NOAA's program of long-term monitoring of the vertical distribution of ozone
with electrochemical concentration cell (ECC) ozonesondes has undergone a
number of changes over the 50-year record. In order to produce a homogenous
data set, these changes must be documented and, where necessary, appropriate
corrections applied. This is the first comprehensive and consistent
reprocessing of NOAA's ozonesonde data records that corrects for these
changes using the rawest form of the data (cell current and pump temperature)
in native resolution as well as a point-by-point uncertainty calculation that
is unique to each sounding. The reprocessing is carried out uniformly at all
eight ozonesonde sites in NOAA's network with differences in sensing solution
and ozonesonde types accounted for in the same way at all sites. The
corrections used to homogenize the NOAA ozonesonde data records greatly
improve the ozonesonde measurements with an average one sigma uncertainty of
Soon after the discovery of ozone in the atmosphere by Schönbein in 1840 (Bojkov, 1986) the first semi-quantitative measurements of ozone were made by exposing starch/iodide test papers to outdoor air using the Schönbein “ozonometer” developed in 1845 (Bojkov, 1986; Graedel, 1993). The scientific interest over this new form of oxygen resulted in a broad range of studies that focused on the role ozone plays in the atmosphere and refining measurement techniques. Accurate measurements of ozone by wet-chemical methods using a bubbler and aqueous potassium iodide (KI) were developed. A. Levy, using a bubbler/titration technique, began daily surface ozone measurements at the Montsouris Observatory in France (Volz and Kley, 1988) that continued for 34 years from 1876 to 1910. The wet-chemical method based on the fast reaction of ozone and iodide in a neutral buffered KI solution remained a standard measurement method up through the 1970s, when ozone studies focused on air quality in cities along the California urban corridor. By the 1980s, ultraviolet photometry had become the new standard for measuring surface ozone (Oltmans, 1981). However, the neutral potassium iodide method remained a useful technique for balloon-borne vertical profile measurements of ozone. A number of balloon-borne techniques were tested and employed to measure the ozone vertical profile. Early ozonesondes included optical (Külke and Paetzold, 1957; Kobayashi et al., 1966), chemiluminescent (Regener, 1964) and electrochemical (Brewer and Milford, 1960) sensors. Each of these methods exhibited limitations in terms of making an accurate quantitative measurement of the ozone profile as well as somewhat cumbersome preparation procedures (Moreland, 1960). The electrochemical concentration cell (ECC) ozonesonde eventually emerged as a widely used, relatively simple method to measure accurate ozone profiles from the surface to 30–35 km above sea level when the sensing instrument is interfaced with a balloon-borne meteorological radiosonde (Komhyr et al., 1969, 1995a).
Ozonesondes have played an important role in monitoring the stratospheric ozone layer where harmful solar ultraviolet radiation is absorbed by ozone, thus protecting the biosphere (Stolarski, 2001). Although ozonesonde sites around the globe are relatively sparse and not uniformly distributed, selected long-term data sets have been compared and analyzed for trends. Ozone trend estimates at selected altitude intervals were first reported by Logan (1985, 1994), Tiao et al. (1986), London and Liu (1992), and Oltmans et al. (1998) using data from several ozonesonde sites that had compiled 2 or more decades of data with 1–2 balloon flights per week. More recently, ozonesonde data have been used in developing ozone climatologies (Tilmes et al., 2012; Hassler et al., 2013; Sofieva et al., 2014) and validating satellite tropospheric retrievals (Verstraeten et al., 2013; Martins et al., 2015; Thompson et al., 2012; Hubert et al., 2016). Ozonesonde data have been used for analyzing long-range transport of tropospheric ozone (Cooper et al., 2011) and stratospheric/tropospheric exchange events (Terao et al., 2008; Langford et al., 2012). Ozonesondes have shown the characteristic view of the zero ozone depletion layers during the Antarctic ozone hole monitoring (Hofmann et al., 2009; Hassler et al., 2011) as well as revealing Arctic stratospheric ozone loss rates (Rex et al., 2002). An important question at this time is how stratospheric ozone responds to climate variability in the future (Harris et al., 2015).
Data homogenization is necessary for long-term ozone profile records that have gone through instrument and operating procedure changes in order to provide consistent data with reduced uncertainties and offsets. The framework for addressing global data quality and consistency from all ozone profile measurement techniques (Hassler et al., 2014) came from the SPARC/IO3C/IGACO-O3/NDACC (SI2N) initiative designed in 2011 (SPARC – Stratosphere-troposphere Processes And their Role in Climate; IO3C – International Ozone Commission; IGACO-O3 – Integrated Global Atmospheric Chemistry Observations – Ozone; NDACC – Network for the Detection of Atmospheric Composition Change). In order to directly address the quality of ozonesonde data records, a subsection of the SI2N initiative presents the OzoneSonde Data Quality Assessment (O3S-DQA) and homogenization of the balloon-borne ozonesonde records. The O3S-DQA report by Smit and the O3S-DQA panel (2012) outlined the following goals: (a) produce a fully homogenized ozonesonde data set from selected long-term sites by removing biases from known changes in instruments and applying transfer functions for sensor solution changes in operating procedures; (b) clearly document the process and address quality of the individual ozonesonde profiles; (c) reduce uncertainty from 10–20 to 5–10 %; and (d) include uncertainty in the reported ozonesonde flight data.
