Potter T.D., Colman B.R. (co-chief editors). The handbook of weather, climate, and water: dynamics, climate physical meteorology, weather systems, and measurements
Подождите немного. Документ загружается.
number of remote stations are involved. In addition, stations may be powered by
batteries and solar panels, thereby requiring low-power radio transmitters. As
satellite communi cations technology evolves, communication restraints will be
eased, which will lead to vastly improved meteorological networks, especially for
the mesoscale.
Power Source Electrical power consumption of a measurement system is often a
vital consideration; the primary concern is cost. Where commercial power is
available, cost is not usually a problem. However, many systems are required to
be portable or to operate in locations where commercial power is not available. In
these cases, the power source is usually batteries, perhaps supplemented with solar
panels. Such systems must operate on a severely limited power budget and that
constraint affects the selection of components and the overall system design.
Battery-powered systems are constrained to select sensors with low power consump-
tion and=or to switch the sensors on only as needed to conserve power. Heaters
generally cannot be used and local computational capability may be severely limited.
Therefore, all components must be rated for operation over the expected temperature
range.
Unique methods are sometimes used to power remote meteorological stations and
are often used out of necessity. The following is an example of what can be done
when no traditional power source is available.
The year-long SHEBA (Surface Heat Budget of the Arctic Ocean) field experi-
ment used PAMII 1 (the third-generation Portable Automated Mesonet) remote
meteorological stations and was located well above the Arctic Circle. PAMIII
stations require 1.5 to 2.5 A at 12 V dc for continuous operation and, for most
deployments, the power system consists of photovoltaically charged deep-cycle
storage batteries with nominal capacity of 3 to 6 days of operation in the absence
of sunshine. Obviously, modifications were required for SHEBA due to the lack of
sunshine during the winter months and the extreme cold. Three separate power
sources were eventually used for differing reasons and at different times of the year.
The primary source of power for each station was a small propane-fueled 12 V dc
electro-thermal generator (TEG) and a single battery. To mitigate the effects of
extreme cold, the generator, 2 propane bottles, the battery, and PAIII electronics
were all placed inside an insulated container mounted on a sled for transportation
between sites. Two 33-lb propane bottles allowed for 15 to 20 days of unattended
operation. Heat from the TEG kept the electronics and more importantly the propane
and battery warm when outside temperatures dropped below 15
C. In the spring
two photovoltaic panels were installed at each station and paralleled with the TEG to
augment system power. In addition, two small wind generators were added in spring
to the least accessible remote stations. The wind generators operated at 12 V dc that
allowed parallel operation with both the TEG and photovoltaics through the
PAIvIIIiI battery charging and monitoring system.
Some problems were encountered with the TEG, caused by poor fuel quality.
However, the system described above was in general successful and was a unique
solution to a difficult situation.
806 DESIGN, CALIBRATION, AND QUALITY ASSURANCE NEEDS OF NETWORKS
3 CALIBRATION
The calibration needs of a network depend on, among other factors, the desired
accuracy of the field measurements. Sufficiently accurate laboratory standards are
readily available for most meteorological sensors, but acquiring this same level of
accuracy in the field is often very difficu lt if not impossible. For example, it is
relatively easy to calibrate temperature sensors in the lab to an accuracy of
0.1
C or even 0.01
C. However, measuring atmospheric air temperature with
this same accuracy requires great care and knowledge of the measurement system.
Therefore, extremely accurate laboratory calibrati ons may not be required in some
situations. Nevertheless, the total error associated with an instrument in the field is
the square root of the sum squared errors (assuming the errors are independent) and,
therefore, the total measurement error can be reduced if a careful laboratory calibra-
tion is performed.
Calibration standards are maintained by standards laboratories in each country
such as the National Institute of Science and Technology (NIST) in the United States.
Standards for temperature, humidity, pressure, wind speed, and for many other
variables are maintained. The accuracy of these standards is more than sufficient
for meteorological purposes. Every organization attempting to maintain one or more
measurement stations should have some facilities for laboratory calibration including
transfer standards. These are standards used for local calibrations that can be sent to
a national laboratory for comparison with the primary standards. This is what is
meant by traceability of sensor calibration to NIST standards. Ideally the calibration
of all sensors can be traced back to such a standards laboratory.
