Access to this full-text is provided by Springer Nature.
Content available from Scientific Reports
This content is subject to copyright. Terms and conditions apply.
Scientic Reports | (2020) 10:18867 |
www.nature.com/scientificreports
A Bayesian Belief Network
model to link sanitary inspection
data to drinking water quality
in a medium resource setting
in rural Indonesia
D. Daniel*, Widya Prihesti Iswarani, Saket Pande & Luuk Rietveld
Assessing water quality and identifying the potential source of contamination, by Sanitary inspections
(SI), are essential to improve household drinking water quality. However, no study link the water
quality at a point of use (POU), household level or point of collection (POC), and associated SI data in
a medium resource setting using a Bayesian Belief Network (BBN) model. We collected water samples
and applied an adapted SI at 328 POU and 265 related POC from a rural area in East Sumba, Indonesia.
Fecal contamination was detected in 24.4 and 17.7% of 1 ml POC and POU samples, respectively.
The BBN model showed that the eect of holistic—combined interventions to improve the water
quality were larger compared to individual intervention. The water quality at the POU was strongly
related to the water quality at the POC and the eect of household water treatment to improve the
water quality was more prominent in the context of better sanitation and hygiene conditions. In
addition, it was concluded that the inclusion of extra “external” variable (fullness level of water at
storage), besides the standard SI variables, could improve the model’s performance in predicting the
water quality at POU. Finally, the BBN approach proved to be able to illustrate the interdependencies
between variables and to simulate the eect of the individual and combination of variables on the
water quality.
Water quality has a prominent place in the Sustainable Development Goal 6.11, because it has been recognised
that unsafe drinking water is responsible for high numbers of diarrheal morbidity and mortality among chil-
dren below the age of ve1. Water quality analysis becomes important because supplied water, especially in low
and middle-income countries (LMICs), is oen contaminated, even though it is categorised as an improved
water source2. Groundwater, which is considered safer than surface waters, is also found contaminated in many
locations3. In Addition, high levels of contamination has been found at the household level in LMICs and water
quality oen deteriorates aer collection4–6.
To tackle this, the World Health Organization (WHO) and International Water Association (IWA) launched a
Water Safety Plan (WSP) concept, which is a comprehensive risk assessment and management covering all steps
in water supply from catchment to consumers7. e goal is to minimise the risk of contamination and provide
safe drinking water to people. Identifying potential sources of contamination is part of the risk assessment and
one of the critical elements in WSP.
In order to assess potential sources of contamination in a water supply system, systematic observation,
called sanitary inspections (SI), are performed. SI variables record potential sources of contamination based on
“on-site inspection and evaluation by qualied individuals of all conditions, devices, and practices in the water-
supply system that pose an actual or potential danger to the health and well-being of the consumer”8. SI have
the advantage to be easy to implement, not expensive, can be adapted to the local context, and can give a quick
snapshot of potential causes or pathways of contamination. However, SI are not a substitute for drinking water
quality testing, but identify contamination source in the system, especially in the context of risk management,
and can be used to design appropriate actions to change the situation9. erefore, it has been recommended to
accompany drinking water quality testing with SI10.
OPEN
Department of Water Management, Delft University of Technology, Delft, The Netherlands. *email: d.daniel@
tudelft.nl
Content courtesy of Springer Nature, terms of use apply. Rights reserved
Vol:.(1234567890)
Scientic Reports | (2020) 10:18867 |
www.nature.com/scientificreports/
Conducting drinking water quality testing in LMICs, however, can be challenging, especially because of
limited resources such as laboratory facilities or infrastructure11. Bain etal.12 summarised all available micro-
bial water quality tests for low and medium resource settings and they classied the resource settings into low,
medium, and high resource settings. A low resource setting has been characterised as having no laboratory equip-
ment and 24h electricity; the medium one having at least a basic laboratory or clean space with 24h electricity;
while the high resource setting is equipped with reliable 24h electricity and a modern laboratory. Researchers
are able to choose relevant water quality tests according to local context or situation.
Attempts have been made to link SI data to drinking water quality in order to be able to judge the reliability of
the system. e most common approach has been to analyse the SI and drinking water quality by using statistical
analyses, e.g., bivariate correlation or regression analyses, especially in high resource settings6,10,13–16.
Bayesian Belief Network (BBN) is another alternative to analyse factors responsible for the water quality17,18.
BBN oers benets compared to other statistical methods, such as the ability to integrate quantitative and qualita-
tive information in the model and an intuitive visualisation of the hypothetical causal relationships that can aid
stakeholders with less technical knowledge in understanding the system19.
However, the application of BBN in analysing water quality at the household level [mentioned as a point of
use (POU)] and at water source or point of collection (POC) is very limited. Hall and Le20 utilised BBN to predict
the faecal contamination of drinking water by household’s socio-economic characteristics as predictor variables,
however not using SI variables. To the authors’ knowledge , the present study is the rst to link drinking water
contamination at the POU with a combination of water quality at POC, the hygiene conditions in the household,
water handling, and household water treatment (HWT) practices in a medium resource setting. is study aims
to delineate the microbial water quality and general sanitary conditions in POC and POU in the district of East
Sumba, Indonesia.
Methods
Study setting. A cross-sectional study was conducted in July–August 2019 in the district of East Sumba,
Province East Nusa Tenggara, Indonesia (Fig.1). is study is the continuation of a previous household water
treatment study conducted in the same area21. A total of 328 households in nine villages in four sub-districts
were revisited during this study. is area is known as one of the poorest areas in Indonesia where open defeca-
tion is still common and there is high prevalence of children’s malnutrition22. e topography of the area is hilly.
Furthermore, about 40% the total populations in East Sumba relied on wells as their main water source and
only 18% had access to piped distribution system in 201723. No water treatment is conducted in the rural piped
distribution systems in this area.
Figure1. Map of the study location. ere were nine villages visited in four sub-districts. e map is drawn
using QGIS24.
