VOL. 18 NO. 4 (2006)
ARTICLE
Real-time detection of possible
harmful events using UV/vis
spectrometry
G. Langergraber,a J. van den Broeke,b,* W. Lettlb and A. Weingartnerb
a
Institute for Sanitary Engineering and Water Pollution Control, BOKU - University of Natural Resources and
Applied Life Sciences, Vienna, Muthgasse 18, A-1190 Vienna, Austria. E-mail: [email protected]
b
s::can Messtechnik GmbH, Brigittagasse 22-24, A-1200 Vienna, Austria. E-mail: [email protected]
Introduction
Operators of water or wastewater
­treatment plants benefit when changes
of concentration and/or composition of
inlet water can be detected and, therefore, a possible failure of the treatment
plant performance can be avoided.
When monitoring drinking water, either
at the source or in the distribution
system, identification of low probability/
high impact events that might compromise water quality and, as a consequence, public health is an important
goal. Early identification is the prerequisite for an effective response that
reduces or entirely prevents the adverse
impacts of such a contamination. For
wastewater treatment plants, the early
detection of, for example, inhibiting
substances in the influent is of great
importance and can prevent negative
impact on the biological degradation
processes.
Systems that facilitate such early
detection and identification are referred
to as early warning monitoring systems
or alarm systems. Typical causes of
high impact contaminations are natural events (for example, flooding, anoxia,
algal blooms), accidental anthropogenic events (for example, inadvertent discharges, spills) or intentional
discharges (for example, vandalism,
terrorism). As the number of potentially
harmful compounds is enormous, only
a fraction of these can be monitored, as
the development of specific sampling
strategies for all contaminants would be
an impossible task.
www.spectroscopyeurope.com
The development of new variables,
such as alarm parameters, that allow for
an integrated assessment of changes in
water quality using surrogate or aggregate variables, instead of searching for all
specific contaminants individually, can be
a useful strategy to cover a much broader
range of potential threats than possible
with conventional monitoring solutions.1
As spectral data, and their evolution over
time, are very rich in information, the
loss in compound-specific information
is compensated for by a much better
overall picture of change in water quality and the possibility to detect changes
that will not be picked up by conventional, single contaminant directed,
monitoring programmes. Furthermore, it
can be expected that alarm parameters
developed from spectral data are accurate, selective and transferable amongst
applications to an extent that is beyond
the state-of-the-art. This contribution
describes how alarm parameters can be
derived from spectral data.
Methodology
The instrument
For the development of alarm parameters an s::can Messtechnik GmbH
sub­mersible ultraviolet/visible (UV/
vis) spectro­photometer, spectro::lyser,
was used. This measures absorbance
of ultraviolet and visible light from 200
to 750 nm.2 The instrument is built as a
compact sub­mersible sensor, enabling
measurement of UV/vis spectra with
laboratory quality directly in liquid media.
For high robustness and long-term stabil-
ity, even under harsh conditions, the
spectrometer is equipped with an autocleaning system using pressurised air.
Furthermore, the long-term stability of the
measurements it guaranteed by a twobeam-design, which allows zero-compensation for changes in the instrument as
well as for the detection of instrumental
failures, see pre­ceding ­article in this issue
of Spectroscopy Europe.
Alarm parameters derived
from spectral measurements
Alarm parameters are in many ways
different from measurements of welldefined chemical substances. 3 Most
importantly, they must show a quick
response to suspect quality changes,
which implies the need for real-time
measurements. Furthermore, a high
sensitivity and, at the same time, low
probability for false alarms, as well as a
broad-band response to diverse contamination sources are prerequisites. In many
cases this will mean that application- and
location-specific, tailor-made, parameters
will need to be developed. Less important for alarm parameters are their parameter selectivity and their compliance to
analytical laboratory norms; when monitoring surrogate parameters, the results
of the measurements will be difficult
to compare with the results of regular
analytical methods.
Broadband optical instruments, like
spectrometers, allow the development
of such alarm parameters. Spectroscopy
fulfils most of the requirements, because
it allows:
SPECTROSCOPYEUROPE VOL. 18 NO. 4 (2006)
ARTICLE
1. Direct and selective
­spectrometric ­measurements
Single substances or substance groups
that are detectable in the UV/vis
­spectrum can be identified using spectral algorithms evaluating the spectra.
Parameters derived from spectral information include surrogates and single
substances.4 Changes in the low ppb
range can be monitored using on-line
spectrometry. However, the selectivity is
rather low and single-substance identification below 100 ppb is difficult, if not
impossible, for many substances.
