FIELD APPLICATION OF A MULTIPHASE METER BASED ON

NEURAL NET CHARACTERISATION OF FLOW CONDITIONS

Haluk Toral and Shiqian Cai Petroleum Software Limited United Kingdom
Dasline Sinta and Meramat Tajak Sarawak Shell Bhd Malaysia

Abstract

The paper presents measurements taken with a multiphase flow meter based on neural net
methodology at the National Engineering Laboratory (NEL) and Institute Francais Petrole (IFP)
multiphase test loops and an offshore production platform operated by Sarawak Shell Berhad
Malaysia (SSB). The paper examines potential sources of error inherent in different multiphase
metering technologies. The method described in the paper, named Expert System for Multiphase
Metering (ESMER) provides the basis of a general approach to multiphase flow metering which
can be used under a wide range of operating conditions in oil and gas production lines and in the
process industries. The meter was calibrated at NEL and IFP flow loops by characterizing the
turbulence characteristics of the flow by pattern recognition techniques and relating these
characteristics to the flow rates of individual phases by means of supervised neural nets. Field
performance of the meter since November 2002 is reviewed. Limitations, sources of uncertainty
and sensitivity to various parameters are discussed.

1. Introduction
[1]
There are two different approaches in multiphase measurement technology . In the first
approach (lets call this the conventional approach), phase velocities and cross sectional
fractions (or related quantities) must be measured individually (six unknowns requiring a total
of five measurements for three phase flow as another equation is provide by the simple
arithmetic sum of three phases adding to 1). The conventional approach can be illustrated by
the schematic diagram shown in Figure 1.

Conventional meters have to overcome a number of problems.

The first problem arises from the difficulty of measuring the velocities of individual phases. A
number of work arounds have been attempted including
- measure velocity of one phase and estimate the other from empirical “slip” correlations
- homogenise to alleviate slip
- separate phases to remove the problem at source

Each approach introduces more problems than it removes.

First work around has two difficulties to surmount. A technique must be found for measuring
the velocity of one of the phases. The common approach has been to employ the cross
correlation method to measure the flow rate of the liquid phase but this approach can only
work under certain flow regimes, such as slug and plug flow, but quite likely to fail in stratified,
annular and dispersed flows. (Cross correlation technology is used in slurry metering where
flow structures are transported across finite differences without loosing their character. In
multiphase flow structures are very rapidly dissipated). Second, any empirical correlation for
estimation of slip is likely to fail once taken beyond the laboratory conditions for which it is
tuned up

Second work around, homogenisation, can only be effective across a very narrow range of
conditions where liquid is the continuous phase (eg bubbly flow).

A related side effect which affects many commercial conventional meters is that they have
opted out for a vertical orientation to achieve optimum homogenisation. This orientation gives
rise to problems at lower liquid rates when the liquid starts falling back and puts a restrictive
lower limit. For example, for a 4” MPFM meter, a minimum of 2000 bpd liquid flow rate may
[2]
be required .
Third work around, separating the phases is a highly intrusive approach which appears to
defeat the original objective of an in-line multiphase flow meter (why not stick to a test
separator?)

Even if we assume that all the problems mentioned above are solved, there remains one last
hurdle in front of the conventional approach which simply cannot be overcome by any amount
of (currently available technology). The last hurdle can be quantified through simple
arithmetic as follows.

Let the fraction of oil (? oil), water (? water) and gas ((? gas ) to be calculated from the fraction
models and the water-in-liquid ratio or water cut is defined as:

WLR = ? water / ( ? oil+? water )

Let the individual phase flow rates Qoil, Qwater and Qgas to be computed from the following
equations:

Qoil = A*V* ? oil

Qwater = A*V*? water
Qgas = A*V*? gas

Where A is the cross sectional area of the pipe and V is the total velocity of the flow.

Then, it can be shown that the uncertainty in the measurement of the liquid flow rate will be given
by:

Liq_err = Sqrt[(V_err)**2+(GVF_err/(1-GVF))**2]

Where V_err is velocity relative error and gvf_err is absolute error.

