Evaluation of Tipping Bucket Rain Gauge Performance and Data Quality

w w w.environment-agency.gov.
uk
Evaluation of Tipping Bucket Rain

Gauge Performance and Data
Quality
Science Report: W6-084/SR

The Environment Agency is the leading public body protecting and
improving the environment in England and Wales.
It’s our job to make sure that air, land and water are looked after by
everyone in today’s society, so that tomorrow’s generations inherit a
cleaner, healthier world.
Our work includes tackling flooding and pollution incidents, reducing

industry’s impacts on the environment, cleaning up rivers, coastal
waters and contaminated land, and improving wildlife habitats.
This report is the result of research commissioned and funded by

the Environment Agency’s Science Programme.
Published by: Author(s):

Environment Agency, Rio House, Waterside Drive, Aztec R A Hodgkinson, T J Pepper and D W Wilson
West, Almondsbury, Bristol, BS32 4UD
Tel: 01454 624400 Fax: 01454 624409 Dissemination Status:
www.environment-agency.gov.uk Publicly available
ISBN: 1 844 323 242 Keywords:

tipping bucket raingauge, rainfall, literature review,
© Environment Agency September 2004 temperature, intensity, dynamic calibration,
aerodynamic,
All rights reserved. This document may be reproduced
with Prior permission of the Environment Agency. Research Contractor:
ADAS Consulting Ltd, Gleadthorpe Research Centre
The views expressed in this document are not Meden Vale, Mansfield, Notts NG20 9PF
necessarily those of the Environment Agency. Tel: 01623 844331 Website: www.adas.co.uk
This report is printed on Cyclus Print, a 100% recycled Environment Agency’s Project Manager:
stock, which is 100% post consumer waste and is totally David Brown, North West Region, Richard Fairclough
chlorine free. Water used is treated and in most cases House, Warrington
returned to source in better condition than removed.
Research Collaborator:
Further copies of this report are available from: The Met Office – www.meto.gov.uk
The Environment Agency’s National Customer Contact
Centre by emailing enquiries@environment- Science Project Number:
agency.gov.uk or by telephoning 08708 506506. W6-048
Product Code:
SCHO0904BIFW-E-P
ii Environment Agency Evaluation of Tipping Bucket Rain Gauge Performance and Data
Quality
Science at the Environment
Agency
Science underpins the work of the Environment Agency, by providing an up to date

understanding of the world about us, and helping us to develop monitoring tools
and techniques to manage our environment as efficiently as possible.
The work of the Science Group is a key ingredient in the partnership between
research, policy and operations that enables the Agency to protect and restore our
environment.
The Environment Agency’s Science Group focuses on five main areas of activity:
• Setting the agenda: To identify the strategic science needs of the Agency to
inform its advisory and regulatory roles.
• Sponsoring science: To fund people and projects in response to the needs
identified by the agenda setting.
• Managing science: To ensure that each project we fund is fit for purpose and
that it is executed according to international scientific standards.
• Carrying out science: To undertake the research itself, by those best placed to
do it - either by in-house Agency scientists, or by contracting it out to
universities, research institutes or consultancies.
• Providing advice: To ensure that the knowledge, tools and techniques
generated by the science programme are taken up by relevant decision-makers,
policy makers and operational staff.
Professor Mike Depledge Head of Science
Environment Agency Evaluation of Tipping Bucket Rain Gauge Performance and Data iii
Quality
ACKNOWLEDGEMENTS
The project manger is pleased to acknowledge the commitment and enthusiasm of the
Environment Agency hydrometry staff Gary Kinsella and Billy Fleming in ensuring the
success of the field testing phase. The project manager would also like to acknowledge
the technical support and commitment of the staff at Eskdalemuir, particularly Ian
Dawson, whose input to the project also greatly improved the quality of the data
collected.
iv Environment Agency Evaluation of Tipping Bucket Rain Gauge Performance and Data
Quality
EXECUTIVE SUMMARY
This report covers four key areas:
1. An in depth literature review of the issues and problems pertaining to the use of
tipping bucket raingauges (TBRs)
2. A survey of current Environment Agency practice pertaining to the use of TBRs.
3. Laboratory testing of 8 types of TBR to determine the best method for calibration
and compare gauge reliability and accuracy
4. Field evaluation of 8 types of TBR to assess, reliability in the field, relative
performance in terms of catch compared to the standard Meteorological Office 5"
The need for the work arose from the fact that the Environment Agency routinely
collects rainfall data for flood risk prediction, flood warning and water resource
planning activities. Discrepancies between the records obtained from tipping bucket
raingauges (TBRs) and conventional storage gauges. The problem that has been
identified, namely under-recording by TBRs, has particular implications for work
relating to water resources since, whilst the error may be relatively small on any one
event, the significance may increase as rainfall totals are accumulated over longer
periods.
The literature review found general agreement among the authors of the papers
reviewed that one of the most important factors affecting raingauge performance is
exposure to wind and that the effect increases with height of the gauge rim above the
ground surface. Evidence was found in the literature that use of gauges with an
aerodynamic shape (champagne glass) were measurably better than standard shapes
(cylindrical). The other most important effect identified in the literature reviewed relates
to the relationship between TBR calibration error and rainfall rate. This was reported to
be most marked during heavy rainfall, when water flowing in to the buckets while they
are tipping can represent an error of up to 10 – 30% compared to a co-located storage
gauge.
The laboratory investigations compared the two methods, constant rate and burette,
currently used by the Environment Agency for calibration of TBRs. It was concluded
that constant rate method should provide a calibration coefficient that is more robust at
higher intensities than one measured by the burette method. It was also concluded that
gauges calibrated by the constant rate method will underestimate at higher rainfall
intensities and better precision could be obtained by use of a full dynamic calibration
(i.e. calibration at a number of rates).
The influence of temperature on the calibration methods themselves was concluded to

be small and can be corrected for if required The calibration coefficient of tipping
bucket rain gauges varies with temperature although in most gauges the change were
found to be small. The frequency at which gauges were found to need calibration was
annual in general although biannual calibration might be beneficial in high rainfall
areas.
Field testing of the gauges was undertaken at an experimental site located at

Eskdalemuir Observatory in Dumfries and Galloway, which is situated in the Southern
Environment Agency Evaluation of Tipping Bucket Rain Gauge Performance and Data v
Quality
Uplands of Scotland. The National Grid Reference of the site is NT 32356026 and the
station is located at an altitude of 242 m. Long-term average annual rainfall for the site
is 1567mm. Eight different types of tipping bucket rain gauge were tested by
comparison against a control gauge (meteorological office 5″ gauge read daily) at a
vertical height of 12″. To allow greater experimental robustness, three of each gauge
type were deployed, making a total of 24 TBR gauges at the test site. Supplementary
control data were obtained from 2 extra meteorological office 5″ gauges ; one pit
mounted and one behind a turf wall.
Under the conditions encountered at the study site it was found that a TBR caught at
least 5% rainfall less than a standard meteorological office 5" gauge and that under
reading could be has large as ca. 20% with the worst performing gauges. Under the
conditions encountered at the study site considerable differences in rainfall catch were
observed between different types of TBR. A key finding was that under the conditions
encountered at the study site an aerodynamic gauge achieved the catch that was closest
to a standard meteorological office 5" gauge. No definitive link could be shown
between variations in either temperature or wind speed and the rainfall catch of a
specific gauge although on a qualitative basis the results suggest that there is a seasonal
link, with a 4-5 % reduction in TBR catch (for all gauge types) observed in the winter
months. Due to the low rainfall intensities encountered at the study site implementation
of a correction using dynamic calibration did not significantly increase measured catch
although benefits may well be obtained at other sites.
vi Environment Agency Evaluation of Tipping Bucket Rain Gauge Performance and Data
Quality
CONTENTS
ACKNOWLEDGEMENTS iv
EXECUTIVE SUMMARY v
LIST OF FIGURES viii
LIST OF TABLES ix
1. INTRODUCTION 1
2. METHODOLOGY 2
2.1 Literature Review 2
2.2 Laboratory testing of gauges 2
2.3 Field testing of TBR gauges 2
3. LITERATURE REVIEW 3
3.1 Objective 3
3.2 Introduction 3
3.3 Tipping Bucket Errors and calibration 3
3.4 Tipping Bucket Raingauge Construction and Use 7
3.5 Tipping Bucket Raingauge data Manipulation and
Comparison of Data sets 7
3.6 Comparison of TBR types 8
3.7 Comparison with other Automatic Systems 9
3.8 Discussion 11
4. LABORATORY STUDIES 13
4.1 Introduction 13
4.2 Detailed Methodology 14
4.3 Results 17
4.4 Discussion 22
4.5 Overview of Calibration Methods 24
4.6 Specific Conclusions arising from laboratory tests 25
5. FIELD TESTING OF GAUGES 26

5.1 Introduction 26
5.2 Materials and Methods 27
5.3 Results 32
5.4 Discussion 45
5.5 Conclusions 47
6. NEED FOR FURTHER WORK 48
REFERENCES 49
Appendix A: Title Parameters describing regression relationships between

relative rainfall catch and other climatic parameters 52
Environment Agency Evaluation of Tipping Bucket Rain Gauge Performance and Data vii
Quality
LIST OF FIGURES
Figure 4.1: Percentage error in rainfall measurement as a consequence of the
difference between calibration temperature (20°C) and operational temperature due to
changes in density of water.
Figure 4.2: Influence of calibration method on amount of rainfall required for 7500
tips
Figure 4.3: Relationship between % error and rainfall intensity for gauges calibrated
at 10-12 mm hr-1.
Figure 4.4: Change in % error in measurement due drift in calibration with time
(based on simulated rainfall.
Figure 5.1: Location Plan

Figure 5. 2: Layout of trial site
Figure 5.3a: Comparison of tipping bucket totals (1 June 2002 to 31 May 3002)
against meteorological office 5" gauge. total-block 1
Figure 5.3b: Comparison of tipping bucket totals (1 June 2002 to 31 May 3002)
against meteorological office 5" gauge total-block 2
Figure 5.3c: Comparison of tipping bucket totals (1 June 2002 to 31 May 3002)
against meteorological office 5" gauge total-block 3
Figure 5.4a: Percentage of rainfall recorded by pit and turfwall gauges – all year
Figure 5.4b: Percentage of rainfall recorded by pit and turfwall gauges – non winter
Figure 5.4c: Percentage of rainfall recorded by pit and turfwall gauges – winter only
Figure 5.5: Relationships between relative catch (%) and average wind speed when
raining
Figure 5.6: Comparison between no dynamic correction and correction at 5 minutes

or time of
viii Environment Agency Evaluation of Tipping Bucket Rain Gauge Performance and Data
Quality
LIST OF TABLES
Table 4.1: Summary of temperature testing based on all 3 blocks
Table 4.2: Relationship between Rainfall intensity and % error compared to

calibration at 10 mm hr-1.
Table 5.1: Randomisation of treatments
Table 5.2: Overall catches (June, 2002 to May, 2003 inclusive) from manually read
gauges
Table 5.3: Summary of overall mean gauge performance
Table 5.4: Rank order of gauge catches
Table 5.5: Mean catch for each gauge type expressed as a proportion of the pit/turf
wall gauges
Table 5.6a: Duncans test – Probabilities for Post Hoc tests - whole season
Table 5.6b: Duncans test – Probabilities for Post Hoc tests – winter
Table 5.6c: Duncans test – Probabilities for Post Hoc tests – non-winter
Table 5.7: Allocation of gauges to significantly different populations using Duncans

test, in order of decreasing performance
Table 5.8: Mean catch for each gauge type expressed as a proportion of the
pit/turf wall gauges
Table 5.9: Relative catches (%) of individual gauges (whole year data)
Table 5.10: Changes % in measured rainfall due to implementation of a dynamic

correction factor at a 5 minute timestep
Table 5.11: Cumulative period (minutes in year) during which intensities were over
10mm hr-1 in any 5 minutes maximum duration possible per year
525600 hours.
Table 5.12: Speed of snow melt January 3, 2003
Environment Agency Evaluation of Tipping Bucket Rain Gauge Performance and Data ix
Quality
1 INTRODUCTION
The Environment Agency routinely collects hydrometric data for flood risk prediction,
flood warning and water resource planning activities. As part of this work they have
identified discrepancies between the records obtained from tipping bucket raingauges
(TBRs) and conventional storage gauges. The problem that has been identified, namely
under-recording by TBRs, has particular implications for work relating to water
resources since, whilst the error may be relatively small on any one event, the
significance may increase as rainfall totals are accumulated over longer periods.
Work is needed to determine whether the discrepancy between the two systems is
systematic and can be linked to specific factors, or occurs randomly. If the error is
systematic, and is linked to some specific variable, data correction may be possible.
The environmental factors that are likely to be most critical are rainfall intensity, wind
speed and raindrop size although temperature may also play a part. Raingauges supplied
by different manufacturers may perform differently and performance may also change
with time, depending for example, on the robustness of the equipment and the
environmental conditions under which it is used. Calibration of gauges is thus also a
key issue relating to data quality. Variation between different types of dataloggers used
to collect and store the data is considered less likely to be a contributing factor, given
reliable supporting programming. However, detection of false signals due to inductive
effects in the cable connecting the gauges to the logger can occur. For example,
spurious rainfall data have been detected which were traced to the effect of an electric
cattle fence running parallel to the data collection cable.
1 Environment Agency Evaluation of Tipping Bucket Rain Gauge Performance and Data
Quality
2 METHODOLOGY
This section describes the overall approach used in each of the sections, however, the
detailed methodologies pertaining to individual tests will be found in the specific
chapters relating to this topic area
2.1 Literature Review
The objective of the review was to provide a synopsis of the most recent International
publications addressing TBR deployment, data quality, calibration methodology and
measurement errors. The approach used was to evaluate all publications with relevant
keywords cited on the Key databases DIALOG, CAB etc and thus identify all relevant
references published within the last twenty years. All known UK manufacturers of
meteorological equipment were contacted and asked to contribute any relevant in-house
publications to the review. The review also made use of available information from
International standards pertaining to the use of raingauges. Data was also sourced
relating to comparative trials of raingauges undertaken by the UK Meteorological
office, for which purpose a member of its staff was retained as a project consultant.
2.2 Laboratory Testing of Gauges
Eight different patterns of TBR were tested covering a wide range of materials, shapes
and cost. The work in this area focussed on a number of specific key issues:
1. To determine whether there are specific advantages in calibrating tipping bucket

rain gauges using either:
a. the burette method
b. the constant rate method (or one point dynamic calibration
method)
c. the full dynamic calibration method
2. To determine the influence of the temperature at which calibration is undertaken and

rainfall intensity.
3 To determine the long–term reliability of different makes of tipping bucket rain

gauge
4 To identify specific operational problems relating to specific makes of tipping

bucket raingauges.
2.3 Field Testing of TBR gauges
A field trial was undertaken at Eskdalemuir observatory with the objective of

comparing in the field the relative performance of the same 8 different patterns tipping
bucket rain gauges evaluated in the laboratory. The specific objectives were to look at:
(a) Reliability in the field
(b) How the "rainfall catch" as measured by the TBRs compares to that achieved by
a standard Meteorological Office 5" gauge.
(c) How rainfall intensity and wind speed interact with the rainfall catch obtained
from different designs of tipping bucket rain gauge.
Environment Agency Evaluation of Tipping Bucket Rain Gauge Performance and Data 2
Quality
3 LITERATURE REVIEW
3.1 Objective
To compile a synopsis of the most recent international publications addressing Tipping

