Hydrology-From Measurement To Hydrological Information (9)



Quality assurance programmes, such as the
ISO 9000 series (see ISO, 2000), have been adopted
by many agencies, putting in place practices that
aim to formalize and standardize procedures, from
data collection, through processing, to extensive
data checking. Quality assurance typically encompasses training, best practices, recording errors and
malfunctions, corrective action, checking and other
quality control, and independent audit of the entire
operation. This chapter will cover quality control,
considered here as the process of checking and validation of the data, but not the wider topic of quality
Following their capture on some medium, whether
paper, punched-tape or electronic digital form,
hydrological data are converted to a form suitable
for archiving and retrieval. In addition, at various
stages, data undergo a range of checks to determine their accuracy and correctness. As computer
archiving has become a standard practice in most
countries, the processing will involve the data
being converted to the required format early in
the process.
Data are collected and recorded in many ways,
ranging from manual reading of simple gauges to
a variety of automated data-collection, transmission and filing systems. With accelerating
developments in technology, it is now more important than ever that data-processing and quality
control systems be well-organized and understood
by the people involved in collecting and using
By way of example, a fl ow chart of a relatively
simple system is depicted in Figure I.9.1.
It is noted that quality assurance encourages the
adoption of recognized best practices and advances
in data validation. It is recommended that, subject
to the availability of resources, Hydrological Services
should consider the adoption of a quality management programme such as that described in ISOOnce this has been achieved, organizations
usually employ an accredited certifi cation agency
to provide independent verifi cation and advice on
developing the programme (Hudson and others,
As disciplines, hydrology and climatology have
followed the “rules” of good science, in that data
collection and its use should always use recognized good practices and be scientifically
defensible by peer review. These principles require
that a conservative attitude be taken towards
altering data, making assumptions and accepting
hypotheses about natural processes that one
would quite possibly understand less about than
one assumes.
9.2.1 Conservatism, evidence and
The hydrologist has a duty to be conservative in
carrying out any correction of data. In 9.7.2, it is
suggested to use strict criteria for altering or adding
data values. This must always be done using assumptions based on evidence rather than any element of
guesswork. Where an element of guesswork is
involved, this should be left to the user to carry out,
although all information that may be of use in this
process should be available, normally by way of
fi led comments or by being fi led separately in the
Another important convention is that any alteration made to data should be recorded in such a way
that others can follow what has been done, and
why. It should not be necessary to refer to the
persons who made the alteration for an explanation. An audit trail should be available, such that
with the documented procedures the process can
be tracked through and checked. This traceability is
also a requirement for a quality system.
9.2.2 Data accuracy standards and
A Hydrological Service or equivalent recording
agency should formulate data standards in terms
of resolution and accuracy for each parameter.
This process should be done in conjunction with
international standards such as detailed in the
Guide to Climatological Practices (WMO-No. 100),
and with consideration of the present and, more
importantly perhaps, the likely future needs of the

Figure I.9.1. Data-processing flow chart

When formulating data standards, it is important
to distinguish between resolution, accuracy, errors
and uncertainty:
(a) The resolution of a measuring device or technique is the smallest increment it can discern.
For instance, a data logger and pressure transducer will often resolve a stage measurement to
1 mm, but the accuracy may be less than this
due to errors resulting from drift or hysteresis
in the transducer;
(b) The accuracy of a measurement relates to how
well it expresses the true value. However, as the
true value is often unknown, the accuracy of
a hydrological measurement usually has to be
expressed in terms of statistical probabilities.
Accuracy is a qualitative term, although it is
not unusual to see it used quantitatively. As
such, it only has validity if used in an indicative
sense; any serious estimate should be in terms
of uncertainty (below);
(c) The error in a result is the difference between
the measured value and the true value of the
quantity measured. Errors can commonly be
classified as systematic, random or spurious;
(d) Uncertainty is the range within which the true
value of a measured quantity can be expected
to lie, at a stated probability (or confidence
level). The uncertainty and the confidence
level are closely related; the wider the uncertainty, the greater the confidence that the true
value is within the stated range. In hydrology, a
commonly used confidence level is 95 per cent,
which, in a normal distribution, relates to two
standard deviations. For further information,

The level of uncertainty that users of the data
require is normally the prime consideration. Once
this has been established, the uncertainties of the
methods, techniques and instruments should be
considered. Often there will be a compromise
between the uncertainty desired and the accuracy
and resolution of the instruments due to costs,
practicalities and limitations of techniques.
The usefulness of data is, to a great degree, dependent on its completeness, and targets should also be
set for reasonably expectable performance measures
such as the percentage of missing records. It is
recommended that agencies, rather than putting
great effort into fi lling missing data with estimates,
allocate resources (including training) to avoid the
need for this.

9.3.1 General
A database will necessarily contain various code
fields as well as the fields containing data values
(also 2.3.2). This is because of the requirement for
various descriptors to give meaning to the data, and
these generally need to be codes so that the fi les are
more compact and less ambiguous. A prime example is a code number for each recording station.
Some codes, such as station number, will be a database key, but others will be codes that describe,
among others, standard methods, data quality,
measurement units and parameters. Depending on
the database structure, codes may be required to
signify what parameter a variable represents or,
alternatively, this may be defined by the file
Coding systems should be comprehensive and fl exible, and data collectors should be encouraged to
make full use of the options available. In addition
to the application of codes to guide the processing,
comments should be included at this stage. These
comments provide a general description of the data
within defi ned time periods and should be available
automatically when data are presented to users.
9.3.2 Developing codes
The steps involved in devising and using codes are:
(a) Defi ne the data that require coding.
These are normally descriptive data items
that are used frequently, for example, the
names of locations, variables, analysis
methods, measurement units and data quality
(b) Decide when coding should be performed.
To satisfy the objective of common recording
and data-entry documents, coding should
be performed at the time of data logging by
the hydrological observer or the laboratory
(c) Consider the adoption of existing (national or
international) coding systems for some data
items. Schedules of variable codes, laboratory
analysis methods and measurement-unit codes
have been developed by several countries. The
adoption of such coding systems facilitates
the interchange of data and reduces the need
to devote resources to developing new coding
(d) Consider possible current or future links to
GIS (9.3.8) when compiling codes. For example, it could be beneficial to derive station and
river numbering codes based on locations keyed
to GIS;
(e) Obtain or prepare coding lists, incorporate
the codes into the reporting and dataentry forms and the computer systems, and
include coding instructions (and relevant
coding lists) into technician instruction
(f) Train observers in the use of codes, monitoring completed forms very closely for the initial
period after introducing or modifying the
coding system. This should be done for several
months to allow technicians to become familiar
with the codes.
Most codes used for hydrological purposes are
numeric. However, different combinations of
alphabetic and numeric codes are also used.
Alphabetic or alphanumeric codes are widely used
for borehole logs and where more descriptive data
are needed, such as soil land-use classifi cation.
The typical usage of codes in hydrological systems
is described below and in the NAQUADAT
Dictionary of Parameter Codes (Environment
Canada, 1985).
9.3.3 Location codes
Codes normally exist for basin or sub-basin, and it
is very useful to incorporate them into the station
description data fi le (Chapter 2). This allows rapid
identifi cation of all stations (or stations measuring
selected variables) in a single basin or group of
basins. For additional information on station
numbering, see 2.5.2.

9.3.4 Variable (parameter) codes
This heading covers the largest group of codes.
The range of hydrological and related variables
that may need to be included in a comprehensive
database is enormous. Fortunately, several hydrological agencies have prepared and published
variable-code lists (Environment Canada, 1985;
United Kingdom Department of Environment,
The code lists normally comprise a four- or fi vedigit code for the variable, a text defi nition of the
variable and possibly some abbreviations or synonyms. One feature that varies between the lists is
whether the measurement units and/or analyses
techniques (particularly for laboratory derived data)
are included in the defi nition or are themselves
coded. Thus, in one system variable code 08102 is
dissolved oxygen measured in mg/l using a
dissolved-oxygen meter, whereas another system
describes the same variable as 0126 (dissolved
oxygen) with measurement unit code 15, where
0126 and 15 are entries in the relevant code lists for
mg/l and m, respectively.
9.3.5 Data qualifi cation codes
With manual data collection it is common to
have a set of codes available for the hydrological
observer and the laboratory technician to qualify
unusual or uncertain data so that future data
usage may be weighted accordingly. There are
basically two groups of qualifi cations – the fi rst
can be viewed as the current status (reliability) of
the data value and the second indicates some
background conditions that may cause a nonnormal status. For both groups, the code used is
normally a single alphabetic character, also
known as a fl ag.
Flags for the status of the data are typically:
E Estimated value, with an implication of a
satisfactory estime;
U Uncertain value, thought to be incorrect but
no means to verify;
G Value greater than calibration or measurement
limit (value set to limit);
L Value less than the detection limit (value set to
V Value outside normally accepted range but has
been checked and verifi ed.
Flags for background conditions may be:
I Presence of ice (or ice damming);
S Presence of snow;
F Presence of frost;
D Station submerged (during fl ood events);
N Results from a non-standardized (quality
controlled) laboratory;
P Results from a partially quality controlled
Flags should be entered if they are present and
stored with the data to which they relate. The
computer data validation procedures performed on
the input data may generate more status fl ags, and
the same codes may be used.
Alternatively, some database systems allow for the
entry of plain-language comments into the database system (usually into a linked text file
9.3.6 Missing data codes
It is extremely important to differentiate between
data that are missing and data that were recorded as
having a zero value. If the data fi eld for a missing
numeric value is left blank, databases may automatically infi ll a (misleading) zero. Since a character
value is not allowed in a numeric data fi eld, this
missing data problem cannot be overcome by
inserting ‘M’ (for missing). One possibility is to
enter the code M as a separate data-status fl ag, but
in systems where fl ags are not used, some physically impossible data value, for example, –999, are
entered in the data fi eld to indicate a missing value
to the processing system. If required, this value may
be decoded to a blank or a “–” on output.
9.3.7 Transmission codes
Many systems for the transmission of data make
use of some form of coding method, the purpose of
which is to ensure that the information is transmitted quickly and reliably. This will require that data
be converted to coded form before being transmitted or processed. Codes will need to be designed so
that they are in a form compatible with both
processing and transmission.
9.3.8 Geographical Information Systems
GIS are fi nding wide application in the fi elds of
operational hydrology and water resources