The recommended guidelines for homogenizing long-term ozone records and standardizing current operational procedures (Smit and the O3S-DQA panel, 2012) are based on several ozonesonde intercomparison projects in the laboratory and field. The laboratory experiments were conducted at the World Calibration Centre for Ozonesondes (WCCOS) in Jülich, Germany. The WCCOS is an environmental chamber capable of simulating various ozone profiles with a UV ozone photometer reference measurement (Proffitt and McLaughlin, 1983). Jülich OzoneSonde Intercomparison Experiments (JOSIE) were conducted in 1996, 2000, and 2009. These experiments had slightly different set ups and goals, but usually focused on comparing different ozonesonde models from the manufacturers (Science Pump Corporation (SPC) and EN-SCI Corp), different sensing solution recipes (Smit, 2007; Smit and Sträter, 2004a), and different standard operating procedures (SOPs) (Smit and Sträter, 2004b). A field test intercomparison of ozonesonde models and sensor solutions was conducted during the World Meteorological Organization (WMO)-sponsored Balloon Experiment on Standards for OzoneSondes (BESOS) campaign held at the University of Wyoming Balloon Facility in Laramie, Wyoming, USA (Deshler et al., 2008). The BESOS balloon gondola carried 12 ozonesondes (6 EN-SCI and 6 SPC) alongside the JOSIE UV ozone reference instrument.
These intercomparison projects showed that when the same sensing solution was used the EN-SCI Corp model ozonesondes measured approximately 5 % higher ozone than SPC ozonesondes; when the same ozonesonde type was used the standard 1 % KI buffered sensor solution measured approximately 5 % higher ozone than the half-percent solution (Smit et al., 2007; Deshler et al., 2008). Deshler et al. (2017) provide linear transfer functions to apply to ozonesonde data to account for changes in ozonesonde model or sensor solutions based on the BESOS and JOSIE intercomparisons and other multi-ozonesonde comparison flights done by individual ozonesonde groups.
Several ozonesonde sites have published results of homogenized records, including the Canadian ozonesonde (Tarasick et al., 2016) and the Southern Hemisphere ADditional OZonesondes (SHADOZ) networks (Thompson et al., 2017; Witte et al., 2017a). Homogenization of sites that switched from Brewer–Mast-type ozonesondes to ECC ozonesondes include Uccle (Lemoine and De Backer, 2001; Van Malderen et al., 2016) and Payerne Aerological Station (Stübi et al., 2008). Dual and multiple ozonesonde flight data at Sodanklyä were used to homogenize data and evaluate trends from different Arctic ozonesonde sites (Kivi et al., 2007; Christiansen et al., 2017).
Here we present the homogenization procedure and results for the NOAA/ESRL/GMD ECC ozonesonde network consisting of eight long-term monitoring sites. While this effort represents the first reprocessing of these data that attempts to account for all known contributors to inhomogeneity and biases in the data in a systematic way, several earlier versions of the data have been available that tried to account for some of the inhomogeneity and biases. Previous versions of the ozonesonde data archived at NDACC, SHADOZ, and the WOUDC accounted for pump efficiency losses, impact of sensing solution composition, unrealistic background current measurements, and ozonesonde manufacturer differences. A better quantification of these factors as well as a number of others is discussed and their incorporation into the reprocessed data set is presented here.
Homogenizing long-term records of ECC ozonesonde data begins with reviewing the upgrades and changes in instrument design and SOPs. Table 1 lists the different models and manufacturing dates of ozonesondes made by Science Pump Corporation and EN-SCI Corporation. There have been seven changes in the manufacturer model design. NOAA's first ozonesondes in 1967 included the earliest version, 1A, ozonesonde. Ozonesonde launches were infrequent until after 1985, when three sites began launching regular weekly ozonesonde flights, eventually using all of the ozonesonde models (1A, 3A, 4A, 5A, 6A, 1Z and 2Z) up to the present time. There were two major design changes in the ozonesonde models. One was the introduction of the more efficient, cylindrical cross-section pump in the 4A ozonesonde. Section 3.2 outlines the method for accounting for this change. The second was moving the position of the thermistor to more accurately measure the true gas temperature flowing through the pump chamber. Section 3.3 outlines the method of applying a correction algorithm for adjusting box temperature to gas temperature in the pump.
Ozonesonde manufacturer, model, years manufactured, and design changes.
The guidelines for preparing an ozonesonde for flight include the manufacturers instruction manuals and the WMO SOPs (Smit and ASOPOS panel, 2014), which are based on workshop reviews of JOSIE and BESOS ozonesonde testing. During the long-term NOAA record there have been adjustments to the guidelines or SOPs that include, for example, how to measure the cell background and changes in the radiosonde interfaced with the ozonesonde. By far, the two changes that have the greatest impact on ozonesonde measurement accuracy are the pump flow rate efficiency correction curve applied and adjustments to composition of the sensor solutions (Johnson et al., 2002; Smit et al., 2007).
Changes in the ozonesonde sensing solution compositions (Table 2) used are a
significant factor that needs to be taken into account since this affects
the chemistry of the ozone iodide reaction stoichiometry. The sensor
solution composition recipes used by the early ECC ozonesondes originated
from the wet chemical, iodometric techniques (Bartel and Temple, 1952;
Littman and Benoliel, 1953; Saltzman and Gilbert, 1959; Boyd et al., 1970). The method involves the absorption of ozone and oxidation of iodide
ions to iodine (I
Amount of each chemical in grams/liter deionized water used in the five commonly used cathode sensing solutions. (1) Komhyr and Harris (1971), (2) 1994 EN-SCI Model 1Z Instruction Manual, (3) 1996 revised EN-SCI Model 1Z & 2Z Instruction manual, (4) Johnson et al. (2002), (5) currently used by NOAA since 2005.
The iodine product can be measured by titration, colorimetric methods, or
coulometry. For example, Saltzman and Gilbert (1959) used kinetic
colorimetric detection of iodine when testing various absorption reagents.