Bucket Raingauge deployment, data quality, and calibration methodology and
measurement errors.
3.2 Introduction
The Tipping Bucket Raingauge (TBR) is widely cited as having been invented as long
ago as the seventeenth century by Sir Christopher Wren (e.g. Biswas 1970). In the
intervening years the TBR has been extensively refined, and is now one of the most
popular recording raingauges in use by many national weather agencies. Although the
basic principle of individual gauge designs are similar, there is no international
standard, and many different models are in use nationally and internationally which may
vary in materials and design. Guidelines for the siting and use of raingauges are
produced by individual national meteorological offices, and also by the World
Meteorological Organisation (WMO). In the UK, in addition to the Meteorological
Office guidelines, there is a British Standard (BS 7843) covering all aspects of
raingauge use. A number of these recommendations/guidelines are incorporated in the
relevant sections of the Environment Agency’s Hydrometric Manual.
The advantages of TBRs are well known, and have been reported extensively. For
instance, Smoot (1971) and Linsley (1973) reported on their high accuracy in recording
low-to-intermediate intensity rainfalls, reliability and suitability for remote recording.
Conversely, researchers have also been aware of potential recording problems for an
equally long period. Problems with under-recording when compared with a storage
gauge are identified in the UK Meteorological Office handbook of meteorological
instruments as early as the 1956 edition. Under-recording during high-intensity rainfall
events was reported by Bruce and Clark (1966), and the use of calibration methods to
remedy this problem by Smoot (1971).
Other possible recording inaccuracies are common to all raingauges, including TBRs,
and include those due to location (aspect and exposure), especially wind effects, which
were extensively reported by Robinson and Rodda (1969).
This report seeks to give an overview of recent (<20 years) published research into the
use of the TBR and alternative automatic gauges, including aspects of data quality and
calibration. The information presented will be applied to the wider review of TBR
laboratory and field performance being undertaken by the Environment Agency.
Reviews of published papers are given in chronological order under each section.
3.3 Tipping Bucket Raingauge Errors and Calibration
3.3.1 Non-specific TBR errors

Errors due to site and exposure common to both storage and recording gauges.
Folland (1988) Reviewed the problem of raingauge exposure (to wind effects) and
developed a quantitative theory of raingauge exposure. He considered wind flow over
Quality
the orifice of a standard 5 inch raingauge using wind tunnel data published by Robinson
and Rodda (1969) and Helliwell and Green (1974). The resulting model was integrated
for a range of raindrop sizes and frequency distributions after Ulbrich (1983) to give a
range of curves describing the percentage of rainfall lost at given wind speeds and
rainfall characteristics. The calculated results compared well to observed losses under
UK conditions, using ground level gauges as a reference. He also suggested a new
design of raingauge (“flat champagne glass gauge”) which aimed to substantially lower
the systematic losses of rainfall volume in the high winds commonly found over
exposed areas of land.
World Meteorological Organisation (1994) Emphasises the importance of site selection,

especially in having all gauges within a given area subject to the same siting criteria.
The suggested aim would be to choose a site to give the minimum possible wind speed
over the gauge orifice, without blocking precipitation by surrounding objects. Sites
subject to turbulent wind, including sloping sites and those with windbreaks, should be
avoided, although some uniform shelter from all directions is desirable. The
recommended method of reducing exposure where natural shelter is not available is to
install the gauge in a pit, or slightly less effective, within a circular turf wall of radius
1.5m. An alternative solution was to fit a windshield around the instrument. WMO
estimate loss of catch due to wind effects in the range 2-10% of the actual rainfall.
Other errors considered include wetting losses from internal walls of collectors and
measuring containers (2-10% loss), and loss from evaporation (0-4% loss). A number of
correction factors for different wind speeds are given.
BS 7843 2.1 (1996) Identifies over-exposure (to wind) as the most important factor
contributing to loss of catch by all raingauges. As the wind speed is known to increase
with height above the ground surface, the recommended method of reducing wind
effects is by mounting the raingauge in a pit with the collecting rim at ground surface.
This method can increase rainfall catch at lowland sites by 3% to 6%, and at windy
upland sites by up to 20% compared to a standard above-ground gauge. Other suggested
methods of reducing errors due to exposure include the construction of a turf wall round
the gauge to the height of the collection rim, at a distance of five times the gauge height.
The use of wind breaks, or the use of natural hollows are also considered acceptable,
although objects with an angle of elevation over 26.5° from a gauge location will lead to
over-sheltering, and reduced rainfall catch.
Seibert et al. (1999) Sought to reduce systematic errors due to aerodynamic effects and
wetting losses known to bias point measurements of precipitation. In the NOPEX
project a rain gauge with a new type of windshield and a special weighing construction
was used to minimize these errors. The windshield consisted of a flange surrounding the
gauge at the level of the orifice. The idea was to screen the area above the orifice from
the disturbance of the wind field by the gauge. At different locations the measured
precipitation amounts were compared with the amounts caught by standard gauges. The
analysis showed that the catch of the new gauge was higher than that of the standard
gauges. A difference of about 3% was related to reduced wind-induced losses, while a
difference of about 0.25 mm per event was explained as elimination of wetting loss. At
one location the differences were related to wind speed and rainfall intensity to evaluate
the effect of the windshield. The relative differences were largest (20%) for events with
low intensity and high wind speed.
Quality
3.3.2 Solid precipitation
Goodison and Louie (1985) Describe the use of a snowgauge to measure the water
equivalent of snowfall in Canada, where one third of the mean precipitation for the
country occurs as snow. The Canadian Nipher shielded snowgauge gave an average
catch within 10% of the ground true (determined using snow boards) at wind speeds of
up to 5.5 ms-1.
WMO (1994) and BS 7843 2.1 (1996) Both identify that raingauge exposure errors are
magnified when measuring snowfall, where an error in the range 10-50% of actual
deposition can be expected. Sources of error include “Blow out” in low temperatures
where small snow particles are sucked out of the collector by wind eddies above the
collector. “Blow in” when drifting of snow occurs, and burial in major snow events.
Problems other than exposure effects include “arching” when snow adheres to the
funnel rim and bridges the gauge orifice when temperatures are close to melting. And
“over topping” of the gauge. Specialized snow gauges designed to melt snow in the
collector are much more effective, although are rarely used in the UK.
3.3.3 Errors specific to TBRs and calibration techniques

Calder and Kidd (1978) considered that the dynamic calibration of TBRs was required
to achieve both high resolution and accuracy. They state that in general, a non-linear
relation exists between flow rate and the tipping rate of a tipping-bucket gauge. Other
methods have been used either to avoid or correct for this non-linearity, including the
use of a siphon incorporated between receiver and bucket which is designed to empty at
a constant flow rate into the bucket. The method proposed in this report was to
determine the gauge parameters V and t by measuring the tipping rate of the bucket at
known flow rates. If T, the time between tips, is plotted graphically against the
reciprocal of the flow rate, the volume of the bucket (V) is given by the slope of the
line, and the tipping time (t) is given by the intercept on the T- axis. Graphs were
produced to show the typical calibration results of a 1.2-l, tipping-bucket flowmeter and
of a 0.1-mm, tipping-bucket raingauge. These calibrations showed that the tipping time
did not change significantly either with different gauges of the same type or with
moderate adjustments to the volume of the gauge. This result was important because it
allows an individual dynamic calibration to be obtained from one determination of T at
a given flow rate.
Adami and Da Deppo (1986) Investigated systematic errors associated with TBRs. The
research focused on the accuracy of the tools used for operational use by the National
Hydrological Service of the Italian Ministry of Works. Positive errors (to 2-3%) were
shown at intensities lower than the calibration standard of 60 mm/hr and negative errors
(to 7%) were seen at higher intensities. Non-negligible sources of error, other than those
referring to external effects of wind, splashing, wetting, evaporation and blowing snow,
are singled out, e.g., horizontal misalignments or poor upkeep. A new type of weight
pluviograph (raingauge) aimed at overriding the mentioned shortcomings was
conceived and developed. A brief description of the structure of the instrument was
given, including the details of the collecting mouth and the electronic scaling. The
foremost feature of the tool, which can inserted at ease within centralized data
acquisition systems, lies in its high sensitivity in recording very intense precipitation.
Niemczynowicz (1986) Carried out the dynamic calibration of three types of TBR in use
in the Nordic countries (the LTH, PLUMATIC and RIMCO gauges). It was found in all
Quality
gauges tested that the volume of water required to tip the bucket was not a constant
characteristic for the gauge but depended upon rainfall intensity. Thus, to avoid errors, a
calculation of the rainfall intensity or of rainfall volume from tipping bucket
registrations must go through an empirical, usually non-linear calibration function. The
procedure involved in the dynamic calibration of the tipping bucket rain gages is
described. Examples of typical calibration curves are provided. The magnitude of
errors, with regard to measured rainfall intensity, which occur when linear gauge
calibration is used ranged from 2% across the measuring range for the RIMCO gauge
(siphon system delivers constant flow to bucket) to 10% at 5mm min-1 for the
PLUMATIC raingauge.
Hsu Sheng (1990) Tested the rainfall rate error associated with a TBR under various
rainfall intensities. It was found that the rate-error could be as large as a 35%
underestimation of the actual rainfall amount at an upper measurable limit of the gauge,
and close to 8% for less extreme events. However, if a laboratory-based regression
equation was applied to normalize the rate-error on each of 10-minute records, the error
of underestimation could be reduced to as low as 0.8%. Other possible errors were
observed in the laboratory. These were generally small, and difficult to quantify. They
included surface tension effects, and the time taken for the bucket to drain totally. It was
estimated that these effects would amount to no more than 2% of the rate error.
Simic and Maksimovic (1994) Considered the dynamic characteristics of using a siphon
that can be applied to control the movement of rainwater from the collecting funnel of a
raingauge to the compartments of the tipping bucket. Since commercially available
tipping bucket raingauges suffer from non-linearity, a modified siphon was developed.
The dimensions of its elements were designed to meet linearity requirements up to a
certain maximum rainfall intensity. The modified raingauge was shown to achieve
linearity for the chosen range of intensities.
Humphrey et al. (1997) developed an automated method for the dynamic calibration of
TBRs. The system consisted of a programmable pump, datalogger, digital balance, and
computer. Calibration was performed in two steps: 1) pump calibration and 2) rain
gauge calibration. Pump calibration ensures precise control of water flow rates
delivered to the rain gauge funnel; rain gauge calibration ensures precise conversion of
bucket tip times to actual rainfall rates. Calibration of the pump and one rain gauge for
10 selected pump rates typically requires about 8 h. Data files generated during rain
gauge calibration were used to compute rainfall intensities and amounts from a record
of bucket tip times collected in the field. The system was tested using 5 types of
commercial TBRs (15.2-, 20.3-, and 30.5-cm diameters; 0.1-, 0.2-, and 1.0-mm
resolutions) and using 14 TBRs of a single type (20.3-cm diameter; 0.1-mm resolution).
Ten pump rates ranging from 3 to 154 ml min-1 were used to calibrate the TBRs and
represented rainfall rates between 6 and 254mm h-1 depending on the rain gauge
diameter. All pump calibration results were very linear with R2 values greater than
0.99. All rain gauges exhibited large non-linear underestimation errors (between 5%
and 29%) that decreased with increasing rain gauge resolution and increased with
increasing rainfall rate, especially for rates greater than 50mm h-1. Calibration curves of
bucket tip time against the reciprocal of the true pump rate for all rain gauges also were
linear with R2 values of 0.99. Calibration data for the 14 rain gauges of the same type
were very similar. The developed system can calibrate TBRs efficiently, accurately, and
virtually unattended and could be modified for use with other rain gauge designs.
Quality
3.4 Tipping Bucket Raingauge Construction and Use
Cornish and Green (1981) Constructed a TBR from mainly plastic materials more
economically than the equivalent commercially available gauges. Over a two month
trial, 24 h rainfall of 1-69 mm was recorded with an accuracy of 2% when compared to
a storage gauge.
Hughes, Strangeways and Roberts (1993) tested a prototype “champagne glass” and a
production model “cone” gauge. He found that these “aerodynamic” models had the
capacity to improve rainfall catch, and could be exposed well above ground in windy
conditions while still giving reasonable results.
Hanna (1995) Notes that there was to date no international agreement for raingauge
design, only a compromise in the form of the World Meteorological Organisation
(WMO) Interim Reference Precipitation Gauge which was considered to be markedly
different from most national gauge designs. The author suggests that wind turbulence is
likely to statistically outweigh any instrumental errors that may be specific to the TBR,
and that special importance must be attached to more rigorous and quantitative rather
than qualitative site selection. As windshields were considered to be demonstrably of
limited use, a certain amount of natural shielding was suggested to be preferable. Apart
from the wind turbulence problem, reduction of the discrete sampling error of the TBR
was discussed. While there was as yet no absolute standard for rainfall measurement,
there was clearly a need to standardise a suitably effective and practical type of TBR.
The Weighing Tipping Bucket Raingauge (WTBR), a direct-rate measuring instrument,
was believed to offer the greatest future potential once practical versions could be more
widely implemented, while taking care to maintain historical continuity of precipitation
records.
Hellin and Haigh (1998) Reported the performance of a TBR located in southern
Honduras during Hurricane Mitch. The Gauge recorded 698mm in a 41 hour period,
with peak rates approaching 60mm h-1.
3.5 Tipping Bucket Raingauge Data Manipulation and Comparison of Data