2. Monitoring of ­indicator
parameters instead of
­contaminants
Indicator parameters can be used instead
of monitoring contaminants directly. An
example would be to use nitrate measurements as an indicator for pesticides in
groundwater.
3. Monitoring of matrix
­ arameters instead of
p
­contaminants
The whole UV/vis spectrum can be used
as a fingerprint of the water composition (“matrix”). For the monitoring of
matrix changes, evaluation of these
fingerprints allows the detection of very
small changes. When using derivatives
of the fingerprints, most of the natural
deviations of the baseline spectrum are
eliminated and, therefore, a better identification of abnormal spectral features
is possible. An example of such an
application is the tracing of differences
between spectra over time or space.
Often, the detected changes are not
directly related to known substances but,
nevertheless, provide a sensitive alarm
parameter.
Time-resolved delta
spectrometry
There are several ways to evaluate the
information provided by the fingerprints,
including the qualitative interpretation of
spectral deviations from a site-specific
reference spectrum (for example, peaks,
shoulders, gradients, analysis of the
derivative spectra etc.), the comparison
of spectral differences between measuring points in a measurement network, i.e.
SPECTROSCOPYEUROPE
spatially-resolved delta spectrometry and
the evaluation of changes of the ­spectral
features over time, i.e. time-resolved
delta spectrometry. The latter will be
discussed in more depth in the following paragraphs.
Time-resolved delta spectrometry 5
tracks changes in the shape of the
fingerprint over time, as well as the
speed with which changes occur. As
changes due to extreme natural events
and anthropogenic changes are typically
faster than gradually occurring natural
changes, such as seasonal changes in
water composition, it is possible to identify unusual water compositions solely
on the basis of changes in time. This
means that contaminants that do not
provide very distinct signals can still be
detected, because they still cause a (nondistinct) change in the absorption spectrum. It also means that in cases where
no water body can serve as a reference
for ­unimpaired water quality, for example due to continuous fluctuations, the
use of UV/vis spectrometry nevertheless
provides the possibility to detect irregularities on the basis of the size and speed
of changes in the fingerprint. The multidimensional information provided in UV/
vis spectra offers a much greater information potential on top of the “normal”
baseline compared to single parameters
such as turbidity or dissolved organic
carbon��������������������������������
(DOC���������������������������
), for which time-resolved
data analysis will not allow such an identification of abnormal changes.
For definition of spectral alarm parameters, the following procedure has to be
followed.
■ Learning period: during the learning period (usually some months)
­spectra are collected. After this period
the baseline spectrum (shape and
features) is established.
■ Abnormality definition: absorption
spectra or their 1st- and 2nd-order derivatives can be used to identify deviations from “normal” spectral features.
Detection of deviations is possible
even with a continuously changing
baseline, as in this case the trigger
values for abnormal changes will be
set outside the limits of the normal
deviations or at wavelengths little
affected by the normal fluctuations.
■ Alarm definition: alarm definition
is based on the concept of virtual
contaminants. Since one never knows
in advance which contaminant may
be the next to enter the system,
several groups of “virtual” contaminants are defined. These virtual
contaminants are designed in such
a way that by monitoring this limited
set of parameters, the whole range
of contaminants visible in the UV/vis
spectrum is covered.
■ Sensitivity definition: sensitivity of the
virtual contaminants to changes in the
water matrix has to be adjusted individually with respect to risks involved
and acceptable false alarm levels. This
is based on empirical and statistical
evaluation.
■ Definition of actions to be taken in
case of alarm. Examples for possible actions in the case of an alarm
would be: alarm level 1—slight deviation from normal situation: automatic
sampling triggered, waiting for laboratory analysis, no additional action
needed; alarm level 2—strong abnormalities: automatic sampling plus
urgent action at site needed.
Applications
Detection of (unusual) changes
in wastewater composition6
For the operation of wastewater treatment plants, the early detection of
changes in wastewater influent quality
is a necessity to prevent a possible failure of the treatment plant performance.
Quality changes include both changes
in concentration and composition of the
wastewater. Changes of the waste­water
composition—especially changes of
industrial inlets—can lead to operational
problems and/or a possible inhibition of
the biochemical degradation processes.
In one example, the spectrometer was
operated in the influent of a waste­
water treatment plant for several weeks.
During most of the monitored period the
concentrations showed only typical daily
variations. During some periods, however,
strong fluctuations occurred (Figure 1).