assessment. Their ability to assimilate and present
data in a spatial context is ideal for many purposes,
ranging from the provision of base maps to the
operation of catchment or multicatchment models
for runoff mapping and flood or drought
In network planning and design, the ability to map
quickly and display surface water and related
stations enables a more effective integration to take
place. Network maps, showing basins or stations
selected according to record quality, watershed or
operational characteristics, can be used for both
short-term and long-term planning. The essential
features of complex networks can be made very
The term data capture is used to cover the processes
of acquiring data from written, graphic, punched
media, analogue or digital electronic form to a
medium whereby these data can be further processed, stored and analysed. In recent years this
medium has almost always been a computer,
perhaps a mainframe, but usually a personal computer (PC) possibly connected to a network.
9.4.1 Key entry
Data collected as written notes, whether in notebooks or on forms designed for the purpose, will
need to be entered into the computer manually.
While some form of scanning with optical character recognition may be possible, it is usually
preferable to avoid this unless the process is proven
to be free of error.
Where observers are required to write data values on
paper, it is recommended that they be issued with
standardized forms (possibly in a book) that set out
the required entries in a clear and logical sequence.
Such forms can be produced on a computer wordprocessing package, and their issue and use should
be controlled as part of the organization’s data
handling process. Examples of hydrometric data
that may be written and later entered by hand
include manual water-level readings, rainfall and
other climate observations, and current meter
streamfl ow measurements. There will also be secondary data entry in this category, such as short pieces
of edited record and stage-discharge curves.
It is recommended that key entry of data be
decentralized such that those persons responsible
for data collection are also responsible for its
entry and the initial data validation stages. Thus,
data fi les will normally be produced on a PC,
which will not need to be online or networked
(except for the ease of data backup and transfer).
As the size of the data fi les produced will be small
relative to the capacity of memory or storage
devices such as diskettes, data can be readily
backed up and transferred for archiving according to the data handling and verifi cation processes
in place.
The minimum verifi cation process for keyed data
should be for a printout of the keyed data to be
checked, entry by entry, with the original written
record by a person other than the one who entered
them. Checking will be enhanced if suitable
graphs of the data can be made available. Even
simple print-plots may be adequate. Where
possible, and especially where there are large
volumes of data to be entered, it can be useful to
use automated data verifi cation, such as range
checks and previous and/or following value
comparisons included in the data entry program.
9.4.2 Capturing chart data
Analogue records of parameters such as water
level and rainfall were commonly collected in the
past, and this technology still endures owing to
the advantages of rapid interpretation, simplicity
and the costs of changing instrumentation
Capture to computer records can be done by
manual reading and keying, or digitizing from
either a digitizing tablet or a scanner. Manual
reading normally involves a person reading the
series of values at an appropriate time interval,
and transferring these to a form from which the
values are later keyed into a computer fi le as
described in 9.4.1.
Tablet or fl at-bed digitizing is the most common
method and relies to some extent on the skill of
the operator not to introduce errors of precision
into the record. The use of a scanner with software to interpret the trace is a more recent
development, but is not widespread owing to the
trend towards electronic data loggers.
Whatever the method, charts should be stamped
with the prompts for date, time and water-level
readings at chart off and chart on. As these
indicate the adjustments to the original
observations, the need for clear annotation is
obvious. The most useful tool for verification of

chart data is to produce a plot of the data once
they are captured for comparison with the original
plot. Preferably, these plots will be printed at the
same scales as the original, so that the two
can be overlaid (for example, on a light table)
and any errors identified and subsequently
9.4.3 Punched-tape data
Electromechanical recording instruments were
widely used from the 1960s to the 1980s, but have
largely been replaced. These machines usually
punched binary coded decimal (BCD) values at
each time interval on to a roll of paper tape, and
were the fi rst commonly used machine-readable
recorders. They could be read relatively rapidly by
an optical tape reader and captured to computer
fi le.
Data-processing operations were similar to those
used for the later solid-state data logger, and
verification processes developed for them are
the basis of those used today for electronic
9.4.4 Electronic data logging
The use of electronic memory for storing values
from sensors with various forms of electrical
output has been common since the 1970s and
became increasingly widespread during the last
two decades of the twentieth century. As costs fall
in at least real terms, these instruments have
become more like computers and more easily
connectable to them.
As data capture to electronic code is one of the data
logger’s primary functions, this step of data processing has become simpler. At the same time this
technology has tended to make it easier for errors to
be more serious and widespread so that quality
control needs to be at least as rigorous as for other
Unlike charts, forms and tapes, electronic data
fi les do not exist in a tangible form to enable
them to be easily identifi ed, tracked and show
evidence of how they have been modifi ed. Batches
of data, whether time series, point or sample data,
need to be tracked and their processing managed
through a register for each set of data. These registers, for reasons of simplicity, integrity and ease
of use, are usually paper-based folders of forms.
However, they may also take the form of electronic fi les, such as spreadsheet or database fi les,
if these criteria can be satisfi ed.
9.5.1 General
Primary processing is defi ned here as the processing
steps required to prepare the data for storage in the
repository or archive from which they will be available for use in the medium term and/or long term
(Figure I.9.2). There will usually be some quality
control and verifi cation steps employed during this
process; these are described separately in later
Depending on the type of data, the primary processing stage may involve several steps and some labour,
for example, with chart processing or simple fi le
translation with little or no editing, such as for a
precisely set-up data logger.
Data may also be used prior to this processing, for
example, with telemetered water levels; however,
users should be made aware that the data are unverifi ed and may contain errors.
Secondary processing is regarded here as the steps
necessary to produce data in a converted, summarized or reduced form, for example, daily rainfall
from event data or daily mean discharges from stage
and rating data. This is covered further in 9.7.
9.5.2 Preliminary checking of data
The difference between preliminary checking and
error detection is rather arbitrary. Procedures
included under preliminary checking in one country may be thought of as error detection in another.\

Figure I.9.2. A two-stage processing/updating procedure for hydrological data

Also, the extent of the use of computers in processing the data may change the definition of
preliminary checking. For instance, for data
collected manually and later captured to computer
fi les (perhaps by key entry or optical scanning), the
term preliminary checking will be used to cover
those procedures performed prior to transcribing
the data into machine-readable form. For data
collected directly to computer-readable fi les, there
may be little prior checking other than verifying
the proper identifi cation of the batch and its associated manual readings (identification of site of
collection, proper beginning and ending dates of
this unit of data, and proper identifi cation of the
type of data involved, such as items sampled and
frequency of sampling).
For data collected manually, preliminary checking
should generally include the following steps:
(a) Log in data to a register at the time of the receipt
of the report form;

(b) Ensure completeness and correctness of the
associated information, that is, dates, station
name and station identifi cation, if required in
subsequent machine processing;
(c) Ensure completeness of the data;
(d) Check the observer’s arithmetic, if any;
(e) Compare the observer’s report with the recorded
In many countries, this last step will be taken
following computer plots of the data.
Corrections on the forms should be entered legibly
and in ink of a different colour from that used in
the completion of the original form, making sure
that the original entry is not erased or made illegible. It is usually best if these corrections are also
dated and signed by the person carrying them out.
Certain preliminary checks should also be applied
to data from chart recorders. The times recorded at
the beginning and end of the chart, and at any time
check in-between, should be checked against the
timescale on the chart to determine if time corrections need to be applied, or to determine the
magnitude of the time correction. An attempt
should be made to determine whether a time correction is due to clock stoppage or whether it can
reasonably be spread over the chart period. All
manual readings should be written on the chart in
a standard format.
For data-logger data, preliminary checks on time
and comparison with manual readings will
normally be part of the fi eld data capture routine.
Depending on the software available, checks may
also include plotting the data on screen before
leaving the station to confi rm that all is operating
correctly. Where this is possible, it should be
included as part of standard procedures.
Upon return to the offi ce, preliminary checks will
include registering the data and the associated
information, and appropriate backups of the fi les in
at least three independent places (for example, PC
hard drive, portable disk and network drive).

For telemetered digital data, there may be little or
no preliminary checking before they are handed
over to a user. In such situations, the user should
be made well aware of the unverifi ed condition of
the data and use them accordingly. Even when
there is an automated checking process in operation, it will only be able to check certain aspects of
the data (for example, range, spikes, steps or missing values) and the user should be made aware of
the limitations.
Usually, and this is recommended procedure, telemetered data will also be fi led at the station or another
secure part of the system and will only be updated
to the archive and/or verifi ed status after full preliminary checks (as for data-logger capture above) plus
error detection and validation. Quality codes, if
used, may assist with communicating these issues
to users.

9.5.3 Traceability and processing
Hydrological data are valuable in that they are relatively expensive to collect and irreplaceable, and
potentially have very high value following certain
events. To realize and maintain their value, a means
of verifying the accuracies and giving assurance
that errors are largely absent must exist. Thus the
traceability of the data and the methods used to
collect and process them must be available in a
readily followed form. Hydrological agencies should
establish procedures aimed at achieving this traceability, in conjunction with the efficient processing
of the data while preserving and verifying their
A data-processing system should include provisions
(a) Registering the data after collection to confirm
their existence and tracking their processing;
(b) Keeping backups of the data in original form;
(c) Positively identifying the individual batches of
data at the various stages of processing;
(d) Identifying the status of data as to their origin
and whether they have been verified as fi t for
(e) Presenting and storing evidence of any modifi –
cations to the data;
(f) Filing all fi eld observations, log books, forms,
etc., which verify the data;
(g) Controlling the amount and type of editing
which can be performed and providing the
authorization to do so;
(h) Presenting the data in a number of ways for
checking and auditing by trained persons who
are to some extent independent of the process.
9.5.4 Data registers and tracking
As soon as data reach the fi eld offi ce (whether by
telemetry, as computer fi les, charts or on handwritten forms), they should be entered into one of a set
of registers, usually classifi ed by to station and data
type, and in chronological order.
These registers are usually on hard copy in a folder
(but may be electronic) in the hydrological offi ce
and are updated daily as batches of data arrive.