Initial calibration of all instruments should include a full range test if at all
possible. Even when instrumentation is new from the manufacturer, all instruments
should be tested prior to use in the field. Testing each instrument can be time
consuming and expensive but to not do so is a mistake that will inevitably cost
many times more. Even seemingly simple instruments require a laboratory check
before field use. For example, a tipping bucket rain guage has an output that is a
function of rain rate and not the same for all instruments; therefore, a different
calibration curve may have to be developed for each sensor. In addition, slight
variations in bucket volumes can result in differences in measured rainfall and can
be accounted for. As each rain guage is ‘‘calibrated,’’ sensor deficiencies (e.g., faulty
bearings or closure switches) may be detected prior to field deployment where such
problems could go undetected.
The importance of checking, if not calibrating, each sensor prior to field use
cannot be overemphasized. Instrument manufacturers, no matter how reputable,
may not be able to maintain the desired stand ards. Even if sensor specifications
meet or exceed the required accuracy, instruments can drift or be damaged during
shipment and storage or electronic components can fail. Therefore, to maintain the
highest level of confidence in a network’s data, sensor performance should be
verified prior to use in the field.
Fortunately, in many ca ses it is not necessary to purchase expensive calibration
chambers to achieve the level of accuracy required for meteorological measure-
3 CALIBRATION 807
ments. Simple yet effective ‘‘calibration’’ met hods that require a minimal investment
can be developed for many meteorological instruments (calibration is p ut in quotes
to indicate that true calibrations, which require a sensor with an order of magnitude
more accuracy, may not be occurring). For example, rain guage calibration can be
performed with a reasonably accurate electronic scale with RS-232 or analog output,
a simple water reservoir, a simple flow valve, and a data collection system like that
used in the network. The scale output is connected to a computer (or datalogger) and
the rain guage is connected to the data collection system. The water reservoir is
placed on the scale and water is siphoned off into the rain guage. The scale output
(weight of water remaining in the reservoir) is easily related to the water input to the
rain guage and the siphon rate can be adjusted to simulate different rain rates . This
calibration does not account for many factors that can affect the rain guage accuracy
but does provide a simple yet effective check of the instrument performance.
An equally simple setup can be developed to test temperature sensor s
(Richardson, 1995). Require d equipment includes a temperature transfer standard
with errors less than about 0.030
C, a temperature chamber (e.g., a small freezer), a
bath, and a stirrer. The use of a temperature standard means it is not necessary to
maintain precise chamber temperatures nor is it necessary to hold the chamber
temperature constant; it is only necessary to control the rate of change of tempera-
ture. The rate must be low enough that errors induced by spatial gradients between
the temperature standard and the test sensors and the errors induced by the time
response of the sensors are small compared to the acceptable calibration error.
Dynamic errors are fairly easy to detect if the test sensors lag the reference sensor
during increasing temperature, and lead it when the temperature is falling, the rate of
change is too high. The response of the sensors will be a function of the kind of bath
(air or liquid) and the amount of stirring. The bath also affects spatial gradients.
These can be detected by correlating errors with sensor position.
4 QUALITY ASSURANCE
Quality assurance methods and=or final data format may depend on the primary end
user of the data, e.g., will they be used for research or by the general public? For
example, relative humidity (RH) sensor inaccuracy specifications allow for the RH
reported by the sensor to be in excess of 100% (as high as 103%). A sensor reporting
an RH of 102% may be operating correctly but could cause problems for under-
informed users. For example, modelers ingesting RH may need to be aware that
values greater than 100% are possible and do not necessarily indicate supersaturation
conditions. In addition, those unaware of instrument inaccuracy specifications may
think the data is incorrect if RH is above 100%.
A more complete data quality assurance program can be developed for a network
of stations than for a single measurement station, so this discussion will address the
issue of quality assurance for a network. A single-station assurance program would
be a subset of this. In designing a measurement system, it is useful to consider the
impact of automation on data quality.
808 DESIGN, CALIBRATION, AND QUALITY ASSURANCE NEEDS OF NETWORKS
Schwartz and Doswell (1991) claim that automation sometimes has been accom-
panied by a decreas e in data quality but that this decrease is not inevitable. When a
measurement system is automated, some source s of human error are eliminated such
as personal bias and transcription mistakes incur red during reading an instrument, to
name only two. However, another, more serious form of error is introduced: the
isolation of the observer from the measurement process.
