FINAL REPORT ON STATISTICAL ASSOCIATIONS OF EXTERNAL PARAMETERS WITH COLIFORM FAILURES IN DISTRIBUTION SYSTEMS USING STATUTORY MONITORING DATA
Report No FR0436
Identifying associations between the presence of coliforms and
fluctuations in external parameters will provide information on how
samples with coliforms differ chemically, physically and
biologically compared to those without coliforms.
This information will form the foundation for developing a model to
predict (from fluctuations in controllable or known parameters) the
increased probability of coliform failures within distribution
It will also assist in the identification of root causes and origins
of failures so that operational factors and remedial treatments may
be developed to minimise the problem.
To identify statistical associations between the occurrences of
coliforms in water samples and fluctuations in chemical, physical and biological
factors within the water distribution supply using statutory water quality
monitoring data provided by the water companies.
Coliform failures in 100 ml volume water samples collected from
distribution systems under statutory monitoring (water quality) legislation may
occur for a variety of reasons such as problems at the treatment
works, infiltration, release from biofilms and sediments, and
regrowth. Coliform failures tend to occur intermittently. Before
this contract, little work had ever been performed using the
statutory monitoring water quality data to identify associations
between the presence of coliform organisms in such samples and
fluctuations in other physical, chemical and biological factors
within the distribution. This would provide information on whether
intermittent coliform failures reflected real changes in the local
environment within the distribution system or, alternatively, were
mainly due to the chance `capture' of a coliform from a
homogeneously distributed population during sampling. Identifying
such associations would also facilitate implementation of
operational processes and remedial treatments to minimise coliform
Furthermore, the associations identified may form the foundations
for further work to develop a model to predict coliform concentrations within
particular parts of the distribution supply and hence impending coliform
failure. At present the only precautionary method available is to
monitor large volume water samples (up to 10 litres) for coliforms;
the presence of one coliform in a litre volume suggesting a 10% risk
of coliform failure in a 100 ml statutory monitoring sample. A model
to predict coliform concentrations of above one in a litre volume
from chlorine, temperature and plate counts (perhaps measured from
10 ml samples) would be beneficial to the water supply companies and
could be integrated into chlorine concentration modelling packages
such as WATNET.
Statutory water quality monitoring data, representing some 36 000
water samples (number of total chlorine records), have been analysed
statistically. These data were supplied by nine water companies, A
to I, located across the UK. The main conclusions are:
- Water samples which contain at least one coliform organism per 100
ml differ in many ways from samples without coliforms. The
differences may be chemical, physical or biological. Thus, water
samples with coliforms may have higher plate count concentrations,
higher temperatures, lower nitrate ion concentrations, higher
nitrite ion concentrations, lower total chlorine concentrations, and
higher iron (III) or phosphate ion concentrations compared with
samples without coliforms.
- These associations suggest that coliforms are not homogeneously
distributed within the distribution supply. Indeed, the distribution
supply system would appear to be heterogeneous with respect to
chemical, physical and microbiological factors. Coliform failures
are linked to real changes within the distribution system and are
not merely the result of random capture during sampling. Registering
a coliform failure in a 100 ml volume water sample is not a random
process such as Russian roulette and it may be possible to predict
increased probabilities of coliform failures using fluctuations in
values of measurable or controllable parameters, such as those
- Association of coliforms with higher three day (or seven day)
plate count concentrations appears to be strong and universal
throughout the distribution supply systems of all the water
companies so far analysed.
- Associations of coliforms with external factors vary between
different water companies suggesting that there is not one single
mechanism for the occurrence of coliforms within the UK distribution
supply. Three day plate count concentrations were on average between
1.5x to 25.5x higher for samples with coliforms compared to those
without coliforms depending on the water company. For some
companies, total chlorine and temperature appear to be important
factors. However, total chlorine concentration does not appear to be
an important factor for Water Company C and temperature may not
important for Water Company H. In the distributions of Water
Companies C, G, and I, coliforms are linked to higher iron (III) ion
concentrations, while in Water Company A operational hydrants the
opposite relationship may exist. Furthermore, while Water Companies
A and possibly B and I, show associations of coliforms with lower
nitrate ion concentrations, Water Companies C and G do not. Perhaps
there are two mechanisms for coliform failures operating which are
related to iron (III) and nitrate in opposite ways. There is
evidence that higher nitrite concentrations may be associated with
coliform failures in Water Company A fixed point taps, but not in
random taps from Water Companies B, C and G.
- Different mechanisms for coliform failure may operate within
different tap types. Thus for Water Company A, temperature was not
an important factor for random statutory tap failures but lower
chlorine concentration was. In contrast for the operational hydrant
samples , in which total chlorine concentrations were considerably
higher, total chlorine was not an important factor in coliform
failure but temperature was. Temperature was also important for
fixed point taps in Water Company A. On average, however, three day
plate count concentrations were 1.5x higher for samples with
coliforms in fixed and random statutory taps and operational
hydrants within Water Company A.
- Different mechanisms for coliform failure may operate within
different zones or groups of zones in the same water company.
Analysing pooled data from all the zones from a water company may
fail to detect important associations localised to small numbers of
zones or to individual coliform positive samples. Thus, coliforms in
operational hydrant samples from a group of 11 zones in Water
Company A in 1991 showed strong associations with lower total
chlorine concentrations. This association was lost using pooled
total chlorine data (1990 to 1992 inclusive) from operational
hydrant samples in 55 zones.