Initially this involves noting the starting and ending
times of the data batch, and continues with confi rmation of editing, checking and updating the
database. Each step should be signed with the staff
member’s initials and date, in order that staff take
responsibility for, and gain ownership of, their work
and its progress. The registers will thus contain a
verifi ed chronological record of data-processing
activities in the fi eld offi ce.
9.5.5 Identifi cation and retention of
original records
All data are to be permanently identifi ed with the
station numbers and other required codes. Charts
require a stamp or label to prompt for the dates and
times, manual readings, etc., for both chart on and
chart off. Forms, cards and other hard copy should
be designed with fi elds that prompt staff for the
required information.
This material should be archived indefi nitely in
suitably dry and secure storage, with an adequately
maintained indexing system to enable retrieval of
specifi c items when required. Some safeguards
will be required to prevent undue deterioration,
for instance against mould, insects, vermin or
birds. Some media may gradually become unreadable for other reasons; for instance, foil-backed
punched-tape will bind to itself if tightly wound,
and will also become diffi cult to read as the tape
readers and their software become obsolete and
unable to be run. In such instances it may be
advisable to use imaging capabilities to store this
information in an electronic format. Such a decision will need to be based on a cost–benefit
Electronic data should have an adequate fi le-naming
system and an archive of the unaltered original data
fi les. It is permissible, and normally advisable, for
these fi les to be transformed into a durable computer-readable format that will include the station and
other identifi ers, and in future not depend upon
software or media that may become obsolete. It is
recommended that data-processing and database
managers pay attention to this issue when developing and updating their systems.
It is recommended that an original record of electronic data be retained on hard copy in the form of
a suitably large-scale plot or perhaps a printout of
the values if the data set is small. This is fi led initially
in the processing offi ce, both as a backup of the
data and as the record of processing if all transformations, modifi cations and other steps are written
on it, and signed and dated by the processor. Other
documents, such as plots following modifi cations,
should be attached, and if adequate comments are
added, this can become a simple but complete
record of the processing. Some of the more advanced
software packages may include provisions to
compile such a record electronically, and some
offi ces can confi gure proprietary database software
to do this; however paper-based systems are generally simpler, arguably more foolproof and easier to
understand and use.
9.5.6 Data adjustments for known
These are the errors reported by the fi eld technician
or those persons responsible for manual quality
control of the incoming data sets. Corrections for
these errors must be made before the data are
submitted for validation. The errors may be associated with a gradual drift of the clock, sensor device
or recording mechanism, but may also be caused by
discrete events, for example, clock stoppage or electronic fault. In the former case, a time-series
database may automatically perform the required
adjustment using linear or more complex scaling of
recorded values. In the latter case, it may be possible for manual estimates of missing data to be
inserted under certain conditions (9.4 above), if the
affected period is not too long and suffi cient background information is available.
Adjustments may also be required to compensate
for more complex phenomena, such as the presence of ice at river-gauging stations. In this case, it
is almost certain that the corrected stage values
would be manually computed for the affected
period. Again, this needs to be strictly controlled as
to which assumptions are permissible. Reporting of
errors should use standard procedures and standard
forms. The form may be used for noting corrections
to stage or fl ow. An essential feature of the correction process, whether performed manually or by
computer, is that all modifi ed data should be suitably fl agged and/or comments fi led to indicate all
adjustments that have been made.
Where adjustments are to be made to either time or
parameter values in each batch, the following
points must be answered before proceeding:
(a) Are there valid reasons for the adjustments?
(b) Are the adjustments consistent with the previous batch of data?
(c) Is any follow-up fi eldwork necessary, and have
the fi eld staff or observers been notifi ed?
Answers to (a) and (b) must be fully documented in
the processing record.

9.5.7 Aggregation and interpolation of
Many variables, because of their dynamic nature,
must be sampled at relatively short periods, but are
only utilized as averages or totals over longer periods. Thus, for many hydrological applications,
climatological variables may be required only as
daily values, but must be sampled with greater
frequency to obtain reliable daily estimates.
Temperature and wind speed are good examples,
but the same can be true for water-level and riverfl ow data. In the past, when computer storage costs
were more signifi cant, the levels of data aggregation
were sometimes different for data output and data
storage purposes. Modern time-series databases,
however, will generally have effi cient storage and
retrieval capabilities that will enable the archiving
of all data points. Data at a high level of aggregation, for example, monthly and annual averages,
will be used for reporting and publications or kept
as processed data files for general reference
9.5.8 Computation of derived variables
Derived variables are those parameters that are
not directly measured but need to be computed
from other measurements. The most common
examples are runoff and potential evapotranspiration. However, the full range of derived variables
is very broad and includes many water quality
One important database management decision
is to determine whether derived variables need
to be stored after they have been estimated and
reported. It is obviously not essential to occupy
limited storage space with data that may readily
be recomputed from the basic data held. For
example, it is not usual to store sediment and
dissolved salt loads, because these are used less
frequently and may be very rapidly computed by
the multiplication of two basic time series, flow
and concentration.
Two differing examples illustrate this; in the United
States, the Water Data Storage and Retrieval
(WATSTORE) system (Hutchinson, 1975; Kilpatrick,
1981) keeps online daily average fl ows, while in New
Zealand, the Time Dependent Data (TIDEDA) system
(Thompson and Wrigley, 1976) stores only stages in
the original time series and computes fl ows and
other derived variables on demand. The only fi xed
rule is that whatever subsequent values are derived,
the original data series should be preserved, in an
offl ine, stable, long-term storage facility.
Most modern time-series databases will generally
have capabilities such that recomputation is not
an issue. With the deployment of data loggers
featuring signifi cant onboard programming and
processing capability, a more signifi cant issue is
whether such variables should be computed by
the data logger prior to primary data processing.
It is recommended that they not be computed, in
order to control and achieve standardization of
methods; it is far easier to verify the methodology
in a data-processing system rather than within
the programs of a number of instruments that
will inevitably become disparate over time and
9.5.9 Data status
The status of the data must be carefully monitored to determine whether they require validation
or editing, or are in their fi nal form and ready for
use. Some database systems add a code to signify
this; others provide a means of restricting access
for manipulation and editing according to status
level. For example, the United States Geological
Survey’s Automated Data Processing System
(ADAPS) has three status levels: working, inreview and approved. Database systems without
these facilities will require operating rules specifying which particular members of staff can have
access to the various working and archive directories, and other privileges such as the write
protection on fi les. It is recommended that, where
possible, only one person operate this type of
system for any one database. An alternative
person, preferably at a senior level, should also
have privileges in order to act as a backup, but
only when essential.
The above general procedures may be applied to
various hydrological data types to differing
degrees, and it is necessary to identify some of
the specific procedures commonly practised.
Several WMO and FAO publications (for
instance, WMO-No. 634) deal directly with many
of the procedures described below, and reference
to the relevant publications will be made
frequently. These texts should be consulted for
background theory and the formulation
of techniques, primarily for manual
processing. This section presents some additional information required to extend such

9.6.1 Climatological data [HOMS H25]
For hydrological applications, apart from precipitation, the most signifi cant climatological variables
are temperature, evaporation and evapotranspiration, in the order of progressive levels of processing
complexity. Before reviewing the processing tasks,
it is useful to consider the means by which most
climatological data are observed and recorded,
because this has a signifi cant impact on subsequent operations. The wide range of climatological
variables and their dynamic nature have resulted
in the situation whereby the majority of the
primary data are obtained from one of two sources
– permanently occupied climate stations and packaged automatic climate (or weather) stations. The
implication of the fi rst source is that the observers
tend to be well trained and perform many of the
basic data-processing tasks on site. Since the
processing required for most parameters is quite
simple, fi eld processing may constitute all that is
required. Even where more complex parameters
need to be derived, observers are usually trained to
evaluate them by using specially constructed
monograms, or possibly computers or electronic
logging and data transfer devices. Thus, computerrelated primary processing, if performed at all,
largely comprises the verifi cation of the manual
The implication of the use of automatic
climatological stations is that there exists a
manufacturer-supplied hardware and software
system capable of performing some of the data
processing. Indeed, many climatological stations
are designed specifi cally to provide evaporation
and, normally, Penman-based, evapotranspiration
estimates (Chapter 4). However, it should be
carefully considered whether such variables should
be computed by the data logger prior to primary
data processing. This is not recommended in order
to control and achieve standardization of methods.
If variables do need to be computed, for instance
where these data are used in near-real-time, it is
preferable if the raw data are also entered into the
database for computation using a more controlled
and standardized method. Care must be exercised
in the use of automatic climate-station data because
the range of quality of sensors is highly variable by
comparison with most manual climate stations.
Further details on the processing of climatological
data can be found in the Guide to Climatological
Practices (WMO-No. 100).
There are several climatological variables that need
to be transformed to standard conditions for storage and/or application. For example, wind speeds
measured at non-standard heights may need to be
transformed to a standard 2- or 10-m height by
using the wind speed power law. Similarly, pressure
measurements may be corrected to correspond to a
mean sea-level value, if the transformation was not
performed prior to data entry. Evaporation and evapotranspiration
observations [HOMS I45, I50]
Where direct measurement techniques are used,
the computer may be used to verify evaporation
estimates by checking the water levels (or lysimeter
weights), and the water additions and
To compute lake evaporation from pan data, the
relevant pan coeffi cient needs to be applied. In
some cases, the coeffi cient is not a fi xed value,
but must be computed by an algorithm involving
other climatological parameters, for example,
wind speed, water and air temperature, and
vapour pressure. These parameters may be represented by some long-term average values or by
values concurrent with the period for which pan
data are being analysed. Pan coeffi cients, or their
algorithms, must be provided in the station
description fi le (Chapter 2). If an algorithm uses
long-term average values, these too must be
stored in the same fi le.
Details on the estimation of evaporation and
evapotranspiration are discussed in Chapters 2 and

  1. Existing computer programs for solving the
    Penman equation are provided in HOMS component I50. Precipitation data [HOMS H26]
    Data from recording precipitation gauges are
    frequently analysed to extract information relating to storm characteristics, whereas data from
    totalizing (or storage) gauges serve primarily to
    quantify the availability and variation of water
    Before analysing any data from recording gauges, it
    is necessary to produce regular interval time series
    from the irregular series in which the data are
    usually recorded. If the data have been subjected to
    a previous stage of validation, this time-series
    format conversion may already have taken place.
    The computer program used for conversion should
    allow the evaluation of any constant interval time
    series compatible with the resolution of the input
    data. The selection of a suitable time interval is
    discussed below.