They found that the 1 % KI with neutral phosphate buffering gave the best
result, close to the ideal 1
The ECC ozonesonde is interfaced with a radiosonde to transmit the ozone
data to the surface and have an accurate measurement of atmospheric
conditions, most importantly ambient pressure, temperature and relative
humidity. VIZ radiosondes were used during the analog era (1967–1991) and
gave a data resolution of approximately 1 min or 300 m. The VIZ
radiosondes used a hypsometer for pressure measurements at altitudes above
The RS-80 radiosonde manufactured by Vaisala was used by NOAA from 1991
until 2009, when transition to the iMet-1 radiosondes began. The RS-80s
allowed for digital data acquisition when paired with an electronics board
attached to the ozonesonde. The TMAX electronics board was used to couple
the ozonesonde to the RS-80 radiosonde and was capable of measuring and
transmitting data every 7 s. The V2 electronics board introduced
in 1998 improved the electronic components and increased the time resolution
to 1 s data. The current radiosonde being used by NOAA is the iMet-1
manufactured by International Met Systems. i-Met radiosondes are equipped
with a GPS receiver. Comparing the geometric altitude of the GPS to the
geometric altitude calculated from the pressure, temperature, and relative
humidity from the radiosonde allows for an accurate pressure offset to be
applied to the pressure sensor. The geometric altitude is only used for
correcting the pressure sensor; the geopotential altitude is reported in all
data files. A majority of flights conducted using RS-80 radiosondes did not
have a GPS receiver attached. Several techniques were employed to evaluate
possible errors in the pressure reading and make corrections. Radiosonde
pressure readings at the surface were compared with an accurate surface
pressure measurement. Testing of a number of RS-80 radiosondes in an
altitude chamber showed that the pressure offset at 7 hPa was on average
75 % of the pressure offset observed at the surface. This method of
determining the pressure offset was used for all RS-80 radiosondes from 2008 to 2011 (approximately 1200 profiles in the NOAA long-term network). Before
2008, the RS-80s pressure sensors were new and thus more accurate. A new
data acquisition and processing software called SkySonde was developed to
facilitate the implementation of the corrections associated with the data
quality assessment project. The SkySonde processing software allows for
comparing temperature profiles from nearby meteorological soundings to the
temperature profile measured by the RS-80. RS-80s with large pressure
offsets (
NOAA followed the WMO reprocessing recommendations and guidelines when
applicable (Smit and the O3S-DQA panel, 2012). However, NOAA uses a unique
sensor solution recipe and measured its own pump efficiencies which
necessitated deriving corrections for these unique cases for the NOAA and
many of the SHADOZ ozonesonde data records (Thompson et al., 2012, 2017).
The ozonesonde equation for calculating the ozone partial pressure
(
The first term is a constant consisting of the universal gas constant (
In order to homogenize the NOAA ozonesonde data record and account for changes in ozonesonde types and sensing solutions, a two-step approach was taken. First, the variables that can be quantified directly were treated consistently through the entire record. Individual ozonesonde data profiles were quality-controlled by correcting or flagging erroneous measurements in the measured cell current, pump temperature, and radiosonde pressure. Failed ozonesonde flights were screened out or data were cut off at altitudes where the ancillary data such as battery voltage or pump temperature indicated a failure. Profile altitude errors from radiosonde pressure offsets (before GPS geometric altitude became available) were fixed by applying corrections to the pressure sensors as noted earlier. Erroneously measured variables such as cell current backgrounds were fixed systematically, changes in how variables are measured such as pump temperature were accounted for, and climatological or average values were assumed in instances where a variable was not used in historic data such as for pump flowrate corrections. Second, the ozone sensor efficiency was determined for the different sensing solution and ozonesonde types from the comparisons of the ozonesonde and the reference UV photometer at JOSIE. The ozone sensor efficiencies were then applied appropriately to all data files to create a consistently calculated, homogenous data set. This approach homogenizes the data to the ozone photometer at WCCOS for each solution type and ozonesonde type by applying a unique ozone sensor efficiency. This is in contrast to the approach of homogenizing the record to one of the ASOPOS standard ozonesonde type/solution type/pump efficiency pairing and using transfer functions to adjust for changes in the record.
Figure 1 shows the many changes to the NOAA ozonesonde record. The changes in solution, ozonesonde type, digital-to-analog data acquisition, and an observed change in the cell current backgrounds led to a logical division of NOAA's ozonesonde data record into five eras.
The eight long-term NOAA ozonesonde stations with latitude, longitude, number of profiles, and launch period.
Era 1 is the earliest portion of the analog era from 1 January 1967 to
1 June 1982, which primarily used 1A and 3A ozonesonde types and the
1.0 % KI, 1.0
The historic JOSIE data sets were valid in quantifying the ozone sensor efficiency for these different eras because the ozonesonde measurements taken at JOSIE were consistent with the ozonesondes, solutions, and standard operating procedures being used by NOAA at the time.
Before homogenizing the NOAA ozonesonde network all of the necessary metadata that were available were collected and added to the digital data files, and all data files were converted to a common, editable file type which includes the rawest form of the data (cell current and pump temperature). This allows the SkySonde software to read all data files and calculate all ozone values from the raw cell current and pump temperature regardless of the data acquisition system or file format previously used. This was a time- and labor-intensive process. The 1 min analog data were read from chart records and digitized. It was common to only calculate significant and designated levels in the analog chart record data. However, NOAA digitized every 1 min data point for all 1179 analog data files in the NOAA ozonesonde record. In the analog data, the commutator was powered by the pump motor. Changes in the motor speed resulted in changes in the time resolution of the data. With careful consideration, the changing motor speed was accounted for by multiplying the cell current by a motor speed correction factor.