Sets
Maksimovic, Buzek and Petrovic (1991) Presented a methodology for correcting TBR
data to take account of non-linearity in rainfall catch at differing rainfall intensities.
They also investigated errors due to “double tipping” and bucket volume, and suggested
corrections.
Sevruk (1996) Compares the 6-year mean annual difference in precipitation values as
measured using the Hellmann (storage) gauge and the tipping-bucket recording
precipitation gauge at the same open site at the Airport of Geneva. The tipping-bucket
gauge was shown to record less precipitation by a mean of 14%. The percentage
differences in daily precipitation amounts depended on wind speed and intensity of
precipitation. The non-linear prediction model developed from these data was based on
576 daily values during the period 1980-1985, and showed that there was a threshold
value of precipitation intensity for each interval of wind speed. This threshold value
increases with increasing wind speed. Below the threshold value a sharp increase in
percentage difference exists with decreasing intensity. It was suggested that
Quality
relationships of a similar nature could have a general application for corrections of the
wind-induced error of precipitation measurement or for adjustments of precipitation
values from one gauge type to the other
Yu et al. (1997) Discuss discrete sampling errors when using TBRs to develop
infiltration – time relationships in small catchments. An analytic expression was derived
for sampling errors associated with a constant rainfall intensity or runoff rate. The
analytic results were then generalized to include real precipitation events when both
rainfall intensity and runoff rate are highly variable. The implications on equipment
design and assessment of model performance were considered.
3.6 Comparison of TBR Types
Marsalek (1981) Carried out laboratory calibrations on three TBRs available in Canada.
– The Canadian Atmospheric Environment Service TBR, The Stevens TBR (Leupold
and Stevens Inc) and Texas Electronics model 6118-1. These were calibrated in the
laboratory by adjusting the volume required to tip the bucket and by correcting the
raingauge output readings. In the volumetric calibration, the effects of raingauge
installation, the wetting of buckets, and the surface tension of the liquid used were
considered. To calibrate the raingauge output, the recorded rainfall intensities were
compared to the actual intensities calculated from the rate of inflow to the raingauge
receiver. Recorded intensities were typically smaller than actual ones, in extreme cases
by as much as 10%. This underestimate was explained by the loss of water during the
bucket rotation. Estimates of these losses were made by timing the bucket movement
for various rainfall intensities. The movement of the bucket from rest to the central
position was found to take 0.3 – 0.6 seconds, depending on rainfall intensity. Finally,
the sensitivity of the TBR output to the variations in the basic design parameters of the
raingauge were studied numerically using the analytical expression derived for the
recorded intensity.
Muller and Van Londen (1983) Compared the accuracy of three raingauges in use in the
Netherlands. The Thies raingauge, the electrical Royal Netherlands Meteorological
Institute (KNMI) rain gage, and the standard KNMI rain gage were compared with each
other and with World Meteorological Organisation (WMO) requirements. The Thies
drop counting system did not meet WMO requirements and was worse than the KNMI
rain gages for the measurement of rainfall amounts. The Thies tipping bucket system
results were so strongly dependent on the bucket cleanliness that errors of 5% were
common. The use of such a system was not recommended. The electrical KNMI rain
gage with float did not meet WMO requirements. The KNMI standard is simple, but an
extra calibration is required if this gauge is used for comparative measurements
Met office (2000) The UK Met office carried out a trial to test the performance and long
term reliability of some of the most common TBRs used in the UK. This was carried out
at the Eskdalemuir observatory between January 1998 and June 2000. Gauges from
Casella, Munro, Didcot instruments and Environmental Measurements were compared
with the Met Office MK5 TBR and a standard five inch check gauge; 90 –95% of daily
values produced by the gauges tested were within 10% of the check gauge reading. The
Munro and Didcot gauges had the least scatter, with 80% of daily readings within 5% of
the check gauge value. A possible seasonal variation in the results from the Casella
gauge when compared to the check gauge was identified. It was suggested that this
Quality
might have been due to the plastic used for the buckets having a different co-efficient of
expansion than that used in the other gauges.
3.7 Comparison with other Automatic Systems
Kreuels and Breuer (1986) Evaluated both a TBR and a drop size disdrometer (device
for the optical detection and measurement of precipitation particles) on days with
different wind speed and gust conditions. Both instruments were operated with a 1-min
time resolution; and were carefully calibrated and system corrected. It was assumed that
differing results would be due to environmental parameters. In the measurements of
heavy storms with 1-min intensities as great as 90 mm/hr, deficits in water amount for
the rain gage were found to be between 0 and 35%. Besides the wind, all other
meteorological parameters were identical for both instruments. The drop size
disdrometer was less susceptible to wind in regard to its installation. The only
explanation for the water deficits was considered to be the influence of wind. To
characterize this influence, it was suggested that the velocity of wind gusts rather than
the mean velocity should be considered. It was suggested that new criteria should be
developed to address the problems identified.]]
Hewston and Sweet (1989) Describe the use of a Weighing Tipping Bucket Raingauge
(WTBR) developed by the Operational Instrumentation Branch of the UK
Meteorological Office. In this device, the tipping bucket mechanism is suspended by a
strain wire between the poles of a magnet. This wire is excited electrically at its
resonate frequency, which is dependent on the tension in the wire due to the mass of the
tipping bucket mechanism. As rain water from the collector enters the bucket, the
tension on the wire increases, and there is a proportional increase in the resonate
frequency of the wire. By measuring this change the rate of rain water accumulation can
be calculated. The sharp change in resonant frequency as the bucket tips is taken into
account in these calculations. The instrument was compared with a standard tilting
siphon gauge on performance requirements for weapons systems trials, where accurate
measurement of rainfall rate is required at a all intensities and at sampling periods as
short as 10 seconds. The WTBR proved to be close to the tilting siphon instrument in
long term accuracy, but was far superior in resolving rainfall intensity down to 10
second increments, allowing accurate determination of rainfall intensity peaks.
Schonhuber, H.E. et. al. (1995) Compared data collected by a 2D video disdrometer and
a standard TBR during winter storms in Austria. In an event of 20mm rainfall, in which
intensity was up to 10mm h-1 the disdrometer recorded 7% less than the TBR. However,
the disdrometer was able to resolve light drizzle several hours before the TBR collected
the first bucket of water.
Nystuen et al. (1996) Rainfall data from six different types of automatic rain gauge
systems were collected for a set of summer rain events and for a set of winter rain
events at Miami, Florida. The rain gauge systems included three types of collection
gauges: weighing, capacitance, and tipping bucket; two gauges that inherently measure
rainfall rate: optical scintillation and underwater acoustical inversion; and one gauge
that detects individual raindrops: the disdrometer. All of these measurement techniques
produced rainfall estimates that were highly correlated to one another. However, each
method had limitations. The collection gauges were affected by flow irregularities
between the catchment basin and the measurement chambers. This affects the accuracy
Quality
of rainfall-rate measurements from these instruments, especially at low rainfall rates. In
the case of the capacitance gauge, errors in 1-min rainfall rates could exceed 10mm/h.
The rainfall rate gauges showed more scatter than the collection gauges for rainfall rates
over 5mm/h, and the scatter was relatively independent of rainfall rate. Changes in drop
size distribution within an event could not be used to explain the scatter observed in the
optical rain gauge data. The acoustical inversion method can be used to measure the
drop size distribution, allowing rainfall classification and estimation of other rain
parameters--for example, reflectivity or liquid water content--in addition to rainfall rate.
The acoustical inversion method had the advantage of an extremely large catchment
area, resulting in very high time resolution. The disdrometer showed a large scatter
relative to the other rain gauge systems for low rainfall rates. This was consistent with
the small catchment area for the disdrometer system.
Grossklaus et al. (1997) described a new optical disdrometer optimized for use in high
wind speed. The minimal detectable size of droplets is 0.35 mm. Each drop is measured
separately with regard to its size and residence time within the sensitive volume. From
the available information, the drop size distribution can be calculated with a resolution
of 0.05mm in diameter either by evaluation of the residence time of drops or by drop
counting knowing the local wind. Rain rates can be determined from the droplet spectra
by assuming terminal fall velocity of the drops according to their size. Long-term
simultaneous measurements of the disdrometer and a conventional rain gauge have been
used to validate this procedure.
Nystuen (1999) Evaluated six different types of automatic rain gauges, including tipping
bucket, weighing, capacitance, optical, disdrometer, and acoustical sensors.
These were deployed for 17 months (September 1993-January 1995) at the Atlantic
Oceanographic and Meteorological Laboratory in Miami, Florida. Different rainfall
conditions encountered during the experiment ranged from winter frontal rainfall to a
tropical storm (Tropical Storm Gordon). Overall, all of the rain gauges performed well,
with inter-correlation’s of the order 0.9 or better using 1-min rainfall rates and biases of
less than 10%; however, each showed limitations under different rainfall situations. In
particular, under extremely heavy rainfall rates (over 100 mm/h), the disdrometer and
tipping bucket rain gauges biased low, while the optical rain gauge biased high. Under
light rainfall rates (under 2 mm/h), the capacitance and tipping bucket rain gauges
showed significant instrument noise using a 1-min sampling interval. The optical gauge
was sensitive to the relative proportion of small to large raindrops within the rain. The
raindrop distribution parameter N0, the coefficient of the exponential fit to the drop size
distribution, could be used to predict the optical gauge bias. When N0 is large (relatively
more small drops), the optical gauge biases high, and when N0 is small (relatively more
large drops), the optical gauge biases low. The acoustic rain measurement showed
significant variability when compared to the other gauges. The acoustic measurement is
very sensitive to the presence of very large raindrops (over 3.5mm diameter) as these
raindrops are extraordinarily loud underwater and prevent the smaller drop size
populations from being heard and accurately counted when they are present. While the
range of wind speeds encountered during the experiment was limited, wind did affect
the performance of several of the gauges. At higher wind speeds (over 5ms-1), the
disdrometer and acoustic rain gauges biased low and the instrument noise of the
capacitance gauge increased significantly.
Quality
3.8 Discussion
There is general agreement among the authors of the papers reviewed that one of the
most important factors affecting raingauge performance is exposure to wind. This effect
increases with height above the ground surface, such that a gauge installed at 305mm
above the ground surface may have a reduced catch of between 3% (lowland site and
20% (upland site) compared to a gauge installed at ground level (BS 7843 1996). This
factor may give rise to differences in catch between TBRs and check gauges when there
is a relative difference in height between them. The British Standard, Met Office and
WMO all suggest recommended methods of siting gauges to reduce exposure. These
involve reducing or modifying wind flow over the collector by either locating at ground
level (“pit gauge”), or constructing an earth (turf) bank to smooth wind flow. The
relationship between exposure error (reduction in rainfall catch) and observed
conditions is complex, although several researchers have developed models to describe
the process (Robinson and Rodda 1969, Folland 1988). Recommended solutions for
data correction are also given by the WMO. Design factors may reduce the effect of
exposure, with evidence from several researchers (Folland 1988, Hughes et al. 1993)
that “aerodynamic” gauges are measurably better than standard shapes. Commercially
available raingauges have been produced (e.g. Environmental Measurements Ltd) using
these designs.
A number of factors specific to the TBR have been identified. These include errors in
recording caused by evaporation from the bucket, time of tipping and “bouncing” of the
bucket. All of these parameters are likely to produce only small errors, and many can be
reduced in a well-designed gauge. The most important effect identified in many of the
reports reviewed relates to the relationship between TBR calibration error and rainfall
rate. This is most marked during heavy rainfall, when water flowing in to the buckets
while they are tipping can represent an error of up to 10 – 30% compared to a co-
located storage gauge (BS 7843 1996 and others). The nature of this rainfall/rate error
relationship has been described as non-linear (Calder and Kidd 1978, Humphrey et al.
1997), but in the British Standard is described as linear between 0 and 100mm/hour.
The solution to this problem favoured by some national meteorological services (UK
Met Office, Canadian Atmospheric Environment Service), has been to apply a
correction factor derived from the ratio of mean daily TBR readings compared to check
gauge readings. This approach relies on the availability of a daily read check gauge, and
would not suitable for correcting event based sampling where accuracy within small
time steps is required. The alternative, outlined by a number of researchers (e.g. -
Calder and Kidd 1978, Niemczynowicz 1986, Humphrey et al. 1997), is to carry out a
dynamic calibration of the TBR in order to quantify the error over a range of rainfall
intensities. A number of the papers reviewed suggested numerical methods describing
this relationship.
Alternative automatic systems reviewed include developments of the TBR such as the
weighing TBR (Hewston and Sweet 1989). This instrument has the potential to achieve
a much higher resolution than the standard TBR, although retains the possible
inaccuracies associated with the tipping mechanism. Cost may preclude its use in
extensive networks, as commercial models are three times as expensive as standard
TBRs (C Addis 2001, personal communication). Optical/video measurement of
precipitation using a disdrometer to measure and evaluate individual particles has been
developed in the last 20 years, often in conjunction with weather radar systems (to
Quality
provide “ground truth” information). Several researchers have shown good correlation
between these systems and direct measuring raingauges including TBRs (e.g.
Schonhuber, H.E. et al. 1995 Nystuen 1999). Disdrometers have the added advantage of
detecting and differentiating the nature of precipitation (rain, snow, hail etc) as well as
its quantity, and can operate in all wind conditions. At present disdrometers are used as
tools for investigating the nature of precipitation, and are usually close to the necessary
data processing facilities.
Quality
4 LABORATORY STUDIES
4.1 Introduction
flood warning and water resource planning activities. As part of this work they have
identified discrepancies between the records obtained from tipping bucket raingauges
(TBRs) and conventional storage gauges. The problem that has been identified, namely
under-recording by TBRs, has particular implications for work relating to water
resources since, whilst the error may be relatively small on any one event, the
significance may increase as rainfall totals are accumulated over longer periods. The
Agency standard gauge has a calibration coefficient of 0.2 mm per tips as standard (i.e.
1 tip = 0.2 mm). This bucket 'calibration coefficient' of 0.2 mm is used for each tip over
the whole range of rainfall intensities measured; i.e. they do not apply a dynamic
calibration coefficient. When dynamic calibration is used a range of calibration
coefficients are determined and applied to better reflect the volume of water per tip at
different intensities.
Work was needed to determine whether the discrepancy between the two systems was
systematic and could be linked to specific factors, or occurs randomly. If the error was
systematic, and was linked to some specific variable, such as a deficiency of the
calibration method, data correction may be possible. The factor that was thought to
most likely to be critical to calibration was rainfall intensity, although temperature may
also play a part. It has been reported in the literature (Calder and Kidd, 1978) that
dynamic calibration of TBRs was required to achieve both high resolution and accuracy.
They stated that in general, a non-linear relationship existed between flow rate and the
tipping rate of a tipping-bucket gauge. This difference arose from the fact that it took a
finite but fixed time for the bucket to tip. As a consequence of this, an amount of
rainfall could fall into an already full bucket during the time taken to tip. This amount
increased with the rainfall intensity and as a consequence rainfall is underestimated.
The use of a calibration curve covering a range of intensities has been recommended by
a number of workers, for example, Niemczynowicz (1986). Other methods have also
been proposed to either avoid or correct for this non-linearity, including the use of a
siphon that is incorporated between the receiver and the bucket and is designed to
empty at a constant flow rate into the bucket. However, such solutions, while providing
a better measurement of total rainfall, smooth out the measurement of peak intensity.
Raingauges supplied by different manufacturers may perform differently and

performance may also change with time depending, for example, on the robustness of
the equipment and the environmental conditions under which it is used. Depending
upon the type of material used to construct the gauge, expansion and contraction due to
changes in operating temperatures may also influence the accuracy of measurement. In
this context, it may be useful to consider the effect of changes in temperature on the
method used for calibration. The main effect, however, will be derived from the change
in density of water with temperature. As the density of water decreases, a greater
volume will be required to achieve the critical mass necessary for the bucket to tip. The
impact of change in temperature on rainfall measurement is shown in Figure 4.1 for a
gauge calibrated at 20°C.
Quality
0.2
0.15
% Error in calibration due to expansion of water
0.1
0.05
0
0 5 10 15 20 25 30
-0.05
-0.1
-0.15
-0.2
-0.25
-0.3
Temperature (°C)
(Plot derived from information on changes in water density with temperature reported in
Weast, 1971)
Figure 4.1: Percentage error in rainfall measurement as a consequence of the

difference between calibration temperature (20°C) and operational
temperature due to changes in density of water.
4.2 Detailed Methodology
This was undertaken using 24 new rain gauges of 8 different types which are identified
by numbers 1 to 8 to ensure confidentiality.
General Gauge Characteristics

type 1 – conventional metal TBR manufactured in UK
type 3 – conventional metal TBR imported from outside Europe
type 4 – stainless steel aerodynamic TBR manufactured in UK
type 5– conventional metal and plastic TBR made within Europe
type 6 – all stainless steel conventional metal TBR imported from outside
Europe
type 7 – plastic aerodynamic TBR manufactured in UK
4.2.1 Evaluation of calibration methods

Each rain gauge was unpacked, assembled according to instructions (where provided),
mounted on a level base in a 20°C constant temperature room and then the calibration
checked using the two different methodologies.
Quality
Burette Method
The buckets on each gauge were labelled A and B and the volume required to make
both the A and B bucket tip was measured by dripping in water from a 20 ml burette
following the standard procedure (BS 7843 2.1, 1996).
Constant Rate Method