Content courtesy of Springer Nature, terms of use apply. Rights reserved
Vol.:(0123456789)
Scientic Reports | (2020) 10:18867 |
www.nature.com/scientificreports/
Approximately 100ml of drinking water sample, i.e., from the drinking water storage container, was taken
at each household. e households were asked to give water in the same way as for drinking water. e water
samples were put in Nasco Whirl–Pak bags and kept inside a thermos during the transport to the eld lab. All the
samples were analysed within six hours aer collection. We only analysed the microbial water quality and used E.
coli as an indicator bacteria for fecal contamination in water25. We took 1ml of sample using a 1ml sterile pipette
and placed it on a Nissui Compact dry EC plate (CDP) and incubated for 24h at 35 ± 2°C26. Aer incubation, we
counted the colony forming units (CFU) of E. coli in the CDP and reported in concentration units (CFU/1ml).
e process was conducted as sterile as possible to prevent contamination from sample processing, e.g., using
hand gloves and sterile pipette tips when processing the sample, avoid touching the inside of the whirl-pack
bag when collecting and processing the sample, and working in a stable and clean space. e sample processing
was conducted by two master students from Del University of Technology who were familiar with microbial
water quality analyses. According to the classication of Bain etal.12, our analysis was categorised as medium
resource setting, e.g., there was neither distilled water and proper disinfection for laboratory equipment. Data
were collected during the dry season with temperature in that area ranging from 25 to 26°C.
For the SI, we used the Open Data Kit (ODK) soware on a smartphone, and the data were transferred to
a computer for analysis. We did SI at POCs and POUs. Information taken at a POC and POU can be found in
Table1. Participation was voluntary and a written informed consent was obtained from all participants. e study
was approved by the Human Research Ethic Committee of Del University of Technology and the Agency for
Promotion, Investment, and One-Stop Licensing Service at the district level. All experiments were conducted
in accordance with relevant guidelines and regulations.
Bayesian Belief Network (BBN). BBN is a directed acyclic graph showing a hypothetical causal relation-
ship between “causal” variables (where the arrow start; called “parent nodes” in BBN) and an “aected” variable
(called “child node”)27. e strength of the relationship between parent and child node is shown by the values
in the Conditional Probability Tables (CPT) of the child node. e CPT values are showing the probability of a
child node will be in a particular state or category, given all possible combination of the states of its parent nodes.
e CPT values can be obtained from expert or stakeholder judgment or elicitation, the output of other models
or calculations, or by direct measurement. Cain19 provides a good and clear explanation of using a BBN in the
water sector.
Data analysis. A BBN’s structure is oen inspired by a conceptual theory or framework or by consensus
between experts in that eld28. ere are some conceptual frameworks from previous water, sanitation, and
hygiene (WASH) studies that can be adapted into a BBN’s structure29,30, including the well-known F-diagram 31.
According to those frameworks, there are four main clusters of determinants of water quality at POU: (1) Sur-
rounding environment–hygiene condition, (2) HWT, (3) (the water quality at) POC, and (4) the water storage
conditions (see Fig.2). All variables for these four cluster are oen included in a standard SI form8.
However, Navab-Daneshmand etal.29 argues that fecal contamination at the household level in LMICs is
complex. is implies that there might be other variables, besides SI variables, that could correlate with the
household drinking water quality, such as container material, duration of storing water, inappropriate extrac-
tion water from storage, etc32–34. However, all these “external” factors are not included in the standard SI form8.
Based on the above mentioned literature, we created a conceptual model of potential factors that could inu-
ence the water quality at the household level (Fig.2). e conceptual model includes multiple contamination
Table 1. Information used for the analysis. a e sentence inside the [ ] were the questions in the sanitary
inspection and the italic words were the variable/node name in the BBN. b Based on water quality testing.
c External variable besides standard SI variables.
Point of collection (POC)bSurrounding environment–hygiene condition Water storage condition and HWT
Type of POC [Which source do you use for drinking water
purpose right now?]bStill practise open defecation [What types of toilet do you
have?] Storage covered [Is the water storage being covered (at that
time)?]
Livestock nearby [Is there livestock near the point of collec-
tion (POC), 10m?] Livestock nearby [Is there livestock around the house?] Storage cracked [Is the container cracked?]a
Prone to erosion [Is the area uphill from the source visibly
eroded or prone to erosion?] Floor cleanliness [How is the cleanliness of the house
oor?]a
Place of storage [When not in use, is the storage container
kept in a place where it may become contaminated? E.g.,
can be reached by animal easily; open space (risk by ies),
etc.]
Excreta / garbage nearby [Is excreta or garbage found within
10m of the tap stand/water source?] Faeces around [Is there human or animal faeces in the yard
(or even inside the house)?]afullness level of water at storage [How full is the water
storage?]a,c
Proper fencing [Is there proper fencing or a barrier around
the well to prevent contact with animals?] Garbage around [Is there garbage around the house?] Household water treatment [Is the water in the storage
treated?]
Latrine within 10m [Distance to the nearest latrine (m)] Flies around [Could you see ies around the water storage
container?]
Cracked structure [Are there any damages/cracks in the
system/source?]
E. coli detected at POC/wellb
Content courtesy of Springer Nature, terms of use apply. Rights reserved
Vol:.(1234567890)
Scientic Reports | (2020) 10:18867 |
www.nature.com/scientificreports/
pathways in a system35 and was used to create the BBN’s structure by clustering SI variables based on those ve
clusters.
Because some houses used the same POC, we could make pairs of 271 POCs–POUs (Fig.3). 49 POU did not
have POC samples, i.e., POC samples were not taken, mostly due to long distance walk (> 30min return trip).
However, these 49 POU samples were included in the BBN analysis, since the EM algorithm compensated for
the missing information with the available data36.
Four BBN models of the water quality at the POU were created (Fig.3). BBN model 1 (A and B) and 2 (A and
B) dier in terms of the variables used in the cluster of POC. For BBN model 1 we added node Type of POC as
a parent node for E. coli detected at POC (Figs.4, 5). But for BBN model 2 we used information of the SI at the
POC as parent nodes of E. coli detected at POC, but we modelled only one type of POC: well (Figs.6, 7). at
is because the SI information that we collected at POC were only relevant to the well’s characteristics. For BNN
model 1, we had in total of 328 samples and for BNN model 2 was only 89 well samples (Fig.3).
In addition, we added one extra variable, fullness level of water at storage, on top of both models and com-
pared the model’s performance, i.e., BBN model 1A vs 1B and model 2A vs 2B. is variable could indicate the
duration of storing water, because water quality could deteriorate over time4. us, BBN model 1A and 2A were
Figure2. e conceptual model of ve clusters of the determinants of water quality at a point of use (POU).
Red arrows indicate that the variables are oen included in a standard SI form and white arrow is not included
in the standard SI form.
Figure3. Overview of the datasets and analysis.