This equation shows that the liquid flow rate uncertainty will deteriorate strongly with increasing
GVF. For example, Figure 2 shows the variation of Liq_err with varying GVF_err for V_err=+/-5%
(lets assume V_err is measured at this accuracy as an optimistic assumption). One can see that
the liquid measurement uncertainty will be catastrophic above 90% GVF. The inverse is true for
the gas flow rate, which will suffer with reducing GVF and can turn into a catastrophe for
GVF<30%.

We started saying that there are two different approaches to multiphase metering and seen the
high likelihood of the conventional approach ending in disaster (or coming up with a very limited
solution across a very tight operating envelope) with our analysis above.

The second approach is described as follows in a recent review article (1): “…parameters of the
flow are measured that are functions of the three flow rates. For example a pressure drop across
a venturi, the attenuation of a gamma beam and the impedance of the mixture can be determined
and relationships established between these measurements and the flow rates of the respective
phases, three independent measurements are required to establish the three flow rates. These
relationships cannot be predicted theoretically, therefore, they must be established by calibration.
…”

The ESMER flow meter takes a variation of this latter approach (Figure 3). ESMER establishes the
non-linear relationships described above by pattern recognition / neural net training rather than by
theoretical modeling (which is quite impossible as admitted by the review authors). The end result
is that ESMER flow meters can measure the flow rates of individual phases in oil production lines
without the need for separation or complex sensor. ESMER does not require a-priori empirical
models or knowledge of slip, does not employ cross-correlation, does not determine flow rates
from the set of equations shown above (so its accuracy is flat across 1-99% GVF) and it is
oriented vertically without a mixer (and therefore can work at very low flow rates).

ESMER depends on the naturally occurring multiphase flow patterns in the pipeline which it
characterises by neural net analysis of high frequency signals emitted by simple and standard
sensors. The characterisation process starts in the laboratory (factory calibration) and adapted to
the field against separator reference measurements (field tuning).
2. ESMER Hardware

ESMER spool comprises two modular sub-spools: the Pressure Spool and the Impedance Spool.
The Pressure Spool is essentially an orifice (but any differential device would in fact be
acceptable) equipped with the usual complement of simple sensors related to differential pressure
measurement (one differential pressure sensor, one absolute pressure sensor and a temperature
sensor). The Impedance Spool comprises a capacitance sensor for oil-continuous applications, a
conductance sensor for water-continuous applications or both for full water cut range applications.
The spool is normally installed horizontally and the flow passes through the spool in a straight line.

The PC contains a data acquisition board for sampling and digitising the sensor signals at a
relatively high frequency (to capture a range of fluid dynamic turbulent fluctuations and features at
a range of time and length scales). The digitised signal is processed by neural network algorithms
and flow rates of individual phases are identified by the ESMER system on-line with the frequency
of once a minute. The neural nets are originally trained at the factory and tuned up under field
conditions. At intervals chosen by the operator, field data is transmitted to the Server running at
base for re-tuning the neural nets (running on-line). The new calibration is downloaded to the
Client remotely by means of an update disk or electronic download. An alert is automatically
raised on deviation of field conditions beyond a pre-set threshold.

ESMER is suitable for use in oil, gas condensate and wet gas production lines under a wide range
of conditions including the full water cut range. The present system was trained up to 98% GVF,
but in principle the range can be extended beyond this. The expected level of accuracy will be in
the range ? 10% to ?15% for liquid and gas flow rates in relative terms and ?5% water cut in
absolute terms. The accuracy of the system will improve if field-based neural training against a
test separator is possible.

A photograph of 3-inch ESMER T3 meter installed at SFJT - C is shown in Figure 4

The operating envelope for SFJT-C field conditions is shown in Figure 5.

3. Flow Loop Test s

ESMER T3A multiphase flowmeter was tested at the Institute Francais Petrole’s Solaize
Lyon Laboratory (IFP) using domestic fuel and nitrogen gas. The operating envelope covered
liquid flowrates range from 373 to 6,038 bpd, GVF from 60 to 99% at 0% water cut under the
operating pressure in the range 100 psig (+-5psi).

The factory calibration (neural networks) of the meter was based on reference measurements
taken at NEL and IFP during 1999 and 2000. Performance of the meter was evaluated against
reference measurements for varying flow rates across a matrix of 21 measurement points. The
details of the results are presented in Fig. 6 – Fig. 7. RMS average error and proportion of points
meeting specification are shown in Table 1.