Once all data files were in a common format and included the rawest form of the measurement, corrections could be applied in batch. This first step was a major achievement and paved the way for quickly and easily making changes to the entire data set. This will also allow for future reprocessing of the data if additional information on the characteristics of the ozonesondes (and perhaps radiosondes) performance are obtained.
Early on when a TMax interface board was used, the data acquisition software did not output the cell current. In order to include cell current in the data files, a reverse calculation of cell current was performed. Careful consideration is required to back-calculate cell current correctly. All of the necessary variables needed to back-calculate the cell current from the ozone partial pressure were available in the data file. Thus, this calculation was carried out with negligible error.
The measured cell current is the electrical current that is produced by the ozone sensor cell and measured by the electronics board throughout the flight. The time resolution and acquisition systems have changed over the record, but the variable has not. The background cell current is the residual current produced by the ozonesonde when ozone-free air is sampled and is determined during the flight preparation. A detailed analysis of the source of ozonesonde background current revealed that it was not oxygen dependent. Thornton and Niazy (1982, 1983) and Vömel and Diaz (2010) demonstrated that cell current background declines for up to 90 min when ozone-free air is sampled after exposure to ozone, as well as the importance of the background in the very low ozone observed in the tropics. It is theorized that this long decaying background is related to the slow side reactions of the phosphate buffer.
Current recommended SOPs call for three cell current background measurements to be recorded. Ib0 is recorded after the ozonesonde has been sampling ozone-free air for 10 min before ozone exposure, Ib1 is recorded after sampling ozone-free air for 10 min after ozone exposure, and Ib2 is recorded directly before launch with the goal of achieving a low and constant reading (Smit and the ASOPOS panel, 2014). Historically, NOAA has always used Ib2 for calculating ozone. For portions of the early record, Ib2 was the only cell current recorded. References to the cell current backgrounds in this work are to Ib2. NOAA's SOPs were to use an ozone destruct filter at the launch site to establish the background current of the cell. These filters degraded over time, especially in humid marine environments, causing many erroneous background measurements. When an ozone-free air source is used, the background is dependent primarily on the solution type and also on the ozonesonde type. Additionally, cell current backgrounds decreased substantially around 1991 (Smit and the O3S-DQA panel, 2012). These facts align well with the eras since they are primarily based on solution type changes. The drop in backgrounds in 1991 led to grouping Era 1 and Era 2 together and leaving Eras 3, 4, and 5 separate for the cell current background analysis as seen in Fig. 2. In Fig. 2a and b, the large number of backgrounds greater than the scale of the histograms are attributed to erroneous measurements due to the degraded ozone destruct filters. To correct the erroneously high background measurements, a background reduction system was created based on an average cell current background and standard deviation for each era. If the measured cell current background was greater than the average background plus 1 standard deviation, the background measurement was replaced by the average value.
Histogram of all cell current backgrounds from Boulder, South Pole
and Hilo broken into four time periods:
The three longest running stations (Boulder, South Pole, and Hilo) have had
the most consistent and highest-quality ozone preparation and documentation.
Figure 2 shows the histograms of the originally measured backgrounds after
exposure to ozone, Ib2, at these three sites. When these histograms are
compared to the backgrounds taken at intercomparisons, it is clear that
Era 1/2 and Era 3 were measuring a large number of erroneous backgrounds. The
statistics on the backgrounds in these eras (Fig. 2a and b) are
not indicative of the actual backgrounds and thus are not used for the
background reduction. Instead, the mean and standard deviation found at
intercomparisons where high quality background measurements were taken are
used. For Era 1 and 2, the mean background was taken as 0.09
All ozonesonde pump flowrates were measured with a 100 mL bubble flow meter
at the station by averaging five stopwatch measurements. The measured flow
rate must be corrected for two issues. A correction must be applied to
account for the humidification of air being measured and the cooling of the
air from the pump temperature to the temperature of the air being measured
in the bubble flow meter (Smit and ASOPOS panel, 2014). Second, a correction
must be applied to the volumetric pump flowrate to account for the loss of
efficiency of the ozonesonde pump at pressures below 300 hPa. The volumetric
pump flow rate (
The pump flowrate correction for the surface measurement (
The correction for the humidification effect (
During the flowrate measurement, the ozonesonde samples the filtered air
exiting the test unit. The volume of air being measured becomes saturated
with water vapor as it is bubbled through the sensor solution and travels
along the wetted walls of the bubble flow meter. The correction for the
humidification effect (
The volume of water vapor added to the air being measured is dependent on
the ambient pressure in the flow meter, the vapor pressure of the air in the
flow meter, and the relative humidity of the air entering the ozonesonde
pump, which is assumed to be the relative humidity of the air exiting the
test unit (RH
The correction for the pump temperature and air temperature in flow meter
difference is assumed to be adiabatic compression and is approximated by
Eq. (6):
The pump temperature during the flowrate measurements (
Since 2010, NOAA has used a Drierite air purifier/desiccant filter rather than canister ozone destruct filters to produce a zero ozone air source at the Boulder, Trinidad Head, and Fiji sites. The desiccant strips the air of all water vapor. With a stable lab temperature and pressure and zero humidity air being sampled for the flowrate measurement, the flowrate correction becomes nearly constant.
Figure 3 shows the different climatological flowrate corrections
(
Monthly climatological volumetric pump flowrate corrections for surface measurement.