Content courtesy of Springer Nature, terms of use apply. Rights reserved
Vol.:(0123456789)
Scientic Reports | (2020) 10:18867 |
www.nature.com/scientificreports/
the BBN models with SI variables only and BBN model 1B and 2B were the BBN models with SI variables plus
variable fullness level of water at storage. e results of validation tests, i.e., AUC value, indicated the model’s
performance. e predictive inference tests were then conducted using BBN models with the best performance.
Moreover, Since it is not recommended to have many parent nodes in BBN19, we needed to reduce the BBN
structure as much as possible. Clustering the SI variables reduces the parent nodes of the outcome node, e.g. water
quality at the POC. All variables in the SI for POC were grouped as one cluster and the variables in the SI related
to water storage were grouped as another cluster. In the latter case, e.g., three variables related to the condition
of the water storage, Storage covered, Storage cracked, and Place of storage, were connected to an intermediate
node Chance of (re)contamination from water storage (red node in Fig.4).
Figure4. BBN model 1A (type of POC as a parent node of “E. coli detected at POC”). Blue nodes: data
obtained from SI; green nodes: data obtained from water quality testing; red nodes: intermediate nodes were
obtained by summation of the value in the outer nodes. e percentages in each node indicate the probability of
a node being in a certain state, e.g., 56% of the households perform household water treatment.
Figure5. BBN model 1B (type of POC as a parent node of “E. coli detected at POC” and adding node “fullness
of water at storage” as one of the parent nodes of “E. coli detected at POC”).
Content courtesy of Springer Nature, terms of use apply. Rights reserved
Vol:.(1234567890)
Scientic Reports | (2020) 10:18867 |
www.nature.com/scientificreports/
Since we did not have the information on intermediate nodes in our datasets, the CPT corresponding to this
node was populated manually. First, we gave score 1 to the best situation in each variable, e.g., score 1 if “yes” in
variable storage covered and score 1 if “no” in variable storage cracked. en we created a simple index by summing
all the scores of the three parent nodes. Finally, we categorised it as “low” if the total score was 0–1, “moderate”
if the total score was 2, and “high” if the total score was 3. In the same way, another intermediate node Chance
of (re)contamination from environment was created by six variables (six parent nodes of this variable, see Fig.4).
We categorised it as “low” if the total score was 0–2, “moderate” if the total score was 3–4, and “high” if the total
score was 5–6. Dierent from the other intermediate nodes, we used the results of water quality testing to ll
the information of node E. coli detected at POC (see Fig.4; green nodes). BBN requires discrete or categorical
information for the analysis. erefore, we discretised and categorised the number of E. coli into E. coli detected
or non-detected.
We used soware GeNIe 2.2 (https ://www.bayes fusio n.com) to perform the BBN analysis. e soware
uses the Expectation Maximization (EM) algorithm to estimate the CPT values36. We performed validation
Figure6. BBN model 2A (SI variables at well as parent nodes of “E. coli detected at POC”).
Figure7. BBN model 2B (SI variables at well as parent nodes of “E. coli detected at POC” and adding node
“fullness of water at storage” as one of the parent nodes of “E. coli detected at POC”).
Content courtesy of Springer Nature, terms of use apply. Rights reserved
Vol.:(0123456789)
Scientic Reports | (2020) 10:18867 |
www.nature.com/scientificreports/
tests using the same soware to assess the model’s performance. We used the ten-fold cross-validation and the
performance was reected by the value of area under the ROC curve (AUC): AUC of 0.5 indicates poor model,
AUC between 0.5 and 0.7 is a “less accurate” model, 0.7 < AUC ≤ 0.9 is a “moderately accurate”, 0.9 < AUC < 1 is
a “highly accurate” model, and AUC = 1 is a perfect model37.
We also conducted a “predictive inference” in BBN, to nd inuential nodes that help us to prioritise actions
to improve the water quality of POU in that area. We performed that by setting the state of a specic node to
100% and observe the updated probability in the output node. For example, if we wanted to observe the inuence
of HWT on POU’s water quality, we set the probability of node Household water treatment being “yes_treat” to
100% and observed the updated probability of E. coli detected at POU being “detected”. We did that to all states
in all nodes.
Finally, we simulated the “best scenario”, i.e., targeting all SI variables or potential source of contaminations in
the system, by setting the best situation of all SI variables (outer nodes) at all clusters, including node Household
water treatment being “yes_treat” and node E. coli detected at POC being “not_detected”. By setting node E. coli
detected at POC being “not_detected”, we assumed that all types of water source that household use are safe.
Results
Socio‑demographic characteristics of the respondents. When asked about the education of the
household’s head, 12.5% of them had no formal education, and 57.3%, 11.9%, and 18.3% nished primary, sec-
ondary, and higher school, respectively. In terms of housing condition, 87.6% did not have permanent walls, e.g.,
wood or bamboo, 7.5% did not have a permanent roof, i.e., straw, and 71.4% still had a natural oor, i.e., com-
pacted soil. Moreover, 45.3% of the respondents had no electricity. About 32.7% of the respondents practised
open defecation. Based on observations, households either had simple pit latrines or pour-ush latrines, some
were communal and some were in respective households. Tap water (from a small-scale distribution network)
was used by 31.8% of the respondents, followed by wells 27.2%, water trucks 19.6%, and spring water 17.4%,
respectively. Remaining respondents used river water, rainwater, or rell potable water stations. Boiling was used
to treat the drinking water.
Description of the sanitary inspection and water quality results. e general hygiene situation of
the respondents is depicted in the BBN model, i.e. the outer nodes in Fig.4 (in blue colour). For example, 23%
of the respondents did not cover their drinking storage and only 30% of the respondent’s houses were free from
ies. From the cluster of surrounding environment–hygiene condition, we found that 66.7% of the respondents
kept their livestock near the house, resulting in 60% of the respondents had animal faeces around the house. In
addition, 89% and 70% of the respondents had garbage and ies around the water storage or house, respectively.
ese conditions led to only 15% respondents had low chance of contamination from the surrounding environ-
ment and hygiene condition.
e general condition of the cluster water storage condition indicated that 37% of the respondents had a low
chance of contamination from “bad condition of water storage”, i.e., comply to all three criteria: storage with
cover, without cracking, and proper-safe place. About 77% and 96% of the storages were found to be covered
and without cracking, but 51% of the storages were put in a place that can be prone to (re)contamination, e.g.
on the oor.