Whether the data are derived from recording
or totalizing gauges, first priorities are the
apportionment of accumulated rainfall totals and
the interpolation of missing records. Accumulated
rainfall totals are common in daily precipitation
records when, for example, over a weekend, a gauge
was not read. However, they are also common with
tipping-bucket gauges that report by basic systems
of telemetry. If reports of bucket tips are not received
during a rainfall period, the fi rst report received
after the gap will contain the accumulated number
of bucket tips. The difference between this accumulation and that provided by the last report must be
apportioned in an appropriate manner. The techniques for apportioning accumulated totals and
estimating completely missing values are essentially
the same. Apportioned or estimated precipitation
values should be suitably fl agged by the process
that performs these tasks. Exactly the same techniques may be applied to shorter interval data from
recording gauges; however, estimates of lower quality will be obtained because there will usually be
fewer adjacent stations and because of the dynamic
nature of short-term rainfall events. Precipitation
may also be measured using different instruments
and at non-standard heights. Therefore, data may
need to be transformed to a standard gauge type
and height for consistency. Further details on the
processing of climatological data can be found in
the Guide to Climatological Practices (WMO-No. 100),
and in Chapter 3.
9.6.2 Streamfl ow data
[HOMS H70, H71, H73, H76, H79]
There are several processing steps required to
produce streamfl ow data. The fi rst deals with the
water-level series data, the second covers the fl ow
measurements, the third incorporates the gauged
fl ows into rating curves and the fourth involves
the fi nal step of applying the rating curves to the
series data in order to compute flows.
Comprehensive details of techniques for flow
computation are presented in the Manual on Stream
Gauging (WMO-No. 519), but several time-series
databases do this as a standard process. Water-level series data
As for other series data, water-level series data
should fi rst be verifi ed that the start date, time and
values for the batch match up with the end values
for the previous batch. It is useful if the initial
processing provides the maximum and minimum
gauge heights to enable an initial range check. Any
unusually high or low values should be fl agged to
check that the values are in context.
The batch should be plotted to a suitable scale and
examined to detect such problems as the following:
(a) Blocked intake pipes or stilling wells, which
will tend to show up as rounded-off peaks and
recessions that are unusually fl at;
(b) Spikes or small numbers of values which are
obviously out of context and incorrect due to
an unnaturally large change in value between
adjacent data points. Such situations can occur,
for instance, with errors in the digital values
between a sensor and a data logger input port;
(c) Gaps in the data, for which there should be
known explanations, and a note as to the remedial action taken;
(d) Errors induced by the fi eld staff, such as pumping water into a well to fl ush it;
(e) Restrictions to the movement of the fl oat/
counterweight system or the encoder (perhaps
caused by a cable of incorrect length);
(f) Vandalism or interference by people or animals;
(g) Debris caught in the control structure, or other
damming or backwater condition.
Note that such problems, if not detected during the
fi eld visit, must be investigated at the earliest opportunity. In cases where the cause has not been
positively identifi ed, full processing of the data
should be delayed until it can be investigated on
Following examination of the plot, a description of
the problems identifi ed should form part of the
processing record together with any printout showing data adjustments, editing and any other work.
As well as any comments, the following should also
be included:
(a) The date the data was plotted;
(b) The signature of the processor;
(c) Notes of all corrections to the data, plus any
subsequent actions carried out, which change
the data in the fi le from that which is plotted
(for example, removal of spikes or insertion of
manual data resulting from silting). Normally, if
any corrections are applied, another plot is added
to show their effects, and all evidence is fi led.
Charts should be stamped with the prompts for
date, time and water-level readings at chart off and
chart on. As these indicate the adjustments to the
original observations, the need for clear annotation
is obvious. Flow measurements
The computation of fl ows from current-meter gauging data is normally done in the fi eld offi ce or in
the fi eld depending on the instrumentation. Other

methods, such as volumetric, dilution or acoustic
moving-boat methods, will have a variety of computation methods that will also normally be performed
in the fi eld or fi eld offi ce. Further processing will
include a computation check and possibly some
post-processing if any adjustments are found necessary, for example to instrument calibrations, offsets,
or edge estimates.
Primary processing will also include recording in
the gauge register, and plotting on the rating curve,
if applicable, and entry of the results into the database. Depending on the method and associated
software, it may also include fi ling of the complete
raw data fi le in the appropriate part of the hydrological database.
Owing to their infl uence on subsequent fl ow estimates, it is recommended that fl ow measurement
data be submitted to suitable verification. This
should include the calculation of statistical uncertainty using recognized methods, for example, in
ISO 748 (ISO, 1995). If available from the techniques
and software used, the process of verifi cation should
also include an examination of plots of the crosssection as well as of the velocities measured to check
for gross errors and inconsistencies. If warranted by
experience, provision should also be made to perform
corrections for excessive defl ection of sounding lines
for suspended meters, and cases where velocities are
not perpendicular to the gauged section (Manual on
Stream Gauging (WMO-No. 519)). Rating curves
Rating curves defi ne the relationship between stage
and flow. This relationship is determined by
performing many river gaugings over a wide range
of fl ows and by using the stage and discharge values
to defi ne a continuous rating curve. While control
structures may have standard, theoretical ratings, it
is a recommended practice to rate structures in the
fi eld.
Traditionally, rating curves have been manually
fi tted to the plotted measurements. In many cases
the curve may be fitted more objectively by
computer methods. If necessary, weights may be
assigned to each discharge measurement to refl ect
the subjective or statistical confi dence associated
with it. However, because some sections have
several hydraulic control points, many hydrologists
prefer to keep the defi nition of rating curves as a
manual procedure. Many factors have an impact on
the quality of the rating curve. It is thus imperative
that a fl ow-processing system be able to identify
and locate the correct rating curve and be aware of
its limits of applicability. Of particular note is the
importance placed on preserving historic rating
curves to allow fl ows to be recomputed.
There are two forms in which rating curves may be
stored in the computer: functional and tabular. The
tabular form is still the most common, and the
table is prepared by manually extracting points
lying on the rating curve. The extraction is
performed so that intermediate points may be interpolated, on a linear or exponential basis, without
signifi cant error in fl ow estimation. The functional
form of the rating curve has one of three origins:
(a) A theoretical (or modifi ed) equation for a
gauging structure;
(b) A function fi tted by the computer to the gauged
points, that is, automation of the manual curvefi tting process;
(c) A function fi tted to the points of the table
prepared as described in the previous paragraph,
that is, a smoothing of the manually-fi tted
curve. A time-series database with hydrological
functions will normally use this method. Flow computation
Modern time-series database software will include a
standard process to apply fl ow ratings to water-level
series. Whether minor changes of bed level are
applied as new ratings, or as shifts that apply a
correction or offset to the water-level data, will
depend on the capability of the software.
For either method, the rating to be applied must
cover the range of water levels for the period, and if
necessary should have been extrapolated by a recognized and defensible method and be valid for the
period. Stage values should have been validated
following any required corrections for datum,
sensor offset and timing error. Where rating curves
are related to frequently changing artifi cial controls,
for example, gates and sluices, a time series of
control settings may be needed to guide computer
selection of the relevant rating curve.
Although the compilation of rating curves is simple
in theory and is a standardized procedure, there is
often some interpretation and decision-making
required. This is due to the likelihood that the
number and timing of fl ow gaugings will be inadequate because of practical diffi culties in getting this
work done. It is possible that the knowledge and
experience of the hydrologist will be necessary to
cope with such questions as:
(a) Which fl ood among those that occurred
between two successive gaugings has caused a
rating change or shift?

