Technical Note: A new global database of trace gases and aerosols from multiple sources of high vertical resolution measurements

A new database of trace gases and aerosols with global coverage, derived from high vertical resolution profile measurements, has been assembled as a collection of binary data files; hereafter referred to as the "Binary DataBase of Profiles" (BDBP). Version 1.0 of the BDBP, described here, includes measurements from different satellite- (HALOE, POAM II and III, SAGE I and II) and ground-based measurement systems (ozonesondes). In addition to the primary product of ozone, secondary measurements of other trace gases, aerosol extinction, and temperature are included. All data are subjected to very strict quality control and for every measurement a percentage error on the measurement is included. To facilitate analyses, each measurement is added to 3 different instances (3 different grids) of the database where measurements are indexed by: (1) geographic latitude, longitude, altitude (in 1 km steps) and time, (2) geographic latitude, longitude, pressure (at levels ~1 km apart) and time, (3) equivalent latitude, potential temperature (8 levels from 300 K to 650 K) and time. In contrast to existing zonal mean databases, by including a wider range of measurement sources (both satellite and ozonesondes), the BDBP is sufficiently dense to permit calculation of changes in ozone by latitude, longitude and altitude. In addition, by including other trace gases such as water vapour, this database can be used for comprehensive radiative transfer calculations. By providing the original measurements rather than derived monthly means, the BDBP is applicable to a wider range of applications than databases containing only monthly mean data. Monthly mean zonal mean ozone concentrations calculated from the BDBP are compared with the database of Randel and Wu, which has been used in many earlier analyses. As opposed to that database which is generated from regression model fits, the BDBP uses the original (quality controlled) measurements with no smoothing applied in any way and as a result displays higher natural variability.