A special rig was set up whereby eight gauges could be calibrated simultaneously. The
rig used a Watson Marlow 505U peristaltic pump fitted with two 4-channel pump heads
attached to the drive shaft. This pump/head configuration is the standard equipment
used by the Environment Agency for calibration of raingauges using the constant rate
method. The target range for the rainfall intensity at which the gauges were calibrated
was between 10 and 12 mm hr-1. Since the volume required to tip the bucket for each
gauge is a function of funnel diameter, a different flow rate was required for each gauge
to meet the intensity required for calibration. As only a limited number of pipe sizes
were available for the peristaltic pump, each type of gauge was calibrated at a slightly
different rate. The pipe sizes used were selected such that for the chosen speed of
rotation (48 rpm) all gauges received water at a rate that fell within, or very close to, the
target range. The rotation speed was chosen as, combined with the available pipe
diameters, it allowed calibration of all eight gauges simultaneously at an intensity
within or very close to the target rate. Each calibration test was run for two hours.
The gauges were set up on a specially designed bench such that the delivery from each
bucket could be collected in a beaker. All the gauges were connected to a NEWLOG
datalogger set to record the number of tips in each 5 minute time interval.
The specific procedure used was as follows:

Prior to testing, each gauge was wetted by running 0.5l of water through it. Each intake
pipe from the pump was placed in an individual reservoir and the pump was then run
until all air had been expelled from the system. The pump was then stopped and the
reservoir refilled and weighed. A clean labelled and pre-weighed beaker was then
placed under the collection point below each bucket. The datalogger was then started
and the pump set to run for two hours using a digital time switch. At the end of the test
the reservoir and collection beakers were reweighed. Gauges were adjusted and
calibration checked repeatedly until they met the criteria that the calibration coefficient
was within 2% of 0.2mm, and the amounts of water required to tip each bucket were
within 5% of each other. A further two runs were then made, without adjustment and
the mean of these three runs was taken to be the calibration coefficient for the gauge
concerned. This procedure produced, on occasions, mean calibration coefficients that
fell slightly outside the acceptable range of +/- 2% error of rainfall measurement. The
individual calibration coefficients determined for each gauge will be applied to the data
from each gauge when deployed in the field study.
4.2.2 Effect of temperature at time of calibration on calibration coefficient

obtained using the constant rate method.
Experimental design
The 24 rain gauges were divided into three gauge sets (blocks) each containing one
example of each type of gauge. The first and third sets of gauges were set up in the
constant temperature at laboratory at 20°C and calibrated using the constant rate method
(as detailed above). As previously stated, once a gauge had come into calibration, two
Quality
further measurements were made to provide three values that could be used to produce a
mean calibration coefficient for the gauge.
Statistical analysis of data

The statistical analysis was undertaken in accord with the experiment design, which can
be regarded as a criss-cross design, in which rows of the data can represent temperature,
and columns can represent gauges. The triplicate readings were nested in the column-
row structure and are not replicates of gauges but replicate estimations of the response
of a single gauge in each block. Thus the temperature, gauge, and gauge interaction had
1,7 degrees of freedom (when 8 gauges were used) and 7 degrees of freedom
respectively. The gauges, temperature and interaction have their separate residual errors
and associated p-values and the within-gauge variation was represented by the mean
square error of the blocks/row/column/rep stratum. Of particular importance in data
interpretation, was the gauge/temperature interaction. If there was a statistically
significant interaction then this should be scrutinized and any temperature effect
questioned. Since each gauge had been set to a unique, though similar coefficient, a
significant gauge effect has no relevance. This approach could only be applied to the
combined data from sets one and three where the direction of temperature changes had
been the same. An analysis using the data from all 3 blocks was undertaken where the
block 2 data were extracted for the sequence 20° to 4° degrees Celsius and merged with
similar data from block 1 and 3 data; this just compared the difference between the two
temperatures. Analysis of variance was undertaken using GENSTAT (Alvey et at,
1972) and regression analysis using STATISTICA.
4.2.3 Effect of rainfall intensity on gauge calibration coefficient

The relationship between calibration coefficient and the simulated rainfall intensity at
which it was measured were determined for each gauge type. The rainfall intensities at
which calibrations were undertaken encompassed the range 1- 100 mm hr-1 with target
intensities of 100 mm hr-1, 50 mm hr-1, 40 mm hr-1, 30 mm hr-1, 20 mm hr-1, 10 mm hr-1,
5 mm hr-1 and 1mm hr-1. In each case, the pump settings used and the actual rainfall
rate achieved were recorded. All dynamic calibrations were undertaken in an
environment where the temperature was maintained at a constant 20°C. Measurements
at each rainfall intensity were replicated three times for each gauge and three gauges
from each manufacturer were calibrated. The numbers of tips were recorded, as was the
outflow from each bucket. Tests were either carried out using the Watson Marlow
eight channel pump or, at intensities that required a flow rate in excess of 30 ml minute-
1
, using an Altec single channel peristaltic pump. The test procedure used was identical
to that described in section 5.2.1.
4.2.4 Long-term reliability tests

Two replicates of each gauge were subjected to a simulation which generated the
equivalent number of tips to that which would be obtained from 10 years use in a
climate with 1500 mm per annum rainfall. A special rig was used for the long-term
tests, which utilized a range of peristaltic pumps manufactured by Autoclude. This
allowed water to be input to each gauge at a constant and pre-defined rate. The rate
selected was as close to 75 mm hr-1 of rainfall as could be obtained given the limitation
imposed by the available pump speeds and tube diameters. The rate of 75 mm hr-1 was
chosen to minimize the time take for a single years run. This enabled the completion of
the test within the timescale required so that the gauges could be deployed for field
testing by January 2002.
Quality
Prior to starting the tests each pump/tube combination was run in for minimum of 12
hours. The pumps were controlled by a time switch that stopped the pump after periods
equivalent to 1 year, 2 years and 10 years rainfall so that spot checks on the calibration
of the gauge could be carried out. The number of times each tipping bucket tipped was
recorded on a Newlog datalogger running at a 5 minute time interval. A record was kept
of any breakages that occurred or faults that developed. If any gauge failed to operate,
for a reason that required repair to the gauge itself, rather than from external influences,
then it was deemed to have failed testing and testing of that gauge ceased.
4.2.5 Observations on gauge performance and ease of use

A record was kept of specific problems and difficulties encountered when calibrating
the gauges;
4.3 Results
4.3.1 Evaluation of calibration methods

A comparison of the results of applying the two calibration methods to the gauges under
test is shown in Figure 4.2. The comparison uses the measured calibration coefficient s
to calculate the amount of rainfall needed to achieve 7500 tips, which if the calibration
coefficient is 0.2 should be 1500mm. This data shows that not only doe the two
calibration methods clearly give a different result, but also the considerable variability
in actual calibration of new gauges as received from the manufacturer. Data for gauge
type 8 are not presented due to an error in assembly which compromised the results of
this test for this gauge.
1600
1575 Burette method

Constant rate method
1550
1525
1500
Rainfall needed to cause 7500 tips
1475
1450
1425
1400
1375
1350
1325
1300
1275
1250
Type 1 Type 2 Type 3 Type 4 Type 5 Type 6 Type 7
Figure 4.2: Relationship of rainfall required for 7500tips and calibration method
Quality
4.3.2 Effect of temperature at time of calibration on calibration coefficient
obtained from constant rate method
The effect of changing temperature (20°C to 4°C and 4°C to 20°C) and movement on
measured calibration coefficients of the gauges are summarized in Table 4.1.
Differences between the calibration coefficient measured at 20°C and that measured at
4°C are small in most cases.
Table 4.1: Summary of temperature testing based on all 3 blocks

Gauge type
No 1 No 2 No 3 No 4 No 5 No 6 No.7 No. 8
Mean calibration
coefficient at 0.191 0.189 0.197 0.196 0.201 0.226 0.201 0.208
20°C
Mean calibration
coefficient at 0.196 0.196 0.195 0.195 0.193 0.196 0.199 0.202
4 °C
% Difference
between 2.91 3.84 0.91 0.17 4.14 15.23 0.95 2.85
4°C and 20°C
If moving the gauges had no effect, then returning the gauges to their starting
temperature should return the calibration coefficients to their original values; this was
clearly not the case, but in most cases differences due to this factor were small. Despite
the minimal impact of temperature on measured rainfall; the data presented have been
corrected to take account of the changes in the density of water that occur between 20°C
and 4°C. The changes represented here thus represent the impact of changing
temperature on the instruments themselves, or of moving the instruments. Clearly gauge
type 6, stands out as the being the most affected of the gauges. After the gauge type 6,
the next most affected by temperature changes was type 5. It is interesting to note that
the type 6 gauges were constructed entirely of stainless steel and the mechanism of type
5 was predominantly plastic.
Analysis of variance, using a criss cross design, of the combined data obtained from
gauge sets 1 &2 showed that the difference in calibration coefficients that occurred as a
consequence of moving the gauges (initial compared to end) was not statistically
significant (P>0.5). However, the difference due to changes in temperature were (P<
0.05). When the data relating to the type 6 gauges were excluded from the analysis the
difference caused by temperature changes was no longer statistically significant. Thus
although there were clearly differences arising from both factors in the case of all of the
gauges, the variation due to movement falls within the repeatability of measurement of
the calibration coefficient (for the gauges).
Analysis of variance using all three sets of data, each as the blocks, and looking at the
difference between first temperature and second temperature i.e. 20-4°C and 4-20°C
showed that there was a significant gauge-temperature interaction (P=0.022). When the
data from the type 6 gauge were excluded from the analysis, the significance level
remained similar (P=0.017). Examination of the means suggest that this significant
interaction arises from the divergent behaviour of the gauges e.g. type 1 calibration
coefficient goes from 0.196 at 20°C to 0.191 at 4°C, whereas the type 5 goes from 0.193
Quality
at 20°C to 0.201 at 4°C. This divergence amplifies the gauge interaction causing the
difference to be significant. The actual magnitudes of the changes for the gauges
(+0.005 to -0.008), other than the type 6 are similar to the standard error of difference
(0.004899) and this suggest that these differences are not truly significant. Thus only in
the case of type 6 gauges was calibration significantly influenced by temperature
changes.

The % errors generated by applying a calibration coefficient determined at 10-12 mm
hr-1 over a range of simulated rainfall intensities are presented in Figure 4.3. The data
from all the gauge types shows a similar trend, with some overestimation of total
rainfall at low intensities and then underestimation as rainfall intensity goes above that
at which the gauge was calibrated. Individual gauge specific relationship, between %
error and rainfall intensity, can be calculated (Table 4.2) although the strength of these
relationships varies considerably. A single generalized relationship was also calculated
using the data from all the gauges and this is shown with the data in Figure 4.4
Table 4.2: Relationship between Rainfall intensity and % error compared to

calibration at 10 mm hr-1
Gauge type Polynomial Regression R2

1 Y = 0.164741 - 0.0891X + 0.000204X2 0.228
2 Y = 0.338803 - 0.102475X + 0.000163X2 0.649
3 Y = 0.937434 - 0.128398X + 0.000294X2 0.775
4 Y = 0.522824 - 0.123834X + 0.000565 X2 0.397
5 Y = 1.51539 - 0.100015X + 0.000055X2 0.687
6 Y = 0.765467 - 0.159311X + 0.000687X2 0.628
7 Y = 2.28896 - 0.339171X + 0.00229EX2 0.384
8 Y = -0.24 - 0.0210X -0.000667X2 0.615
Y = 0.523 - 0.107X + 0.000254X2 0.519
All gauges
Quality
15
0 20 40 60 80 100 120 140 160
10
-5
% Error
-10
-15
-20
-25 Outlier ignored
-30
Simulated rainfall intensity (mm hr-1)
Figure 4.3: Overall Relationship between % error and rainfall intensity for gauges
calibrated at 10-12 mm hr-1.

The mean error in the measurement of rainfall as a consequence of drift in calibration
coefficient over a period of 10 years simulated rainfall is shown in figure 5.3. No
breakages or failure of any kind were recorded on any gauge during the test. It is
important to note that in a number of cases most of the drift occurred within the first
year.
On average the drift in the calibration did not result in an error of measurement of more
than +/- 4.5%. Drift was not linear with time and with some gauges the impact on
percentage error of measured rainfall was ca. 3% after only one-year and then
subsequently decreased again from 2-10 years. In the worst individual case (replicate 2
of Gauge 6), the error in measurement increased by nearly 8% in the first year. Some
gauges (type 5 & 7) were particularly resistant to drift. However, as only two gauges of
each type were tested it would be inappropriate to recommend that these gauge types
can be used in the field for longer than the others without calibration
Quality
5
% Error compared to "as calibrated"

3
2 Type 1
Type 2
Type 3
1
Type 4
Type 5
0
Type 6
0 2 4 6 8 10 12
Type 7
-1
Type 8
-2
-3
-4
Simulated years
Figure 4.4: Change in % error in measurement due drift in calibration with time
(based on simulated rainfall)

All of the gauges, with the exception of types 5 and 7, could be placed directly on the
calibration rig without any modification. Neither type 5 nor 7 were constructed in such
a way whereby the water could be collected easily individually from each of the
buckets, thus making verification of the balance difficult. In the case of the type 7
gauges this was relatively easily overcome by drilling additional holes in the base of the
raingauge. However, type 5 did not lend itself to this modification. The only way in
which it was practicable to collect individually the water from each of the buckets was
to dismantle the gauge and rotate the tipping bucket mechanism on its mounting
pedestal so that it was a no longer located immediately above the base plate. This is not
regarded as particularly satisfactory solution as it resulted in the gauge being calibrated
in a configuration other than that in which it was being used.
type 1
The small adjustment nuts were fiddly, making fine adjustment difficult.
type 2
This was provided with 5mm adjustment bolts which were easy to adjust, making it one
of the first gauges to be brought into calibration. However, gauge number 2 was found
to lock up in central position with water pouring either side of the bucket divider. When
the gauges was examined by the supplier’s engineer, following failure under test, it was
found that the reed switch and magnet had been set too closely together during the
manufacturing process.
type 3
Too many adjustment fasteners (three) but a very good bucket design, water empties
completely, due to small wire on the bucket.
type 4
Very easy to adjust but very shallow bucket angle plus a rough texture of the surface
leads to water retention after tip.
Quality
type 5
Easy to adjust, but no locking nuts. Some collection losses due to splashing and had to
be modified to allow calibration.
type 6
This gauge was very difficult to adjust due to poor quality plastic screws that wore very
rapidly. The bucket empties well but when filling, water almost runs over central divide
before tipping, which could present problems if the bucket balance starts to drift during
use.
type 7
This gauge appeared good to adjust at first but it was difficult to get the calibration
exactly right. Responded erratically to adjustment this could be due to asymmetry in
hemispherical stops. Good visual markers on stops aided adjustment. The bucket holds
a lot of water after tipping which could be lost due to evaporation causing small error if
subsequent tip is sometime later.
type 8
This was very easy to adjust, however, in the absence of an instruction manual; the
bucket can be incorrectly inserted, even by technical staff used to setting up TBRs,
causing it to be a long way out of calibration. One of the gauges supplied was found to
have a defective reed switch. The replacement of this was found to be quite difficult.
4.4 Discussion
4.4.1 Effect of temperature at time of calibration on gauge calibration coefficient

and its impact on rainfall measurement
The type 6 gauges (all stainless steel construction) were particularly susceptible to
temperature changes and the calibration coefficients of these gauges also changed
considerably due to movement. The impact of the change in calibration coefficient due
to temperature would have the effect of causing over 8% error in rainfall measurement
from a gauge calibrated at 20°C but deployed at lower temperatures (4°C). The exact
cause of this change is uncertain but it must assumed to be due to expansion changing
the amount of force required to tip the buckets. It may be that the all stainless steel
design makes this gauge more susceptible to changes. The changes in the other gauges
due to temperature were of a much smaller magnitude and had a much smaller impact
on the error of measurement. However, even changes as small 1% which are observed
with most gauges, could be sufficient to effectively take a gauge out of the allowed
calibration tolerance. The error due changes in calibration coefficient due to
temperature will be added to that inherent in the use of a standard 0.2mm per tip
coefficient to convert tips to rainfall. Thus if the actual calibration coefficient of a gauge
is for example deviates from 0.2 by -1.8% at 20°C and there is a further drift of -1% due
to temperature change from 20° to 4°C then the total error will be -2.8% at 4°C. Thus if
gauge specific calibration coefficients were used to convert tips to mm of rainfall rather
than a standard 0.2mm, the total error will be reduced since the inherent error due to
difference between actual and assumed calibrations will be removed. An alternative
approach would be to ensure that the inherent error was always in the opposite direction
to any temperature generated errors, i.e. if temperature error for specific gauges is
Quality
negative calibrate gauges to between 0 and +2% rather than +/- 2%. In this way the
errors would cancel each other out.
The impact on error of measurement was inverse when the gauges were calibrated at
4°C and used at 20°C. Thus an alternative approach to error reduction might be to
calibrate at the most common temperature at which rainfall occurs when deployed in the
field. However, as this temperature would vary spatially and temporally this approach
might prove to complex to implement. Consideration of the range of temperatures
encountered in the UK suggests that 20°C should certainly be the maximum
temperature at which gauges are calibrated if errors are to be minimized.