As the ambient pressure decreases during flight, the efficiency of the
ozonesonde pump begins to decline due to leakage, the dead volume in the
piston, and the back pressure exerted on the pump by the sensor solution
(Komhyr and Harris, 1971, Steinbrecht et al., 1998, Johnson, 2002). Smit and
the ASOPOS panel (2014) recommends using the Komhyr (1986) or Komhyr et
al. (1995) pump efficiency corrections. This recommendation was based on the
observed agreement of the ozonesonde to the reference ozone photometer at the
JOSIE and BESOS intercomparison campaigns when the Komhyr (1986) and Komhyr
et al. (1995) pump efficiencies were paired with a 1 % KI, 1.0
The Komhyr (1986) pump efficiencies were measured with a similar apparatus as
Torres (1981) with the assumption that the hydrostatic back pressure from the
sensing solution and the pump dead volume were responsible for the loss of
efficiency of the ECC pump. The Torres (1981) apparatus used the ozonesonde
pump to pressurize a chamber to the expected hydrostatic back pressure at
varying pressure levels. The Komhyr (1995) pump efficiencies assumed that the
pump efficiency of an ozonesonde pump was 100 % at all pressures if no
back pressure was applied to it. The apparatus used to measure the
Komhyr (1995) pump efficiencies used two competing ozonesonde pumps (one
pumping into a sensing cell with 3 cm
This agreement led to using the Johnson et al. (2002) “all average” for the 1A and 3A pump efficiencies in this work. The PCF averages for 5A, 1Z, and 2Z ozonesonde types were all within 1 standard deviation up to 10 hPa. The 6A average fell outside of 1 standard deviation. Due to this fact, 6A ozonesondes were processed with the Johnson et al. (2002) 6A average PCFs. All other ozonesonde types are processed with the “all” average PCFs. An updated and more detailed study of the ozonesonde pump efficiency could provide reduced uncertainty in the pump flowrate and improved confidence in the consistency of the pump performance over time.
An accurate measurement of the pump temperature is required to calculate the
volume of air passing through the ECC pump. The location of the pump
temperature measurement has changed multiple times. In the NOAA ozonesonde
record, there are three possible configurations. For 1A, 3A, and 4A
ozonesonde types, a rod thermistor at the base of the ozonesonde body was
used. For the 5A ozonesonde type, a thermistor was epoxied to the surface of
the pump block. For 6A, 1Z, and 2Z ozonesonde types, the thermistor was
mounted inside a hole drilled in the pump block. In order to account for
these changes, the WCCOS conducted experiments comparing old pump
measurement configurations to the new configuration and the new
configuration to the internal piston temperature (Smit and the
O3S-DQA panel, 2012). All temperatures used in calculating ozone are in
kelvin. The pump temperature (
For 1A, 3A, and 4A ozonesondes, the correction for the difference in the
temperature measured by the rod thermistor at bottom of the ozonesonde and
the temperature inside the pump block (
This set of transfer functions increases the pump temperature by 4.5–7.0 K. For the 5A ozonesonde, the correction for the difference in the temperature measured by the thermistor epoxied to the pump base and the temperature inside the pump block is estimated by Eqs. (11) and (12) (Smit and the O3S-DQA panel, 2012):
For 6A, 1Z, and 2Z ozonesonde types the temperature measured is the
temperature inside the pump block. For all other ozonesonde types, the
measured temperature was corrected to the temperature inside the pump block
by Eqs. (8–12). To obtain the best estimate of the pump temperature, the
difference in the temperature inside the pump block and the internal piston
temperature (
After these pump temperature corrections are applied, the pump temperature used in Eq. (2) for all ozonesonde types has been transferred to the internal piston temperature, making the pump temperature measurements homogenous.
The ozone sensor efficiency (
These variables are difficult to measure directly and independently, so they are measured and accounted for by comparing to an ozone photometer. The past JOSIE experiments are of great value in quantifying the ozone sensor efficiency for the different eras and ozonesonde type–sensing solution configurations. In order to accurately measure the ozone sensor efficiency by this comparison, the previously discussed variables used to calculate ozone partial pressure that can be quantified directly must be treated identically in ozonesonde data record and the JOSIE comparison. For example, the pump flowrate efficiency used to calculate the partial pressure of ozone for the JOSIE experiments must be the same efficiencies used in the data record. Otherwise, the comparison and derived ozone sensor efficiency will be invalid. The ozone sensor efficiency is determined by iteratively minimizing the percent difference in the ozonesonde and the ozone photometer for a given ozonesonde type–sensing solution pairing. Figures 4 and 5 show these differences. The differences seen in the ozonesonde and the ozone photometer at the JOSIE campaigns cannot be attributed to just one of these efficiencies. Therefore, the derived ozone sensor efficiency is accounting for both the absorption and conversion efficiency. The ozone sensor efficiency is believed to be dominated by not only the stoichiometry of the reaction but also the ozonesonde type. Therefore, deriving the ozone sensor efficiency for each era is the logical approach.
Percent difference in ozone partial pressure between the
ozonesonde and the reference ozone photometer with 1 % KI, 1.0
Percent difference in ozone partial pressure between the
ozonesonde and the reference ozone photometer with 2 % no buffer solution
and 2Z EN-SCI
The absorption efficiency is a measure of how much of the gaseous ozone in
the air pumped into the sensing solution is absorbed in the liquid phase.
Davies et al. (2003) showed that when 3.0 cm
The conversion efficiency is a measure of how much of the ozone molecules that are dissolved into the cathode solution are converted into electrons. A conversion efficiency of one would follow the stoichiometry of Eq. (1), where one ozone molecule is converted into two electrons. Different sensing solutions and ozonesonde types result in different conversion efficiencies; the positive bias from the phosphate buffers is believed to cause the largest deviations in the conversion efficiency. There may be other unknown processes besides the stoichiometry that affect the conversion efficiency. These efficiencies are accounted for by measuring the ozone sensor efficiency.