Introduction
Ozone is a greenhouse gas and as such past and future changes in ozone drive changes in radiative forcing of the climate system. To incorporate these changes in radiative forcing by ozone, global climate models require ozone boundary conditions that span the atmosphere from the surface to the lower mesosphere (0-70 km), from pole to pole, and at high vertical resolution, e.g. to resolve changes in ozone close to the tropopause where the effect on radiative forcing is largest (Forster and Shine, 1997). Furthermore, if zonal asymmetries in the ozone changes can be included, this results in a more accurate representation of changes in radiative forcing.
A vertically resolved ozone database with sufficient density to detect changes in 10 ozone as a function of latitude, longitude and altitude, is valuable for attributing past changes in ozone e.g. zonal asymmetry in ozone trends may be indicative of the influence of changes in dynamics. Such a database is also valuable for the evaluation of chemistry-climate models and in particular for validating the ability of these models to reproduce the latitude-altitude structure in past ozone changes. 15 There are two commonly used vertical ozone profile databases currently available (Fortuin and Kelder, 1998;Randel and Wu, 2007). However, both report monthly mean zonal mean data only, with no longitudinal resolution. The BDBP is not a zonal mean database and individual measurement sets are archived. A measurement set is a group of measurements made at the same date, time, latitude, longitude, altitude and 20 from the same instrument. As a result, the BDBP can be used to extract ozone profiles for a specified location.
In Randel and Wu (2007) no changes in tropospheric ozone are reported and ozonesondes are used only from Syowa and Resolute. These ozonesonde measurements are used to infer changes in ozone poleward of 60 • latitude. Because Syowa 25 (69 • S) is close to the Antarctic vortex edge, meridional movements of the vortex result in Syowa ozonesondes sampling air from both inside and outside the vortex and therefore decreases in ozone over the Antarctic are likely underestimated. 7659 a considerably longer time period for this database could be achieved. In addition, the high vertical resolution of the BDBP allows more detailed analyses of vertical ozone structures compared to the 19 pressure levels in the Fortuin and Kelder database.
The BDBP has been implemented in a flexible and extensible data file format structured for rapid extraction of data. Three different instances of the database have been 10 created where measurements are indexed by: (1) geographic latitude, longitude, altitude (in 1 km steps) and time, (2) geographic latitude, longitude, pressure (at levels ∼1 km apart) and time, (3) equivalent latitude (Butchart and Remsberg, 1986), potential temperature (8 levels from 300 K to 650 K) and time (see Sect. 2). A detailed description of the different data sources from which measurements have been added 15 to Version 1 of the BDBP is given in Sect. 3. By including data from as many sources as possible, dense coverage of the globe, at high temporal resolution, is achieved. The spatial and temporal coverage of the database is quantified in Sect. 4. Monthly mean 2 • zonal mean ozone concentrations were extracted from the BDBP and are compared with the data set of Randel and Wu (2007) in Sect. 5. Section 6 then shows some 20 examples from the BDBP of O 3 , NO 2 and H 2 O for one (or more) specified level(s) and a defined latitude region. Finally, in Sect. 6, the advantages of the BDBP over other existing databases are highlighted and suggestions are made for possible applications of the BDBP. In all cases one of the dimensions is time. The other two dimensions are: in Grid I, geographical latitude and altitude, in Grid II, geographical latitude and pressure, and in Grid III, equivalent latitude and potential temperature. The data are stored within the grids as "measurement sets". A measurement set is a collection of measurements made at the same date and time, latitude, longitude, and altitude, from the same instru-5 ment, which are stored together with the source of the measurement set (e.g. "SAGE2 V6.2"). Each grid contains the same source data but gridded in three different ways to provide different meridional slices of the database. For example, in Grid III, binning the data by equivalent latitude and potential temperature preserves the steep meridional gradients in any zonal means calculated from the database, e.g. close to the vortex 10 edge. In the other two grids these would be smeared out as a consequence of averaging data inside and outside the vortex on lines of constant latitude. For each grid, the data are stored in 90 files, each of which span 2 • in geographic latitude or, for Grid III, 2 • of equivalent latitude, for convenience. Grid I has 70 altitude levels extending from 1 km to 70 km in 1 km steps, and the measurements have been interpolated to these levels 15 (as discussed further below). Grid II has 70 pressure levels spaced approximately 1 km apart given by: Interactive Discussion on a given isentropic level is calculated by taking the PV at the measurement latitude and longitude, and, using the meridional profile of PV vs. equivalent latitude at the nearest 6 h mark, the equivalent latitude is linearly interpolated using the PV value. Grid III in version 1.0 of the BDBP has no data before 1978 since equivalent latitudes before 1978 were not available. 5 Each measurement comprises a value, an error in percent, and a data descriptor (e.g. "Ozone"). The length of each measurement set varies according to the number of measurements available. In this way, the grids are kept compact since no null values need to be stored. If the source data are in the form of vertical profiles, values are interpolated onto the pre-defined vertical levels with the result that the profile is no 10 longer kept as a single entity within the grid. Each measurement set has its own unique time stamp and therefore the measurement sets at a given altitude and latitude bin need not be equally spaced in time (see Fig. 1).
The two primary target variables for this version of the BDBP are ozone and temperature and the data sources used have been selected to optimize the spatial and 15 temporal coverage for these variables. Where other coincident measurements (e.g. NO 2 or H 2 O) are available from the data sources, these have been added to the grids. The variables included in the BDBP, together with their data sources, are listed in Table 1. 20 Criteria for the selection of source data for this version of the BDBP were:

Data sources
1. Only profile data are considered and only profiles with high vertical resolution (i.e. better than ∼1.5 km). Two data sources fulfilling these criteria are solar occultation satellite-based instruments and ozonesondes. Lidar and aircraft profile measurements would also have been suitable but were not included in this version of the 3.1 SAGE I and II

SAGE I and II instrument and data information
Both Stratospheric Aerosol and Gas Experiment (SAGE) instruments were built and launched by NASA (see for example McCormick et al., 1989;Cunnold et al., 1989;Nazaryan and McCormick, 2005;Liu et al., 2006  Interactive Discussion included in the data files were not retrieved parameters, but were provided by the National Meteorological Center (NMC). Nevertheless they were used to determine the levels at which the data were inserted into Grid II and Grid III, and were also included as measurements within each measurement set. Since both SAGE I and SAGE II measure trace gas profiles with the solar occulta-5 tion method, an additional identifier specifying whether the measurement was made at sunrise or sunset was included in each SAGE measurement set.

Altitude correction for SAGE I data
There is a known altitude error in SAGE I observations (Veiga et al., 1995). Wang et al. (1996) discuss and analyse this error in detail. To correct for this error in the SAGE I 10 data, an altitude correction based on Fig. 3 of Wang et al. (1996) was performed. Altitudes for every profile measured between 60 • N and 60 • S were shifted upwards by an offset dependent on latitude but independent on altitude i.e. the measured profiles are shifted rigidly upwards. Although Wang et al. (1996) only discuss the altitude error for ozone profiles, the correction was applied to all measurements. Pressure and temper- 15 ature values were assumed to be correct since they were provided from NMC directly (Wang et al., 1996) and therefore the pressure profiles were not shifted.