Of all the POU samples, 56.5% of the respondents claimed to treat water at the time of visit. 75% of households
who abstracted water from river treated their drinking water, followed by 68.5% and 59.4% from households
who used well and piped system, respectively.
Of all the POU samples, 56.3% of our respondents claimed to treat water at the time of the visit. For the water
quality, we did not detect E. coli in the 1ml samples in 195 (75.6%) of the POC samples and 270 (82.3%) of the
POU samples. E. coli was not detected in almost 90% of the piped and spring samples, while 42% and 83% of
well and river samples, respectively, were detected with E. coli.
Comparison of the BBN models’ performance. e four BBN models are shown in Figs.4, 5, 6 and
7. We rst compared the performance of BBN models with SI variables only and SI variables plus extra variable
fullness level of water at storage. e validation tests of these four BBN models gave AUC value: 0.55, 0.69, 0.71,
and 0.84 for model 1A (Fig.4), 1B (Fig.5), 2A (Fig.6), and 2B (Fig.7), respectively. According to the classica-
tion of Greiner etal.37, model 1A and 1B were classied as “less accurate” and model 2A and 2B as “moderately
accurate”.
e addition of variable fullness level of water at storage, which is not part of “standard” SI variables, improved
the model’s performance. erefore, we decided to use BBN model 1B (Fig.5) and 2B (Fig.7) for further BBN
analyses, because model 1 and 2 dier in structure (Fig.3).
Predictive inference of the BBN models. Node E. coli detected at POC was the most inuential node
(see ∆P = 21 in Table2—le) for the model 1B (type of POC as one of the outer nodes), i.e., the better the water
quality at POC, the better the water quality at the household level or POU. Node Type of POC and Fullness level of
water at storage appeared as the second most inuential nodes (∆P = 17 in Table2—le). e intermediate node
Chance of (re)contamination from the water storage was the third most inuential node (∆P = 10 in Table2—le).
e probability of not detected E. coli at POU was 75% for households who used both Piped and Spring,
considering other information in the BBN model. e fuller the level of water in the storage, the better the water
quality at POU was: the probability of E. coli contamination at POU was 58% for Almost empty compared to 74%
for Full. Among all three outer nodes in the cluster (re)contamination from water storage, node storage covered
(∆P = 5 in Table2–le) was the most inuential node.
Content courtesy of Springer Nature, terms of use apply. Rights reserved
Vol:.(1234567890)
Scientic Reports | (2020) 10:18867 |
www.nature.com/scientificreports/
BBN model 1B: with type of POC as one of the outer nodes BBN mode 2B: with SI at well as one of the outer nodes
Var iable Probability of E. coli not-detected at POU (%) ∆PaVariabl e Probability of E. coli not-detected at POU (%) ∆P
Point of collecti on Point of collecti on
Type of
POC
Piped Well Spring River Other 17 Cracked
struc-
ture
Yes No 2
75 69 75 58 72 69 71
E. coli
detected
at POC
Yes No 21 Live-
stock
nearby
Yes No 1
56 77 70 71
Household water treatment Proper
fencing
Yes No 0
House-
hold
Water
treat-
ment
No Yes
6
70 70
69 75 Excreta/
garbage
nearby
Yes No 0
(re)contamination from environment–hygiene condition 70 70
Still
practise
open
defeca-
tion
Yes No
2Prone to
erosion
Yes No
1
71 73 71 70
Live-
stock
nearby
Yes No 1Latrine
within
10 m
Yes No 0
72 73 70 70
Floor
cleanli-
ness
Dirty Clean 1E. coli
detected
at POC
Yes No 19
72 71 59 78
Faeces
around
Yes No 1Household water treatment
72 73 House-
hold
water
treat-
ment
No Yes
13
Garbage
around
Yes No 060 73
72 72 (re)contamination from environment–hygiene condition
Flies
around
Yes No
0
Still
practise
open
defeca-
tion
Yes No
4
72 72 67 71
Chance
of con-
tamina-
tion
from the
environ-
ment
High Moderate Low
7Live-
stock
nearby
Yes No
5
68 75 70 68 73
(re)contamination from water storage Floor
cleanli-
ness
Yes No 2
Storage
covered
Yes No 570 72
74 69 Faeces
around
Yes No 5
Storage
cracked
Yes No
4
67 72
69 73 Garbage
around Yes No 1
Place of
storage
Easy to contaminated Not easy to contaminated 370 71
71 74 Flies
around
Yes No 1
Chance
of con-
tamina-
tion
from
water
storage
High Moderate Low
10
70 71
64 74 74
Chance
of con-
tamina-
tion
from the
environ-
ment
High Moder-
ate Low 22
Fullness level of water at storage 57 67 79
Continued
Content courtesy of Springer Nature, terms of use apply. Rights reserved
Vol.:(0123456789)
Scientic Reports | (2020) 10:18867 |
www.nature.com/scientificreports/
e households who claimed to do HWT have a higher chance of not to be contaminated by E. coli than
households who claimed not doing HWT, i.e., PNot_detected = 75%, PNot_detected = 69%, respectively.
In model 2B, intermediate node Chance of (re)contamination from the environment was the most inuential
node among households who used a well as their water source (∆P = 22 in Table2—right). Node E. coli detected
at POC was the second most inuential nodes (∆P = 19 in Table2—right), followed by node Household water
treatment (∆P = 13 in Table2—right). In addition, the inuence of node Fullness level of water at storage and
the intermediate node Chance of (re)contamination from the water storage was not large, compared to model 1B
(both had ∆P = 4 in Table2—right).
e eect of HWT to improve the water quality was larger in model 2B (∆P = 13 in Table2—right), compared
to model 1B (all types of POC; ∆P = 6 in Table2—le). If we compare the situation of intermediate nodes Chance
of (re)contamination from the environment and Chance of (re)contamination from the environment in model 1B
(Fig.5) and 2B (Fig.7), the hygiene situation was better in model 2B. e probability of being “high” in both
intermediate nodes in model 2B was lower than in model 1B, e.g., 24% in model 1B compared to 13% in model
2B for the intermediate node Chance of (re)contamination from the environment.