The performance of the meter can be described as very good when compared against its
“specification accuracy” of ? 10% relative for liquid flow rate and +-15% relative error for gas flow
rate with 90% uncertainty. In the flow loop test, all measurement points met the specification
accuracy criteria as shown in Table 1.

Total Number of Proportion of points RMS average error

Point=21 meeting specification

Liquid flowrate 100% 3.90%

Gas flowrate 100% 3.45%
Table 1 ESMER T3A Summary of Flow Loop Measurements
4. Field Test

ESMER T3A multiphase flowmeter was commissioned on the SFJT-C platform (South Furious
Jacket – C off-shore platform of Shell Sarawak Berhad - Malaysia) in November 2002. South
Furious Jacket off-shore platform is located some 35 km away from the Kota Belud shore. The
SFJT-C, although a slim production jacket platform, is a highly-automated platform, with around
800 instrument/measurement points and 3 “smart” wells. The platform was installed in September
2001 and produces from one of the first smart fields in the world. Apart from a few manually-
controlled valves, most of the equipment and pumps can be controlled and remotely accessed
through the DCS system. For example, the operator can line up wells and run well tests from the
main platform (SFP-A) or from onshore. All the process variables can be measured through the
DCS and accessed from the main platform or the head office at Lutong in Miri (some 200 km
away). Gas lifting to the well heads and the casing pressures can also be remotely controlled.

The ESMER meter was installed as a ‘drop in’ spool piece on a 3 inch horizontal section just prior
to the test separator. There is a vertical bend 1.5m upstream and 0.5m downstream of the meter.
The platform is equipped with a gas – liquid separator, various temperature, pressure, level, valve
control systems, PD meter for liquid measurement and gas (orifice plate) flow meters. Flow rates
of each single phase stream are measured at the outlet of the separator. Flow, pressure and
temperature (instantaneous) measurements are transmitted to the network via the DCS.

Measurements were obtained from ESMER with a frequency of 2 minutes and from the single
phase reference meters at the exit leg of the separator (PD meter for liquid and orifice for gas) with
a frequency of 1 minute (average of measurements taken every second). Typical time series of the
measurements from ESMER and references during the course of a well test is shown in Figure 8.
The correlation between the time series is remarkably good despite the inevitable time lag
between the inlet (ESMER) and outlet (references). Another factor that needs to be taken into
account is that the separator liquid exit leg pump is controlled by the tank level controller.

In early field tests conducted after tune up, comprising a total of seven well tests across a range
of flow conditions, it has been observed that liquid flow rates measured on-line by the ESMER
system agreed within +/-5% of the test separator and gas flowrates agreed within +/- 5-10%. Low
watercut readings (1-3%) detected by the meter were undetected vi sually in the manual samples.
SSB will take samples for reconfirmation of the water cut at the onshore laboratory by using the
more sophisticated Karl Fisher Method for analysis. The ESMER multiphase flowmeter has been
connected into the Shell computer net work and being monitored from North Sabah control center
as well as from Shell Miri head office, which is some 200 km away.

The summary of well tests conducted during SAT is listed in Table 2.

START END ESMER Separator Liquid Gas GVF

WELL NO. DATE TIME TIME Liquid Gas GVF Liquid Gas GVF Accuracy Accuracy Accuracy
bpd mmscfd % bpd mmscfd % % % %
303 11/11/02 10:12 14:52 2232.09 0.876 90.23 2281.95 0.862 89.82 -2.185 1.624 0.42
301 12/11/02 15:00 23:00 453.8 0.606 96.90 433.3 0.651 97.20 4.731 -6.912 -0.30
305 13/11/2002 10:30 15:30 2329.16 1.578 92.82 2409.02 1.607 92.71 -3.315 -1.805 0.11
307 14/11/2002 10:30 14:30 467.27 0.548 96.63 467.85 0.502 96.21 -0.124 9.163 0.42
305 11/16/02 10:30 13:30 2537.68 0.945 89.73 2633.43 0.991 88.83 -3.636 -4.642 0.90
305 11/16/02 14:32 15:04 2500.51 0.922 89.52 2590.11 0.898 88.90 -3.459 2.673 0.63
307 11/16/02 16:00 05:32* 460.63 0.534 96.75 499.6 0.426 95.38 -7.800 25.352 1.37
305 11/17/02 10:30 14:15 2702.67 0.62 N/A 0.77 N/A -19.481 N/A
301 11/17/02 14:50 22:36 428.96 0.592 N/A 0.578 N/A 2.422 N/A
Table 2 Summary of Well Tests

These results are shown graphically on Figures 9 ( ESMER measurement versus reference
measurement) and Figure 10 accuracy of the measurement versus GVF. No particular trend or
deterioration of measurement was noticed in the range of GVF from 88.9 to 97.2%
RMS average error and proportion of points meeting specification are shown in Table 3.