From these data it is quite clear that
the variations in the total organic carbon
equivalent (��������������������������
TOCeq���������������������
) concentration were
caused by variations of the dissolved
www.spectroscopyeurope.com
nearby industrial discharges into the sewer system were conducted. Figure 3 compares the 1st-derivative
of measured fingerprints at the wastewater treatment plant influent at 04:52, 06:44, 06:52 and 07:02
hours, the period when the rapid changes of water quality occurred, with 1st derivative fingerprints of
the baseline fingerprint and one measured directly at an industrial discharge. The changing contribution
monitored simultaneously with a high measurement frequency (in this case the interval between the of the industrial inlet to the water composition can be clearly identified in the wavelength region
measurements was two minutes).
between 255 nm and 295 nm. However, also other sources contribute to the rapid changes of water
Figure 1: Detailed picture of strong fluctuations (August 13, between 02:00 and 08:00 hours).6
quality, e.g. represented by a peak at about 400 nm at 06:44 hours, at 350 nm at 04:52 hours and at 300
In Figure 2 the fingerprints of the measurements marked in Figure 1 (full red circles) are shown and nm at 07:02 hours, respectively, that had to be identified in a consecutive study.
compared with a typical fingerprint of the plant that shows a shape typical for communal wastewater. Figure 3: 1st- derivative fingerprints of measured wastewater treatment plant influents at different times,
The three selected fingerprints were recorded within 18 minutes around 07:00 hours and clearly show the baseline fingerprint and an industrial discharge.
different features from the baseline fingerprint. Two peaks, one at 270 nm and another at 360 nm can Development of alarm parameters for a sewer network [5]
be observed at 06:44 and 06:52 hours. At 07:02 hours – only 10 minutes later – both peaks have This second example shows the application of time resolved delta spectrometry for the development of
alarm parameters. Figure 4 shows a 3-D spectral picture of a water body over a period of 1 day. Peaks,
disappeared and another peak at 280 nm is visible, indicating a pronounced change in water quality.
troughs and shoulders can be clearly identified. For this water body a three-month learning period was
Figure 2: Baseline fingerprint and fingerprints for the times shown in the inset.
To investigate the origin of these rapid changes of the wastewater composition some measurements at needed to cover all variations that occurred naturally. For evaluation of the baseline spectrum, only the
nearby industrial discharges into the sewer system were conducted. Figure 3 compares the 1st-derivative fingerprints from the periods between the visible peaks were utilised.
of measured fingerprints at the wastewater treatment plant influent at 04:52, 06:44, 06:52 and 07:02 Figure 4: 3D-UV/vis spectra plotted against time [5].
hours, the period when the rapid changes of water quality occurred, with 1st derivative fingerprints of The concept of virtual contaminants as described above was used for the definition of the alarm
the baseline fingerprint and one measured directly at an industrial discharge. The changing contribution parameters generating several groups of virtual contaminants. Occurrence probabilities of the virtual
of the industrial inlet to the water composition can be clearly identified in the wavelength region contaminants are calculated off-line and plotted against the normalised spectral derivatives. Figure 5
between 255 nm and 295 nm. However, also other sources contribute to the rapid changes of water shows the occurrence probabilities and the defined alarm levels for 3 virtual contaminants (Parameter 1
and 2, and colour). E.g. for the parameter "Colour" the first and second alarm levels are reached at an
monitored simultaneously with a high measurement frequency (in this case the interval between the quality, e.g. represented by a peak at about 400 nm at 06:44 hours, at 350 nm at 04:52 hours and at 300 occurrence probability of 99.3 % and > 99.9 % respectively.
nm at 07:02 hours, respectively, that had to be identified in a consecutive study.
measurements was two minutes).
VOL. 18 NO. 4 (2006)
ARTICLE
4
Spectral TSSeq (mg/l)
800
700
600
500
400
300
200
100.0%
-1
Spectral DOCeq (mg/l)
99.9%
-1
Spectral TOCeq (mg/l)
900
Absorbance - 1st der. (abs.m .nm )
1000
99.8%
0
99.7%
99.5%
Industrial inlet
13.08. 03:00
13.08. 04:00
13.08. 05:00
13.08. 06:00
13.08. 07:00
13.08.2002 04:52
600
Colour
99.2%
13.08.2002 07:02
290
340
390
440
99.1%
490
99.0%
0
0.2
0.4
Wavelength (nm)
0.6
0.8
1
1.2
1.4
Value
Figure 3: 1st- derivative fingerprints of measured wastewater treatment plant influents Figure 5: Definition of “Alarm level 1” and “Alarm level 2” for three alarm
st the baseline fingerprint and an industrial discharge.
parameters [5].
at different times,
Figure 1: Detailed picture of strong fluctuations (August 13, between 02:00 and 08:00
hours).6
Development of alarm parameters for a sewer network [5]
Figure 1. Detailed picture of strong fluctuations (13 August, between 02:00 and 08:00
hours).