(b) Over what period of, for instance, a fl ood,
should a progressive change from one rating to
another be applied?
(c) Should a high fl ow gauging carried out in bad
conditions or using a less accurate technique
be given less weight when it plots further from
the curve than the velocity-area extension
These and similar questions can lead to rating
curves being subject to considerable scrutiny and
sometimes revision following the completion of
additional flow gaugings, especially at high
fl ows.
A problem frequently encountered when applying multiple rating curves is that they can
produce abrupt changes in flow rates at the time
of change-over. If the processing system does not
have the capability to merge ratings over time,
then some means of manual adjustment of flows
during the transition period will be required. If
the stage values are being altered instead of the
rating curve (not recommended), then these
shifts can be varied over the time period to
achieve this.
Shifting the stage values instead of producing a new
rating is not recommended because:
(a) With original data being given a correction that
is in fact false, considerable care and resources
need to be applied to ensure that the correct
data are safeguarded, and that the shifted data
are not used inappropriately;
(b) The processes of determining, applying and
checking shifts add considerable complexity to
(c) Quality control processes (such as plotting
gauging stages on the stage hydrograph or
plotting gauging deviations from the rating)
become more diffi cult to implement and use.
With modern hydrometric software, the effort
required to compile and apply new rating curves is
much reduced, thus avoiding the need to alter stage
data as a “work-around”.
9.6.3 Water-quality data
There are four main areas of activity in the primary
processing of water quality data:
(a) Verifi cation of laboratory values;
(b) Conversion of measurement units and
adjustment of values to standard reference
(c) Computation of water quality indices;
(d) Mass–balance calculations.
Verifi cation of laboratory results may comprise the
re-evaluation of manually-computed values and/or
consistency checks between different constituent
values. These operations are essentially an extension of data validation techniques.
The standardization of units is important in
obtaining consistency of values stored in the database. The operations necessary comprise the
conversion of measurement units used, such as
normality to equivalence units, or correction of
values to match a reference standard, for example,
dissolved oxygen and conductivity values transformed to corresponding values at a standard water
temperature of 20°C.
Water quality indices are generally based on empirical relationships that attempt to classify relevant
characteristics of water quality for a specifi c purpose.
Thus, indices exist for suitability, such as drinking,
treatability, toxicity or hardness. Since these indices
are derived from the basic set of water quality data,
it is not generally necessary to store them after they
have been reported. They may be recomputed as
Some indices have direct signifi cance for water
management. For example, empirical relationships
of key effl uent variables may be used as the basis of
a payment scheme for waste-water treatment – the
higher the index, the higher the charges.
Mass-balance calculations are performed to monitor pollution loadings and as a further check on the
reliability of water quality data. Loadings are calculated as the product of concentration and fl ow (or
volume for impounded water bodies). By computing loadings at several points in a river system, it is
possible to detect signifi cant pollution sources that
may otherwise have been disguised by variations in
fl ow. Obviously, mass-balance calculations must be
performed after fl ows have been computed. Mass–
balance calculations may be performed easily for
conservative water quality constituents, that is,
those that do not change or that change very slowly
with time.
Non-conservative constituents, for example,
dissolved oxygen and BOD, may change extremely
rapidly and quite sophisticated modelling techniques are required to monitor their behaviour.
Additional information and techniques can be
found in the Manual on Water Quality Monitoring:
Planning and Implementation of Sampling and Field
Testing (WMO-No. 680) and the Global Environment
Monitoring System (GEMS) Water Operational Guide

Secondary processing is regarded here as the steps
necessary to produce data in a converted, summarized or reduced form, for example, daily rainfall
from event data or daily mean discharges from stage
and rating data. It also covers the secondary editing
following more complex validation, and the insertion of synthetic data into gaps in the record.
In addition, the regrouping of data and additional
levels of data coding may be performed and measurement units may be converted to the standards
adopted in the database. The conversion of irregular to regular time series is also one of the operations
necessary in many cases. There are many options
regarding the way in which data may be compressed
for effi ciency of storage, but modern database software and hardware are reducing the need for
9.7.1 Routine post-computation tasks
A signifi cant task for all data processing particularly
relevant to fl ow data is that of performing the necessary housekeeping operations on data sets. These
operations implement decisions on which data sets
should be retained, and discard any data sets that
are superfl uous or incorrect and could possibly be
mistaken for the correct data set. It is advisable to
save only the essential basic data (and security
copies) and, possibly, depending on the database
software capabilities, any derived data that are very
time-consuming to create. For example, in some
systems, daily mean fl ows are time-consuming to
derive and thus these data sets are retained as prime
data. On the other hand, some agencies use software (packages such as TIDEDA of New Zealand and
Time Studio of Australia) that compute these data
sets rapidly from the archived stage and rating data,
and thus there is no point in storing them beyond
the immediate use. (It is inadvisable to do so,
because these systems make it easy to update ratings
when additional data become available, and thus
the most up-to-date fl ow data will automatically be
available.) (Details on the HOMS component,
TIDEDA, are available on the WMO website at
Depending on the systems used, and for guidance
purposes, the following fl ow-related data should
usually be preserved:
(a) The stage data in their original unmodifi ed
form (data-logger fi les should appear where
station and date/time information are embedded or attached);
(b) Field data relating to time and stage correction,
and the corrections that have been applied
in primary processing, along with the name
of the person who made the corrections and
(c) The adjusted stage data, that is, the time series of
water levels corrected for datum, gauge height
and time errors. A working copy and at least
one security copy should be held (offl ine);
(d) The gaugings in their original form, whether
written cards or computer fi les such as ADCP
(e) Rating curves, also in their original form,
whether paper graphs or graphical editor fi les;
(f) Any associated shift corrections;
(g) If relevant daily average fl ows;
(h) Basin water-use data used to obtain naturalized
fl ows, if these are calculated.
Generally, most other data sets will be transient or
may be readily derived from these basic data sets.
It is vital that all electronic data sets be backed up
offl ine and offsite. It is advisable to keep multiple
backups at different frequencies of rewrite, with
that of longer frequency stored in a different town
or city. Original paper records should be kept in
designated fi re-rated and fl ood/waterproof purposebuilt rooms, with access strictly controlled, or
alternatively, if cost-effective, imaged and stored

9.7.2 Inserting estimates of missing
The usefulness of data is, to a great degree, dependent on its completeness. However, fi lling missing
data with estimates can severely compromise its
value for certain purposes, and as future purposes
may not be apparent when the data are collected
or processed, this should be done with great
caution and restraint. It should also be traceable,
so that the presence of fi lled-in data is apparent to
the user and the process can be reversed if
As mentioned in 9.2, the hydrologist has a duty
to be conservative in carrying out any correction
of data. An agency should formulate strict criteria for altering or adding data values, and this
work must always be done using assumptions
based on evidence rather than any element of
The following are some suggested criteria relating
to water-level and rainfall data as used for the
national Water Resources Archive in New Zealand

(National Institute of Water and Atmospheric
Research, 1999, unpublished manual):
(a) No alterations shall be made unless justifi cation for the assumptions made is scientifi cally
defensible, and recorded, as below;
(b) Such alterations must have the explanation
recorded in the processing record, which may
be a plot of the original data in the station’s
register or as a comment on the database;
(c) As a general guideline, gaps due to missing
records will not be fi lled with synthetic data
or interpolated. Any approximate data shall be
made available to users by inclusion or reference to them in a comment in the database.
Exceptions to the non-use of synthetic data and
interpolation are in (d) to (e) below;
(d) A gap in a water-level record may be fi lled with
a straight line or curve as applicable, if all of the
following conditions are fulfi lled:
(i) The river is in a natural recession with the
water-level lower (or the same) at the end
of the period;
(ii) It has been ascertained that no signifi cant
rain fell in the catchment during the time
of concentration that would relate to the
gap period;
(iii) The catchment is known to be free of
abstractions and discharges that modify
the natural fl ow regime, for example,
power station and irrigation scheme;
(iv) The resulting plot of the data shows
consistency with the data on either side;
(v) In some situations (for example, power
stations), an adjacent station may measure the same data or almost the same
data. In the former case, the record can be
fi lled in as if it were a backup recorder. In
the latter, the data may be fi lled in if the
uncertainty is less than that in the standard or if the correlation between stations
for that parameter and that range can be
shown to be 0.99 or greater. A comment
containing the details of the relationship
must be fi led;
(vi) The station is not a lake that normally has
seiche or wind tilt (these are often studied,
and a synthetic record will not be able to
recreate the phenomena);
(e) Where the conditions do not meet these criteria,
but trained personnel were on site for the whole
period (for example, for desilting the stilling
well) and recorded manual observations, the gap
may be fi lled with these values and interpolated
(f) Filling a gap in the original data record with
synthetic data derived by correlation is not
permissible. These can be used, however, for
supplying data requests where the user is
informed of the uncertainty involved. Such
data must be carefully controlled to ensure they
are not erroneously fi led on either the local or
central archives;
(g) A gap in a rainfall record may be interpolated
only if it can be established that no rain fell
during the period, by means of correlation with
other gauges inside or outside the catchment
area for which there is an established correlation and with a correlation coeffi cient of 0.99
or higher.
A professed reluctance to archive data that do not
meet strict standards has the advantage of focusing
an organization on taking steps to reduce the
amount of missed data. As many root causes of
missed data are preventable, a culture where people
strive to improve performance in these areas makes
a tangible difference to overall data quality.
Where it is necessary to fi ll gaps left by missing
records, as it inevitably will be for some types of
analysis, time spent on the estimation during the
preprocessing stage may pay large dividends when
the fi nal data are used or analysed. It is also appropriate that these fi rst estimates be made by the data
collector with the benefi t of recent and local knowledge. It is often the case, however, that reconstructing
faulty records is time-consuming or that recovery
requires access to processed data from another
source covering the same period. A decision must
be made as to whether the onus for the initial estimation of the missing record lies with the collector,
or whether it could be synthesized more effi ciently
later in the process by using tertiary-processing
Some attempt is normally made to fi ll data gaps by
cross-correlation with nearby gauging stations,
particularly those in the same river system. In the
absence of reliable cross-correlation relationships,
rainfall-runoff models, including the use of conceptual catchment models, may be used. All estimated
data should be suitably fl agged or stored in a separate archive.
Many river systems are signifi cantly affected by
human activities and these effects tend to change
with time. For hydrological and water-resources
studies, it is frequently necessary to try to isolate
these artifi cial effects from the natural catchment
response, that is, to try and obtain a stationary time
series. This process requires extensive background
information on all forms of direct and indirect
diversions, discharges and impoundments in the
catchment. Water-use effects may be aggregated