Proportion of points RMS average error

meeting specification
Total number of well tests: 7
Liquid flow rate 100% 4.2%
Gas flow rate 86% 10%
Table 3 ESMER Performance at SAT

ESMER MPFM has remained in continuous operation since its commissioning in November
2002 to date (June 2004). The meter was interfaced to the company network. Measurements
and diagnostics (comprising stochastic flow features) are routinely gathered at the Miri head
office and sent to the manufacturer (PSL) by internet for analysis and re-tuning. A number of
further validation tests have been conducted during this time. In particular it was noted that
the performance of the meter deteriorated at low flow rates (close to its lower operating
boundary at 400 bpd). The meter had also failed to detect a number of occasions on which
the wells were shut down. The neural nets were re-trained with field data for tuning to the low
flow rate conditions and for detecting shut down and the new calibration was uploaded on the
meter (this can be done remotely from the on-shore control centre).

The minute by minute history of live measurements during Jan – Feb 2003 are shown in Fig.
11 to 12 as a sample. Low / no flow detection failures are marked on the charts. Simulation
tests were run across the same time frame with the new neural nets and it was confirmed that
the new model identified the shut-down conditions correctly.

5. Conclusions

The experience on the SFJT – C offshore platform have confirmed that multiphase meters can be
used for on-line and unmanned well testing under remote control and thereby allow significant
cost savings for offshore operations. In some well tests carried out to date, it has been observed
that the ESMER meter has the potential to deliver well test results as accurately as a
conventional test separator.

The performance of the ESMER meter did not appear to be affected by GVF. The measurement
accuracy was at its best when the flow rates encountered were as close to the centre of the
operating envelope and deteriorated towards the edges.

In addition to its advantages for well testing, on-line measurement of GVF, water-cut and
individual phase flow rates allows operators to monitor changes in the gathering facilities
continuously and provides a real-time view of the status of production and flow lines. Another
advantage of the multiphase meter, which should be realised in the medium term, is that
instantaneous measurements performed during start up and shut down transients allows us to
improve our understanding of the hydrodynamics of the reservoir near the well bore and should
help to optimize gas lifting.

References

[1] G. Falcone, G. F. Hewitt, C. Aliminti and B. Harrison, Multiphase Flow Metering: Current
Trends and Future Developments, SPE JPT April; 2002, 77-84.

[2] L. Scheers, K. Busaidi, M. Parper, M. Halvorsen and T. Wideroe, Multiphase Flow

nd th
Meeting per Well can it be justed? North Sea Flow Measurement Workshop, 22 -25
October, 2002.

[3] H. Toral and T. Darwich (1992) Measurement of Gas/Liquid Flow, United Kingdom Patent,
2221042

[4] H. Toral and T. Darwich (1991) Method and Apparatus for the Measurement of Gas/Liquid
Flow, United States Patent, 5051922.
[5] T. Darwich, H. Toral and J. Archer, A software technique for flowrate measurement in
horizontal two-phase flow, SPE Production Engineering, 1991, 265-270.

[6] S. Cai, H. Toral and J. Qiu, Flow regime identification by a self organizing neural network,
International Conference on Artificial Neural Networks, 1993, Amsterdam, 13-16.

[7] S. Cai and H. Toral, Flow rate measurement in air-water horizontal pipeline by neural
networks, Proc. Int. Joint Conf. On Neural Networks, 1993, Vol. 2, Nagoya, Japan, 25-29
October, 2013-2016.

[8] S.E. Akartuna, Identification of phase flow rates in oil-gas-water flow from turbulent
capacitance and pressure signals, Ph.D. thesis, Imperial College of Science Technology and
Medicine, University of London, 1994.