Screening of SAGE I data
There are few publications, if any, that describe how best to screen SAGE I data to remove outliers. Therefore, in this analysis, data quality controls, similar to those per-20 formed for the SAGE II data (described below), were applied to remove outliers from the O 3 and NO 2 data. Particularly below 15 km, measurements of O 3 and NO 2 can be affected by aerosols. Therefore, measurements of O 3 and NO 2 were removed if the aerosol extinction at 1000 nm was higher than 0.001/km (L.W. Thomason are only available in an unscreened version (although they are provided with quality flags for unreliable data) additional treatment of the data was necessary to ensure the highest possible quality. The screening was performed following the suggestions of Wang et al. (1996) and Rind et al. (2005). Data points or whole profiles were removed if the following checks were true:  For NO 2 specifically: -If the 1020 nm aerosol absorption is greater than 7×10 −4 km −1 the measurement is excluded. For H 2 O specifically: -If the relative humidity exceeds 100% or is less than 0% the measurement is excluded. -If clouds are present anywhere between 6 and 25.5 km (as denoted by the cloud identifier flags) the water vapour measurement at the altitude of the flagged clouds 5 is omitted. -Above the tropopause, if the aerosol absorption at 1020 nm exceeds 4×10 −4 km −1 the water vapour measurement is excluded. -If the optical depth at 1020 nm is large (as denoted by the quality flags) the measurement is excluded.

10
-If the H 2 O slow convergence flag is set, the measurement is excluded.
For all aerosol extinction measurements: -If the "T-C inversion routine failure for the screened aerosol extinction retrieval" is set, the measurement is excluded. -If the "cloud test not successful" flag is set, the measurement is excluded. The climatology and statistics needed to perform the last test were calculated separately for each of the 3 species as area weighted means from the unscreened SAGE II data.

HALOE
The Halogen Occultation Experiment ( Table 3). Although the vertical resolution of the profiles is ∼1.6 km, the data were included in the BDBP since these measurements provide good global coverage and have been used in numerous previous studies (McKenna et al., 2002;Steil et al., 2003;Remsberg and Deaver, 2005). For every measurement a measurement error is also provided in the source data files and these are added to the BDBP. A sunrise/sunset identifier is also included in each 15 measurement set. HALOE data are already screened for cirrus cloud contamination, described in Hervig and McHugh (1999), and therefore no additional screening of the data was done. A detailed analysis of the HALOE ozone data quality (Version 18) is presented in Bhatt et al. (1999). For a more detailed HALOE data description see Russell et al. (1993).  were made between 55 • and 71 • N and between 63 • and 88 • S with a vertical resolution of ∼1 km. The profiles of the different measured species cover an altitude range from 10 km to 50 km (see Table 3). For a more detailed description of the POAM II instrument and the retrievals see Glaccum et al. (1996) and Lumpe et al. (1997), respectively. POAM III was launched after the satellite on which POAM II was located failed. It 5 started measurements in April 1998 and ended in November 2005. Version 4 is the latest available dataset for POAM III. In this version, quality flags for the profiles of O 3 , H 2 O and NO 2 were included to allow screening of lower quality measurements resulting from sunspot activity and aerosol artifacts (Lumpe et al., 2006). Beside these additions, POAM III data sets contain the same species as POAM II, with the measuring channels 10 for the aerosol extinctions slightly shifted (see Table 1). The vertical resolution is also ∼1 km (Randall et al., 2003), and the measurements were made in almost identical latitude bands (from 54 • -71 • N and 62 • -88 • S). The altitude range is slightly bigger for POAM III than for POAM II (see Table 3). For a more detailed description of the POAM III instrument and its retrieval algorithms see Lucke et al. (1999) and Lumpe et 15 al. (2002), respectively.

