://doi.org
https://doi.org/10.2166/ws.2013.138. /10.1061/(ASCE)WR.1943-5452.0000696.
Arregui, F. J., E. Cabrera, R. Cobacho, and J. Garcia-Serra. 2006. “Reduc- Kendall, M. 1975. Rank correlation methods. Granville, OH: Charles
ing apparent losses caused by meters inaccuracies.” Water Pract. Tech- Griffin.
nol. 1 (4): wpt2006093. https://doi.org/10.2166/wpt.2006.093. Lafare, A. E., D. W. Peach, and A. G. Hughes. 2016. “Use of seasonal trend
Arregui, F. J., R. Cobacho, J. Soriano, and R. Jimenez-Redal. 2018. “Cal- decomposition to understand groundwater behaviour in the Permo-
culation proposal for the economic level of apparent losses (ELAL) in Triassic sandstone aquifer, Eden Valley, UK.” Hydrogeol. J. 24 (1):
a water supply system.” Water 10 (12): 1809. https://doi.org/10.3390 141–158. https://doi.org/10.1007/s10040-015-1309-3.
/w10121809. Mann, H. B. 1945. “Nonparametric tests against trend.” Econometrica J.
Awty-Carroll, K., P. Bunting, A. Hardy, and G. Bell. 2019. “An evaluation Econom. Soc. 13 (3): 245–259. https://doi.org/10.2307/1907187.
and comparison of four dense time series change detection methods Miller, C. 2019. “What’s in the box?! Towards explainable machine learn-
using simulated data.” Remote Sens. 11 (23): 2779. https://doi.org/10 ing applied to non-residential building smart meter classification.” En-
.3390/rs11232779. ergy Build. 199 (Sep): 523–536. https://doi.org/10.1016/j.enbuild.2019
Cleveland, R. B., W. S. Cleveland, J. E. McRae, and I. Terpenning. 1990. .07.019.
“A STL: A seasonal-trend decomposition procedure based on loess.” J. Miller, C., and F. Meggers. 2017. “Mining electrical meter data to predict
Off. Stat. 6 (1): 3–73. principal building use, performance class, and operations strategy for
Cordeiro, C. 2016. “stl.fit(): Function developed in cristina et al. (2016).” hundreds of non-residential buildings.” Energy Build. 156 (Dec):
360–373. https://doi.org/10.1016/j.enbuild.2017.09.056.
Accessed May 21, 2020. https://github.com/ClaraCordeiro/stl.fit.
Moahloli, A., A. Marnewick, and J. Pretorius. 2019. “Domestic water meter
Cordeiro, C., A. Borges, and M. R. Ramos. 2020. “Script, dataset and
optimal replacement period to minimize water revenue loss.” Water SA
functions used in a strategy to assess water meter performance.” Ac-
45 (2): 165–173. https://doi.org/10.4314/wsa.v45i2.02.
cessed December 23, 2020. https://github.com/ClaraCordeiro
Monedero, I., F. Biscarri, J. I. Guerrero, M. Peña, M. Roldán, and C. León.
/StrategyWaterMeterPerformance.
2016. “Detection of water meter under-registration using statistical al-
Criminisi, A., C. Fontanazza, G. Freni, and G. L. Loggia. 2009. “Evaluation
gorithms.” J. Water Resour. Plann. Manage. 142 (1): 04015036. https://
of the apparent losses caused by water meter under-registration in inter- doi.org/10.1061/(ASCE)WR.1943-5452.0000562.
mittent water supply.” Water Sci. Technol. 60 (9): 2373–2382. https://doi
Mutikanga, H. E., S. K. Sharma, and K. Vairavamoorthy. 2011. “Assess-
.org/10.2166/wst.2009.423. ment of apparent losses in urban water systems.” Water Environ. J.
Cristina, S., C. Cordeiro, S. Lavender, P. Costa Goela, J. Icely, and A. 25 (3): 327–335. https://doi.org/10.1111/j.1747-6593.2010.00225.x.
Newton. 2016. “Meris phytoplankton time series products from the SW Ncube, M., and A. Taigbenu. 2019. “Assessment of apparent losses due to
Iberian Peninsula (sagres) using seasonal-trend decomposition based on meter inaccuracy—A comparative approach.” Water SA 45 (2): 174–
loess.” Remote Sens. 8 (6): 449. https://doi.org/10.3390/rs8060449. 182. https://doi.org/10.4314/wsa.v45i2.03.
FEMS. 2021. “Àgua distribuìda/consumida por habitante. Onde se utiliza, Ohana-Levi, N., S. Munitz, A. Ben-Gal, A. Schwartz, A. Peeters, and Y.
em média, por pessoa, mais e menos água canalizada?” Accessed Feb- Netzer. 2020. “Multiseasonal grapevine water consumption—Drivers
ruary 17, 2020. https://www.pordata.pt/Municipios/%C3%81gua and forecasting.” Agric. For. Meteorol. 280 (Jan): 107796. https://doi
+distribu%C3%ADda+consumida+por+habitante-484. .org/10.1016/j.agrformet.2019.107796.
Fontanazza, C. M., V. Notaro, V. Puleo, and G. Freni. 2015. “The apparent Oviedo-Ocaña, E., I. Dominguez, J. Celis, L. Blanco, I. Cotes, S. Ward, and
losses due to metering errors: A proactive approach to predict losses and Z. Kapelan. 2020. “Water-loss management under data scarcity: Case
schedule maintenance.” Urban Water J. 12 (3): 229–239. https://doi.org study in a small municipality in a developing country.” J. Water Resour.
/10.1080/1573062X.2014.882363. Plann. Manage. 146 (3): 05020001. https://doi.org/10.1061/(ASCE)
Fourie, R., A. L. Marnewick, and N. Joseph. 2020. “An empirical analysis WR.1943-5452.0001162.
of residential meter degradation in Gauteng Province, South Africa.” Pacheco, V. M. F., R. E. Valdés, E. B. Gil, A. N. Manso, and E. Á. Álvarez.
Water SA 46 (4): 645–655. 2020. “Techno-economic analysis of residential water meters: A
caused by water meter inaccuracies at ultralow flows.” J. Am. Water Zeileis, A., F. Leisch, K. Hornik, and C. Kleiber. 2002. “Strucchange: An
Works Assoc. 102 (5): 123–132. https://doi.org/10.1002/j.1551-8833 R package for testing for structural change in linear regression models.”
.2010.tb10115.x. J. Stat. Software 7 (2): 1–38. https://doi.org/10.18637/jss.v007.i02.