[9] H. Toral, N. Beg and J. Archer, Multi-phase flow metering by software, Int. Conf. On Basic
Principles and Industrial Applications of Multiphase Flow, London, 1990.

[10] N. Beg and H.Toral, Off-situ calibration of a two-phase pattern recognition flow meter,
1993, Int. J. Multiphase Flow, 19, 999-1012.

[11] S. Cai, An artificial neural network method for three phase flow measurement, Ph.D.
Thesis, Imperial College of Science Technology and Medicine, University of London, 1995.

[12] H. Toral, S. Cai, E. Akartuna, K. Stothard and A. Jamieson, Field tests of the Esmer
multiphase flowmeter, The North Sea Flow Measurement Workshop, 26-29 October 1998,
Perthshire, Scotland.
 Impedance, or
Gamma ray or
microwave Fraction
Models
Water-cut
Venturi (dp) Flow Rate Oil flow rate
Models

Water flow rate

Gas fraction

Gas flow rate

X-correlation (velocity)

Figure 1. Schematic diagram of the conventional MPFM Concept

Liquid Uncertainty Distribution at Total Velocity Uncertainty at 5%

100.0
90.0
GVF=+/-0.02%
80.0 GVF=+/-0.05%
70.0
Uncertainty, %

GVF=+/-0.1%
60.0 GVF=+/-0.2%
50.0 GVF=+/-0.5%

40.0 GVF=+/-1%
GVF=+/-2%
30.0
GVF=+/-5%
20.0
GVF=+/-10%
10.0
0.0
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 98
GVF

Figure 2 – Liquid Flow Rate Error with GVF for an assumed Total Drift Velocity Error =+/-5%

variable fuzzy on/off

pipe diameter, density,

viscosity, salinity,etc

turbulence

neural net mass rates

modelling

phase
distribution

Figure 3. Schematic diagram of the ESMER concept

 Figure 4 – Photograph of ESMER T3A installed at SFJT – C offshore platform

FAT Test
GVF 50% 75% 90%
10000 Mandhane Map

SF30 Well Test Data

Liquid Flow Rate (m3/d)

1000 GVF 99%

100 GVF 99.90%

ESMER against Separator

Liquid: +/- 5 - 10%
10 Gas: +/- 5 - 15%
10 100 1000 10000 100000 GVF: +/- 0.5 - 1.5%

Actual Gas Flow Rate (m3/d)

Figure 5 – ESMER T3A operation envelope

 3

2.7

2.4
Meter Liquid Velocity m/s
2.1
Oil Cont.
1.8
0% Rel. Error
1.5 -10% Rel. Error
1.2 +10% Rel. Error

0.9

0.6

0.3

0
0 0.3 0.6 0.9 1.2 1.5 1.8 2.1 2.4 2.7 3
Reference Liquid Velocity m/s

20

18

16
Meter Gas Velocity m/s

14
Oil Cont.
12
0% Rel. Error
10 -10% Rel. Error
8 +10% Rel. Error

0
0 2 4 6 8 10 12 14 16 18 20
Reference Gas Velocity m/s

100.00%

90.00%

80.00%
Meter Gas Volume Fraction

70.00%
60.00%

50.00% Oil Cont.

40.00%

30.00%

20.00%

10.00%

0.00%
0.00% 10.00 20.00 30.00 40.00 50.00 60.00 70.00 80.00 90.00 100.00
% % % % % % % % % %
Reference Gas Volume Fraction

Figure 6 – Flow Loop Test. ESMER measurement against reference.

 10.0

8.0

6.0

4.0
Relative Error %

2.0

0.0

-2.0

-4.0

-6.0

-8.0
7.7 8.0 8.8 10.4 13.3 13.6 15.7 16.0 20.3 20.4
Liquid Rate m3/hr

8.0

6.0

4.0

2.0
Relative Error %

0.0

-2.0

-4.0

-6.0

-8.0

-10.0
83.1 86.4 89.0 110.1 117.6 169.8 173.9 201.0 210.0 262.4
Gas Rate m3/hr

Figure 7 – ESMER Liquid and gas measurement error distributions in FAT tests
 % Mmscfd Bpd