One tends to let computers handle the data under the mistaken assumption that
errors are not being made or that they are being controlled. Unless programs are
specifically designed to test the data, the computer will process, store, transmit, and
display erroneous data just as efficiently as valid data. Automatic transmission of
data tends to isolate the end user from the people who understand the instrumenta-
tion system. Utilization of computers in a measurement system allows data to be
collected with finer time and space resolution. Even if the system is designed to let
observers monitor the data, they can be overwhelmed by the sheer volume and
unable to effectively d etermine data quality.
Automation, or the use of computers in a measurement system, can have a
beneficial impact on data quality if the system is properly designed. Inclusion of
data monitoring programs that run in real time with effective graphical displays
allow the observer to focus on suspect data and to direct the attention of technicians.
The objective of the data quality assurance (QA) system is to maintain the highest
possible data quality in the network. To achieve this goal, data faults must be
detected rapidly, corrective action must be initiated in a timely manner, and ques-
tionable data must be flagged. The data archive must be designed to include provi-
sion for status bits associated with each datum. The QA system should never alter
the data but only set status bits to indicate the probable data quality. It is probably
inevitable that a QA system will flag data that are actually valid but represent
unusual or unexpected meteorological conditions. Flagged data are available to
qualified users but may not be available for routine operational use.
The major components of a QA program are the measurement system design,
laboratory calibrations, field intercomparisons, real-time monitoring of the data,
documentation, independent reviews, and publication of data quality assessment.
Laboratory calibrations are required to screen sensors as they are purchased and
to evaluate and recalibrate sensors returned from the field at routine intervals or
when a problem has been detected. Field intercomparisons are used to verify perfor-
mance of sensors in the field since laboratory calibrations, while essential, are not
always reliable predictors of field performance. QA software is used to monitor data
in real time to detect data faults and unusual events. A variety of tests can be used
from simple range and rate tests to long-term comparisons of data from adjacent
stations. Documentation of site characteristics and sensor calibration coefficients and
repair history is needed to answer questions about possible data problems.
Independent reviews and periodic publication of data quality are needed since
people close to the project tend to develop selective awareness. These aspects of the
overall QA program must be established early and enforced by project leader s.
The whole QA program should, ideally, be designed before the project starts to
collect data.
4 QUALITY ASSURANCE 809
Laboratory Calibrations
As discussed previously, laboratory calibration facilities are required to verify instru-
ment calibration and to obtain a new calibration for instruments that have drifted out
of calibration or have been repaired. However, a laboratory calibration is not neces-
sarily a good predictor of an instrument’s performance in the field. This is because
laboratory calibrations never replicate all field conditions. For example, a laboratory
calibration of a temperature sensor does not include all the effects of solar and Earth
radiation nor is it likely to simulate poor coupling with the atmosphere due to low
wind speeds. This type of error is called ‘‘exposure error’’ and is a result of inade-
quate coupling between the sensor and the environment. Exposure errors can be very
large (e.g., radiative heating of temperature sensors can result in air temperature
errors in excess of 5
C; Gill, 1983) and are not accounted for in lab calibrations. In
addition, the manufacturer does not specify this type of error because it is not under
its control. Therefore, knowledge of the measurement system and instrument design
is crucial to minimize measurement error.
Field Intercomparisons
Two types of field intercomparisons should be performed to help maintain data
quality. First, a field intercomparison station should be established and, second,
when technicians visit a station, they should carry portable transfer standards and
make routine compariso n checks.
The field intercomparison station should be comprised of operational sensors and
a set of reference sensors (higher quality sensors). Both should report data to the
base station but the reference station data should be permanently flagged and not
made available for operational use.
Portable transfer sensors can be used to make reference measurements each time a
technician visits a station. This method can detect drift or other sensor failures that
could otherwise go undetected. These sensors can include a barometer and an
Assmann psychrometer. In addition, technicians can carry a lap-top computer to
read current data, make adjustments to calibration coefficients when sensors are
changed, set the data logger clock ( if it cannot be set remotely), and reload the
data logger program after a power interruption.