Parameter 2
99.3%
13.08.2002 06:44
13.08.2002 06:52
13.08. 08:00
In Figure 2 the fingerprints of the measurements marked in Figure 1 (full red circles) are shown and
compared with a typical fingerprint of the plant that shows a shape typical for communal wastewater.
The three selected fingerprints were recorded within 18 minutes around 07:00 hours and clearly show
6 the baseline fingerprint. Two peaks, one at 270 nm and another at 360 nm can
different features from
be observed at 06:44 and 06:52 hours. At 07:02 hours – only 10 minutes later – both peaks have
disappeared and another peak at 280 nm is visible, indicating a pronounced change in water quality.
Parameter 1
99.4%
Baseline FP
-8
0
13.08. 02:00
Alarm level 2
99.6%
-4
-12
240
100
Alarm level 1
Figure 3. 1 derivative fingerprints of
measured wastewater treatment plant influents at different times, the baseline fingerprint and an industrial discharge.
Figure 5. Definition of “Alarm level 1” and
“Alarm level 2” for three alarm parameters.
Experience has shown that the occurrence probabilities of alarm parameters derived from spectral data
5 In
This second example shows the application of time resolved delta spectrometry for the development of are in most cases different from occurrence probabilities obtained from conventional parameters.
alarm parameters. Figure 4 shows a 3-D spectral picture of a water body over a period of 1 day. Peaks, addition peaks for different contaminants are occurring at different times [5]. This enables the detection
troughs and shoulders can be clearly identified. For this water body a three-month learning period was of events that cannot be detected using conventional parameters.
needed to cover all variations that occurred naturally. For evaluation of the baseline spectrum, only the
fingerprints from the periods between the visible peaks were utilised.
Absorbance (abs.m-1)
500
13.08.2002 06:44
13.08.2002 06:52
13.08.2002 07:02
Baseline FP
organic carbon equivalent��������������
(DOCeq�������
). The
were conducted. Figure 3 compares
concentration of the ­particulate matter
the 1 st derivative of measured finger(TSSeq—total suspended solids equivprints at the wastewater treatment plant
alent) was rather constant until about
influent at 04:52, 06:44, 06:52 and
07:00 hours when a peak occurred.
07:02 hours, the period when the rapid
Wavelength (nm)could only be
This specific behaviour
changes of water quality occurred, with
Figure 2: Baseline fingerprint and fingerprints for the times shown in the inset.
observed
because
several
parameters
1st derivative fingerprints of the baseline
To investigate
the origin of these
rapid changes of the
wastewater composition
some measurements at
nearby industrial discharges into the sewer system were conducted. Figure 3 compares the 1 -derivative
of measured fingerprints at the wastewater treatment plant influent at 04:52, 06:44, 06:52 and 07:02
are
with
a of fingerprint and one measured directly
hours, the
periodmonitored
when the rapid changes ofsimultaneously
water quality occurred, with 1st derivative
fingerprints
the baseline fingerprint and one measured directly at an industrial discharge. The changing contribution
of the high
industrial inlet
to the water composition can frequency
be clearly identified in (in
the wavelength
measurement
thisregion at an industrial discharge. The changing
between 255 nm and 295 nm. However, also other sources contribute to the rapid changes of water
case the interval between the measurecontribution of the industrial inlet to the
ments was two minutes).
water composition can be clearly identiIn Figure 2, the fingerprints of the
fied in the wavelength region between
measurements marked in Figure 1 (full
255 nm and 295 nm. However, other
red circles) are shown and compared
sources also
�����������������������������
contribute
������������������������
to the rapid
with a typical fingerprint of the plant that
changes of water quality, for example,
shows a shape typical for communal
represented by a peak at about 400 nm
waste­water. The three selected fingerat 06:44 hours, at 350 nm at 04:52
prints were recorded within 18 minutes
hours and at 300 nm at 07:02 hours,
around 07:00 hours and clearly show
respectively, that had to be identified in
different features from the baseline
a ­consecutive study.
fingerprint.
peaks,
one
monitored
simultaneously with Two
a high measurement
frequency
(in thisat
case 270 nm
the interval between the
measurements was two minutes).
and another at 360 nm can be observed monitored
Development
of alarm
simultaneously with a high measurement
frequency (in this case the interval between the
measurements was two minutes).