The ozone sensor efficiency for Eras 1, 2, and 3 are treated the same as
they all used 1.0 % KI, 1.0
This bias in the ozone sensor efficiency is assumed to primarily be due to
the secondary reaction involving the buffer and is dependent on the amount
of cumulative ozone exposure seen by the ozonesonde up to a given pressure
(or altitude) level. The ozone sensor efficiency was estimated using the
total accumulated column ozone as a measure of the exposure and is
represented by Eq. (15):
The cumulative column ozone is in units of atm
Era 4 was subdivided because Deshler et al. (2008, 2017) and JOSIE 2000 (Smit and Sträter, 2004b) showed an ozonesonde type bias between 6A SPC ozonesondes and Z En-Sci ozonesondes when all other variables were constant. Era 4a used the 2 % KI, no buffer solution unique to NOAA and SHADOZ with En-Sci Z ozonesondes. This sensing solution/ozonesonde type pairing exhibits a negative bias in ozone when compared to a UV photometer (Smit and Sträter, 2004a). It is believed that the lack of potassium bromide (KBr) and a buffering agent in the solution recipe caused this bias. Figure 5a shows the comparison of three ozone profile simulations at JOSIE 2000 for this ozonesonde/solution configuration. The ozone sensor efficiency for 2 % KI, no buffer solution with En-Sci Z ozonesondes is 0.98 throughout the entire profile. Era 4b also used the 2 % KI, no buffer solution, but with SPC 6A ozonesondes. The 6A SPC ozonesondes have been shown to measure 4 % less than EN-SCI ozonesondes up to 30 hPa, increasing to 10.3 % at 10 hPa (Deshler et al., 2017). Deshler et al. (2017) did not account for a difference in pump efficiencies for the 6A and Z ozonesondes. The pressure dependence of the bias is partially accounted for by the difference in the Johnson et al. (2002) 6A average and the Johnson 2002 all-average pump correction factors used in this work. For Era 4b, the ozone sensor efficiency was estimated to be 0.94 through the entire profile as seen in Fig. 5b. The difference between the 2Z and 6A ozonesondes observed by Deshler et al. (2008), JOSIE 2000 (Smit and Sträter, 2004b) and in Fig. 5 have led NOAA to apply an ozone sensor efficiency of 0.96 to all 6A ozonesondes in addition to any needed ozone sensor efficiency for a buffered solution.
Era 5 uses the 1.0 % KI, 0.1
Ozonesonde type–sensing solution pairings with corresponding ozone sensor efficiency.
One of the primary objectives of the ozone data homogenization project was
to estimate and calculate the uncertainty of the ozonesonde measurement. The
partial pressure of ozone is a function of the measured cell current
(
A robust and accurate estimation of the ozone partial pressure uncertainty will be particularly beneficial when conducting trend analyses on this data set.
The uncertainty in the measured cell current is a function of the errors and
uncertainty of the electronics used for the measurement of the measured cell
current. To estimate the uncertainty for the different digital interface
boards, a reference current ranging from 0.025 to 7.5
Piece-wise functions for the relative uncertainty in the measured cell current of each interface board type.
If the cell current background was not reduced and remained the measured
background, the estimated uncertainty in the background is 1 standard
deviation or 0.02
To estimate the uncertainty in the volumetric flow rate of the pump, the
uncertainty in the measurement of the flowrate using a 100 cm
The uncertainty in taking the flow rate measurement with the stop watch and
bubble flow meter (
Ozone partial pressure and the relative uncertainty with the relative uncertainty of each variable vs. altitude for an ozone sounding in Boulder, CO.
Ozone partial pressure and average relative uncertainty for each era vs. altitude for Boulder, CO, and South Pole for April and October.
Ozone partial pressure and average relative uncertainty for each era vs. altitude for Hilo, HI and Pago Pago, American Samoa, for April and October.
Boulder Dobson vs. ozonesonde total column ozone comparison.
South Pole Dobson vs. ozonesonde total column ozone comparison.
Hilo Dobson vs. ozonesonde total column ozone comparison.
The relative uncertainty of the pump efficiency is taken as the 1 standard deviation of the pump efficiency average. Older rectangular cross-section Teflon pumps used in earlier ECC ozonesonde models (1A and 3A) have not had the pump efficiency measured using the techniques in Johnson et al. (2002). As discussed earlier in Sect. 3.2.2, measurements of the 3A pump efficiency (Komhyr and Harris, 1971) using a bag inflation method determined 3A pump efficiencies not to dissimilar to those measured for the cylindrical cross-section pumps by Johnson et al. (2002). Measurements of the pump efficiency using the same technique for both pump configurations found that the rectangular cross-section pumps were less efficient than the cylindrical cross-section pumps (Torres, 1981). Taking this into account, the uncertainty for the pump flowrate efficiency at low pressures for 1A and 3A ozonesondes were doubled to account for this difference.
The uncertainty of the temperature of the pump is estimated by adding in
quadrature the uncertainty of the thermistor and the electronics measuring
the temperature, the uncertainty of the pump temperature difference to the
temperature of the base of the pump, and the uncertainty of the correction
for the temperature of the base of the pump to the internal piston
temperature and is represented by Eq. (18):
The uncertainty of the measurement of the pump temperature (
The uncertainty in the ozone sensor efficiency is obtained by adding in
quadrature the uncertainty in the absorption efficiency and the uncertainty
in the conversion efficiency Eq. (19):
The absorption efficiency is assumed to be 1 with an estimated uncertainty of
Quantifying the uncertainty of each variable used in the ozonesonde equation (Eq. 2) on a point-by-point basis was one of the key goals of the data quality assessment project. Figure 6 shows the uncertainties of each variable as well as the total uncertainty for an example ozone profile from Boulder, CO. The relative uncertainties of each variable in Fig. 6 are added in quadrature to obtain the total uncertainty as shown in Eq. (16). Every profile in the NOAA long-term ozonesonde record now has a unique uncertainty estimate similar to this.
