The United States government has operated Dobson ozone spectrophotometers at various sites, starting during the International Geophysical Year (1 July 1957 to 31 December 1958). A network of stations for long-term monitoring of the total column content (thickness of the ozone layer) of the atmosphere was established in the early 1960s and eventually grew to 16 stations, 14 of which are still operational and submit data to the United States of America's National Oceanic and Atmospheric Administration (NOAA). Seven of these sites are also part of the Network for the Detection of Atmospheric Composition Change (NDACC), an organization that maintains its own data archive. Due to recent changes in data processing software the entire dataset was re-evaluated for possible changes. To evaluate and minimize potential changes caused by the new processing software, the reprocessed data record was compared to the original data record archived in the World Ozone and UV Data Center (WOUDC) in Toronto, Canada. The history of the observations at the individual stations, the instruments used for the NOAA network monitoring at the station, the method for reducing zenith-sky observations to total ozone, and calibration procedures were re-evaluated using data quality control tools built into the new software. At the completion of the evaluation, the new datasets are to be published as an update to the WOUDC and NDACC archives, and the entire dataset is to be made available to the scientific community. The procedure for reprocessing Dobson data and the results of the reanalysis on the archived record are presented in this paper. A summary of historical changes to 14 station records is also provided.
Diagram of the Dobson instrument, with cover omitted from view (some components shown are actually mounted in the cover).
The Dobson ozone spectrophotometer was designed in the 1920s and is still in
use today. The instrument is fully described elsewhere (Dobson, 1931, 1968)
but, briefly, it measures the relative intensity of solar radiation between
selected wavelength pairs in the range of 300–350 nm. These pairs are named
A (305.5 and 325.4 nm), C (311.5 and 334.4
Distribution of cumulative differences between results from direct
sun (ADDS) compared to zenith measurements on the same day. The frequency of
compared zenith and ADDS total ozone (
The relative intensity of wavelength pairs measured if instrument was operated outside the Earth's atmosphere is referred to as the
extraterrestrial constant (ETC). The instrument's ETC is determined either through a Langley plot method (Langley, 1884) or by direct
comparison with a standard Dobson instrument (Komhyr and Evans, 2008). The concept of measurement of the Dobson spectrophotometer
exploits the change in the ratio of solar light intensities measured at the respective wavelength pairs as caused by the passage of UV
light through the ozone layer. The light enters the instrument (Fig. 1), and the right side of the optics produces a spectrum projected
on a slit arrangement containing slits S
The
The Dobson instrument has limitations in the accuracy of measurements at certain observing conditions (Basher, 1982). Internal stray light is one such limitation. Moreover, each Dobson instrument has unique optical components that result in an instrument-specific level of the stray light. The quality and aging stability of the individual wedge construction has improved over time, especially for instruments within the NOAA network, which had optical components replaced with those of a more robust design during instrument rebuilding in the 1980s.
Data reduction algorithms are fully discussed in Sect. 7 of the Operations Handbook (Komhyr and Evans, 2008;
Current stations in the NOAA network using Dobson ozone spectrophotometers.
There are measurements of TOC in the USA prior to 1960 made by university and federal organizations (Brönnimann, 2003), but the
development of a coherent network of observing sites within the US Weather Service started in the 1960s under the guidance of
Walter Komhyr. The network was transferred to NOAA's Global Monitoring for Climate Change (GMCC) in the early 1970s and is currently
operated by NOAA's Earth System Research Laboratory's Global Monitoring Division (ESRL/GMD). As many as 16 stations comprised the
network since its establishment. One station was closed; another was transferred to another parent authority. Table 1 displays the
stations reporting at end of 2015. Originally, observations using the Dobson instruments were recorded with pen or pencil on forms
designed to assist manual calculations (
TOC is normally archived as a single representative value of TOC selected for
each day. This not an average value, but the result of the “best”
observation during the day. As the exact instrumentation and observational
scheduling vary from station to station, the number of observations made
daily also vary. The full record of
observations is available per request from NOAA Dobson network personnel
listed at
To convert measurements to TOC values, calibration
The set of computer programs used for the NOAA processing were written in the FORTRAN language and, by the 2010s, were difficult to use and maintain due to changes in computer hardware and personnel. The decision was made to convert the NOAA processing to processing using the WinDobson software package, as the fully automated instruments were updated to a modern system based on this software. Developed by personnel of the Japan Meteorological Agency (Miyagawa, 1996), WinDobson is a software package for operations, data analysis and quality assurance of Dobson spectrophotometer observations. The algorithm for the reduction of ozone from DS observations with the Dobson is the standard method used by the NOAA software, but the ZS observations are reduced with a method described later in this paper. For the NOAA application, new components were developed. These new components are available from NOAA to other users of WinDobson. It is applicable for both TOC and Umkehr (ozone vertical profile) measurements. As this software has a different statistical method for the reduction of the zenith measurements, and set of rules (see Sect. 2.4) for determining the representative value of total ozone for each day with observations, the entire data record of each operational station was reprocessed in the WinDobson system to minimize the effect of the change when future data are placed in the archive. In the development of the data files and calibration information for WinDobson processing, the entire record of observations, repair and calibration checks of each station was investigated and re-evaluated. This investigation allows for correction of past errors.