Furthermore, keeping the house free from livestock (PNot_detected = 73%) and faeces (PNot_detected = 72%) seemed
important to reduce the probability of fecal contamination at the household storage among households who
used a well as their water source. Respondents who practiced open defecation had a larger probability of fecal
contamination at the POU than they who did not, i.e., PNot_detected = 67%, PNot_detected = 71%, respectively (∆P = 4).
e inuence of HWT to reduce the chance of contamination was prominent in model 2B, i.e., PNot_detected = 73%
for households who treated their drinking water and PNot_detected = 69% for not treating water.
e ∆P of intermediate nodes in both model 1B and 2B were bigger than their outer (parent) nodes. For
example, in model 2B, the ∆P of 6 outer nodes in the cluster of surrounding environment–hygiene condition had
less variation (range ∆P = 1–5) compared to the intermediate node Chance of (re)contamination from the environ-
ment (∆P = 22), whereas the intermediate nodes were the sum of the values in outer nodes.
For simulating the best scenario, i.e., combination of variables, model 2B was used to simulate all respond-
ents (Fig.8). e updated probability of outcome node E. coli detected at POU being “not_detected” was 91%,
compared to the 70% in the baseline situation (Fig.7). Given the same scenario in model 1B, the updated prob-
ability of the outcome node was 92%, compared to the 72% in the baseline (Fig.5), which suggests the same
pattern as model 2B.
BBN model 1B: with type of POC as one of the outer nodes BBN mode 2B: with SI at well as one of the outer nodes
Var iable Probability of E. coli not-detected at POU (%) ∆PaVariabl e Probability of E. coli not-detected at POU (%) ∆P
Fullness
level of
water at
storage
Almost
empty One quarter Half ree quarter Full 17 (re)contamination from water storage
58 64 70 75 74 Storage
covered
Yes No 0
70 70
Storage
cracked
Yes No 0
70 70
Place of
storage
Easy to contami-
nated Not easy to contaminated 2
71 69
Chance
of con-
tamina-
tion
from
water
storage
High Moder-
ate Low
4
68 72 68
Fullness level of water at storage
Fullness
level of
water at
storage
Almost
empty One
quarter Half ree
quarter Full 4
70 73 69 70 71
Table 2. Predictive inference, measuring the eect of changes in the states of each node on the output node of
BNN models: E. coli detected at POU (drinking water storage). e value under each category corresponding
to a node as displayed in the rst column is the updated probability of the output node being “Not_detected”
given that all households maintain this state. e le side of the table was for the BBN model 1A (Fig.5) and
the right side was for BBN model 2B (Fig.7). a ∆P is the dierence between the lowest and highest value of the
updated probability of output node: E. coli detected at POU being “Not_detected”, in %. Examples of how to
read the table: (a) row 4–5 BBN model 1B: if the type of POC is piped, the Probability of E. coli not-detected
at POU (%) is 75%; (b) row 6–7 BBN model 1B: if E. coli is detected at POC (“yes”), the Probability of E. coli
not-detected at POU (%) is 56%; (c) row 4–5 BBN model 2B: if there is a cracked in the structure (“yes”), the
Probability of E. coli not-detected at POU (%) is 69%.
Content courtesy of Springer Nature, terms of use apply. Rights reserved
Vol:.(1234567890)
Scientic Reports | (2020) 10:18867 |
www.nature.com/scientificreports/
Discussion
BBN model’s performance. Since there is no BBN study which links SI and water quality data, we com-
pared our models’ performance with statistical analysis. Snoad etal.13 utilized logistic regression to predict the
fecal contamination by SI and their AUC values were low (range 0.41–0.64). Other authors also used multiple
statistical analyses and found that SI variables could not explain well the water quality10,14,16, which imply that
our models (with AUC values of 0.69 and 0.84) were slightly better in predicting the water quality at POU, using
SI data.
However, we found that an “external” factor, besides standard SI variables, increased the model’s performance,
in our case we used the level of water fullness inside the storage, as also found to be relevant in other studies32–34,
suggesting the need to extend the standard SI with external factors for better model performance. In addition,
BBN models with SI variables at well (AUC for model 2A and 2B are 0.71 and 0.84, respectively) perform bet-
ter than BBN models with dierent types of POC (AUC for model 1A and 1B are 0.55 and 0.69, respectively).
Since the same type of POC, e.g., well, can have varying conditions, detailed information of the POC condi-
tions can better explain the water quality than the information on the type of POC itself. is may explain why
BBN models with SI variables as explanatory variables perform better than BBN models with types of POCs as
explanatory variables.
Sanitary inspection, water quality, and BBN predictive inferences. To the authors’ knowledge,
this is the rst study that links SI data with water quality in a medium resource setting. e BBN approach
allowed the inclusion of all factors inuencing the water quality at POU and grouping them in relevant clusters
and pathways, as implied by other conceptual frameworks29–31. Furthermore, we were able the analyse the water
quality at POU by considering not only the water management and hygiene situation at home, but also the
broader scope, such as the situation at the water source. Moreover, the conventional statistical analysis methods,
e.g., bivariate correlation or regression analyses, oen quantify the eect of the individual variable on water qual-
ity, but not a combination of variables or pathways6,10,16. e BBN approach was able to simulate both the eects
in one model and can then help to prioritise the interventions that improve the water quality at household level,
i.e., either targeting one variable or combination of multiple variables.
e BBN approach also enabled the portrayal of interdependencies vividly among variables, while this inter-
dependency have attracted the attention of WASH practitioners and experts over the past years35. For example,
SI results revealed that there were some hygiene challenges related to livestock ownership. e majority of the
respondents (67%) kept livestock in the surroundings of the house, which could be the reason why many ies
(70%) and faeces (60%) were detected in our respondents’ houses (see Fig.5 cluster (re)contamination from
environment–hygiene condition). A study of Ercumen etal.38 found that the presence of animals is related to
fecal contamination, and the presence of animal faeces is associated with diarrhea and stunting39. is could be
the reason why this area was reported as one of the locations with the highest stunting levels in Indonesia40. To
tackle these conditions is challenging, since in East Sumba livestock is a symbol of social status41.
Our BBN models (1B and 2B) showed that the water quality at POCs critically aected the water quality at
the POU in the study area, which has also been found by others6,42. We also found that types of water source
used by the households determine the drinking water quality that they have at home, similar to the ndings in
rural Honduras43. ese data suggest that the fecal contamination at POU due to poor water quality at the water
Figure8. e best scenario of water and hygiene management at households level using BBN model 2B (SI at
well as one of the outer nodes, SI variables, and fullness of water at storage).