All of the gauges tended to increasingly underestimate rainfall at higher intensities,
which is in line with the findings of other workers. Niemczynowicz (1986) tested three
different types of gauges and found, for example, under reading of 10% at 300 mm hr-1
with the PLUMATIC raingauge. Similarly, Marsalek (1981) carried out laboratory
calibrations on three TBRs available in Canada and found that recorded intensities were
typically smaller than actual ones, in extreme cases by as much as 10%. The
mechanism by which this under recording occurs is that due to the finite time it takes
for a bucket to tip some water will fall into a full, but already moving, bucket (Calder
and Kidd, 1978). Relationships between intensity and percentage error, for both
individual gauges and the overall data set, were non-linear. This is in accordance with
the findings of Calder and Kidd (1978) who stated that in general, a non-linear relation
exists between flow rate and the tipping rate of a tipping-bucket gauge. Further support
for the non-linearity of the relationship comes from the work of Humphrey et al (1997)
who tested 5 types of commercial TBRs and found that all exhibited large non-linear
underestimation errors (between 5% and 29%) that decreased with increasing rain gauge
resolution and increased with increasing rainfall rate, especially for rates greater than
50mm h-1.
The impact of this intensity dependence of calibration on rainfall records will vary
according to the frequency with which higher rainfall intensities occur at the location
where a specific gauge is being used. The variability in the data suggests that a large
number of determinations will be required to develop a robust relationship between
intensity and percentage error for a specific type of gauge. However, given the
relatively small number of different gauge types used by the Environment Agency this
should not prove an insurmountable problem. Should a decision be made to adopt error
correction, to offset the affect of the intensity dependence of the calibration, then it is
recommended to undertake a pilot study on a single gauge type. A carefully determined
correction factor could then be used to correct the data to take account of intensity
effects. This test should be possible on one of the datasets where the agency have
identified a discrepancy between the check gauge and tipping bucket totals.
An alternative solution to the problem could be to use TBRs with buckets with a larger
volume per tip, for example 0.5 mm. This would decrease the tipping rate and hence
reduce the error. However, adopting this solution would tend to increase evaporation
losses during summer due to the increased residence time in the bucket following small
amounts of precipitation.
Quality
The fact that the long-term reliability tests produced no failures suggests that it is
corrosion in the outdoor environment that is the key cause of failure rather than just
repeated tipping. However, the tests have provided useful data supporting the need for
regular calibration of tipping bucket raingauges if first class accuracy is to be
maintained. Based on the rate of drift, an annual calibration should be adequate in areas
where rainfall is less than 750 mm per annum. In areas where there is higher rainfall, or
if a high degree of precision were required, then biannual calibration would provide a
better control as in many cases most drift occurred during the first year's simulation.

Unless a special calibration rig can be obtained from the manufacturer the use of the
type 5 gauge is not recommended due to the need to calibrate in a different
configuration to that in which it is used. The gauge would also be substantially
improved by fitting with more robust adjusting screws. In general manufacturers need
to look to design gauges to make adjustment easy, thus an easy to use gauge is one that
that has the largest adjusters. Other things to be avoided are any asymmetry in the
shape of the stops, which control the bucket volumes as this causes erratic calibration.
The gauges where the bucket surface was rough tended to retain more water after the
bucket had tipped than those where it was smooth. The effects of this will be to slightly
increase the volume required at the next depth if sufficient time passes to allow this
small volume of water to evaporate. However, the amounts of water concerned are so
small that the impact of this "increased wetting up error" on the rainfall record will be
minimal.
4.5 Overview of a Calibration Methods
The continuous rate method has a significant advantage in that it provides an approach
which is closest to the mode of operation of the gauge during rainfall and the calibration
coefficient at least takes account of losses that occur during tip time at the intensity at
which it is calibrated. As a consequence, underestimation of rainfall due to losses
during time it takes the bucket to tip are likely to be greater with gauges which are
calibrated by the burette method. However, it could be argued that the burette methods
would provide a better calibration for very fine low intensity rain than calibration at ca.
10-mm hr-1.
The standard Environment Agency constant rate method is based on filling a calibration
rig with a known volume of water from a volumetric flask and then running the pump,
which supplies water to the gauges, until it is empty as described in the Environment
Agency, Hydrometric Manual. However, using this method, it is difficult to be sure of
the exact amount of water that has been put into the gauge since the residues retained in
the system may be variable. The method used by ADAS in this study (section 3.2)
overcomes this problem by making a direct measurement of the water input which can
be compared against the amount collected from the buckets with confidence.
Neither the burette method nor the constant rate method themselves would be expected
to be particularly affected by temperature changes. In the case of the constant rate
method this is because a direct measurement of the mass of water delivered to the gauge
is made. Any changes in pump delivery rate that might occur due to temperature
Quality
changes are irrelevant unless the errors become gross and this is unlikely. In the case of
the burette method the burettes themselves are normally calibrated for use at 25 degrees
C. Decreasing temperature will tend to reduce slightly the volume delivered. However,
to some extent this will be balanced by the change in the water density. In any case it is
a quite simple matter to calibrate the mass of water delivered by the burette against
temperature.
The greatest scope for error in the constant rate method is splashing losses between the
buckets and the collection vessels. However, controls can be put in place such that this
is detected and kept within acceptable limits. If larger diameter gauges are tested, such
as the FSS500, it is suggested that a larger volume of water be used for the test to
increase the number of tips counted. However, this will also increase the time taken for
the test, which may have logistical implications. If the same of volume of water is used
with a large diameter gauge as a small one, then the number of tips counted is
significantly reduced. The greater the number of tips counted, the more precise the
estimate of the calibration coefficient is likely to be. If it were feasible to set the gauges
up to run unattended on the calibration rig, then an increase in the volume of water used
would be desirable for all gauges, as this would increase the number of tips counted.
4.6 Specific Conclusions arising from Laboratory Tests
The constant rate method of calibration should provide a calibration coefficient that is
more robust at higher intensities than one measured by the burette method.
Gauges calibrated by the constant rate method will underestimate at higher rainfall
intensities and better precision could be obtained by use of a full dynamic calibration
The calibration coefficient of tipping bucket rain gauges varies with temperature
although in most gauges the change is small
The influence of temperature on the calibration methods is likely to be small and can be
corrected for.
Tipping bucket rain gauges need at least annual calibration and possible biannual in
high rainfall areas.
Quality
5 FIELD TESTING OF GAUGES
5.1 Introduction
flood warning and water resource planning activities. As part of this work
discrepancies have been identified between the records obtained from tipping bucket
raingauges (TBRs) and conventional storage gauges. The problem that has been
identified, namely under-recording by TBRs, has particular implications for work
relating to water resources since, whilst the error may be relatively small on any one
event, the significance increases as rainfall totals are accumulated over longer periods.
The Agency standard gauge has a calibration coefficient of 0.2 mm per tips as standard
(i.e. 1 tip = 0.2 mm) which is measured at a rainfall rate of ca. 10 mm hr-1. This
measured bucket 'calibration coefficient' of 0.2 mm is used for each tip over the whole
range of rainfall intensities measured; i.e. they do not apply a dynamic calibration
coefficient. When dynamic calibration is used a range of calibration coefficients are
determined and applied to better reflect the volume of water per tip at different
intensities.
Work was needed to determine whether the discrepancy between the two systems was
systematic and could be linked to specific factors, or occurred randomly. If the error
was systematic, and was linked to some specific variable, such as e.g. a deficiency of
the calibration method or rainfall intensity, data correction may be possible.
The factor that was thought to most likely to be critical to calibration was rainfall
intensity, although temperature may also play a part. It has been reported in the
literature (Calder and Kidd, 1978) that dynamic calibration of TBRs was required to
achieve both high resolution and accuracy. They stated that in general, a non-linear
relationship existed between flow rate and the tipping rate of a tipping-bucket gauge.
This difference arose from the fact that it took a finite but fixed time for the bucket to
tip. As a consequence of this, an amount of rainfall could fall into an already full
bucket during the time taken to tip. This amount increased with the rainfall intensity and
as a consequence rainfall is underestimated. The use of a calibration curve covering a
range of intensities has been recommended by a number of workers, for example,
Niemczynowicz (1986). Other methods have also been proposed to either avoid or
correct for this non-linearity, including the use of a siphon that is incorporated between
the receiver and the bucket and is designed to empty at a constant flow rate into the
bucket. However, such solutions, while providing a better measurement of total
rainfall, smooth out the measurement of peak intensity.
The other main factor thought to be likely to influence catch was over-exposure (to
wind). This is identified as the most important factor contributing to loss of catch by all
rain gauges (BS 7843 2.1, 1996). Similarly, the World Meteorological Organisation
(1994) states that the suggested aim when selecting a site for a gauge would be to
choose a location that gives the minimum possible wind speed over the gauge orifice,
without blocking precipitation by surrounding terrain. The impact of exposure to wind
on rain gauge catch has been well documented in the literature e.g. Folland (1988) who
suggested a new design of raingauge (“flat champagne glass gauge”) which aimed to
substantially lower the systematic losses of rainfall volume in the high winds commonly
found over exposed areas of land. Tests of a prototype “champagne glass” and a
Quality
production model “cone” gauge found that these “aerodynamic” models had the
capacity to improve rainfall catch, and could be exposed well above ground in windy
conditions while still giving reasonable results (Hughes, Strangeways and Roberts,
1993). Therefore, as “champagne glass” tipping bucket rain gauges are commercially
available and the Environment Agency have purchased a number model for use in high
exposure situations there was also a need to understand what impact the use of this type
of gauge may have on the rainfall record. In addition the Agency own a number of
other makes of gauge and needed to identify the key characteristics that make a TBR
function well
Therefore a field trial was established to compare, in the field, the performance of eight
different patterns of TBR against that of a standard meteorological office 5" gauge.
5.2 Materials and Methods
5.2.1 Experimental site

The experimental site was located at Eskdalemuir Observatory in Dumfries and
Galloway, which is situated in the Southern Uplands of Scotland (figure 5.1). The
Observatory, which is 3½ miles NNW of Eskdalemuir Church, occupies a site of about
11 acres on a rising shoulder of moorland, bounded, on the East, by the Ettrick and
Selkirk road, on the West by a small stream (Davington Burn) and on the South by the
village of Davington.
The hillside in the vicinity of the Observatory slopes generally from NW to SE, the
mean height above sea level being 800ft or 244 metres. Cassock Hill, slightly more than
a mile distant to the NW, is (367m) while the benchmark at Davington School, ¼ mile
to SE is 699ft (213m) above MSL. To the East, the ground slopes fairly rapidly to the
valley bottom. The height at the River White Esk, at a point due East of the Observatory
and about mile away being 675ft(208m) above MSL. Beyond the river, Dumfedling
Hill rises to a height of nearly 1200ft (366m) above MSL. Some 4-5 miles away to the
North is a high ridge, the highest point of which is Ettrick Pen (to NNW)
2270FT(700M) above MSL. Rather more than half a mile to the West, the ground rises
to 1040ft (317m) and reaches nearly 1200ft(366m) a further half mile on. To S and SSE,
The Observatory commands a view of the White Esk Valley as far as Hartmanor, 4
miles away. Beyond, the upper slope of Cauldkine Hill, about 10 miles distant, is
visible.
The National Grid Reference of the site is NT 32356026 and the station is located at an
altitude of 242 m. Long-term average annual rainfall for the site is 1567mm.
Quality
Image produced from the Ordnance Survey <a href="http://www.ordnancesurvey.co.uk/getamap">Get-a-map</a> service. Image
reproduced with kind permission of <a href="http://www.ordnancesurvey.co.uk/">Ordnance Survey</a> and <a
href="http://www.osni.gov.uk/">Ordnance Survey of Northern Ireland</a>.
Image produced from the Ordnance Survey <a href="http://www.ordnancesurvey.co.uk/getamap">Get-a-map</a> service. Image
reproduced with kind permission of <a href="http://www.ordnancesurvey.co.uk/">Ordnance Survey</a> and <a
href="http://www.osni.gov.uk/">Ordnance Survey of Northern Ireland</a>.
Figure 5.1: Location plan
Quality
5.2.2 Experimental Design
Eight different types of tipping bucket rain gauge were tested, which for the purpose of
experimental design were considered as the treatments. These are described as types 1
to 8 for reasons of confidentiality. These were compared against a control gauge
(meteorological office 5″ gauge read daily) at a vertical height of 12″. Supplementary
control data will be obtained from 2 extra meteorological office 5″ gauges, one pit
mounted and one turf wall. The treatments were as follows:
Treatment 1 type 4 (stainless steel aerodynamic gauge)

Treatment 2 type 7 (plastic aerodynamic gauge)
Treatment 3 type 1 (conventional gauge)
Treatment 8 type 6 (conventional gauge, all stainless steel)
Treatment 9 Control
The gauges were laid out in a randomised block design with each of the treatments and
controls replicated three times (Figure 5.2 and Table 5.1)
Table 5.1: Randomisation
Gauge position No Block No Treatment no

1 1 7
2 1 8
3 1 4
4 1 1
5 1 5
6 1 Control
7 1 2
8 1 3
9 1 6
10 2 3
11 2 Control
12 2 1
13 2 2
14 2 6
15 2 5
16 2 7
17 2 4
18 2 8
19 3 7
20 3 2
21 3 8
22 3 3
23 3 6
24 3 Control
25 3 4
26 3 5
27 3 1
Quality
Old tower –where anemometer located
Upslope
N
19
Fibre
20
optic
cable
21
10 22
23
11
12 24
1 Block 3
Block 1 13 25
2
14 26
3
15 27
4
16
5
17 Block 2
6
18
7
30
8
10m
9
Downslope
Figure 5.3: Layout of trial site
Each tipping bucket gauge was mounted on a precast concrete paving slab, which itself
was bedded on a sand and cement mixture to ensure stability. All gauges were installed
following standard Environment Agency procedures and also taking account of any
manufacturer specific instructions. Each block of raingauges was connected to a
separate Newlog datalogger that recorded the exact time when a tip was measured on
each of the gauges.
5.2.3 Measurements
A continuos record was kept, on a NEWLOG datalogger, of each tip recorded on each
raingauge. The rainfall recorded by the standard meteorological office 5″ gauges,
including exposed and those in pit and turf wall installations, was measured at 1400
Quality
hours each day every day using standard meteorological office procedures. Data
describing wind speed (average and gust) and direction, at hourly intervals were
provided by the close by climate station (a 2m high tower 15m from block 1) operated
by the Meteorological Office. Temperature and relative humidity data were provided
from the measuring facilities some 300m away at the observatory
The performance of the gauges following snowfall was compared using a simple scale
was used to assess the amounts remaining in all the gauges: (0=Clear, Tr=Trace,
S=Slight, M=Moderate, L=Large). A record was also kept of any faults that developed
with individual gauges or specific operational problems encountered
5.2.4 Statistical Analysis