Statistics of the overall differences between WOUDC and NDACC records and WinDobson record (WinDobson–WOUDC, NDACC).
The NOAA processed data were converted to “long line format” (LLF) files. These files are actually the image of the information sent to printers in the 1990s version of the data stream. The select values for the WOUDC and NDACC archives were originally produced from these files, using a process of both machine and inspection by personnel. Programs were developed to convert the LLF and dayfiles into formats compatible with the WinDobson data stream. Files with instrument, station and calibration information (parafiles) were also developed to complete the structure of the WinDobson system. Connections to other sources of TOC information (satellite data records, for example) were developed so that comparisons with these values could be performed using tools internal to WinDobson. Reference lamp values were extracted from the LLF records for time periods prior to 1995 and from the dayfiles afterwards. By the end of 2015, all operational stations' data were being processed in WinDobson.
Initially, the datasets of only ADDS
(fundamental wavelength pairs) observations from the two processing streams
were compared with the expectation that the results should agree within
Displayed is the cumulative agreement in percent for specific ZS and CDDS results compared to ADDS results on the same day. For example, an agreement of 2 % occurs in 91 % of the cases for ADZB observations. Displayed are the average of 12 stations in the NOAA network (Barrow, Fairbanks, Caribou, Bismarck, Haute Provence, Boulder, Wallops Island, Mauna Loa, Tutuila, Perth, Lauder and South Pole). Definitions of all abbreviations are in Sect. 1.
The probability of a daily value changing by a particular percentage
for each station. The red line is for ADDS-type observations; the blue for
all other types; the horizontal line is the percent of ADDS in the range
Graphic representation of the changes in the Mauna Loa Observatory, Hawai'i, USA (19
Graphic representation of the changes in the South Pole, Antarctica (90
Graphic representation of the changes in the Bismarck, North Dakota, USA (47
Graphic representation of the changes in the Caribou, Maine, USA (47
Graphic representation of the changes in the Nashville, Tennessee, USA (36
Graphic representation of the changes in the Fairbanks, Alaska, USA (65
Graphic representation of the changes in the Boulder, Colorado, USA (40
Graphic representation of the changes in the Wallops Island Flight Center, Virginia, USA (38
Graphic representation of the changes in the NOAA/ESRL/GMD Observatory, Barrow, Alaska, USA (71
Graphic representation of the changes in the NOAA/ESRL/GMD Observatory, American Samoa (14
Graphic representation of the changes in the Fresno and Hanford, California, USA (36
Graphic representation of the changes in the Observatoire de Haute-Provence, France (44
Graphic representation of the changes in the Perth Airport, Western Australia, Australia (32
The individual station records are archived as daily values in the World Ozone and Ultraviolet Radiation Data Centre (WOUDC) in Canada
( There are data in the WOUDC dataset for some stations that were reported by earlier organizations. The processing and selection
rules for these data are unknown. The older (1995) processing included time periods of special processing to attempt to account for specific problems in the older
optics of specific instruments. This was accomplished by a modification to the reference lamp correction used in the data
processing. The lamp corrections for the pre-1995 processing were extracted from the LLF format and applied to the WinDobson data
process to introduce the correction applied in the earlier processing. In some cases, the full correction was not possible to
reproduce, so special The so-called mu dependency (Komhyr et al., 1995), where DS results are lower at low sun angles. As this effect is dependent on the
intensity of the input solar beam, and thus on the TOC, no attempt was made to account for this effect in WinDobson processing. This
problem is related to the internal scattered light in the instrument, which is difficult to evaluate. Drifts in the shape of “wedge” calibration. It is unclear how the drift correction was actually performed in earlier