0
500

10.00
15.00
20.00
25.00
30.00
35.00
40.00
45.00
50.00

0.00
5.00
1000
1500
2000
2500
3000
3500
4000
4500

0.000
0.200
0.400
0.600
0.800
1.000
1.200
1.400
1.600
1.800
16-Nov-02 10:30 16-Nov-02 10:30 16-Nov-02 10:30
16-Nov-02 10:41 16-Nov-02 10:40 16-Nov-02 10:40
16-Nov-02 10:53 16-Nov-02 10:50 16-Nov-02 10:50
16-Nov-02 11:05 16-Nov-02 11:01 16-Nov-02 11:01

16-Nov-02 11:17 16-Nov-02 11:11 16-Nov-02 11:11

16-Nov-02 11:29 16-Nov-02 11:21 16-Nov-02 11:21

16-Nov-02 11:41 16-Nov-02 11:32 16-Nov-02 11:32

16-Nov-02 11:53 16-Nov-02 11:42 16-Nov-02 11:42

16-Nov-02 12:04 16-Nov-02 11:53 16-Nov-02 11:53

water composition at intervals of two minutes

16-Nov-02 12:03 16-Nov-02 12:03

Date & Time

16-Nov-02 12:16
Date & Time

Date & Time

16-Nov-02 12:13 16-Nov-02 12:13
16-Nov-02 12:28
16-Nov-02 12:24 16-Nov-02 12:24
16-Nov-02 12:40
16-Nov-02 12:34 16-Nov-02 12:34
16-Nov-02 12:52
16-Nov-02 12:44 16-Nov-02 12:44
16-Nov-02 13:04
16-Nov-02 12:55 16-Nov-02 12:55
16-Nov-02 13:16
16-Nov-02 13:05 16-Nov-02 13:05
16-Nov-02 13:28
16-Nov-02 13:16 16-Nov-02 13:16

0
5
10
15
20
25
30
35
40
45
50
0
0

20
40
60
80
20
40
60
80

100
120
140

%
100
120
140

Psia Psia

ESMER
ESMER
ESMER

Pressure
Pressure

Separator
Separator

Wcut Sample

Figure 8 – Typical well test results showing the variation of the liquid rate, the gas rate and the
 5000

4500

4000

3500
ESMER bpd 3000

2500 Oil Cont.

2000

1500

1000

500

0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Separator bpd

1.8

1.6

1.4
ESMER mmscfd

1.2

1 Oil Cont.

0.8

0.6

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Separator mmscfd

100

90

80

70
ESMER GVF %

60

50 Oil Cont.

40

30

20

10

0
0 10 20 30 40 50 60 70 80 90 100
Separator GVF %

Figure 9 - ESMER against separator references during Field Testing (see Table 2)
 10

Liquid Relative Error %

4

-2

-4

-6

-8

-10
88.8 88.9 89.8 92.7 95.4 96.2 97.2
GVF %

25

20

15

10
Gas Relative Error %

-5

-10

-15

-20

-25
88.8 88.9 89.8 92.7 95.4 96.2 97.2
GVF %

2.0

1.5

1.0
GVF Absolute Error %

0.5

0.0

-0.5

-1.0

-1.5

-2.0
88.8 88.9 89.8 92.7 95.4 96.2 97.2
GVF %

Figure 10 – ESMER Error vs GVF in Field Testing ( see Table 2)

 3500
Should have
3000
identified as no
2500
flow

Oil, bpd
2000
1500
1000
500
0

2000
Should have
1600 identified as no
Gas, mscfd

1200
flow

800

400

0
1/01 5/01 10/01 15/01 20/01 25/01 30/01

Should have
80.0
identified as no
flow
60.0
Wcut, %

40.0

20.0

0.0

20.0

15.0
Pressure, bar

10.0

5.0

0.0

Figure 11 – ESMER live measurement in Jan. 2003

Should have
4000
identified as no
flow
3000
Oil, bpd

2000
Should have
1000 identified as no
flow
0

2000

1600
Gas, mscfd

1200

800 Should have

identified as no
400
flow
0

1/02 5/02 10/02 15/02 20/02 25/02 28/02

80.0

Should have
60.0
identified as no
Wcut, %

40.0 flow

20.0

0.0

20.0

15.0
Pressure, bar

10.0

5.0

0.0

Figure 12 – ESMER live measurement in Feb. 2003