1: Detailed picture of strong fluctuations (August 13, between 02:00 and 08:00 hours).
at 06:44 and 06:52 hours. At 07:02 Figure
parameters
for a sewer
In Figure 2 the fingerprints of the measurements marked in Figure 1 (full red circles) are shown and
compared with a typical fingerprint
5 of the plant that shows a shape typical for communal wastewater.
three
selected fingerprints were recorded within 18 minutes around 07:00 hours and clearly show
hours—only 10 minutes later—both The
network
different features from the baseline fingerprint. Two peaks, one at 270 nm and another at 360 nm can
observed at 06:44 and 06:52 hours. At 07:02 hours—only 10 minutes later—both peaks have
peaks have disappeared and another bedisappeared
This
second
shows
thein appliand another
peak at 280 example
nm is visible, indicating
a pronounced change
water quality.
Figure 2: Baseline fingerprint and fingerprints for the times shown in the inset.
To
investigate
the originof
of thesetime-resolved
rapid changes of the wastewater composition
measurements at
peak at 280 nm is visible, indicating a nearby
cation
deltasomespecindustrial discharges into the sewer system were conducted. Figure 3 compares the 1 -derivative
of measured fingerprints at the wastewater treatment plant influent at 04:52, 06:44, 06:52 and 07:02
hours,
the
period
when
the
rapid
changes
of
water
quality
occurred,
with
1st
derivative
fingerprints of
pronounced change in water quality.
trometry for the development of alarm
the baseline fingerprint and one measured directly at an industrial discharge. The changing contribution
of the industrial inlet to the water composition can be clearly identified in the wavelength region
To investigate the origin of these rapid betweenpara­meters.
Figure
4sources
shows
3-D
spec255 nm and 295 nm. However,
also other
contributea
to the
rapid changes
of water
quality, e.g. represented by a peak at about 400 nm at 06:44 hours, at 350 nm at 04:52 hours and at 300
respectively,of
that had
be identified
in a consecutive
study.a period
changes of the wastewater composition, nm at 07:02
tralhours,
picture
a towater
body
over
1 - derivative
wastewater
plant influents
some
nearby
of3:atone
day.fingerprints
Peaks,of measured
troughs
andtreatment
shoulders
Figure
1: Detailedmeasurements
picture of strong fluctuations at
(August
13, betweenindus02:00 and 08:00 Figure
different times, the baseline fingerprint and an industrial discharge.
hours).
Development of alarm parameters for a sewer network [5]
In Figure
2
the
fingerprints
of
the
measurements
marked
in
Figure
1
(full
red
circles)
are
shown
and
trial discharges into the sewer system This second
canexample
beshows
clearly
identified.
this
water
the application
of time resolved delta For
spectrometry
for the
development of
compared with a typical fingerprint of the plant that shows a shape typical for communal wastewater.
400
300
200
100
0
230
280
330
380
430
480
530
580
630
st
1000
Spectral TOCeq (mg/l)
900
Spectral DOCeq (mg/l)
Spectral TSSeq (mg/l)
6
800
700
600
500
400
st
300
200
100
0
13.08. 02:00
13.08. 03:00
13.08. 04:00
13.08. 05:00
13.08. 06:00
13.08. 07:00
13.08. 08:00
st
6
alarm parameters. Figure 4 shows a 3-D spectral picture of a water body over a period of 1 day. Peaks,
The three selected fingerprints were recorded within 18 minutes around 07:00 hours and clearly showtroughs and shoulders can be clearly identified. For this water body a three-month learning period was
different features from the baseline fingerprint. Two peaks, one at 270 nm and another at 360 nm canneeded to cover all variations that occurred naturally. For evaluation of the baseline spectrum, only the
be observed at 06:44 and 06:52 hours. At 07:02 hours – only 10 minutes later – both peaks havefingerprints from the periods between the visible peaks were utilised.
disappeared and another peak at 280 nm is visible, indicating a pronounced change in water quality.
600
Absorbance (abs.m-1)
500
13.08.2002 06:44
13.08.2002 06:52
13.08.2002 07:02
Baseline FP
400
300
200
100
0
230
280
330
380
430
480
530
580
630
Wavelength (nm)
Figure 2: Baseline fingerprint and fingerprints for the times shown in the inset.