A multiple comparison test (Dunnet, 1970) was used to determine whether the
treatments were significantly different from the controls and Duncans multiple range
test (Duncan, 1955) was used to determine whether the differences between pairs of
means were significant. Regression analysis was undertaken to determine whether a
relationship between gauge catch and other variables could be found. Analyses were
undertaken using either Statistica or Minitab packages.
5.2.5 Summary of climate data recorded at the observatory

The climate data recorded outside the main observatory building are reported as these
can be compared to long term averages and provide a contextual background against
which the data from the trial can be considered.
Rainfall recorded at the Observatory

Trial year 1583.8mm
1911 - 2001 Average 1567.6mm
'Wet Days' (Number of days with 1.0mm or more)

Trial year 191 days
1911 - 2001 Average 180 days
Intensity (mm per 'Wet Day')

Trial year 8.29mm
1911 - 2001 Average 8.71mm
Snowfall
Days with sleet or snow reported
Trial year 21
1911 - 2001 Average 56
Sunshine
Trial year 1290.1 hours
1911 - 2001 Average 1190.2 hours
Mean Temperature
Trial year 7.97°C
1911 - 2001 Average 7.10°C
Quality
The year of the trial period was unremarkable when compared to the past 12 years or so,
with above average temperatures and reduced snow frequency. The most significant
rainfall event was undoubtedly 21st/22nd October, when between 65 and 70mm were
recorded by the check gauges in 24 hours. In all, the period 20th - 27th October
produced about 170mm.
Wind direction was in line with long term averages, being predominantly from the
Southwest. Similarly, wind speeds were very much in line with the past few years, that
is somewhat lower than historically due to the shelter effects of local afforestation.
5.3 Results
5.3.1 Overall performance of gauges

The total rainfalls measured over the study period (1/06/2002 to 31/05/2003) by the
check gauges in each block are presented below in table 5.2. The largest catch of
rainfall was measured by the turf wall gauge, with the pit gauge recording 1.5% less.
The mean catch recorded by the 3 unsheltered meteorological office 5" gauges (control
gauges) was 4.4% less than the turf wall.
Table 5.2: Overal catches (June, 2002 to May, 2003 inclusive) from manually read
gauges
Block 1 Block 2 Block 3 Turf Wall Pit

(Lower) (Middle) (Upper)
1556.6 1551.0 1547.1 1622.6 1598.6
The total rainfall recorded at the observatory was less than the total of 1622 mm
recorded at the turf wall check gauge on the trial site. The figure for the trial site would
have been considerably higher still had it been possible to measure the accumulation of
snow which fell in early February.
To allow a comparison of the raw catch data from the TBRs against total catch by the
check gauges individual totals were calculated for the study period. A total was
produced for each TBR for the days when that specific gauge was operational, a total
was also produced for same time period using the meteorological office 5" gauge
(control gauge) in the same block. The totals derived in this way are presented in
figures 5.3a to 5.3c and an overall summary is presented in table 5.3.
Table 5.3: Summary of overall mean gauge perfromance

Gauge type 7 type 4 type 1 type 8 type 5 type 3 type 2 type 6
% 94.6 91.3 83.8 89.8 84.1 76.9 81.3 78.9
Control
catch
recorded
Quality
1600 100%
95%
1200
Rainfall (mm)
90%
800 85%
80%
400
75%
0 70%
type 7 type 4 type 1 type 8 type 5 type 3 type 2 type 6
Figure 5.3a: Comparison of tipping bucket totals (1 June 2002 to 31 May 3002)
against meteorological office 5" gauge total (block 1).
1600 120%
100%
1200
Rainfall (mm)
80%
800 60%
40%
400
20%
0 0%
Figure 5.3b: Comparison of tipping bucket totals (1 June 2002 to 31 May 3002)
against meteorological office 5" gauge total (block 2).
1600 100%
80%
1200
Rainfall (mm)
60%
800
40%
400
20%
0 0%
TBR rainfall Check gauge rainfall TBR as % of check gauge
Figure 5.3c: Comparison of tipping bucket totals (1 June 2002 to 31 May

2003) against meterological office 5" gauge total (block 3).
Quality
The first, and most striking fact that can be noted from this data is that, on average,
none of the TBRs came within 5% of the check gauge totals. It is also immediately
evident that the best performing gauges were those with an aerodynamic shape.
Detailed consideration of the data presented in the graphs was undertaken by ranking
the gauges in order of performance (table 5.4). This showed that while the type 7 had
consistently the highest catch compared to the controls that there was little to choose
between the type 8 and the type 4 in two out of three blocks. The type 3 always stood
out as having a low catch in all three blocks and the type 6 in two out of three. The close
agreement (<1%) between the Met Office 5" check gauges annual totals in blocks 1,2
and 3 should also be noted.
Table 5.4: Rank order of gauge catches

Rank Block 1 Block 2 Block 3 Overall
Gauge % control Gauge % control Gauge % control Gauge % control
catch catch catch catch
1 type 7 94.7 type 7 96.9 type 7 92.3 type 7 94.6
5.3.2 Relative performance of different gauge types

To carry out a true comparison between the gauges a subset of the data was produced
which produced an overall total for daily means, which met the following conditions.
Data available for all three replicates of all gauges
Mean of all three meteorological office 5" gauge > 5mm (5mm selected to avoid
influence of drizzle)
No snow effects noted by observers
The relative performance of each of the gauge types (including the meteorological
office 5″) expressed as mean percentage of the rainfall recorded by the pit and turf wall
gauges (based on the subset described above) are presented in table 5.5 for the whole
year, winter (November to March) and non winter periods. The use of data expressed
as a proportion of the pit and turf wall gauges, rather than as a proportion of the control,
had the additional advantage of allowing the impact of aerodynamic effects on the
meteorological office 5" gauge to be quantified and removed this interaction from any
assessment of the TBRs. The catch measured by the type 7 (plastic aerodynamic gauge)
corresponded most closely to that measured by the standard meteorological office
5″gauge in all three cases, followed closely by the type 4 (stainless steel aerodynamic)
and type 8 gauges. The next highest catch (4th in rank order) was consistently recorded
by the type 5 gauge. Fifth and sixth places in rank order were by the type 1 and type 2
gauges when either the whole year or winter data were considered. However in the case
of the non winter data 6th position was occupied by the type 6 gauge, which shared
lowest rank position with the type 3 based on the whole year data and was the worst
performer over the winter period.
Quality
wall gauges
type 7 type 4 type 1 type type 5 type 3 type 2 type 6 Met.

8 Office 5″
Whole year 92 88 81 87 82 75 79 75 96
November to 90 85 78 85 79 73 78 68 95
March
April to October 94 91 84 90 85 78 81 82 98
The mean catch for each gauge and the 95% confidence intervals are presented in
figures 5.4a, to 5.4c. It can be seen from this that the confidence interval around the
mean catch from the meteorological office 5" gauge is much tighter than around any of
the other gauges. Further, the confidence intervals around the means of all different
types of gauges are larger during the winter period, i.e. in winter the agreement between
gauges of the same type is less good.
Dunnett's test showed that the mean catch over the whole period from the
meteorological office 5" gauge was significantly greater than that from the tipping
bucket gauges (type 7 (p<0.05) and all other gauges (p<0.0001). In the non winter
period the catches from all the tipping bucket rain gauges was less than that from the
meteorological office 5" gauge (p<0.0001). In contrast when the winter period only
was considered the catch from the type 7 gauge was not significantly different from the
meteorological office 5" gauge (p>0.05) whereas the catches from all the other tipping
bucket gauges were significantly less (p<0.0001).
Duncan’s test was used to allow comparison between individual gauges as opposed to a
simple comparison of all gauges against a control. The results of this are presented in
Tables 5.6a to 5.6c. This test allows the gauges to be allocated to groups, or
populations, with catches that are significantly different. For example the test shows
that the standard meteorological office 5" gauge and the type 7 gauges both have mean
catches that are statistically different from each other and the rest of the gauges. The
populations to which gauges belong are shown in table 5.7, in which groups or
populations are placed in order of the magnitude of the mean catch.
Quality
% of mean rainfall recorded by pit/turf wall
100
95
90
85
80
75
70
65
60
type 7 type 4 type 1 type 8 type 5 type 3 type 2 type 6 Met
Office
5"
Figure 5.4a Percentage of rainfall recorded by pit and turfwall gauges – all year
100
95
90
85
80
75
70
65
60
Office
Figure 5.4b: Percentage of rainfall recorded by pit and turfwall gauges – 5"
non winter
100
95
90
85
80
75
70
65
60
Office
5"
Figure 5.4c: Percentage of rainfall recorded by pit and turfwall gauges –
winter only
Quality
Table 5.6a: Duncans test – Probabilities for Post Hoc tests - whole season
Gauge type 7 type 4 type 1 type 8 type 5 type 3 type 2 type 6 Met. Office 5″
Mean 93.97 89.61 82.54 89.23 83.79 76.52 80.97 76.93 97.12
type 7
type 4 0.000009*
type 1 0.000004* 0.000003*
type 8 0.000011* 0.650119 0.000011*
type 5 0.000003* 0.000011* 0.128894 0.000009*
type 3 0.000005* 0.000004* 0.000003* 0.000004* 0.000004*
type 2 0.000004* 0.000004* 0.056145 0.000003* 0.000884* 0.000011*
type 6 0.000004* 0.000004* 0.000011* 0.000004* 0.000003* 0.61469 0.000010*
Met. Office 0.000139* 0.000011* 0.000004* 0.000003* 0.000004* 0.000001* 0.000004* 0.000005*
5″
Table 5.6b: Duncans test – Probabilities for Post Hoc tests – winter
Mean 93.77 87.76 80.94 88.80 82.17 75.41 80.91 70.99 96.42
type 7
type 4 0.000014*
type 1 0.000004* 0.000011*
type 8 0.000107* 0.417353 0.000003*
type 5 0.000003* 0.000020* 0.336903 0.000011*
type 3 0.000004* 0.000004* 0.000032* 0.000004* 0.000003*
type 2 0.000004* 0.000003* 0.981116 0.000004* 0.357351 0.000025*
type 6 0.000005* 0.000004* 0.000003* 0.000004* 0.000004* 0.000558* 0.000011*
Met. Office 0.038016* 0.000003* 0.000004* 0.000011* 0.000004* 0.000005* 0.000004* 0.000001*
5″
Table 5.6c: Duncans test – Probabilities for Post Hoc tests – non-winter
Mean 94.17 91.40 84.10 89.64 85.34 77.54 81.03 82.14 97.73
type 7
type 4 0.005504*
type 1 0.000004* 0.000003*
type 8 0.000019* 0.078328 0.000011*
type 5 0.000003* 0.000011* 0.213703 0.000025*
type 3 0.000005* 0.000004* 0.000003* 0.000004* 0.000004*
type 2 0.000004* 0.000004* 0.003059* 0.000004* 0.000034* 0.000473*
type 6 0.000004* 0.000004* 0.049564* 0.000003* 0.001922* 0.000017* 0.269464
Met. Office 0.000363* 0.000011* 0.000004* 0.000003* 0.000004* 0.000001* 0.000005* 0.000004*
5″
Quality
Table 5.7: Allocation of gauges to significantly different populations using Duncans
test, in order of decreasing performance
Group Whole year data Winter data Non-winter data

1 Met. Office 5” Met. Office 5” Met. Office 5”
2 type 7 type 7 type 7
3 type 4 &type 8 type 4 &type 8 type 4 &type 8
4 type 5/ type 1 type 5/ type 1/type 2 type 5/type 1
5 type 1/type 2 type 3 type 6/type 2
6 type 6/type 3 type 6 type 3
Clearly the type 7 stands out as the gauge that had catch that that was closest to that of
the standard meteorological office 5" gauge, but it caught significantly less (p<0.05).
The type 4 gauges and the type 8 gauges had catches that were not significantly
different whatever period is considered but which have a significantly lower catch again
than the type 7. When the whole year data set is considered group 4 contains type
5/type 1 and group 5 type 1/type 2. This occurs because the confidence intervals around
the means for type 5 and type 1 overlap and type 1 and type 2 overlap but the type 2 and
type 5 do not. In winter there is no significant difference (p>0.05) between the catch of
the type 1, type 2 or type 5 gauges whereas in the non-winter period only the type 5 and
type 1 are indistinguishable (p>0.05). In the non-winter period the catch of the type 2
was not significantly greater than that of the type 6 (p>0.05) forming a group with the
5th lowest catch. However, when considered over the whole year the type 6 and the
type 3 gauges had similar catches (p>0.05) but which were significantly lower than the
type 1/type 2 group (p<0.05). When the winter period was considered separately the
type 6 caught significantly less than the type 3 (p<0.05) with the converse being true in
summer.
In practical terms the groups can reasonably be amalgamated together as follows:
Met. Office 5 ″>type 7>type 4&type 8>>type 1,type 5,type 2>>type 3&type 6.
(94.2%) (91-89%) (84-81%) (78-76%)
5.3.3 Impact of wind and temperature on gauge performance

The data presented earlier in table 5.5 are repeated overleaf in table 5.8 for ease of
reading and show a seasonal comparison of TBR gauge performance compared to the
pit and turf wall gauge totals. What is immediately obvious for all gauges is that there
is a seasonal impact on gauge performance. When the totals for the whole year are
compared with the winter (November to March) and summer (April to October) periods,
the winter totals are observed to be below that of summer, by on average 4-5%. This
trend is also apparent on the standard Met Office 5" gauge, which shows winter
readings 3% less than summer readings. That the trend is observed in the Met Office 5"
control gauges is of considerable importance as the only difference between these
gauges and the Met Office 5" pit and turf wall gauges is there aerodynamic exposure.
Thus the study has shown that in the winter period, which has more months when wind
speeds (Figure 5.5) and gust speeds tend to be higher, the standard Met Office 5" gauges
catch less rain.
This shows that during the study period there was a definite seasonal effect on gauge
performance amounting to on average a 4-5% lower reading for winter than summer.
Quality
wall gauges (Table 5.5 Repeated)
Met. type 1 type 2 type 3 type 4 type 6 type 5 type 7 type