into a single time series of net modifi cations to river
fl ow. When these corrections are applied to the
measured streamfl ows, a naturalized time series is
obtained. Again, any modified data should be
appropriately fl agged.
A somewhat artifi cial distinction has been made
between primary and secondary processing procedures and validation procedures for the purposes of
simplicity in this chapter. Data validation procedures commonly make comparisons of test values
against the input data and will often exist at several
levels in primary processing, data checking and
quality control. They may include simple, complex
and possibly automated checks being performed at
several stages in the data-processing and archiving
path. Some may also be performed on data outputs
and statistical analyses by an informed data user.
As elements of quality control, the aim is to ensure
the highest possible standard of all the data before
they are given to users.
9.8.1 General procedures
While computerized data validation techniques are
becoming more useful and powerful, it should be
recognized that they can never be fully automated
to the extent that the hydrologist need not check
the flagged values. Indeed, to obtain the best
performance, the hydrologist may need to constantly
adjust threshold values in the program and will
need to exercise informed and considered judgement on whether to accept, reject or correct data
values fl agged by the programs. The most extreme
values may prove to be correct and, if so, are vitally
important for all hydrological data applications.
Validation techniques should be devised to detect
common errors that may occur. Normally the
program output will be designed to show the reason
the data values are being fl agged. When deciding
on the complexity of a validation procedure to be
applied to any given variable, the accuracy to which
the variable can be observed and the ability to
correct detected errors should be kept in mind.
It is common to perform validation of data batches
at the same time as updating the database fi les,
normally on a monthly or quarterly basis. Some
organizations carry out annual data reviews that
may rerun or use more elaborate checking processes
in order to carry out the validations after multiple
batches have been updated to the archive. In some
cases these will be run on the entire data set for a
station. Such a system reduces considerably the
error rate of data arriving at the central archive,
where normally further validation is performed.
Perhaps a more signifi cant advantage of having
these procedures in place is that the responsibility
for the major part of the validation process is
assigned to the observers themselves.
There is no doubt that visual checking of plotted
time series of data by experienced personnel is a
very rapid and effective technique for detecting
data anomalies. For this reason, most data validation systems incorporate a facility to produce
time-series plots on computer screens, printers and
plotters. Overplotting data from adjacent stations is
a very simple and effective way of monitoring interstation consistency.
9.8.2 Techniques for automated
In order to review the wide range of techniques
available for automated validation systems, it is
useful to refer to absolute, relative and physiostatistical errors.
Absolute checking implies that data or code values
have a value range that has zero probability of being
exceeded. Thus, geographical coordinates of a
station must lie within the country boundaries, the
day number in a date must lie in the range 1–31
and in a numeric-coding system the value 43A
cannot exist. Data failing these tests must be incorrect. It is usually a simple task to identity and
remedy the error.
Relative checks include the following:
(a) Expected ranges of variables;
(b) Maximum expected change in a variable
between successive observations;
(c) Maximum expected difference in variables
between adjacent stations.
During the early stages of using and developing the
techniques, it is advisable to make tolerance limits
fairly broad. However, they should not be so broad
that an unmanageable number of non-conforming
values are detected. These limits can be tightened as
better statistics are obtained on the variation of
individual variables.
While requiring much background analysis of
historical records, the expected ranges for relative
checks (method (a) above) should be computed for
several time intervals, including the interval at

which the data are observed. This is necessary
because the variance of data decreases with increasing time aggregations. Daily river levels would fi rst
be compared with an expected range of daily values
for the current time period, for example, the current
month. Since there is a possibility that each daily
value could lie within an expected range, but that
the whole set of values was consistently (and erroneously) high or low, further range checks must be
made over a longer time period. Thus, at the end of
each month, the average daily values for the current
month should be compared with the long-term
average for the given month. In a similar way, at
the end of each hydrological year, the average for
the current year is compared with the long-term
annual average. This technique is of general applicability to all hydrological time-series data.
The method of comparing each data value with the
immediately preceding observation(s) (method (b)
above) is of particular relevance to variables exhibiting signifi cant serial correlation, for example,
most types of water-level data. Where serial correlation is very strong (for example, groundwater
levels), multiperiod comparisons could be performed
as described for method (a) above. Daily groundwater observations could first be checked against
expected daily rates of change, and the total
monthly change could subsequently be compared
with expected monthly changes.
Method (c) above is a variation of method (b), but
it uses criteria of acceptable changes in space rather
than time. This type of check is particularly effective for river-stage (and river-fl ow) values from the
same watershed, although with larger watersheds
some means of lagging data will be necessary before
inter-station comparisons are made.
For other hydrological variables, the utility of
this technique depends upon the density of the
observation network in relation to the spatial
variation of the variable. An example is the
conversion of rainfall totals to dimensionless
units by using the ratio of observed values to
some long-term average station value. This has
the effect of reducing differences caused by
station characteristics.
Physio-statistical checks include the use of regression between related variables to predict expected
values. Examples of this type of checking are the
comparison of water levels with rainfall totals and
the comparison of evaporation-pan values with
temperature. Such checks are particularly relevant
to observations from sparse networks, where the
only means of checking is to compare with values
of interrelated variables having denser observation
Another category of physio-statistical checks
involves verification that the data conform to
general physical and chemical laws.
This type of check is used extensively for water
quality data.
Most of the relative and physio-statistical checks
described above are based on the use of time series,
correlation, multiple regression and surface-fi tting
9.8.3 Routine checks
Standard checks should be formulated as part of an
organization’s data-processing procedures, and
applied routinely to test the data. These will usually
involve checking the data against independent
readings to detect errors in time and magnitude.
Instrument calibration tests are also examined and
assessed for consistency and drift. A visual examination is made of sequential readings, and
preferably, of plots of the data in the light of
expected patterns or comparisons with related
parameters that have also been recorded.
On the basis of these assessments, quality codes, if
used, may be applied to the data to indicate the
assessed reliability. The codes will indicate if the
record is considered of good quality and, possibly,
the degree of confi dence expressed in terms of the
data accuracy (9.10 on uncertainty). An alternative
to quality codes is for data to have comments to
this effect attached only if the data fail to meet the
set standards.
At this stage, any detailed comments relating to the
assessment should be attached to the data (or input
to any comment or quality code database) for the
benefi t of future users.
9.8.4 Inspection of stations
It is essential for the maintenance of good quality
observations that stations be inspected periodically by a trained person to ensure the correct
functioning of instruments. In addition, a formal
written inspection should be done routinely,
preferably each year, to check overall performance of instruments (and local observer, if
applicable). For hydrometric and groundwater
stations, this should include the measurement of
gauge datum to check for and record any changes
in levels.

For a stream-gauging station, such inspections
should include the stability of the rating curve, the
inspection duties listed below and a review of the
relationships between the gauges and permanent
level reference points to verify that no movement
of the gauges has taken place. It should also include
a review of the gauging frequency achieved and the
rating changes identifi ed. As pressures on workloads, budgets and resources increase, it is not
uncommon for so-called “discretionary” work such
as gaugings to be neglected. This is an unfortunate,
but understandable and sometimes inevitable,
trend. It is vital, for the quality of data, that resources
for gaugings be allocated and prioritized using rigorous and timely analysis of the probability and
frequency of rating changes.
Every visit for stream gauging should include instrument checks and the rating checks mentioned
above. These should be at an absolute minimum of
two per year, and preferably more often to avoid
the dangers of losing data and/or having data
severely affected by problems such as silting, vandalism or seasonal vegetative growth.
The fi eld programme should also provide for visits
by a well-trained technician or inspector immediately after every severe fl ood in order to check the
stability of the river section and the gauges. If there
is a local observer, this person should be trained to
check for these problems and communicate them
to the regional or local offi ce.
The duties of the inspector or fi eld offi cer should
(a) Noting and recording any change in the observation site (a sketch map and digital photographs are useful);
(b) Making local arrangements for the improvement or restoration of the observation site, for
example, removal of trees affecting raingauge
(c) Checking the instruments and making any
necessary fi eld repairs or adjustments;
and, where applicable:
(d) Inspecting the local observer’s record book;
(e) Instructing the observer on observation procedures and routine instrument maintenance;
(f) Emphasizing to the observer the importance
of promptly fi ling complete and accurate
(g) Briefi ng the observer on any special observations
that may be required, for example, more frequent
readings during storm and fl ood periods.
In order to perform his or her duty effectively (see
(e) above), the inspector must be kept advised of
errors made by observers, especially of any recurring
errors made by a particular observer. Such advice
should be forwarded regularly to the inspector by
the offi cers responsible for the preliminary checking
and error detection procedures. Results of these
inspections should be included in the station
history fi les.
9.8.5 Checking manually collected data
The basis of most quality control procedures for
manually collected data is computer printouts of
(usually) daily data by location or region. From
such tabular arrays, it is easy to detect by sight those
stations at which the data are consistently credited
to the wrong day or subject to gross errors.
However, caution must be exercised in changing
reported data. A study of the original report from the
station, a check against the history of the station (as
to the quality of its record), and an appraisal of the
factors that produced the event (to ensure the data in
question may not be a natural anomaly) are necessary
before an apparent error is corrected. The alteration
should be coded or commented to indicate that a
change to the raw data has been made and that all of
the above information must be documented.
Another method that can be used for checking the
relative fl uctuations of an observed element over a
period is the use of various types of mathematical
relationships, for example, polynomials. The
computed value is compared with the observed value
at the time. If the difference between the two does
not exceed the previously determined tolerance, the
data are considered to be correct. If the limits are
exceeded, then further investigation is warranted.
For data collected manually and later entered into a
computer, errors detected by either preliminary
checking or error-detection procedures should be
dealt with as follows:
(a) The correction should be made legibly on
the original form and initialled by the person
making the correction;
(b) The table or plot containing the erroneous data
should be corrected, and the correction should
be carried through to any other existing copies
of the observation and to data that may have
been derived from the erroneous observations;
(c) The station observer should be advised of the
error. If the error is of a systematic type caused
by the malfunctioning of instruments or the
failure to follow correct observing procedures,
the problem should be remedied through a
visit by the inspector;
(d) A note of the error should be made in a register so that a running check can be kept on