Data Monitoring
Neither laboratory calibration nor routine field intercomparisons will provide a real-
time indicator of problems in the field. In a system that collects and reports data in
real time, bad data will be transmitted to users until detected and flagged. The
volume of data flow will likely be far too great to allow human observers to effec-
tively monitor the data quality. Therefore, a real-time, automatic monitor ing syst em
is required.
The monitor program should have two major compo nents: scanning algorithms
and diagnostic algorithms. The function of the scanning algorithms is to detect
810 DESIGN, CALIBRATION, AND QUALITY ASSURANCE NEEDS OF NETWORKS
outliers while the diagnostic algorithms are used to infer probable cause. The moni-
tor program can analyze the incoming data stream using statistical techniques
adapted from exploratory data analysis, knowledge of the atmosphere, knowledge
of the measurement syst em, and using objective analysis of groups of stations during
suitable atmospheric conditions.
Exploratory data analysis techniques are resistant to outliers and are robust, i.e.,
insensitive to the shape of the data’s probability density function. Knowledge of the
atmosphere allows us to place constraints on the range of some variables such as
relative humidity that would be flagged if it were reported greater than approxi-
mately 103% (due to sensor inaccuracy specifications). Knowledge of the measure-
ment system places absolute bounds on the range of each variable. If a variable
exceeds these limits, a sensor hardware failure may have occurred.
The monitor program must be tailored for the system and should be developed
incrementally. Initially, it could employ simple range tests while more sophisticated
tests can be added as they are developed. The QA monitoring program will never be
perfect; it will fail to detect some faults and it will label some valid data as poten-
tially faulty.
Therefore, the monitor program must not delete or in any way change data but set
a flag associated with each datum to indicate probable quality. The moni tor program
should have a mechanism to alert an operator whenever it detects a probable failure.
Some of these alerts will be false alarms, i.e., not resulting from hardware failure,
but may in fact indicate interesting meteorological events.
Documentation
There are several kinds of documentation needed: Of individual station character-
istics, a station descriptor file, and a sensor database.
Station characteristics can be documented by providing an a rticle describing the
station and its instrumentation. In addition, there should be a file of panoramic
photographs showing the fetch in all directions and the nature of the land. Aerial
photographs can also provide valuable information about a site, as can high-resolu-
tion topographical maps.
As part of the system database, a system descriptor file must include the location
and elevation of each station and the station type (e.g., standard meteorological
station, special-purpose agricultural station, or sensor research station).
It is necessary to maintain a central database of sensors and other major compo-
nents of the system including component serial number, current location, and status.
Some sensors have individual calibration coefficients so there must be a method of
accounting for sensors to ensure that the correct calibration coefficients have been
entered into the appropriate data logger. It is also necessar y to keep records of how
long a component has been in service and where it was used so that components that
suffer frequent failures can be identified. This would help to determine if the compo-
nent was seri ously flawed or if the defect was characteristic of the component design.
This kind of accounting cannot be left to chance; if it is, sensors will inevitably be
matched with the wrong calibration coefficients or put back into service without
4 QUALITY ASSURANCE 811
having been repaired or recalibrated. Some sensors require periodic recalibration,
but it is not feasible to recalibrate them all at once. Therefore a formal database
system should be set up. All technicians should be required to report maintenance
activity including swapping components, and this information shoul d be entered into
the database. The database system should be able to generate reports indicating the
serial number of every component at a station, the number of components awaiting
repair at any given time, the number of spares available, the history of any given
sensor or sensor type, etc.
Independent Review
For much the same reason that scientific proposals and papers are reviewed, periodic,
independent reviews of a network’s performance should be invited. It is always
possible for people in constant close proximity to a project to become blind to
problems; an independent review would help alleviate this at the root cause.
Publication of Data Qual ity Assessment
There will be frequent data faults in any network even with the data quality assur-
ance program outlined above. To assist critics in making a realistic assessment, it is
desirable to publish, periodically, an honest appraisal of the network performance
including all data faults, causes when known, and action taken.
REFERENCES
Brock, F. V., K. C. Crawford, R. L. Elliott, G. W. Cuperus, S. J. Stdaler, H. L. Johnson, and
M. D. Eihs (1995). The Oklahoma Mesonet: A technical overview. J. Atmos. Oceanic
Technol. 12, 5–19.