processing; no attempt was made to account for this effect in WinDobson processing. Newer construction of the optical wedges used in
the instrument have proved to have a much more stable calibration. Drifts in the “extraterrestrial constant” as part of the calibration. This was done in the WinDobson processing of later
data, but with a different scheme – multiple There was a weakness in the NOAA processing in choosing a select value for each day. During the original review of observations,
certain observations were rejected for selection; this rejection was not recorded in the LLF files, and thus rejected observations
appeared in the WinDobson dataset. We scrutinized the record for these discrepancies and amended the results. The results of the zenith measurements changed due to updates to the reduction method, and these types of changes affect all
stations – some of the changes are large and are discussed in the individual station reports. For some stations, it is common for observations to be made throughout the local day but with later observations being on the
next consecutive UTC day. This occurs at Lauder (LDR), Samoa (SMO) and the South Pole (SPO), where UTC dates change during normal
observing period. For SPO, observations on a local day can differ by 22 h; thus the choice of the selected/representative ozone in the
change from NOAA to WinDobson processing and selection may differ by 22 h. At certain times of the year, the TOC can change
appreciatively during this time period at SPO. Data archives sometimes failed to be updated after a calibration drift was detected during an intercomparison with
a standard. This is not necessarily a failure of the internal WOUDC archiving process. NDACC appears to capture these periods more
correctly. The rules for choosing the NOAA selected value for the day were similar to that of WinDobson but were not consistent throughout
the record or across stations, and the documentation of these rules is incomplete. For the WinDobson processing, the same rules are
applied throughout the record and stations and are described in Sect. 2.4. Our investigations of the station and instrument operation history revealed several periods for which different
Often there are multiple observations on an individual day. The observations are given an internal numeric code in WinDobson, based on the observation type and operator input about the observation. The representative value is chosen by the software with the priority groups given below, highest to lowest. These groups are based on Table 2 in the Operations Handbook (Komhyr and Evans, 2008). If there are multiple observations of the highest priority on that day, the observation closest in time to local noon is chosen. After the automatic selection, the daily representative values are reviewed by human inspection with possible intervention to select a different value. The WinDobson software also has quality control routines that rates individual observations as good, questionable (flagged yellow) and likely bad (flagged red), based on internal consistencies of the measurements. If an observation is rejected by the human inspector, the observation is not removed from the data record but flagged as “not included”.
Priority groups are listed here; operator inputs as to sky quality are included in determining priority:
Direct-sun observations using the AD pair combination with or without ground
quartz plate (diffuser) in the instrument's inlet window. Observations with
diffuser have higher priority. Zenith-sky observations using the AD pair combination. Observations on the
clear zenith have higher priority over those on cloudy conditions. Direct-sun observations using the CD pair combination with ground
quartz plate (diffuser) in the instrument's inlet window. Observations without
diffuser have lower priority. Zenith-sky observations using the CD pair combination. Observations on the
clear zenith have higher priority over those on cloudy conditions. Zenith-sky observations using the CC Observations on light reflected from the moon. Observations using AD pair
combination have higher priority. Note these observations are rarely made
other than at the South Pole station during the austral winter.
A discussion of the individual station records and the changes is presented
in the following section.