observational quality at all stations and so that
the fi eld or inspection staff can be advised of
stations with frequent errors.
9.8.6 Checking chart data
The ideal way of checking data captured by digitizing or scanning from charts is to plot a replica of
the chart from the data fi le just before it is archived.
If plotting processes can replicate the axes and
scales, then the two documents could readily be
compared visually on, for example, a light table. It
should be noted that the plot should differ from
the original with any corrections (such as to manual
gauge readings) and other editing that may have
been deemed necessary and valid.
If the processing system cannot provide a close replica
(such as would likely happen with circular charts),
then the plots need to be compared in rather more
detail, with indicative points on each being measured
for comparison, using a scale ruler if necessary.
9.8.7 Checking logged data
Data loggers have few original documents with
which to compare the data they capture. However
as the original, unedited data should have been
plotted and fi led in the station’s processing fi le
(Chapter 2), this document can be used in the same
way as an original chart.
For errors detected in this process, as for initial
checking, the data points should be annotated on
the document(s), which should also be fi led. Again,
the originally recorded fi le should be archived, and
a copy edited and updated to the archive.
Techniques for quality control of data differ for
various elements. The following are examples and
discussion on techniques for several parameters.
9.9.1 Flow data
Since streamfl ow data are continuous in time and
correlated in space, it is possible for the reliability of
the data to be checked by interpolation and statistical methods. Qualitative checks may also be carried
out using a number of techniques, as illustrated by
the following examples:
(a) Rainfall plotted against fl ow or stage to detect
any occurrences of freshets (or fl ood events)
without signifi cant rainfall and the reverse;
(b) Time-series plots of stage (hydrographs) or other
parameters, with overplots of manual readings
(including those from gaugings) during the
(c) Hydrograph plots of fl ow from stage series with
ratings applied, with overplots of fl ow measurements from gaugings (plotted to the same
(d) Cumulative plots (cusums) of annual rainfall overplotted with the monthly means
for the complete record and other double-mass
(e) Detection of steps in the data greater than a
nominated value (that may vary according to
the stage value). This will also normally detect
spikes where a physical or electronic fault has
produced a grossly high or low value (perhaps
maximum possible or zero, respectively);
(f) Detection of missed values (that the software
may otherwise interpolate);
(g) Printouts of periods of straight lines in a stage
record that exceeds certain user-set lengths
(can detect excessive compression or erroneous
interpolation across gaps);
(h) Overplots of the same or related parameters
(fl ow, stage, rainfall, turbidity) from nearby
stations. If there is the opportunity for overplotting stations on the same river system, this
can be particularly useful;
(i) Qualitative assessment by sight from plots, of
the shapes of the hydrograph and their correspondence with normal patterns having regard
to previous values and the given phase in the
regime of the river.
Most of the hydrological database software packages have several of these techniques either built in
or able to be run manually. Some also have the
capability to run the processes automatically as
script fi les (macros).
9.9.2 Stage (water level)
The techniques of tabular and plotted data, range
and rate-of-change checks described above are used
extensively for water-level data. Several of the
plotting techniques can be applied to both stage
and fl ow. However, as fl ow data can have errors
associated with the stage-discharge ratings, it is
important that stage be checked separately (and
normally fi rst).
The following points are the recommended minimum verifi cation techniques for stage:
(a) Checks against the manual readings recorded by
the observer at the beginning and end of each
batch plus any others recorded at intermediate

visits or by a local observer (normally done in
preliminary checking);
(b) Plots of stage with overplots of any other stage
values for the period that have been entered
into the database, such as from fl ow gaugings
or water quality data sets (this will depend on
the database);
(c) A qualitative check of the hydrograph shapes
and events, looking for suspicious features such
as straight lines, steps, spikes or freshets, fl oods
and recessions in conditions where they would
not be expected.
In addition, a number of qualitative checks, as
described above for fl ow, should be carried out. Any
apparent inconsistencies should be investigated to
the extent possible:
(a) First, checks should be carried out to determine
whether there were any comments already
recorded in the database or in the logbook for
the station, or any evidence of processing. The
observer, fi eld technician or data processor may
have already checked this event and/or noted
the actual or apparent cause;
(b) Depending on the inconsistency noted, the
following items could be checked with the fi eld
observer or against the evidence of the data
processing. Some may require specifi c investigation on site at the station, and with previous
and succeeding batches of data:
(i) If there is a “sluggish” peak and recession,
a fi eld check of the stilling well and intake
pipe for silt blockages may be required;
(ii) Steps or spikes in the record may indicate
that the sensor, fl oat, logger or recorder
has malfunctioned or had undergone
some interference;
(iii) A data batch that appears higher or lower
than data on either (or both) sides may
have been wrongly processed with erroneous corrections against manual readings or sensor offset;
(iv) Straight lines in the record can indicate
gaps due to missing data that have been
wrongly interpolated, or instrument or
sensor problems, for example, sticking
fl oat cable or minimum or maximum
range reached;
(v) Increasing fl ows between freshets or
fl oods (uphill recessions) may indicate
wrong stage corrections made, or alternatively weed or sediment build-up on
the channel, the latter indicating work
required on the stage-discharge rating;
(vi) Regular diurnal (daily) fl uctuations may
indicate problems with the sensor (if it is a
pressure type, moisture may be present in
the system), icing on the control (will need
correction to convert to fl ow) or something real, such as channel evaporation or
daily freeze-thaw in the catchment.
Naturally, the best verification techniques are
limited in value, unless there is appropriate investigation of the queries they raise coupled with
appropriate corrective action, including fi ling of
the results. These can be either as comments fi led
with the data or through the assigning of informative quality codes.
Automated methods for carrying out many of these
techniques are available as part of hydrometric software packages or can be developed within them.
Some may be available to run automatically on
near-real-time telemetered data. An example screen
from such a package is shown in Figure I.9.3.
Note: An interesting plotting format is shown in
Figure I.9.3 which, although it depicts fl ow, is equally
valid for water-level data. The plot covers a 13-month
period and is designed to reveal any discontinuities that
may appear between successive annual updates of a
master database.

Figure I.9.3. Time-series plots for checking streamfl ow data
(Source: World Meteorological Organization/Food and Agriculture Organization of the United Nations, 1985:
Guidelines for Computerized Data Processing in Operational Hydrology and Land and Water
Management (WMO-No. 634, Geneva)

9.9.3 Rainfall
As rainfall is a very important and highly variable
hydrological phenomenon, there are many rainfall stations and hence large amounts of data. Most
countries now have well-established systems for
quality control and archiving of rainfall data.
A system used by the Meteorological Offi ce in the
United Kingdom for the processing of daily rainfall
is described in the Guide to Climatological Practices
(WMO-No. 100). The errors occurring in the collection and processing of rainfall data are almost
universal; therefore, this system should serve as a
model for many different environments.
The reliability of a system that uses inter-station
comparisons is related to the network density. In
areas having sparse coverage of raingauges, there is
an increasing tendency to install rainfall radars
(3.7). Areal values derived from such installations
provide excellent data for both validation and rainfall data for areas with no rainfall stations. Another
application of radar data for validation purposes is
encountered in areas subject to intense localized
thunderstorms, for example, most tropical
The event-based nature of rainfall means that there
are a number of ways of plotting and presenting the

data for verifi cation. These include accumulating
the readings over various time intervals and plotting them as separate events or cumulative totals.
The following techniques can be used:
(a) Plot the data as, perhaps, hourly totals and overplot them with stage or preferably fl ow from a
nearby fl ow station. The smaller the station’s
catchment, the more meaningful the comparison is likely to be;
(b) In addition to the above, overplot previous
(c) Plot cumulative daily totals (cusums) for a
period such as the year, and overplot this with
similar plots from adjacent stations, and with
the cumulative totals from the check gauge.
Figure I.9.4 shows a typical double-mass plot;
(d) Overplot cusums as above with cusums of
mean annual weekly or monthly data and thus
compare the current year or season with the
longer-term averages. Also plot the maxima
and minima for comparison;
(e) Plots may also be prepared to allow manual
checking of spatial variation. A simple means
is to plot station positions, together with their
identifi cation numbers and data values. Such
a technique is used widely for monthly and
annual checking of rainfall and groundwater
data on an areal basis. More complex software
can interpolate data in space and plot isolines.
Any apparent inconsistencies in the data should be
investigated to the extent possible:
(a) First, checks should be carried out to determine
whether there were any comments already
recorded in the database or in the logbook
for the station, or any evidence of processing.
The observer, fi eld technician or data
processor may have already checked apparent
problems and noted the actual or apparent
(b) Depending on the inconsistency noted, the
following items could be checked with the
fi eld observer or against the evidence of the
data processing. Some may require specifi c
investigation on site at the station, and with
previous and succeeding batches of data:
(i) If the data show lower precipitation
than expected, this may indicate that
the sensor, logger or connections may
have malfunctioned or undergone some interference; likewise if the sensor appears
to have failed to record some rainfall
(ii) If the rainfall events appear to be attenuated (spread over a longer period), this
may indicate blockages in the gauge due
to debris or interference, or it may indicate accumulations of snow that melt
(iii) A data batch that appears higher or lower
than data on either (or both) sides may
have been wrongly processed with erroneous corrections against manual readings or the wrong units or scaling.
9.9.4 Climatological data
The validation of climatological data by methods of
inter-station comparison can be questionable in
many cases because of the sparsity of the
climatological stations. Thus, the basic validation
techniques applied are range checks, rates of change
checks and, of particular importance, consistency
checks between related variables observed at the
same site.
For example, all reported psychrometric data should
be checked or recomputed to determine whether
the dry bulb temperature exceeds or equals a
reported wet-bulb or dewpoint temperature;
depending on which data are available, the
dewpoint temperature and/or relative humidity
should be computed and checked against the
reported value.
Similarly, empirical relationships between evaporation-pan or lysimeter data and other observed
variables could give broad indications of suspect
data at the validation stage. More sophisticated
adjustments for the evaluation of evaporation and
evapotranspiration are normally made in subsequent primary-processing stages.