Content courtesy of Springer Nature, terms of use apply. Rights reserved
Vol.:(0123456789)
Scientic Reports | (2020) 10:18867 |
www.nature.com/scientificreports/
source, especially wells, is a serious problem in East Sumba, i.e., 40% the total populations in East Sumba used
well as their main water source23.
Since we found that the eect of HWT to improve the water quality was larger in model 2B (POC = well
only) compared to model 1B (all types of POC), we argue that the eect of HWT to improve the water quality
is prominent in the case of better sanitation and hygiene conditions, i.e., the overall condition in model 2B was
“more hygienic” than in model 1B. is result has also been suggested by a previous study44.
Model 1B showed that storage with full water had a better water quality than (almost) empty storage. e
explanation could be that the water inside the empty storage was stored for a longer period than a fuller storage,
resulting in larger risks for recontamination4 and permitting bacteria regrowth45.
Furthermore, we found that the ∆P (the dierence between the lowest and highest value of the updated prob-
ability of output node: E. coli detected at POU being “Not_detected” given the specic condition of a specic
node) of intermediate nodes are larger than the inuence of their outer (parent) nodes. is implies that collective
information of the specic cluster was more meaningful, i.e., more sensitive, to predict the water quality than
individual information of specic node or variable. Additionally, it suggests that our simple index, by summing
the scores of the parent nodes to populate the CPT in some intermediate nodes, was “acceptable”, i.e. simplifying
the BBN structure and the intermediate nodes were related to the output node.
A previous WASH study found that a combined HWT, sanitation, handwashing, and house’s cleanliness
intervention have the same eect as with HWT intervention alone in reducing fecal contamination in house-
hold drinking water46. In contrast to their study, we found that a combined improvement, targeting all potential
contamination sources from the water source until house, had a larger eect in reducing the chance of fecal
contamination in the water storage rather than the improvement of one single condition. is suggests that a
holistic approach or multi-barrier prevention are needed to minimise drinking water contamination at the POU
in rural households7,47. However, considering the costs and time constraint, based on the results on impact of
water quality at POU, it can be suggested to prioritize the improvement of the water quality at the water source,
based on e.g. BNN modelling. Aerwards, WASH behavioural change promotion, e.g., promoting the correct
and sustained use of HWT and safe storage container, could be conducted.
Future water quality studies in that area should analyze and include other external factors that may inuence
the water quality at POC and POU, e.g., type and depth of the well and the types of water containers used by
households. is can improve our understanding of water quality in this area.
Conclusion
is paper introduces an application of BBN to analyse how water quality at the point of use is related to the
water quality at the point of collection and associated sanitary inspection data in the medium resource settings
in low-middle income countries. e model simulations showed that holistic—combined interventions improved
the water quality considerably compared to individual interventions. Moreover, the results demonstrate that
water quality at the POC was, as expected, related to the water quality at the POU and (correct and regular)
household water treatment had a larger eect of improving the storage water quality in the case of better sanita-
tion and hygiene conditions. We also found that the BBN model performance increased by adding an external
variable besides standard SI variables, suggesting that the current SI form should accommodate more (relevant)
variables. Additionally, E. coli was detected in 24.4 and 17.7% of POC and POU samples, respectively, and there
was a hygiene issue related to the ownership and presence of livestock surround the house. Based on the water
quality analysis, tap and spring water are relatively cleaner than other types of water sources and, therefore,
should be prioritised by the households as main drinking water sources. In order to improve the drinking water
quality in this area, reducing the contamination risk at the water source and promoting correct and regular
household water treatment are suggested. From the study it can nally be concluded that the BBN approach
could be considered as an alternative for conventional statistics to link sanitary inspection and water quality
data in low-middle income countries.
Received: 22 May 2020; Accepted: 16 October 2020
References
1. Prüss-Ustün, A. et al. Burden of disease from inadequate water, sanitation and hygiene in low- and middle-income settings: A
retrospective analysis of data from 145 countries. Trop. Med. Int. Heal. 19, 894–905 (2014).
2. Bain, R. et al. Fecal contamination of drinking-water in low-and middle-income countries: A systematic review and meta-analysis.
PLoS Med. 11, e1001644 (2014).
3. Podgorski, J. & Berg, M. Global threat of arsenic in groundwater. Science (80-). 368, 845–850 (2020).
4. Levy, K., Nelson, K. L., Hubbard, A. & Eisenberg, J. N. Following the water: A controlled study of drinking water storage in northern
coastal Ecuador. Environ. Heal. Perspect 116, 1533–1540 (2008).
5. Wright, J., Gundry, S. & Conroy, R. Household drinking water in developing countries: A systematic review of microbiological
contamination between source and point-of-use. Trop. Med. Int. Heal. 9, 106–117 (2004).
6. Daniel, D., Diener, A., van de Vossenberg, J., Bhatta, M. & Marks, S. J. Assessing drinking water quality at the point of collection
and within household storage containers in the hilly rural areas of mid and far-western Nepal. Int. J. Environ. Res. Public Health
17, 2172 (2020).
7. WHO. Water Safety Planning for Small Community Water Supplies: Step-by-step Risk Management Guidance for Drinking-Water
Supplies in Small Communities (WHO, 2012).
8. WHO. Surveillance and Control of Community Supplies, Guidelines for Drinking-Water Quality Vol. 3 (WHO, New York, 1997).
9. Howard, G. et al. Identication and management of microbial contaminations in a surface drinking water source. J. Water Health
5, 67–79 (2007).
10. Misati, A. G., Ogendi, G., Peletz, R., Khush, R. & Kumpel, E. Can sanitary surveys replace water quality testing? Evidence from
Kisii, Kenya. Int. J. Environ. Res. Public Health 14, 152–164 (2017).
Content courtesy of Springer Nature, terms of use apply. Rights reserved
Vol:.(1234567890)
Scientic Reports | (2020) 10:18867 |
www.nature.com/scientificreports/
11. Diener, A. et al. Adaptable drinking-water laboratory unit for decentralised testing in remote and alpine regions. in 1–7 (40th
WEDC International Conference, 2017).
12. Bain, R. et al. A summary catalogue of microbial drinking water tests for low and medium resource settings. Int. J. Environ. Res.
Public Health 9, 1609–1625 (2012).
13. Snoad, C., Nagel, C., Bhattacharya, A. & omas, E. e eectiveness of sanitary inspections as a risk assessment tool for ther-
motolerant coliform bacteria contamination of rural drinking water: A review of data from West Bengal, India. Am. J. Trop. Med.