The relative uncertainty of the measured cell current and background current are the largest contributor to the overall uncertainty in the troposphere, when the difference in the measured and background cell current is the smallest. When the cell current reaches its minimum at the tropopause at approximately 9.5 km in Fig. 6, the uncertainty of the measured/background cell current reaches its maximum of approximately 6.5 %. As the ozonesonde measures higher amounts of ozone through the ozone peak from 10 to 25 km, the difference in the measured and background cell currents becomes larger, making the uncertainty smaller. This is of greater importance at tropical sites where very low ozone values are observed through the troposphere. The measured/background cell current uncertainty is the main contributor to the differences in the average uncertainty observed in the troposphere for Eras 1 and 2 compared to Eras 3, 4, and 5 in Fig. 7. This is because more backgrounds were reduced in Eras 1 and 2, causing a larger uncertainty in the background current and thus a larger average uncertainty. The measured/background uncertainty also plays a large role in the average uncertainty plot for the month of October for South Pole station in Fig. 7. The Antarctic ozone hole forms in September and October, when ozone partial pressure drops to very low values from 12 to 22 km. However, in this case Era 1 and 2 show lower average uncertainties, contrary to the average uncertainties in the troposphere. This is because in Eras 1 and 2 (1967–1982 and 1982–1991, respectively) the ozone hole was not as severe as in the later eras; the ozone partial pressure did not get as low through the ozone peak and therefore the difference in the measured and background cell current did not become very small, leading to a lower average uncertainty in this region for those earlier eras.
The relative uncertainty of the volumetric pump flowrate in Fig. 6 is 1.6 % at the surface and increases with altitude to 2.3 % at 30 km. This increase with altitude is due to the uncertainty of the ozonesonde pump efficiency loss at low pressures. It should be noted that the NOAA ozonesonde records use an average pump efficiency and the uncertainty is taken as 1 standard deviation of many pump efficiency measurements. If the individual measured pump efficiency is used to calculate ozone partial pressure, the uncertainty would be the uncertainty of the pump efficiency measurement. With an accurate and repeatable pump efficiency measurement for individual pumps, the uncertainty in the pump flowrate and thus the total uncertainty can be reduced. Figure 6 is using a climatological pump flowrate correction for the surface measurement. When the pump flowrate correction for the surface measurement is measured during the flight preparation, the uncertainty of the pump flowrate at the surface is reduced to approximately 1.1 %.
American Samoa Dobson vs. Ozonesonde total column ozone comparison.
Percent difference in column ozone between the merged SBUV ozone
data and the ozonesonde data at Boulder, CO. Panels
The pump temperature uncertainty is the smallest contributor to the total uncertainty through the entire record. While the pump temperature uncertainty appears to be constant, it is changing as the pump temperature changes through the flight. The earlier ozonesonde types, 1A, 3A, 4A, and 5A, have a larger uncertainty than the example profile in Fig. 6 because of the added step to homogenize the pump temperature measurement to the inside of the pump block. For 1A, 3A, and 4A ozonesondes correcting to the pump temperature inside the pump block with Eqs. (8), (9), and (10) adds approximately 0.33 % to the pump temperature uncertainty for a pump temperature of 300 K and for 5A ozonesondes the correction to the inside of the pump block with Eqs. (11) and (12) adds 0.17 %. The added uncertainty for correcting the pump temperature from inside the pump block to the internal piston temperature which is applied to all ozonesonde types by Eq. (12) is also 0.17 % for a pump temperature of 300 K. This results in the pump temperature uncertainty being largest for Era 1, 2 and 3.
The uncertainty of the ozone sensor efficiency is consistent for each site and is the same for all eras except Era 4, which was increased due to the unbuffered solution. This difference can be seen in Figs. 7 and 8, where the average uncertainty for Era 4 is larger than Era 3, except in cases where the measured/background uncertainty is dominating the total uncertainty. The ozone sensor efficiency uncertainty is a large contributor to the total uncertainty throughout the profile. Further testing and comparisons at the WCCOS will lead to a better understanding of the ozone sensor efficiency and possibly a reduction in its uncertainty.
For a majority of profiles at the various sites and through the various eras, the total uncertainty in the troposphere is dominated by the measured/background cell current and the ozone sensor efficiency uncertainties. In the stratosphere the largest contributors to the total uncertainty are the ozone sensor efficiency and pump flowrate uncertainty.
Percent difference in column ozone between the merged SBUV ozone
data and the ozonesonde data at Hilo, HI. Panels
Percent difference in column ozone between the merged SBUV ozone
data and the ozonesonde data at Pago Pago, American Samoa. Panels
The total uncertainty has improved over time as the uncertainty is lower for each subsequent era except for Era 4 in some cases as shown in Figs. 7 and 8. To illustrate the uncertainty range from surface to balloon burst, the total column ozone is also given with the uncertainty in Dobson units. This is calculated by multiplying the average relative uncertainty of Eras 2, 3, 4 and 5 to the average ozone partial pressure to obtain the average absolute uncertainty. The average absolute uncertainty is then added to and subtracted from the average ozone partial pressure. The total column ozone is then calculated for the high and low ozone partial pressure profiles and the total column uncertainty is simply half of this range. The average relative total column uncertainty for April in Dobson units as shown in Figs. 7 and 8 are 4.4, 4.2, 4.1, and 4.2 % for Boulder, Hilo, American Samoa, and South Pole, respectively. The total relative uncertainty of ozone with altitude are similar in shape and comparable in magnitude to other recent ozonesonde uncertainty estimates, Van Malderen et al. (2016), Tarasick et al. (2016), and Witte et al. (2017b).