Figure I.9.4. Double-mass plot. Double-mass curve showing the relationship of annual precipitation at station A to the mean of three nearby stations. Note the abrupt change that occurred in 1975.

For all climatological data, station and variable
codes should be tested for validity and, where relevant, sensor-calibration values and ranges should
be output with suspect values.
Comprehensive details of climatological quality control procedures are presented in the Guide to
Climatological Practices (WMO-No. 100).

9.9.5 Snow and ice data
Whereas the water equivalent of falling snow
caught in raingauges may be validated along with
rainfall data, other snow and ice variables are more
diffi cult to treat.
Data on the extent of snow cover may be validated
only by a time-consuming manual synthesis of fi eld
observations, aerial-survey data and satellite
imagery (3.7.4, 3.12 and 3.13). Techniques to
perform automated interpretation of satellite
imagery for snow extent (as well as depth and water
equivalent) are being developed. While these techniques show promise, there are still problems both
with differentiating between snow and cloud cover,
and of insuffi cient image resolution. Further, unless
a GIS is used, data on extent may be stored only as
manually-abstracted catchment-area totals.
Data on snow depth and water equivalent demand
much manual validation and verifi cation by integrating data from snow courses, snow gauges and
conventional precipitation gauges. The large spatial
variation in snow cover makes inter-station comparison diffi cult. However, there are techniques that
can be used to estimate the statistical reliability of
snow-course observations under conditions of melting snow. Degree-day factors are widely used for
correlation purposes and, where snow melt represents a significant proportion of river flow,
established relationships between runoff and snowwater equivalents may be used. Air (and water)
temperature relationships are valuable not only for
the computation of degree-day factors, but also for
the validation of ice-cover and thickness data and
in the forecasting of ice formation and break-up

Table I.9.1. Checking water quality data against physio-chemical laws

Snow and ice data, whether quantitative or qualitative, are important validation data for a wide range
of other hydrological variables. For example, anomalous river-stage data during the winter months
may be explained and possibly corrected if background data indicated the nature and extent of ice
9.9.6 River-gauging data
As each gauging is processed, there are a number of
items that need to be checked, including accuracy
of key entry and correctness of meter calibrations,
as mentioned previously. There are a number of
verifi cation techniques for individual gaugings that
can be applied:
(a) Some computation programs provide overplots of horizontal velocity and measured
depths. While these parameters are not fully
related, most channels show some relationship,
and the person carrying out the gauging should
be able to verify that the plots are sensible and
identify any outliers. This illustrates the desirability of the computation being done as soon
as possible after the gauging and preferably on
(b) Checking of the cross-section area and shape
together with the water level from previously
surveyed cross-sectional data may be feasible at
some stations;
(c) The theoretical uncertainty should be calculated in accordance with ISO 748 in order to
verify that the technique used is capable of
providing the required level of uncertainty.
This will normally be done by the computation
(d) Plotting the gauging on the rating curve may
provide some degree of verifi cation. If it plots
off by a signifi cant margin, then some other
evidence of a likely rating change should be
sought, such as a fl ood event high enough to
trigger a bed change, seasonal weed growth or
(e) With a stage-discharge relationship, the correctness of the stage used to plot the gauging is as
important as the correctness of the discharge
value. Therefore the stage should be verifi ed
against the value recorded by the water-level
recorder (if one exists);
(f) The location of the gauging cross-section
should be verifi ed as appropriate to the data
requirement, with regard to the possibility of
water abstractions, tributaries and artifi cial
discharges, under-bed fl ow, weir leakage, etc.;
(g) For ADCP gaugings, parameters that need to
be checked for correctness include whether
water salinity and density measurements have
been performed, instrument depth off-set has
been accounted for, depth range capability of
instrument is compatible with the depth of the
river, Doppler ambiguity settings are correct,
the presence of moving-bed has been checked,
extrapolation techniques are in accordance
with those required or recommended, ratios
of measured to unmeasured fl ow are suffi cient,
sufficient transects have been measured and
adequate coefficients of variation have been
adopted. It should also be verified that procedures are in accordance with those required or

Table I.9.1. Checking water quality data against physio-chemical laws

For all measurements, the validation program
should check for the use of valid station, instrument and method-of-analysis codes and, where possible, for valid combinations of these. It is also useful for any plots or printouts to contain this
information and any relevant calibration

Additional information on aspects of discharge
measurement is available in the Manual on Stream
Gauging (WMO-No. 519).
9.9.7 Water-quality data
The very wide range of water quality variables has
resulted in the use of relatively simple validation
procedures for water quality data. Such criteria are
normally absolute checks of analysis codes, relative
checks of expected ranges and physio-chemical
checks of determinant relationships. If range checks
are being devised in the absence of historical data,
it should be noted that the valid ranges of many
variables will be associated with the purpose for
which the sample was taken, and the location of
the sampling point. Thus, the levels of dissolved
salts found in water samples taken from drinking
water sources will be less than those found in effluents or in brackish or marine water bodies.
Physio-chemical tests are very effective and, hence,
widely used for water quality data.
Examples of typical physio-chemical tests performed
for normal and specifi c (effl uent) samples are shown
in Table I.9.1.
If some variable values have been determined in
the laboratory and all of the relevant associated
data are available to the computer, they may be
recomputed for verifi cation purposes. All water
quality data and the station, variable and analysis
codes may be checked for validity and, where possible, for validity of their combination.
9.9.8 Sediment data
As with water quality data, mass-balance calculations may be performed if suffi cient data exist. If a
sediment rating curve exists for the section sampled,
the departure of the sampled value from the curve
may be estimated for its statistical significance
and/or plotted for manual scrutiny.
The sediment gaugings and the rating curve should
be examined to determine if there are any changes
in the rating according to the seasons; and if so, the
sampling programme should be reviewed to aim for
at least approximately equal amounts of data from
each season. Similarly, the proportion of gaugings
on the rising and falling stages should be examined
and attempts made to sample in both conditions.
The informed data user will always be concerned
with understanding the accuracy of the data in
question, as this will govern the confi dence that
people can have in the data and the derived information. There are many ways of expressing accuracy,
many of them imprecise and sometimes ambiguous. Statistical uncertainty provides a means of
objectively expressing “accuracy” as a stated range
or percentage range with a given probability of
Several of the ISO standards concerned with
hydrometric techniques cover uncertainty in some
detail as it applies to each topic. The ISO publication
Guide to the Expression of Uncertainty in Measurement
(ISO, 1995) is recommended as a general guide to
the topic. Guidance on the estimation of uncertainty
of discharge measurement is provided in the
Technical Regulations (WMO-No. 49), Volume III,
Annex, Part VIII.
References and further reading
Environment Canada, 1973: NAQUADAT Dictionary
of Parameter Codes. Inland Waters Directorate,
Environment Canada, Ottawa.
Environment Canada, 1985: NAQUADAT Dictionary of
Parameter Codes. Data Systems Section,
Water Quality Branch, Environment Canada,
Hudson, H.R., D.A. McMillan and C.P. Pearson, 1999:
Quality assurance in hydrological measurement. Hydrological Sciences—Journal des Sciences
Hydrologiques, Volume 44, No. 5 (http://www.cig.
Hutchinson, N.E., 1975: WATSTORE User’s Guide. Volume
1, United States Geological Survey Open-File
Report 75-426 (http://www-eosdis.ornl.gov/source_
International Organization for Standardization, 2000:
Quality Management Systems: Requirements. ISO 9001,
International Organization for Standardization, 2005:
Quality Management Systems: Fundamentals and
Vocabulary. ISO 9000, Geneva.
International Organization for Standardization and
International Electrotechnical Commission, 1995:
Guide to the Expression of Uncertainty in Measurement.
ISO/IEC Guide 98, Geneva.
Kilpatrick, Mary C., 1981: WATSTORE: A WATer
Data STOrage and REtrieval System. United States
Government Printing Offi ce publication, 52, United
States Department of the Interior, United States
Geological Survey, Reston, Virginia, pp. 341–618.

National Institute of Water and Atmospheric Research,
1999: TIDEDA—Software for archiving and
retrieving time-dependent data.
Wellington (http://www.niwascience.
Thompson, S.M. and G.R. Wrigley, 1976: TIDEDA.
In: SEARCC 76, M. Joseph and F.C. Kohli (eds.),
Amsterdam, pp. 275–285 (http://www.niwascience.
United Kingdom Department of Environment, 1981:
Hydrological Determinand Dictionary. Water Archive
Manual No. 5, Water Data Unit (http://www.defra.
United Nations Environment Programme, World
Health Organization, United Nations Educational,
Scientifi c and Cultural Organization and World
Meteorological Organization, 1992: Global
Environment Monitoring System (GEMS)/Water
Operational Guide. Inland Waters Directorate,
Burlington, Ontario.
WATSTORE: http://www-eosdis.ornl.gov/source_
documents/watstore.html; http://www.osmre.
gov/h20dbs.htm; http://ak.water.usgs.gov/
World Meteorological Organization, 1980: Manual
on Stream Gauging. Volumes I and II, Operational
Hydrology Report No. 13, WMO-No. 519, Geneva.
World Meteorological Organization, 1983: Guide to
Climatological Practices. Second edition,
WMO-No. 100, Geneva (http://www.wmo.int/pages/
World Meteorological Organization, 1988: Manual
on Water Quality Monitoring: Planning and
Implementation of Sampling and Field Testing.
Operational Hydrology Report No. 27,
WMO-No. 680, Geneva.
World Meteorological Organization, 2006: Technical
Regulations. Volume III – Hydrology, WMO-No. 49,
World Meteorological Organization and Food and
Agriculture Organization of the United Nations,
1985: Guidelines for Computerized Data Processing
in Operational Hydrology and Land and Water
Management. WMO-No. 634, Geneva.