Brock, F. V., and S. J. Richardson (2001). Meteorological Measurement Systems, Oxford
University Press, Inc. New York.
DeFelice, T. P. (1998). An Introduction to Meteorological Instrumentation and Measurement,
Prentice Hall, New Jersey.
DOE (1990). U.S. Department of Energy, Atmospheric Radiation Measurement Program Plan,
DOE=ER 0441, National Technical Information Service, Springfield VA.
Gill, G. C. (1983). Comparison Testing of Selected Naturally Ventilated Solar Radiation
Shields, Report submitted to the NOAA Data Buoy Office, Louis, MS.
Richardson, S. J. (1995). Automated temperature and relative humidity calibrations for the
Oklahoma Mesonetwork, J. Atmos. Oceanic Tech. 12, 951–959.
Schwartz, B. E., and C. A. Doswell III (1991). North American rawinsonde observations:
Problems, concerns and a call to action, Bull. AMS 72, 1885–1896.
Stokes, G. M., and S. E. Schwartz (1994). The Atmospheric Radiation Measurement (ARM)
Program: Programmatic background and design of the cloud and radiation test bed, Bull.
Am. Meteor. Soc. 75, 1201–1221.
812
DESIGN, CALIBRATION, AND QUALITY ASSURANCE NEEDS OF NETWORKS
CHAPTER 41
DATA VALIDITY IN THE NATIONAL
ARCHIVE
GRANT W. GOODGE
1 INTRODUCTION
As we enter a new millennium the United States and indeed the whole world is
becoming increasingly sensitive to various meteorological and climatological
extremes. Given the potential for economic and political disruption as a result of
these anomalies, there has been an increased interest in their prediction. Such
predictions include frequency, magnitude, areal extent, and location. As of this
time one of our best methods of understanding the potential for future climatic
anomalies is to look at the past climatic record.
The integrity of any scientific research depends a g reat deal on the quality of the
data used as well as the user’s understanding of the data and all relevant information
associated with its collection and processing (metadata).
These metadata include, but are not restricted to, instrument type, exposure,
automatic, manual, maintenance=calibration, and the nature of data transfer from
the point of measurement to archive. The final archive for much of the meteorolo-
gical data for the United States and its territories resides at the National Climatic
Data Center (NCDC) in Asheville, North Carolina. A considerable volume of similar
data from other countries and the world’s oceans are also archived at NCDC. The
NCDC is one of several environmental and geophysical data centers in the United
States and is administered by the National Environmental Satellite Data and Infor-
mation Service (NESDIS), which is one of the major components of the National
Oceanic and Atmospheric Administration (NOAA).
Handbook of Weather, Climate, and Water: Dynamics, Climate, Physical Meteorology, Weather Systems,
and Measurements, Edited by Thomas D. Potter and Bradley R. Colman.
ISBN 0-471-21490-6 # 2003 John Wiley & Sons, Inc.
813
The NCDC is not directly involved with the observation of meteorological vari-
ables, sensor maintenance, or observer trai ning. That portion of data collection is
managed by the National Weather Service (NWS), which is also a major component
of NOAA. The NCDC does, however, actively engage in providing relevant feed-
back to NWS managers concerning sensor and=or observer problems that are
discovered durin g the quality assurance processes at NCDC.
Unfortunately, errors in the meteorological databases may be introduced
anywhere between the point of observation and its final storage in the digital archive
or websites. Several examples are observer transcription er rors, radio frequency (RF)
interference or lightning damage to electronically driven sensors, transmission signal
interruption, keying errors of manually observed data during data entry and=or
update during quality assurance, and compu ter program transfer of those files to
the digital archives or websites. Also an increasingly problematic issue in today’s
electronic world is that of missing data. This can be of particular significance with
the transmission of data from the Automated Surface Observing System (ASOS)
where data are stored on site for only 12 h. Obviously, any interruption of the link
between the sites and NCDC that exceeds this perio d will result in the permanent
loss of data from affected sites until the link is restored. We will return to a more
detailed discussion of some of these issues later in this chapter.
Before proceeding further, this writer would like to note that within the
constraints of limited budgets and personnel, NCDC has always endeavored to
produce high-quality databa ses and publications. Any discussion of the weaknesses
in the observational systems and the quality assurance of the resulting observations
are mentioned only to alert the research community to known data problems and
encourage a better understanding of the data prior to use in any study. It is not
intended as an indictment or criticism of those responsible for the collection and
archive of these environmental observations.