Hyg. 96, 976–983 (2017).
14. Robinson, D. T. et al. Assessing the impact of a risk-based intervention on piped water quality in rural communities: e case of
mid-western Nepal. Int. J. Environ. Res. Public Health 15, 1616–1639 (2018).
15. Dey, N. C. et al. Microbial contamination of drinking water from risky tubewells situated in dierent hydrological regions of
Bangladesh. Int. J. Hyg. Environ. Health 220, 621–636 (2017).
16. Ercumen, A. et al. Can sanitary inspection surveys predict risk of microbiological contamination of groundwater sources? Evidence
from shallow tubewells in rural Bangladesh. Am. J. Trop. Med. Hyg. 96, 561–568 (2017).
17. Tang, C., Yi, Y., Yang, Z. & Sun, J. Risk analysis of emergent water pollution accidents based on a Bayesian Network. J. Environ.
Manage. 165, 199–205 (2016).
18. B ertone, E., Sahin, O., Richards, R. & Roiko, A. Extreme events, water quality and health: A participatory Bayesian risk assessment
tool for managers of reservoirs. J. Clean. Prod. 135, 657–667 (2016).
19. Cain, J. Planning Improvements in Natural Resources Management Vol. 44 (UK Centre for Ecology & Hydrology, Wallingford, 2001).
20. Hall, D. C. & Le, Q. B. Use of Bayesian networks in predicting contamination of drinking water with E. coli in rural Viet nam. Trans.
R. Soc. Trop. Med. Hyg. 111, 270–277 (2017).
21. Daniel, D., Pande, S. & Rietveld, L. e eect of socio-economic characteristics on the use of household water treatment via psy-
chosocial factors: A mediation analysis. Hydrol. Sci. J. https ://doi.org/10.1080/02626 667.2020.18075 53 (2020).
22. Sungkar, S. et al. Heavy burden of intestinal parasite infections in Kalena Rongo village, a rural area in South West Sumba, eastern
part of Indonesia: A cross sectional study. BMC Public Health 15, 1–6 (2015).
23. BPS Statistics of East Sumba Regency. Persentase Rumah Tangga menurut Sumber Air Utama yang Digunakan Untuk Minum
di Kabupaten Sumba Timur, 2015–2017. Statistics of Sumba Timur Regency. https ://sumba timur kab.bps.go.id/dynam ictab
le/2018/11/12/50/perse ntase -rumah -tangg a-menur ut-sumbe r-air-utama -yang-digun akan-untuk -minum -di-kabup aten-sumba
-timur -2015-2017.html (2018).
24. QGIS Development Team. QGIS Geographic Information System ver. 2.18.4.https ://downl oad.qgis.org (2017).
25. WHO. Guidelines for D rinking-Wate r Quality: Fourth Edition Incorp orating e First Addendum Vol. 1 (World Health Organization,
New York, 2017).
26. Nissui Pharmaceutical Co. Ltd. CompactDry “Nissui” EC Illustration Manual. https ://www .ni s su i-pharm .co.jp/engli sh/pdf/produ
cts/globa l/illus trati on-manua l_/Compa ctDry Nissu iEC-Illus trati onMan ual.pdf (n.d.).
27. Pearl, J. Probabilistic reasoning in intelligent systems: networks of plausible inference (Morgan Kaufmann Publishers Inc., San Mateo,
1988).
28. Nadkarni, S. & Shenoy, P. P. A causal mapping approach to constructing Bayesian networks. Decis. Support Syst. 38, 259–281 (2004).
29. Navab-Daneshmand, T. et al. Escherichia coli contamination across multiple environmental compartments (soil, hands, drinking
water, and handwashing water) in urban Harare: Correlations and risk factors. Am. J. Trop. Med. Hyg. 98, 803–813 (2018).
30. Cohen, A. et al. Microbiological evaluation of household drinking water treatment in rural China shows benets of electric kettles:
A cross-sectional study. PLoS ONE 10, 1–16 (2015).
31. Wagner, E. G., Lanoix, J. N. & World Health Organization. Excreta Disposal for Rural Areas and Small Communities. Monograph
Series Vol. 39 (World Health Organization, New York, 1958).
32. B oateng, D., Tia-Adjei, M. & Adams, E. A. Determinants of household water quality in the Tamale Metropolis, Ghana. J. Environ.
Earth Sci. 3, 70–77 (2013).
33. Elala, D., Labhasetwar, P. & Tyrrel, S. F. Deterioration in water quality from supply chain to household and appropriate storage in
the context of intermittent water supplies. Water Sci. Technol. Water Supply 11, 400 (2011).
34. Brick, T. et al. Water contamination in urban south India: Household storage practices and their implications for water safety and
enteric infections. Int. J. Hyg. Environ. Health 207, 473–480 (2004).
35. Eisenberg, J. N. S., Trostle, J., Sorensen, R. J. D. & Shields, K. F. Toward a systems approach to enteric pathogen transmission: From
individual independence to community interdependence. Annu. Rev. Public Health 33, 239–257 (2012).
36. Do, C. B. & Batzoglou, S. What is the expectation maximization algorithm?. Nat. Biotechnol. 26, 897–899 (2008).
37. Greiner, M., Pfeier, D. & Smith, R. D. Principles and practical application of the receiver-operating characteristic analysis for
diagnostic tests. Prev. Vet. Med. 45, 23–41 (2000).
38. Ercumen, A. et al. Animal feces contribute to domestic fecal contamination: Evidence from E. coli measured in water, hands, food,
ies, and soil in Bangladesh. Environ. Sci. Technol. 51, 8725–8734 (2017).
39. Penakalapati, G. et al. Exposure to animal feces and human health: A systematic review and proposed research priorities. Environ.
Sci. Technol. 51, 11537–11552 (2017).
40. Local Burden of Disease Child Growth Failure Collaborators. Mapping child growth failure across low- and middle-income
countries. Nature 577, 231–234 (2020).
41. B amualim, A. Livestock production and re management in East Nusa Tenggara. in Fire and Sustainable Agricultural and Forestry
Development in Eastern Indonesia and Northern Australia. Proceedings of an international workshop held at Northern Territory
University, Darwin, Australia, 13–15 April 1999, 69–72 (2000).