The station discussions are accompanied by a referenced graphic of the
time-dependent differences, consisting of either three panels (all stations)
or five panels (NDACC stations). Panel A is the time record of total ozone
measured at the station from the start of observations through 2014 (or until
station was converted to WinDobson processing). Panel B is the percent
difference between daily WinDobson total ozone records compared to the WOUDC
record (WinDobson–WOUDC). The red line is a linear fit. Panel C is the same
as the second but for monthly and yearly averages (based on all the values in
the month in each dataset). The small white circles are averages made from DS
observations only, the red symbols represent averages using all Dobson total
ozone records and the large black open circles are yearly averages of all
observations, based on monthly averages. Large triangle symbols indicate
major calibration or instrument changes that lead to creating the new
Assessment of changes in the WinDobson representative dataset relative to
WOUDC record is analyzed in the form of probability distributions, where
percent differences in TOC are plotted (Fig. 3) as a function of likely change
when the archive is updated. The datasets analyses are separated into ADDS
and other type of measurements. The ADDS curves are symmetric and
indicate that the vast majority of ADDS values will be unchanged. The
“other” curves are less symmetric and are driven by the updated ZS
reduction polynomials. As the overall record average offsets are small
(
Observations at MLO were started in December 1957. The instrument was
damaged in 1961, and thus the calibration is unknown prior to 1963. Before
1984, the primary instrument was D063, with short periods with other
instruments. The data in the archive prior to 1984 were not processed in the
standard method in an attempt to account for instrument calibration drifts
and other instrument problems, which causes larger variation in the
comparison of original to the WinDobson record prior to 1984. The automated
instrument D076 was installed at the station in 1984 after rebuilding in
Boulder. A mirror deteriorated, so the calibration in the period 1990–1995
(indicated by the yearly
The South Pole station was established in 1957. The first Dobson instrument failed due to the extreme cold. Observations started again in 1961 and these results are in the NOAA archive, but the calibration record dates from 1963. The normal routine established in 1985 was to change the instrument every 4 years for calibration checks, but this was not always achieved. This station has the possibility of large changes in reported daily values in the WinDobson, primarily due to the extended daily observation period, and high variation in total ozone during certain periods of the year. The station local day is the same as that of Christchurch, New Zealand, for ease of logistics, but the Dobson observations are reported in the WOUDC in UTC date and hour. The date and time combination often is misleading (for example, in the WOUDC archive, 14 November 1994 has a time of 28 h UTC, which matches the WinDobson and NDACC 15 November 1994 values). The calculation of the astronomical parameters used in the algorithm for reducing reflected moon observations was incorrect in the NOAA program throughout the period of record. Changes in the method of deriving total ozone from ZS observations improved the average with respect to DS averages but creates differences between the old and new archives. There are several periods missing from the WOUDC and NDACC archives (for example, July through December 2002). The difference between the WOUDC and NDACC archive records processed in the NOAA system and WinDobson system is presented graphically in Fig. 5. The exclusion of low TOC values in early October in the archived data (small white circles are outside of the plot range) in some years also produces large percentage differences in the averages (see large deviations in open circles seen in some years in the panels c and e). An example is October 1994, where there are 25 reported days in the WinDobson record but only 18 reported in the WOUDC and only 10 in the NDACC archive. These inconsistencies can produce large percentage differences, especially during low ozone conditions.
The rules for selection and inclusion of days in the archives appear have
been inconsistent in earlier (NOAA) processing and archiving. The NDACC
archive prior to 1999 has TOC expressed as vertical column density
(
The instrument is operated by the US National Weather Service office at
Bismarck Airport. There are observations in the archive from the late
1950s, but the documented record starts in December 1962. The difference
between the WOUDC and NDACC archive records processed in the NOAA system
and WinDobson system is presented graphically in Fig. 6. The periods where
the
The instrument is operated at the National Weather Service office at the
Caribou Airport. There are observations in the archive from the late 1950s,
but the documented record starts in August 1962. The Weather Service office
was rebuilt in the early 2000s, with data gaps during that period of the
record. The difference between the WOUDC and NDACC archive records
processed in the NOAA system and WinDobson system is presented graphically
in Fig. 7. The periods where the
The instrument is operated at the National Weather Service office near Old
Hickory, Tennessee. There are observations in the archive from the late
1950s, but the documented record starts in July 1962. This station record
shows a larger offset (
Observations were started at the Fairbanks airport in 1964 using instrument D076 but ceased in 1972. The values in the WOUDC archive
in the 1964–1972 period do not correspond to the values in the older NOAA internal archive for reasons not determined. Observations
were restarted at the Poker Flat Research Range (65
Dobson observations were started at the University of Colorado east campus in
1966. Earlier observations were made either at the National Center for
Atmospheric Research or at the Table Mountain facility north of Boulder. The
station was moved to the David Skaggs Research Center in 1999. Multiple
instruments have been used here in the record, especially prior to the
automation of Dobson instrument D061 in 1980. The observations made after
1980 automation do not include CC
Dobson observations were started at WIFC in 1967 as support for balloon- and rocket-borne experiments. The station has moved several times to different sites within the facility. Since 1995 only ADDS observations are made to support ozonesonde flights. There are periods in the WOUDC and NDACC archives with either missing data or archived with incorrect calibration. The difference between the WOUDC and NDACC archive records processed in the NOAA system and WinDobson system is presented graphically in Fig. 11.