Office 8
5″
Whole year 96 81 79 75 88 75 82 92 87
November to 95 78 78 73 85 68 79 90 85
March
April to October 98 84 81 78 91 82 85 94 90
9
Monthly average wind speed (kn)
3
be 02
03
2
3
be 02
03
3
02
02
ob 002
00
ug 200
00
00
nu 200
20
20
20
20
20
em 20
,2
,2
,2
2
t,
il,
,
y,
r,
,
,
r,
r,
ay
ly
ry
ch
er
ne
us
be
pr
ar
Ju
ua
M
Ju
ar
A
em
em
br
M
ct
A
Ja
ov
pt
ec
C
Fe
Se
D
N
Figure 5.5: Monthly average wind speed
Further detailed analyses were undertaken to determine relative importance of

aerodynamic effects, rainfall intensity and snow catch performance. Regression
techniques were used to try and determine whether the temporal variation in gauge
performance could be ascribed differences in ambient conditions at the time of
measurement i.e., differences in wind velocity or temperature. The dependant variable
was the relative catch compared to the pit wall and turf gauges and the independent
variable were daily mean wind speed, daily mean hourly gust and daily mean
temperature, all calculated for the time it was raining in each day. This was achieved by
identifying those hours during which tips were recorded on any TBR, and then carrying
froward to the means only climate data associated with those hours. This approach is
necessarily a compromise as rain may not fall for the entire hour, but this approach was
taken, as it seemed unreasonable to use a simple daily wind speed but the resolution at
which climate data were available did not permit a more detailed approach
Quality
The coefficients obtained from the regression analysis, and the R2 values for the
relationships, are presented in Appendix A for whole year, winter (November to March)
and non-winter periods respectively. All regression equations were of the form:
y = c +mx
where
y is the Proportional catch
x is the meteorological variable in question
m is the slope of the regression line
c is a constant
All the intercept values derived by the regression analysis were significantly greater
than zero (p<0.05), but this is as much a reflection of their magnitude as anything else
and some of the slopes were significantly different from zero. However, and more
crucially, all the R2 values for the relation ships were very small (<0.2) which means
that very little inference can be drawn from the equations as they explain very little of
the variation in the gauge catch.
However, it is interesting to note that when windspeed or gust speed are the dependant
variables and winter or all year data are considered only the regression line for the type
7 has a small positive slope. This indicates that the type 7 gauges are catching the same,
or more rain at various windspeeds whereas the other gauges show a decrease in catch
with increasing windspeed). Patterns within the non-winter data, when there is likely to
be less wind, were less clear. Negative slopes dominated the regression lines between
catch and temperature, i.e. as temperature increased catch tended to decrease. The only
positive slopes being that for the type 4 (nearly flat) and the type 6 which was not only
positive but had the largest magnitude of any of the gauges.
Further regression analyses were undertaken using the difference between mean wind
speed and mean gust speed (a possible index of wind variability) and multiple
regression of wind, gust and temperature were undertaken. These relationships were as
weak, or weaker than, the simple regressions and are therefore not presented.
Summary plots of the regression lines based on all three replicates of each gauge are
presented in figure 5.6 for whole year, winter and non-winter datasets respectively, and
the parameters describing these relationships are available in Appendix A. It can be
seen from these graphs that the gauge which appears to be least influenced by wind
speed is the type 7, with even the meteorological office 5" gauge showing a predicted
decline in catch at higher windspeeds. Interestingly, only the relationship between wind
speed and relative catch for the meteorological office 5" gauge had an intercept, which
was close to 100%. Assuming a linear relationship between catch and windspeed the
intercept should tend to 100% for all gauges at zero wind, if wind is the main influence.
However, whether this difference is real or a function of the considerable variation in
response between replicates is open to question. The considerable variation between
replicates is clearly a major cause of the low % variance accounted for by the overall
relationships derived from the observed data. Additionally, the assumption that the
relationship is linear is possibly not valid.
Quality
Whole year
110
100
Relative catch (%)
90
80
70
60
0 5 10 15 20 25
Average windspeed when raining (kn)
Winter period
110
100
Relative catch (%)
90
80
70
60
0 5 10 15 20 25
Non winter period
110
100
Relative catch (%)
90
80
70
60
0 5 10 15 20 25
Figure 5.6: Relationships between relative catch (%) and average wind speed
during hours when rainfall is recorded
Quality
Consideration of the mean relative catches (whole season for each gauge), presented in
table 5.9, identifies that, for a given gauge, in most cases the catch is higher in Block 1
than Block 2 or Block 3 and similarly Block 2 catch tends to be higher than Block 3.
This raises the issue of whether the exposure, and hence wind speed varies significantly
across the site.
Table 5.9: Relative catches (%) of individual gauges (whole year data)
Met. type 1 type 3 type 2 type 4 type 5 type 6 type 7 type 8

office 5″
% catch 97.4 81.7 79.1 87.1 90.3 83.5 79.8 94.5 93.3
block 1
% catch 97.2 83.1 76.9 80.6 89.6 75.7 86.3 96 87.3
block 2
% catch 96.8 83.3 73.9 74.8 89.4 91.1 67.2 91.7 87.2
block 3
5.3.4 Influence of rainfall intensity on catch measured by the TBRs

In order to determine whether the low catches recorded by the TBRs were a
consequence of water lost during the time taken for the bucket to tip the data was
recalculated incorporating a dynamic correction. This used the equation between
rainfall intensity and % error that was determined in the laboratory phase of this study
(Hodgkinson, 2002). The equation used was that developed from combining datasets
from all the gauges. The equation used was Y= 0.779-0.1278X +0.000408X2 where Y
=% error and X = intensity. This was implemented at a 5-minute time step and was run
on the whole years data. Data were only corrected where intensity was such that under
reading was expected due to large uncertainty in the equation at lower intensities. The
change in rainfall recorded by each of the TBRs is presented in table 5.10
Table 5.10: Changes % in measured rainfall due to implementation of a dynamic

correction factor at a 5 minute timestep
Gauge type 7 type 4 type 1 type 8 type 5 type 3 type 2 Type 6

Block 1 0.10 0.11 0.67 0.12 0.12 0.11 0.08 0.14
Block 2 0.13 0.10 0.11 0.14 0.12 0.12 0.07 0.34
Block 3 0.08 0.07 0.07 0.13 0.09 0.08 0.01 0.12
Quality
A comparison was also made, using a short period of data when large rainfall occurred,
and intensities reached ca. 15 mm hr-1, between a dynamic calibration based on a 5
minute time step and correction based on the time of tip. The results (figure 5.7) show
that no significant improvement was obtained by the time of tip approach and that the
differences between the TBRs and the manually read gauges were much greater than
any dynamic corrections
100.00
Rainfall between 18 and 24 October, 2002 (mm)
No dynamic correction
95.00
Dynamic correction at 5
minute timestep
Dynamic correction based on
90.00 time of tip
Met Office 5"
Mean of pit/turf wall

85.00
80.00
75.00
Figure 5.7: Comparison between no dynamic correction and correction at 5

minutes or time of tip
The fact that small only increases in catch were brought about by implementation of the
dynamic calibration were due to the fact that rainfall intensities were only in excess of
the amount (10mm hour) for relatively short periods during the study period (Table
5.11)
Table 5.11: Cumulative period (minutes in year) during which intensities were
over 10mm hr-1 in any 5 minutes maximum duration possible per year
525600 hours.
Gauge type 7 type 4 type 1 type 8 type 5 type 3 type 2 type 6

Block 1 320 325 365 330 300 290 275 300
Block 2 240 295 265 285 270 250 260 310
Block 3 270 200 230 285 275 190 165 320
5.3.5 Performance following Snow

Overnight snow 2nd/3rd, of January 2003 followed by maximum temperature around
zero
Quality
presented an excellent opportunity to compare the length of time unmelted snow
remained in the various types of gauges. The results are presented in table 5.12.
Table 5.12: Speed of snow melt January 3, 2003
Gauge Block 1 Block 2 Block 3 Comments

type 7 L L L Very poor
retains snow longest
type 1 M M M Poor
type 8 S S S Good
type 4 S S Tr Good
type 2 0 0 0 Best performer all gauges
clear
type 6 S M M Average
type 3 Tr Tr S Good
type 5 S M M Average
Reliability and performance in the field
Only a limited number of faults were identified during the field test and no failure of
major components such as reed switches occurred. Gauge specific problems identified
are reported below
type 7.
Leveling found to be difficult due to plastic installation screws and lack of an external
bubble. Algal growth was found to occur on buckets and filters tended to need regular
cleaning
type 4
No specific faults reported
type 1
type 8
type 5
Filter reported to both blow away and retain water at other times.
type 3
Lacquer coating started to come off one bucket by July 02 (deployed in the field Jan.
02)
type 2
Buckets found to lock in central (balanced position)
type 6
Quality
5.4 Discussion
Comparison of the catches from the TBRs with either that of the plain meteorological
office 5" check gauge or the pit/turf wall gauges showed surprisingly large variations
with the TBRs recording consistently less rain fall than the manually read gauges. In the
case of the type 7 and type 4 this contrasts to the findings of (Porter, 2000) who
reported that the catch of the aerodynamic gauges tended to exceed that of the
meteorological office 5" gauge. However, from the standpoint of the other gauges
tested, the data reported by (Porter, 2000) generally supports the finding of this study,
i.e. that TBRs catch less rain. Of the gauges tested by (Porter, 2000) the type 1 gauge
performed particularly poorly being out of agreement with the check gauge by more
than 5% for 36% of the time which is consistent with the findings of this study.
However, in some periods during the summer (Porter, 2000) found that the TBRs
caught more rain than the meteorological office 5" check gauge which is at variance
with the results reported on here. The TBR with the catch closest to the pit/turf wall
gauges or meteorological office 5" check gauge was the type 7 plastic aerodynamic rain
gauge; this tends to suggest an aerodynamic effect as the cause of the differences in
performance. This is supported by the fact that TBR with the next highest catch was the
type 4 also aerodynamic, although its catch was not significantly different from that of
the conventionally shaped type 8 gauge.
A seasonal component to catch performance was observed with performance for all
TBRs observed to be 4 -5 % lower in winter than summer when compared to the turf
and pit gauge controls. The same trend was also observed in the standard Met Office 5”
gauges, with recorded 3 % less in winter then summer when compared to the turf and
pit check gauges. However, it was impossible to identify a statistically significant wind
effect although the slopes of the regression lines suggest that there may be an effect.
The question to answer therefore is why no effect could be demonstrated. It is
suggested that the failure may be due to differences in scale between the measurement
of the wind and what the gauges are exposed too.
Wind velocity was measured at one point (2m) above ground level) whereas each of the
gauges are a slightly different height and at different locations relative to the point at
which the wind speed is measured. A further issue is that, as the predominant wind
comes in to the site it passes across trees, then moorland with say 50cm high vegetation
and cut then grass immediately around the gauges. This raises concerns as to whether
the wind velocity, and turbulence, may change across the site. Expert advice (Lapworth
Pers comm.) suggested that “the change in wind velocity over the whole field is very
sensitive at the height of interest. For instance depending on the actual values of the
upstream and downstream roughness lengths, then at a height of 0.1m the wind speed
might increase by a factor of 2 over the site while at 0.5 metres it might increase by a
factor of 10% over the site.” To determine whether this effect was real or not it would
be necessary implement the actual values of the relevant parameters (i.e. z0 values,
heights of interest, horizontal distances of the rain gauges) into the formula by Panofsky
and Townsend (1964) and obtain the results from the analytic theory. It was considered
that a detailed evaluation of this effect was beyond the brief of the study. However, it
may be that to fully elucidate the interaction between windspeed and catch it may be
necessary to use micro anemometers located close to each gauge and at the same type
8vation to ensure that the measured wind speed is the same as that seen by the gauge.
Quality
Similarly temperature effects could not be statistically confirmed although interestingly
the regression line between type 6 gauge catch an ambient temperature had the greatest
slope. The type 6 gauge was identified in the laboratory-testing phase (Hodgkinson,
2002) as being the gauge most influenced by changes in temperature.
The use of a dynamic calibration approach was not found to increase rainfall catch
greatly at this site. However, the rainfall intensities encountered were not high enough
to frequently trigger the need for dynamic correction, rather it rained steadily for long
periods. The low intensity of rainfall that occurred may in itself be a key to the low
catches measured by the TBRs as low intensities tend to be associated with small drop
sizes which will be more easily deflected by wind or turbulence. However, this does
not explain why the TBRs are more affected than the standard meteorological office 5″
gauges. The difference in gauge performance may simple be a consequence of the
height to width ratio of the respective gauges, with the meteorological office 5″
representing less resistance to the wind than the other gauges and hence causing less
turbulence and thus achieving a greater catch. The noise in the data, and other issues
such as the possibility of changes in wind velocity across the site makes further
evaluation of the effects inappropriate.
That the dynamic calibration did not adequately transform the TBR total to that of the
manual gauge presents a problem as it did not explain the TBR under-read. Impacts of
TBR gauge accuracy, if uncorrected for, will have an impact on the rainfall record as a
record. For example in terms of climate change trend analysis, a 4-5 % step change in
rainfall measurement accuracy will have a considerable impact on a temporal rainfall
analysis of rainfall records.
To overcome this, the following solution is proposed. Every TBR gauge should have a
standard Met Office 5" gauge read on a monthly (or of a different frequency depending
upon rainfall quantities) basis as part of the standard site configuration. For the
equivalent time period, the TBR record should be corrected to match that of the
manually read check gauge. This transformation will be simple and linear, which has
the benefits of being easy to apply. This is approach is supported by the fact that a
dynamic calibration based on rainfall intensity did not make a significant difference to
the record. Correcting the TBR data in this way will minimize the impact on the long
term record of different TBR gauge type characteristics, will ensure BS and Met Office
good practice standards, and can be applied to existing historic TBR records also.
Only the type 2 gauge was found to have serious problems in that it locked in the
balanced position with water passing into both buckets. This problem also occurred in
the laboratory tests (Hodgkinson, 2002) although it was remedied by a visit from a type
2 engineer. This suggest that extra care will need to be taken when setting up these
gauges. This problem was not reported by (Porter, 2000) who also tested the type 2
gauge and found it to be particularly reliable; this suggests that the problem may have
arisen due to change in construction that has occurred recently. Other practical issues
that came to light were greater algal growth on the white plastic gauges (type 7) and the
fact that effectively gauge albedo (colour and material) greatly influenced the time the
gauges remained block after snowfall. It may therefore be worth considering gauges that
maximize the gain of energy for the sun when selecting equipment for locations with
high snowfall.
Quality
5.5 Conclusions
Under the conditions encountered at the study site the best performing TBR gauge
caught only 95% of the rainfall recorded by an unsheltered standard Met Office 5"
gauge. The range of rainfall catch recorded by the TBR gauges varied from 77 to 95 %
of the standard unsheltered Met Office 5" gauges.
The unsheltered standard Met Office 5" gauges caught only 96 % of the rainfall
recorded by the standard Met Office 5" gauge in a turf wall configuration. The standard
Met Office 5" gauge in a pit configuration recorded 1.5 % less rainfall than the turf wall
configuration.
This can be summarized as:

Turf and pit standard > Unsheltered standard Met Office > Tipping Bucket Raingauge
Met Office 5" gauge 5" gauge
100 % 96 % 91% to 73 %
(Percentages are expressed as % of combined turf and pit gauge catch)
Under the conditions encountered at the study site an aerodynamic gauge (ARG 100)
was the best performing TBR gauge with a catch that was closest to the turf and pit
standard Met Office 5" gauge configurations.
A seasonal component to catch performance was observed with overall catch

performance for all TBRs 4-5 % lower in winter than summer when compared to the
turf and pit standard Met Office 5" gauge configurations. This trend was also reflected
in the unsheltered standard Met Office 5" gauge, which recorded 3 % lower in winter
than summer when compared to the turf and pit standard Met Office 5" gauge
configurations.
No definitive link could be shown between variations in either temperature or wind

speed and the rainfall catch of a specific gauge although on a qualitative basis the
results suggest that there is a link.
Due to the low rainfall intensities encountered at the study site implementation of a
correction using dynamic calibration did not significantly increase measured catch. To
overcome the impact on long term records of TBR gauge accuracy, transformation of
the TBR total by a the total measured by a standard Met Office 5" gauge over a monthly
(or similar) equivalent timebase is recommended.
Further work is required to test the performance of TBRs under a wider range of
climatic conditions.
Quality
6 NEED FOR FURTHER WORK
If the dynamic calibration curves obtained in the laboratory testing phase are to be
applied then these will need to be refined by increased replication using a larger number
of gauges to improve confidence in the relationships.
The results obtained from the field trial are only applicable to the climatic conditions
pertaining to Eskdalemuir. For the results to be applicable the work, or at least
sufficient of it to represent those gauges either used or likely to be used by the Agency,
needs to be repeated at a number of locations to represent cross section of range of
climatic conditions encountered in England and Wales.
To fully understand the interaction between wind speed and gauge catch future work
would need to ensure that the wind profile seen by each gauge is fully captured.
Quality
REFERENCES
Adami, A. and Da Deppo, L. On the systematic errors of tipping-bucket recording rain

gauges. Instruments and Observing Methods Report No. 25, April 1986. p. 27-30.
World Meteorological Organisation, Geneva.
Alvey, N., Galwey, N. and Lane, P., 1982. An Introduction to Genstat. Academic
Press, London.
Biswas, A.K., 1970. History of Hydrology. North-Holland Publishing Co., Amsterdam
Bruce, J.P. and Clark, R.H., Introduction to Hydrometeorology. Pergamon, Oxford.
BS 7843 1996. Guide to the acquisition and management of meteorological

precipitation data. BSI London.
Calder, I. R. and Kidd, C. H. R., 1978. Note on the dynamic calibration of tipping-
bucket gauges. Journal of Hydrology, Amsterdam, 39(3/4): 383-386.
Cornish, K. A., Green, G. C. 1981. Economical recording tipping-bucket rain gauge.