42. Cronin, A. A., Breslin, N., Gibson, J. & Pedley, S. Monitoring source and domestic water quality in parallel with sanitary risk
identication in Northern Mozambique to prioritise protection interventions. J. Water Health 4, 333–345 (2006).
43. Trevett, A. F., Carter, R. C. & Tyrrel, S. F. Water quality deterioration: A study of household drinking water quality in rural Hon-
duras. Int. J. Environ. Health Res. 14, 273–283 (2004).
44. Esrey, S. A. & Habicht, J. Epidemiologic evidence for health benets from improved water and sanitation in developing countries.
Epidemiol. Rev. 8, 117–128 (1986).
45. Mellor, J. E., Smith, J. A., Samie, A. & Dillingham, R. A. Coliform sources and mechanisms for regrowth in household drinking
water in Limpopo, South Africa. J. Environ. Eng. (United States) 139, 1152–1161 (2013).
46. Pickering, A. et al. Can individual and integrated water, sanitation, and handwashing interventions reduce fecal contamination
in the household environment? Evidence from the WASH Benets cluster-randomized trial in rural Kenya. https://www.biorxiv.
org/content/https ://doi.org/10.1101/73199 2v1.full.pdf (2019) https ://doi.org/10.1101/73199 2.
47. Gundry, S. et al. A systematic review of the health outcomes related to household water quality in developing countries. J. Water
Health 2, 1–14 (2004).
Acknowledgements
We thank all respondents in the study, all interviewers, and LKP Anugerah Anak Sumba for the support in data
collection. We thank Kirsten van Linden, Ilias Machairas, Dennis Djohan for the hard work during the data col-
lection. We also thank Dr. Doris van Halem, from TU Del Global Drinking Water, and Armand Middeldorp
Content courtesy of Springer Nature, terms of use apply. Rights reserved
Vol.:(0123456789)
Scientic Reports | (2020) 10:18867 |
www.nature.com/scientificreports/
to support us with the eld water quality test equipment. e rst author receives a PhD research funding from
Indonesia Endowment Fund for Education (LPDP) and eld logistics and from the Del University of Technol-
ogy. e second author received a travel fund from TU Del Global Initiative for the data collection.
Author contributions
D.D. and W.P.I. contributed to the experimental design. W.P.I. contributed to the sample collection and process-
ing. D.D., W.P.I., S.P., and L.R. contributed to data analysis and validation. S.P. and L.R. supervised the project.
D.D. prepared the rst dra. All authors reviewed and edited the manuscript.
Competing interests
e authors declare no competing interests.
Additional information
Correspondence and requests for materials should be addressed to D.D.
Reprints and permissions information is available at www.nature.com/reprints.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional aliations.
Open Access is article is licensed under a Creative Commons Attribution 4.0 International
License, which permits use, sharing, adaptation, distribution and reproduction in any medium or
format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the
Creative Commons licence, and indicate if changes were made. e images or other third party material in this
article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the
material. If material is not included in the article’s Creative Commons licence and your intended use is not
permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from
the copyright holder. To view a copy of this licence, visit http://creat iveco mmons .org/licen ses/by/4.0/.
© e Author(s) 2020
Content courtesy of Springer Nature, terms of use apply. Rights reserved
1.
2.
3.
4.
5.
6.
Terms and Conditions
Springer Nature journal content, brought to you courtesy of Springer Nature Customer Service Center GmbH (“Springer Nature”).
Springer Nature supports a reasonable amount of sharing of research papers by authors, subscribers and authorised users (“Users”), for small-
scale personal, non-commercial use provided that all copyright, trade and service marks and other proprietary notices are maintained. By
accessing, sharing, receiving or otherwise using the Springer Nature journal content you agree to these terms of use (“Terms”). For these
purposes, Springer Nature considers academic use (by researchers and students) to be non-commercial.
These Terms are supplementary and will apply in addition to any applicable website terms and conditions, a relevant site licence or a personal
subscription. These Terms will prevail over any conflict or ambiguity with regards to the relevant terms, a site licence or a personal subscription
(to the extent of the conflict or ambiguity only). For Creative Commons-licensed articles, the terms of the Creative Commons license used will
apply.
We collect and use personal data to provide access to the Springer Nature journal content. We may also use these personal data internally within
ResearchGate and Springer Nature and as agreed share it, in an anonymised way, for purposes of tracking, analysis and reporting. We will not
otherwise disclose your personal data outside the ResearchGate or the Springer Nature group of companies unless we have your permission as
detailed in the Privacy Policy.
While Users may use the Springer Nature journal content for small scale, personal non-commercial use, it is important to note that Users may
not:
use such content for the purpose of providing other users with access on a regular or large scale basis or as a means to circumvent access
control;
use such content where to do so would be considered a criminal or statutory offence in any jurisdiction, or gives rise to civil liability, or is
otherwise unlawful;
falsely or misleadingly imply or suggest endorsement, approval , sponsorship, or association unless explicitly agreed to by Springer Nature in
writing;
use bots or other automated methods to access the content or redirect messages
override any security feature or exclusionary protocol; or
share the content in order to create substitute for Springer Nature products or services or a systematic database of Springer Nature journal
content.
In line with the restriction against commercial use, Springer Nature does not permit the creation of a product or service that creates revenue,
royalties, rent or income from our content or its inclusion as part of a paid for service or for other commercial gain. Springer Nature journal
content cannot be used for inter-library loans and librarians may not upload Springer Nature journal content on a large scale into their, or any
other, institutional repository.
These terms of use are reviewed regularly and may be amended at any time. Springer Nature is not obligated to publish any information or
content on this website and may remove it or features or functionality at our sole discretion, at any time with or without notice. Springer Nature
may revoke this licence to you at any time and remove access to any copies of the Springer Nature journal content which have been saved.
To the fullest extent permitted by law, Springer Nature makes no warranties, representations or guarantees to Users, either express or implied
with respect to the Springer nature journal content and all parties disclaim and waive any implied warranties or warranties imposed by law,
including merchantability or fitness for any particular purpose.
Please note that these rights do not automatically extend to content, data or other material published by Springer Nature that may be licensed
from third parties.
If you would like to use or distribute our Springer Nature journal content to a wider audience or on a regular basis or in any other manner not
expressly permitted by these Terms, please contact Springer Nature at
onlineservice@springernature.com
Content uploaded by D. Daniel
Author content
All content in this area was uploaded by D. Daniel on Nov 03, 2020
Content may be subject to copyright.