Dobson observations at the NOAA observatory began in 1973. The instrument was out of operation between 1983 and 1986 due to lack of funding. The difference between the WOUDC archive processed in the NOAA system and WinDobson system is presented graphically in Fig. 12. The station's weather is far cloudier than at other stations, with the station reporting 58 % ZS observations. The change in the method for retrieving TOC from these observations is evident in the variability in the differences between the archives.
Dobson observations were started at the NOAA observatory in 1976. The station is in a warm, humid marine environment, which caused
instrument degradation in the early part of the record. The original processing pre-1995 was not standard and not repeatable. The
periods where the
Dobson observations were started at the Fresno Weather Service Office, California (37
Dobson observations were started at the Observatoire de Haute-Provence (Station Géophysique Gérard Mégie) in 1983 with an automated instrument. This instrument was updated to the WinDobson automation and data processing in 2014. The station and instrument are operated by the Centre National de la Recherché Scientifique (CNRS). The period of 1990 to 1999 was reprocessed to account for calibration drift but has not yet been updated in WOUDC and NDACC. The difference between the WOUDC and NDACC archive records processed in the NOAA system and WinDobson system is presented graphically in Fig. 15. The instrument was damaged several time in its history; inspection within WinDobson resulted in the removal of days from inclusion in the record. This produced several months of higher differences (February 2013, for example).
Graphic representation of the changes in the Lauder, Central Otago, New Zealand (45
Dobson observations were started originally in 1969 at Perth Airport weather radar, Western Australia, then the NOAA automated instrument D081 was installed in 1984. The instrument is operated by the Australian Bureau of Meteorology (BoM). In the late 1990s, the station was moved to the newly constructed weather station. There are periods of missing data in the WOUDC archive. The period after 2012 in the WOUDC archive does not have correct calibration information, as the BoM recalibrated the instrument, and this information was not included in NOAA's database of calibrations. The difference between the WOUDC and NDACC archive records processed in the NOAA system and WinDobson system is presented graphically in Fig. 16.
Dobson observations began in early 1987 at the research station in Central Otago, South Island, New Zealand. The instrument is operated
by New Zealand's National Institute of Water and Atmospheric Research (NIWA). The station's time zone is UTC
NOAA has submitted nearly a half-century's data into the WOUDC and NDACC
archives. Personnel and data processing protocols changed many times
throughout that period, and knowledge of early techniques was slowly lost.
Furthermore NOAA personnel tended to use a larger and more comprehensive
database when performing research, so the accuracy of data within the WOUDC
and NDACC archives was seldom questioned. Our experiences in the
investigation of the long-term archived NOAA Dobson data records should alert
other Dobson data producers of the importance of regular review and
intercomparisons of the archived station's records residing at multiple
archives. The Dobson station data processing procedures and software tend to
change over time after the new knowledge or technology becomes available.
Although the reprocessing of historical datasets is extremely difficult due
to lost documentation or even raw data, the benefit of the investigation is
record's homogenization and adjustment to conform to the WMO/GAW operating
procedure guidance (Komhyr and Evans, 2008). With the advent of WinDobson
software and its newer technique for calculating TOC from zenith observations
and selecting representative observations, we felt it was prudent to
reprocess all previous measurements for the sake of homogeneity. It also
seemed logical to compare and replace data within the WOUDC and NDACC
archives with the newly reprocessed data. The overall changes are small
(
The new WinDobson data are now available from
The authors declare that they have no conflict of interest.
This article is part of the special issue “Twenty-five years of operations of the Network for the Detection of Atmospheric Composition Change (NDACC) (AMT/ACP/ESSD inter-journal SI)”. It is not associated with a conference.
The authors would like to acknowledge the work done in the past by such people as Walter Komhyr, Robert Grass and Kent Leonard in establishing the US Dobson ozone network.
The Dobson observations at Lauder are supported through NIWA's core research funded by the NZ Ministry of Business, Innovation and Employment; at Perth by the Bureau of Meteorology, an executive agency of the Australian government; and at l'Observatoire du Haute-Provence by the National Center for Scientific Research (CNRS), under the responsibility of the French Ministry of Education and Research. Support for updating of the automation at several NDACC sites was provided by the NOAA Joint Polar Satellite System (JPSS) Calibration/Validation program. Edited by: Hal Maring Reviewed by: three anonymous referees