2 TEMPERATURE
In the recent era there has been much public and scientific debate of the issue of
‘‘global climate change,’’ and many of those debates center on the parameter of
temperature. Certainly the fact that the magnitude of atmospheric model predicted
centennial mean global temperature rises are on the same order (1
F) as the required
accuracy (1
F) of NOAA-sponsored instruments indicate the necessity of quality
data. Prior to the 1960s official Weather Bureau thermometers were all compared
against a National Bureau of Standard’s ‘‘certified’’ thermometer, and any mercurial
thermometer having a greater error than 0.3
F (above 0
F) from the certified stan-
dard was not accepted for observational use. During the 1960s most primary
Weather Bureau Stations transitioned from the use of mercurial thermometers to
an HO-60 series temperature sensor that employed the use of a fan-aspirated ther-
mistor exposed in a metal shield with the sensed temperature electrically transmitted
to an electromechanically driven dial in the weather office. Unfortunately, some of
these instruments did not meet the more relaxed standard of 1
F as compared to
the 0.3
F standard of the mercurial thermometers they replaced.
814 DATA VALIDITY IN THE NATIONAL ARCHIVE
In one documented case the mean temperatures, sensed by the HO-62 hygro-
thermometer, at the Asheville, North Carolina, NWS Airport station, migrated
upward more than 4
F (as compared to three surrounding cooperative stations) in
less than 4 years (1965–1969). Since that time there have been other migrations
and=or shifts of 2
F. The most recent of these have been as a result of sensor
changes and=or relocations at the airport. Certainly this station’s record would not be
a good candidate for climate trend determinations. Unfortunately, sensor-driven data
errors and discontinuities for this station and others like it remain unadjusted in the
official databases and archives and in some cases undocumented.
There are several reasons why NCDC does not adjust or correct such errors during
the quality assurance (QA) process. The primary reason is that NWS stations (i.e.,
Chicago, Charlotte, Chattanooga, etc.) are quality assured on a single station basis and
as a result are unable to identify instrument ‘‘drift’’ unless that drift were to exceed
station monthly climatic extremes or cause conflict with another reported value such
as ambient temperature versus dew-point temperature. Second, there are seldom any
coincident data that would allow comparison or correction of the station in question.
Lastly, the historic digital data storage at NCDC has been sequential in nature, which
makes the correction of identified, postprocessing errors very expensive.
It should be noted that a change in a temperature sensor’s exposure can make as
much difference in the observed readings as a change in the type of sensor. Reloca-
tion of meteorological instruments at primary weather stations has been a problem
for climatologists for many years. Some have been moved from rooftops of post
offices in a downtown (urban) environment to rooftops of terminal buildings at
airports and finally from the terminal buildings to airport field level exposures.
After moving to the field level, sensors have continued to be moved due to airport
expansions or alterations. Even if the sensor has not moved, increased extent of
paved surfaces may alter the radiative flux of the area surrounding the sensor.
These relocations of sensors highlight the importance and need for timely and
accurate station history information called metadata. Without it, a user or researcher
is left to guess as to what may have caused the sudden change in the data. Accurate
metadata is just as important in the documentation of observations from the coop-
erative observer network. Even though these stations only record data on a daily
basis, it is still essential that any retrospective user know the type of instruments
used, the exposure, and the time of day when the readings are made. Unf ortunately,
NCDC’s receipt of the B-44 (metadata) forms lagged several months behind the date
of actual station move or instrument change, and thus the publication and databases
improperly indicated the location, instrument type, time of observation, or other
critical station information.
In about 1984 the NWS began replacing the liquid-in-glass thermometers at
cooperative observing sites with an electronic resistance thermometer (hereafter
referred to as the Maxi mum Minimum Temperature System (MMTS)). The sensing
accuracy of these MMTS units are generally compatible with the liquid-in-glass
thermometer they replaced, but the introduction of the MMTS brought several
new undesirable results. First is its susceptibility to electrical interference whether
that be generated by radio frequency (RF) from nearby radio transmitters or induc-
tion surges from nearby lightning strikes. Usually, the display inside the observer’s
2 TEMPERATURE 815