To gauge the efficacy of the ozonesonde homogenization, the total column
ozone values calculated from the ozonesonde were compared to Dobson
spectrophotometers and SBUV satellite measurements. It should be noted that
the data shown before applying the ozone sensor efficiency in Figs. 9–15 and
Figs. S3–10 were never data published or available by NOAA. Rather, they
were used to highlight the effect of the ozone sensor efficiency equations
when all other variables were treated the same. To calculate the residual
total column ozone above balloon burst, the SBUV add-on tables produced by
McPeters et al. (2013) were used. If the balloon burst at a pressure smaller
than 7 hPa, the residual column ozone was calculated from 7 hPa. The Dobson
instruments at Boulder, South Pole, Hilo, and American Samoa are collocated
(within 30 km) with the ozonesonde launch site and taken on the same day as
the ozonesonde profile measurement (Evans et al., 2017). Figures 9, 10, 11,
and 12 show the percent difference (ozonesonde
To gain further knowledge of the accuracy of the shape of the ozone profile,
the ozonesonde data were compared to the SBUV satellite record. The SBUV
satellite record of both total column and stratospheric profile measurements
covers the major portion of the ozonesonde record reprocessed in this work
beginning in 1970. The merged SBUV version 8.6 column ozone record has been
shown to have a consistent time series with offsets not exceeding
At Boulder (Fig. 13), Hilo (Fig. 14), and American Samoa (Fig. 15) with data
prior to 1997 the largest change between the reprocessed and uncorrected
data results from the correction for the sensing solution buffer (1 % KI,
1.0
A thorough homogenization process has improved NOAA's ECC ozonesonde data record in multiple ways. Having all data files in a common file format with all metadata accurately represented and creating the SkySonde Software Package has made the data record more manageable by allowing for fast batch reprocessing of all ozonesonde files. If a better understanding of the less well-quantified variables is realized, NOAA will be well prepared to implement the improved processing techniques. The enhanced plotting capabilities have improved the understanding of the fine details and issues seen in ozonesonde profile measurements, allowing for efficient screening of individual profiles. The reprocessing and homogenization of NOAA's long-term vertical ozone profile record measured by the ECC ozonesondes has greatly improved the agreement of the different ozonesonde types and the different sensing solution types for the five eras shown in this work. The comparison of the ozonesonde data record with the SBUV satellite data record improved in both the total column and pressure layer comparisons. For the first time, a bottom-up, unique, line-by-line uncertainty calculation that accounts for all variables and used in calculating ozone partial pressure has been added to every flight. It is encouraging that the independently calculated uncertainty in total column (4.4, 4.2, 4.1, and 4.2 % – from Sect. 5) is very similar to the standard deviation of the comparison with the Dobson (4.8, 5.5, 4.6, and 5.0 % – from Sect. 6). These uncertainties agree with the total column uncertainties determined for the entire reprocessed SHADOZ data set, which includes our three tropical stations plus 11 additional sites (Thompson et al., 2017; Witte et al., 2017b). Although the uncertainty does not fully capture the Dobson comparison standard deviation, it should be noted that no filtering (besides in the cases of known instrument failures) was conducted on the NOAA ozonesonde record. This allowed for an unbiased look at the processing of the ozonesonde data. The NOAA ozonesonde group is working on developing a screening method that would exclude ozonesonde measurements that do not meet a specific criteria. This will greatly improve the deviation observed in the comparisons. It should also be noted that corrections in this work were not based on comparisons to other long-term ozone data records. This ensures that the ozonesonde data record is independent and non-circular.
This information should make a more robust trend analysis possible narrowing the uncertainties in estimates of long-term changes. There are still questions to be answered, however. The ozonesonde community would benefit from additional published pump efficiency measurements for all ozonesonde types, a deeper look into the cause of the background current, and a continued consistent comparison of ozonesonde type biases. A JOSIE campaign at the WCCOS took place in October and November 2017. JOSIE-2017 focused on comparing ozonesonde profiles with the standard reference UV photometer under several types of tropical profile simulations. This will improve the understanding of the ozonesonde's ability to measure the very low ozone values found in the tropical troposphere and the impact of background cell current.
The NOAA homogenized ozonesonde data archive
is available at
The supplement related to this article is available online at:
The authors declare that they have no conflict of interest.
This article is part of the special issue “Quadrennial Ozone Symposium 2016 – Status and trends of atmospheric ozone (ACP/AMT inter-journal SI)”. It is a result of the Quadrennial Ozone Symposium 2016, Edinburgh, United Kingdom, 4–9 September 2016.
The authors of this work would like to acknowledge Robert Evans, Glen McConville, Dorothy Quincy, Miyagawa Koji, Audra McClure, Irina Petropavlovskikh, and the entire Dobson group at the NOAA Global Monitoring group for their work on the Dobson data that were used to compare ozonesonde and Dobson total column ozone amounts, as well as Richard McPeters, Gordon Labow, and Stacey Frith at the Goddard Atmospheric Chemistry and Dynamics Laboratory for their work on the SBUV overpass data set that was used to compare ozonesonde and satellite vertical ozone profile measurements. Edited by: Mark Weber Reviewed by: three anonymous referees