Figure 2. Baseline fingerprint and fingerprints for the times shown in the inset.
Figure 4: 3D-UV/vis spectra plotted against time [5].
Figure 4. 3D-UV/vis spectra plotted against
time.
To investigate the origin of these rapid changes of the wastewater composition some measurements atThe concept of virtual contaminants as described above was used for the definition of the alarm
nearby industrial discharges into the sewer system were conducted. Figure 3 compares the 1st-derivativeparameters generating several groups of virtual contaminants. Occurrence probabilities of the virtual
of measured fingerprints at the wastewater treatment plant influent at 04:52, 06:44, 06:52 and 07:02contaminants are5calculated off-line and plotted against the normalised spectral derivatives. Figure 5
hours, the period when the rapid changes of water quality occurred, with 1st derivative fingerprints ofshows the occurrence probabilities and the defined alarm levels for 3 virtual contaminants (Parameter 1
the baseline fingerprint and one measured directly at an industrial discharge. The changing contributionand 2, and colour). E.g. for the parameter "Colour" the first and second alarm levels are reached at an
of the industrial inlet to the water composition can be clearly identified in the wavelength regionoccurrence probability of 99.3 % and > 99.9 % respectively.
between 255 nm and 295 nm. However, also other sources contribute to the rapid changes of water
www.spectroscopyeurope.com
body, a three-month learning period
was needed to cover all variations that
occurred naturally. For evaluation of the
baseline spectrum, only the fingerprints
from the periods between the visible
peaks were utilised.
The concept of virtual contaminants,
as described above, was used for the
definition of the alarm parameters generating several groups of virtual contaminants. Occurrence probabilities of the
virtual contaminants are calculated offline and plotted against the normalised
spectral derivatives. Figure 5 shows the
occurrence probabilities and the defined
alarm levels for three virtual contaminants (Parameter 1 and 2, and colour).
For example, for the parameter “colour”,
the first and second alarm levels are
reached at an occurrence probability of
99.3% and > 99.9%, respectively.
Experience has shown that the occurrence probabilities of alarm parameters
derived from spectral data are, in most
cases, different from occurrence probabilities obtained from conventional parameters. In addition, peaks for different
contaminants occur at different times.5
This enables the detection of events that
cannot be detected using conventional
parameters.
Conclusions
In this article, it has been shown that
UV/vis spectrometry is a promising
method to quantify rapid changes in
water quality in the inlet of a treatment
plant. Measurements with real-time UV/
vis spectrometers can, therefore, be
a valuable instrument to monitor the
influent of treatment plants for integrated management, control of municipal sewer networks and monitoring
SPECTROSCOPYEUROPE VOL. 18 NO. 4 (2006)
ARTICLE
of the treatment plant itself, for example, to prevent a possible failure of the
plant. Furthermore, it is possible to identify single industrial discharges and it can
be expected that even changes in the
operational settings of industries can be
monitored.
Time-resolved delta spectrometry
has been applied for developing alarm
parameters. This method to define alarm
levels is based on empirical and statistical evaluation of the fingerprints. It can
be expected that it is possible to detect
events that cannot be detected using
conventional parameters when using
alarm parameters derived from spec-
SPECTROSCOPYEUROPE
tral features, as their occurrence probabilities are, in most cases, different from
occurrence probabilities obtained from
conventional parameters.
References
1. T.H.M. Noij and I. Bobeldijk, Wat. Sci.
Technol. 47(2), 181 (2003).
2. G. Langergraber, N. Fleischmann
and F. Hofstaedter, Wat. Sci. Technol.
47(2), 63 (2003).
3. T.M. Brosnan (Ed), Early warning
monitoring to detect hazardous
events in water supplies. International
Life Sciences Institute, Washington
DC, USA (1999).
4. J. van den Broeke, G. Langergraber
and A. Weingartner, Spectroscopy
Europe, this issue.
5. G. Langergraber, A. Weingartner and
N. Fleischmann, Wat. Sci. Technol.
50(11), 13 (2004).
6. G. Langergraber, N. Fleischmann, F.
Hofstaedter, A. Weingartner and W.
Lettl, in “Proceedings of the 9th IWA
conference on Design, Operation
and Costs of Large Wastewater
Treatment Plants” Ed by I. Ruzickova
and J. Wanner. IWA Publishing,
London, UK, p. 135 (2003).
www.spectroscopyeurope.com