Agricultural Meteorology, Amsterdam, 26(4): 247-253.
Duncan, D. B. 1955. Multiple range and multiple F tests. Biometrics, Vol 11, Pp1-42.
Dunnett, C. W. 1970. Multiple comparison tests. Biometrics 26, 139-141
Folland, C. K. 1988. Numerical models of the raingauge exposure problem, field

experiments and an improved collector design. Q.J.R. Meteorol. Soc. 114. Pp. 1485-
1516.
Folland, C. K.,1988. Numerical models of the raingauge exposure problem, field

experiments and an improved collector design. Q.J.R. Meteorol. Soc. 114. Pp. 1485-
1516.
Goodison, B.E and and Louie, P.Y.T, 1985. Canadian Methods for Precipitation
Measurement and Correction. Workshop on the correction of precipitation
measurements. 1-3 April 1985. Zurich.
Grossklaus, M. et al 1997. An Optical Disdrometer for Use in High Wind Speeds.

JournalAtmospheric and Oceanic Technology: Vol. 15, No. 4, pp. 1051–1059.
Hanna, E. 1995. How effective are tipping bucket rainguages ? a review. Weather. Oct
1995. 50(10): 336-342. Bracknell, England.
Hellin, J. and Haigh, M.J 1998. Rainfall in Honduras during Hurricane Mitch. UK
Department for International Development (DFID) research project.
Helliwell, P.R. and Green,M.J 1974. Raingauge Performance Studies. Final report to the
NERC. Institute of Hydrology. Wallingford.
Quality
Hewston, C. M.; Sweet, S. H. 1989. Trials use of a weighing tipping-bucket rain gauge.
Meteorological Magazine, 118:132-134, June 1989 Bracknell, Eng.
Hodgkinson R A, 2000. Evaluation of tipping bucket rain gauge performance and data
quality - Report on laboratory testing as part of Environment Agency Science Project
W6-048
Hsu, Sheng-I 1990. Revival of the tipping-bucket raingauge. International Conference

on East Asia and Western Pacific Meteorology and Climate, Hong Kong, 6-8 July
1989. p. 199-209. World Scientific Publishing Co. USA.
Hughes, C., Strangeways I.C. and Roberts A.M. 1993. Field evaluation of two
aerodynamic raingauges. Weather 48. Pp66-71.
Humphrey, M.D. et al (1997). A new method for automated dynamic calibration of

tipping-bucket raingauges. Journal of Atmospheric and Oceanic Technology, 14(6):
1513-1519. Boston, MA.
Kreuels, R. K.; Breuer, L. J. 1986.Wind-influenced rain-gauge errors in heavy rain

Instruments and Observing Methods Report No. 25, April 1986. p. 105-107. World
Meteorological Organization, Geneva.
Linsley, R.K.,A manual on collection of hydrologic data for urban drainage design.
Hydrocomp, Inc., Palo Alto, Calif.
Maksimovic, C.; Buzek, L.; Petrovic, J.(1991) Corrections of rainfall data obtained by
tipping bucket rain gauge .Atmospheric Research 27(1-3) 45-53 New York.
Marsalek, J. 1981.Calibration of the tipping bucket raingauge. Journal of hydrology, 53

(1981) 343-354. Elsavier, Amsterdam.
Met office 2000. Internal Met Office report on the trial of tipping bucket raingauges
Muller, S.H. and Van Londen, A. 1983. The evaluation of rain gauges illustrated by the
Thies rain gauge and the type 8ctrical raingauge of the Royal Netherlands
Meteorological Institute. Royal Netherlands Meteorological Institute.
Niemczynowicz J. 1986.Dynamic calibration of tipping-bucket rain gauges. Nordic

Hydrology, Copenhagen, 17(3): 203-214, 1986
Nystuen, J.A. et al 1996. A comparison of automatic raingauges. Journal of

Atmospheric and Oceanic Technology 13(1) 62-73. Boston USA
Panofsky and Townsend 1964. QJ Vol 90 p147
Porter M. 2000. Final report on the trial of tipping bucket raingauges carried out at
Eskdalemuir Observatory between January 1998 and June 2000 Meteorological Office
Unpublished Report.
Quality
Robinson, A.C and Rodda, J.C. 1969. Rain, wind and the aerodynamic characteristics of
rainguages. Meteorological magazine 98. pp 113-120.
Schonhuber, H.E. et. al. 1995. Weather Radar Versus 2D-Vidio-Disdrometer Data.
Proceedings of the III international Symposium on Hydrological Application of
Weather Radars. Aug 20-23 1995. Sao Paulo, Brazil.
Seibert, J. et.al 1999. Reducing systematic errors in rainfall measurements using a new
type of gauge. Agricultural and forest meteorology, 1999,Volume: 98-99 ,341-348 .
Sevruk, B. 1996. Adjustment of tipping bucket precipitation gauge measurements.

Atmospheric Research 42:237-246. Amsterdam.
Simic, M. and Maksimovic, C. 1994. Effect of the siphon control on the dynamic
characteristics of a tipping bucket raingauge. Hydrological Sciences Journal. 39(1) 35-
36. Oxford, England.
Smoot, G.F.,1971. Data Collection for real time systems. Treatise on urban water
systems. Colorado State University Press, Fort Collins, Colorado.
Ulbrich , C.W. 1983. Natural variations in the analytical form of the raindrop size
distribution. J. Climatol. Appl. Meteorology. 22,1764-1775
Weast, R. C., 1971. Handbook of Chemistry and Physics The Chemical Rubber Co
Ohio,
World Meteorological Organisation 1994. Guide to hydrological practices. Data

acquisition and processing, analysis, forecasting and other applications. WMO No 168.
Geneva.
World Meteorological Organisation 1994. Guide to hydrological practices. Data

acquisition and processing, analysis, forecasting and other applications. WMO No 168.
Geneva.
Yu, B., Ciesiolka, C.A.A., Rose, C.W. and Coughlan, K.J. 1997. Note on sampling
errors in the rainfall and runoff data collected by tipping bucket technology.
Transactions of the American Society of Agricultural Engineers. Vol 40 (5) 1305-1309.
Quality
Appendix A
Title Parameters describing regression relationships between relative
rainfall catch and other climatic parameters
Regression analysis based on whole year data
Mean daily wind speed

(when raining)
Intercept Slope p R2
type 7 93.37 0.09 0.56 <0.1
type 4 91.66 -0.28 0.10 <0.1
type 1 89.30 -0.99 <0.000 0.22
01
type 8 91.70 -0.34 <0.05 <0.1
type 5 88.10 -0.62 <0.01 0.10
type 3 80.47 -0.55 <0.01 <0.1
type 2 84.54 -0.50 <0.01 <0.1
type 6 81.37 -0.55 <0.5 <0.1
Met. Office 5" 99.86 -0.39 <0.000 0.17
1
Mean daily gust

(when raining)
type 7 92.16 0.10 0.17 <0.1
type 4 91.85 -0.12 0.13 <0.1
type 1 88.48 -0.32 <0.001 0.12
type 8 90.76 -0.08 0.26 <0.1
type 5 87.10 -0.18 <0.05 <0.1
type 3 80.35 -0.20 <0.01 <0.1
type 2 83.68 -0.14 0.08 <0.1
type 6 80.65 -0.17 0.17 <0.1
Met. Office 5" 99.47 -0.13 <0.01 <0.1
Mean daily temperature

(when raining)
type 7 95.94 -0.25 0.18 <0.1
type 4 89.56 0.02 0.92 <0.1
type 1 83.10 -0.01 0.72 <0.1
type 8 89.77 -0.05 0.76 <0.1
type 5 84.24 -0.05 0.82 <0.1
type 3 77.58 -0.11 0.58 <0.1
type 2 83.96 -0.37 0.09 <0.1
type 6 70.66 0.87 <0.01 <0.1
Met. Office 5" 96.24 0.11 0.31 <0.1
Footnote
All intercepts were significantly different from 0 (p<0.0001)
p values refer to whether the slopes of the regression lines significantly different from zero
Quality
Regression analysis based on winter data

(when raining)
type 7 94.03 0.01 0.95 <0.1
type 4 89.77 -0.27 0.28 <0.1
type 1 89.65 -1.18 <0.01 0.31
type 8 93.02 -0.55 0.02 0.11
type 5 86.88 -0.64 0.03 0.10
type 3 81.88 -0.72 <0.02 0.20
type 2 86.67 -0.76 <0.01 0.19
type 6 76.51 -0.56 0.15 0.04
Met. Office 5" 99.72 -0.44 <0.01 0.16
Mean daily gust

(when raining)
type 7 92.91 0.06 0.56 <0.1
type 4 89.98 -0.11 0.34 <0.1
type 1 89.06 -0.42 <0.01 0.18
type 8 92.23 -0.17 0.13 <0.1
type 5 85.59 -0.18 0.19 <0.1
type 3 80.64 -0.26 <0.05 0.11
type 2 86.22 -0.27 <0.05 0.11
type 6 76.09 -0.20 0.28 <0.1
Met. Office 5" 99.39 -0.15 <0.05 <0.1

(when raining)
type 7 95.94 -0.25 0.18 <0.1
type 4 89.55 0.02 0.92 <0.1
type 1 83.10 -0.09 0.72 <0.1
type 8 89.77 -0.05 0.77 <0.1
type 5 84.24 -0.05 0.81 <0.1
type 3 77.58 -0.11 0.58 <0.1
type 2 83.86 -0.37 0.09 <0.1
type 6 70.77 0.87 <0.01 <0.1
Met. Office 5" 96.24 0.11 0.31 <0.1
Footnote
Quality
Regression analysis based on non-winter data

(when raining)
type 7 92.54 0.20 0.41 <0.1
type 4 92.15 0.12 0.59 <0.1
type 1 87.70 -0.59 0.03 0.08
type 8 89.71 0.02 0.93 <0.1
type 5 88.14 -0.43 0.07 <0.1
type 3 78.95 -0.20 0.47 <0.1
type 2 82.02 -0.12 0.68 <0.1
type 6 82.30 -0.04 0.92 <0.1
Met. Office 5" 99.48 -0.26 0.003 0.15
Mean daily gust

(when raining)
type 7 91.39 0.14 0.16 <0.1
type 4 92.21 -0.05 0.62 <0.1
type 1 86.74 -0.16 0.16 <0.1
type 8 88.84 0.06 0.49 <0.1
type 5 87.28 -0.11 0.28 <0.1
type 3 79.309 -0.10 0.41 <0.1
type 2 81.05 -0.01 0.93 <0.1
type 6 80.98 0.06 0.66 <0.1
Met. Office 5” 99.02 -0.07 0.06 <0.1

(when raining)
type 7 97.79 -0.42 0.06 <0.1
type 4 95.08 -0.39 0.07 <0.1
type 1 87.24 -0.34 0.20 <0.1
type 8 90.79 -0.10 0.59 <0.1
type 5 90.18 -0.90 <0.05 <0.1
type 3 79.85 -0.23 0.39 <0.1
type 2 86.08 -0.51 0.07 <0.1
type 6 79.08 0.32 0.34 <0.1
Met. Office 5" 97.47 0.04 0.66 <0.1
Footnote
Quality

Evaluation of Tipping Bucket Rain Gauge Performance and Data Quality

Uploaded by

Copyright:

Available Formats

Evaluation of Tipping Bucket Rain Gauge Performance and Data Quality

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Evaluation of Tipping Bucket Rain Gauge Performance and Data Quality

Uploaded by

Copyright:

Available Formats

w w w.environment-agency.gov.

Evaluation of Tipping Bucket Rain

Science Report: W6-084/SR

Our work includes tackling flooding and pollution incidents, reducing

This report is the result of research commissioned and funded by

Published by: Author(s):

ISBN: 1 844 323 242 Keywords:

Science underpins the work of the Environment Agency, by providing an up to date

Professor Mike Depledge Head of Science

The influence of temperature on the calibration methods themselves was concluded to

Field testing of the gauges was undertaken at an experimental site located at

5. FIELD TESTING OF GAUGES 26

6. NEED FOR FURTHER WORK 48

Appendix A: Title Parameters describing regression relationships between

Figure 5.1: Location Plan

Figure 5.6: Comparison between no dynamic correction and correction at 5 minutes

Table 4.2: Relationship between Rainfall intensity and % error compared to

Table 5.1: Randomisation of treatments

Table 5.3: Summary of overall mean gauge performance

Table 5.4: Rank order of gauge catches

Table 5.7: Allocation of gauges to significantly different populations using Duncans

Table 5.10: Changes % in measured rainfall due to implementation of a dynamic

Table 5.12: Speed of snow melt January 3, 2003

2.1 Literature Review

2.2 Laboratory Testing of Gauges

1. To determine whether there are specific advantages in calibrating tipping bucket

2. To determine the influence of the temperature at which calibration is undertaken and

3 To determine the long–term reliability of different makes of tipping bucket rain

4 To identify specific operational problems relating to specific makes of tipping

2.3 Field Testing of TBR gauges

A field trial was undertaken at Eskdalemuir observatory with the objective of

To compile a synopsis of the most recent international publications addressing Tipping

3.3 Tipping Bucket Raingauge Errors and Calibration

3.3.1 Non-specific TBR errors

World Meteorological Organisation (1994) Emphasises the importance of site selection,

3.3.3 Errors specific to TBRs and calibration techniques

3.5 Tipping Bucket Raingauge Data Manipulation and Comparison of Data

3.6 Comparison of TBR Types

3.7 Comparison with other Automatic Systems

Raingauges supplied by different manufacturers may perform differently and

Figure 4.1: Percentage error in rainfall measurement as a consequence of the

4.2 Detailed Methodology

General Gauge Characteristics

4.2.1 Evaluation of calibration methods

Constant Rate Method

The specific procedure used was as follows:

4.2.2 Effect of temperature at time of calibration on calibration coefficient

Statistical analysis of data

4.2.3 Effect of rainfall intensity on gauge calibration coefficient

4.2.4 Long-term reliability tests

4.2.5 Observations on gauge performance and ease of use

4.3.1 Evaluation of calibration methods

1575 Burette method

Table 4.1: Summary of temperature testing based on all 3 blocks

4.3.3 Effect of rainfall intensity on gauge calibration coefficient

Table 4.2: Relationship between Rainfall intensity and % error compared to

Gauge type Polynomial Regression R2

-25 Outlier ignored

4.3.4 Long-term reliability tests