- Pooling data from different parts of distribution supply within
the same water company, could also conceivably demonstrate
associations which were an artefact. The strong associations
identified between coliforms and certain external parameters in Water Company I could be artefact arising from natural differences
(e.g. pH, metal ion concentration, and colour) between the source
waters (e.g. spring or surface), with coliform failures arising more
in a particular type of source water because of less efficient
treatment. Thus the associations identified between the presence of
coliforms and lower free or total chlorine concentrations may be the
dominant factor in Water Company I. Regression analysis, however,
performed using coliform positive data would suggest that this is
not the case, at least for colour measurements. Thus regression
analysis showed that coliform concentrations in Water Company I not
only increase with lower total chlorine concentrations, but also
with higher colour measurements.
- Associations between coliforms and other factors are best detected
from statutory water quality monitoring data by comparing
statistically the means from samples with and without coliforms.
This method cannot unfortunately be used per se to predict an
increased probability of coliform failures. The associations
detected, however, will not only be useful in addressing operational
factors and remedial treatments to minimise coliform failures but
also serve as a preliminary screen for developing models to predict
increased probability of coliform failures. A linear regression
approach using coliform positive statutory monitoring data is less
powerful at detecting associations because it uses only a small
proportion of the data, although it provides the possibility for a
predictive model. A more powerful predictive model could be
developed using data from 10 litre samples.
- From the statistical distributions of total coliform
concentrations it is predicted that samples registering 0 coliforms
per 100 ml may contain between 10-1 and 10-9 coliforms per 100 ml.
Samples registering 0 per 100 ml may therefore give a false sense
of security. A predicative model is needed to estimate when and
where in the distribution supply, coliform concentrations are
approaching one per ten litres or one per litre. Although 100 ml
samples may be registering zero coliforms, the risk of a coliform
failure at a coliform concentration of one per litre is 10% and
precautionary action is required.
- Developing models to predict coliform failures from other
measurable or controllable parameters (e. g. chlorine concentration)
may be limited with the statutory water quality monitoring data
available. There are too few samples with coliform concentrations of
one or more per 100 ml, which are applicable to modelling.
Furthermore they represent samples with the highest coliform
concentrations, providing information only on samples which have
failed, and thus may not be representative of the distribution as a
whole. A more pertinent model could be developed by using data from
samples identified as being of high risk of coliform failure, i.e.
samples with 1 coliform per 10 litres or 1 coliform per litre. Such
data would also enable the model to be evaluated at this critical
coliform concentration range.
The present FWR contract (F-1702) ends in March 1994. Work which
relates to coliform failures in the distribution and which is to be
completed and reported by then includes:
- Linear regression analysis to assess feasibility of predicting log
coliform concentrations from values of other factors. This is to be
performed using water quality monitoring data from 100 ml samples
with coliforms as already supplied by the water companies A to I.
- Assessing feasibility of developing a predictive model for
coliform failures using multiple regression.
- Identification (from the associations) of operational factors and
remedial treatments to minimise coliform failures
- Comparing statistical distributions of coliform concentrations
with reported failure rates and commenting on present system of
compliance monitoring based on presence/absence detection of
After the current contract F-1702 has finished the following work
should be undertaken :
- Further identification, using 1993 statutory monitoring water
quality data, of statistical associations between the presence of
coliform organisms and fluctuations in other factors.
- Identification of factors associated with zones which show no
coliform failures. To date associations have been investigated by
comparing results from samples with and without coliforms from the
same zone. To identify conditions specific to zones with no
failures, comparison of external parameters between zones with and
without coliforms is required.
- Development of a model for predicting coliform failures in the
distributing from fluctuations in controllable and measurable
parameters. The existing data bases from statutory monitoring may
not be sufficient to achieve this because samples with 1 or more
coliforms are few in number and do not provide enough information.
It may therefore be necessary to perform detailed analysis of 10
litre volumes to obtain coliform concentrations between 1 and 50 000
per 10 litre. This would increase the data base on which to develop
and evaluate the model and provide data over the range of coliform
concentrations where there is increased risk of detecting a coliform
in a 100 ml volume.
- The present system of monitoring compliance based on the presence
or absence of coliforms in a 100 ml volume is not satisfactory.
First it provides very little information on water quality and
second it falsely exaggerates differences in microbiological quality
between different companies and zones. It is proposed to design and
test an alternative standard based on coliform statistical
distributions which provides a more realistic assessment of water
VI RESUME OF CONTENTS
This Final Report details statistical analyses performed between
April 1992 and November 1993 to identify associations between the
presence of coliforms and fluctuations in external parameters within
the distribution supply. This work is funded by the Foundation for
Water Research and is part of a larger project, 'Water Quality in
the Distribution', (F-1702).
Section 2 describes the graphical presentation and statistical
treatment of data, z"scussing the use of probability plots in
determining the most appropriate and best estimates for the
statistical parameters, namely the mean and standard deviation.
In Section 3, results are presented from statistical analyses for
associations between coliform failures and fluctuations in a variety
of parameters of chemical, biological and physical nature. The
parameters of particular interest include temperature, conc
entrations of plate count organisms (one, two, three and seven day),
and concentrations of total chlorine, nitrate and nitrite, iron
(III) and phosphate.
The findings are put into context for application in the Water
Industry in Section 4, which presents initial studies to develop a
model to predict coliform concentrations within the distribution
supply. Section 4 also introduces the concepts of describing
coliform concentrations in the distribution supply as this is
critical not only for correctly applying regression analysis but
also for defining risks of coliform failures. The concept that
samples registering 0 coliforms per 100 ml give a false sense of
security is developed, Preliminary linear regression relationships
between coliform concentrations and values for external parameters
are also presented in Section 4, which concludes by considering
further information needed to develop the model.
Conclusions are made in Section 5, and recommendations for further
work summarised in Section 6. Details of statistical analyses which
have not previously been reported are presented in Appendix A.
Copies of the Report are available from FWR, price £35.00 less 20% to FWR Members