POAM II and III
Since temperature and pressure are not directly measured with the POAM instruments, both variables are taken from reanalyses either from the UK Met Office (UKMO) or the National Centers for Environmental Prediction (NCEP) and are included in the database. For sorting POAM II and POAM III in Grid II (latitude/pressure) and Grid III 20 (equivalent latitude/potential temperature), the pressure and temperature values from NCEP were used. A sunrise/sunset identifier is also included in each measurement set. Error values are available for both POAM II and POAM III from the original data files and could therefore be added to the database. Ozonesondes are balloon-borne instruments that measure in situ ozone with a wetchemical method: ambient air is pumped through an electrolytic cell containing a buffered potassium iodide solution where ozone oxidizes the iodide into iodine. The 5 resultant current within the cell is directly proportional to the ozone concentration in the cell. There are several different ozonesonde types in use globally, with the most common being the electrochemical concentration cell (ECC) (Komhyr, 1969), the Brewer-Mast (BM) bubbler (Brewer and Milford, 1960) and the carbon-iodine (CI) sonde (Komhyr, 1965). All ozonesondes are flown together with a radiosonde to measure 10 pressure, temperature and relative humidity. Many studies analysing the suitability of ozonesonde measurements for long-term ozone trend detection have been published (Tiao et al., 1986;Bodeker et al., 1998;Logan et al., 1999). Although the quality of the ozonesonde data depends on the sonde preparation, the experience of the measuring team and some sources of error 15 particular to each instrument, the measurement uncertainties are generally small and can be quantified. The ozonesonde measurements are also the only data included in this version of the BDBP that provide coverage in the troposphere. The ozonesonde data were subjected to thorough quality checks (described below) before being added to the BDBP. 20 Seven different sources for ozonesonde data were used for the database: In total, profiles from 136 stations were added to the BDBP spanning 82 • N to 90 • S (see Table A1). As long as the sonde type was detailed in the original data file, that sounding was rated as a potential candidate to be added to the database after passing 10 several quality checks which are described below. Soundings from the following sonde types were accepted: Brewer-Mast, Brewer-GDR (Ronnebeck and Sonntag, 1976), ECC, Carbon-Iodine, Indian (Shreedharan, 1968) and Regener. Where the altitude and/or time after launch was not available in the original data file, or when the values were unrealistic (e.g. a time after launch of 3 h for the first data level), these were 15 (re)calculated from the pressure and temperature measurements assuming a mean ascent rate of 6 m/s. Normalization factors (NFs) are calculated by dividing a total column ozone value derived from the ozonesonde ozone profile by an independent total column ozone measurement available either from a coincident ground-based (Dobson or Brewer spec-20 trophotometer) or satellite-based column ozone measurement. For the ozonesonde flights added to the BDBP, new NFs were calculated as follow. First the ozone column from the surface to the top of the ozonesonde flight was calculated using trapezoidal integration. Then the missing ozone between the top of the flight and the top of the atmosphere was added using the climatology of McPeters et al. (2007). Added to-25 gether these provide the total column ozone estimate from the ozonesonde flight. The independent total column ozone value was extracted from the NIWA combined total column ozone database (Müller et al., 2007)  whether or not to apply the NF to the ozone measurements. While the NF is stored in the BDBP it was not applied to the ozonesonde data.

ACPD
Since the measurements in the BDBP are stored at specific altitude/pressure/theta levels, the values for those levels extracted from the ozonesonde profiles must be interpolated. For an interpolated value to be added to the database at least one mea-15 surement must be within 200 m of the respective level.
The period for which data are available differs from station to station. Flights started in the early 1960s at a few stations distributed globally. At many stations flights were done just for a few years, while other stations have measurements only during some months or during campaigns, and some stations have a continuous time series of 20 ozone profiles up until the present. In the late 1990s, new ozonesonde stations were chosen to cover regions poorly represented at that time, specifically over the equator and Southern Hemisphere (Thompson et al., 2003). Combining all 7 mentioned ozonesonde data sources, flights from the early 1960s to 2006 were added to the BDBP as long as profiles were available and of suitable quality.

Ozonesonde data errors
The ozonesonde data files obtained through the sources detailed above for the most part do not include the measurement errors associated with each measurement, nei-7671 Introduction Interactive Discussion ther for ozone, temperature, relative humidity nor pressure. To estimate the error on the ozone measurement, information about sonde type quality and measurement errors were obtained from Smit and Kley (1996) unless otherwise specified (see below). According to their suggested classification of sonde type and altitude range different ozone error values were applied (see Fig. 2). In addition:

5
-ECC ozonesondes: error values were taken from Komhyr et al. (1985). For Lauder the error profiles for ECC ozonesondes of the 4A and 5A series described by Bodeker et al. (1998) were applied. -Regener ozonesondes: those sondes were not tested in the analysis of Smit and Kley (1996), but it is known that the errors of those sondes tend to be 10 quite high (WMO, 1989), so relatively high error values, commensurate with other ozonesonde types with high errors, were assumed.
Error values for temperature, pressure and relative humidity for the soundings were set as follows: -Temperature: error values for the sonde temperature measurements are set ac- 15 cording to Bodeker et al. (1998). Different error values for unshaded and shaded temperature sensors are prescribed. For most soundings information about the shading of the temperature sensor is not available (the exception being Lauder) so error values for unshaded sensors are used.
-Pressure: depending on the station, different pressure error values are assumed. 20 Since more detailed information about the soundings at Lauder are available, the pressure error value for that station was calculated from a large data set of calibrated pressure sensors and found to be 0.258 hPa. For all other stations error values were assumed to be ∼1 hPa up to a height of 100 hPa, and ∼0.5 hPa from a measurement height higher than 100 hPa, according to personal communication 25 with Vaisala about pressure errors for radiosonde measurements.
- with Vaisala about humidity errors for radiosonde measurements.
It is not always Vaisala radiosondes that are flown with each ozonesonde at every station and the type of radiosonde used is seldom logged in the original ozonesonde data files. However, since Vaisala radiosondes are the most commonly used sondes, for all sonde data added to the BDBP the error values for pressure and relative humidity 5 are set to the Vaisala radiosonde errors.

Quality check
In addition to the error checks described above, several further checks were made to screen and remove poor quality data from the ozonesonde data files, viz.: -Individual ozone records are rejected when negative ozone partial pressures, 10 0.0 mPa ozone partial pressures within 60 • of the equator, or ozone partial pressure ≥25 mPa at altitudes above 30 hPa are measured.
-Entire ozone profiles are rejected when unrealistically low ozone values over the whole profile (profile mean <2 mPa and profile maximum <4 mPa) are found or when more than 33% of all ozone values in the profile are 0.0 mPa.

15
-Individual temperature records are rejected when values in the troposphere >60 • C or in the case of extreme spikes in stratospheric temperature (for one or two consecutive values). The spikes were defined in two different ways according to the resolution of the checked profile. Profiles with more than 10 values per kilometre were categorised as "high-resolution" profiles. For those the mean 20 temperature and standard deviation for every km-layer was calculated and data values were rejected if they exceeded the range of mean temperature ±30 standard deviations. Temperature spikes for "low-resolution" profiles were defined by exceeding a maximum lapse rate. In the troposphere data values were rejected if the lapse rate was higher than 0.06 K/m, for the stratosphere the maximum lapse rate was lowered slightly to 0.04 K/m. -When two identical records at the same pressure/altitude level (not defined as ascending and descending values) are found, one is rejected.
No relative humidity values above the tropopause (as assumed for the application of the error values -see above) were added to the database since errors in humidity measurements from radiosondes are known to increase with decreasing water vapour con-5 tent, temperature and pressure (Elliott and Gaffen, 1991). Stratospheric humidity data from radiosondes are therefore thought to be of no big use (SPARC, 2000). Although Miloshevich et al. (2001) worked out a correction for relative humidity measurements for Vaisala RS80-A radiosondes, no further quality improvements were performed to the sonde humidity data in the database since in most cases the radiosonde type was 10 not known.

Database temporal, latitudinal and longitudinal coverage
By combining measurements from several satellite-based instruments and from ozonesondes it is possible to achieve high temporal and spatial coverage in the BDBP. Table 2 summarizes the temporal coverage of the satellites and these, together with 15 the ozonesondes, are shown graphically in Fig. 3. Note that although the ozonesonde data cover a long time period, the spatial coverage can be poor due to the number and location of the ozonesonde stations. In this version of the BDBP, for 2006, only ozonesonde data are available, and so the spatial coverage for 2006 is poorer than for the preceding years. 20 To quantify the spatial and temporal coverage of the BDBP for a given altitude-/pressure/isentropic level, we have defined factors (hereafter referred to as B-factors) that combine the temporal and spatial coverage into one value. For a given level, a spatial grid is selected (e.g. 2 • latitude by 5 • longitude) and monthly means within each grid cell are calculated (described further in Sect. 5) over some selected time period. Introduction

Back Close
Full Screen / Esc

Printer-friendly Version
Interactive Discussion least N% of the months in the selected time period. The B-Factor does not indicate which part of the globe is not sampled nor which periods within the total period are not covered. Figure 4 shows B-factors for a resolution of 2 • latitude by 5 • longitude for O 3 as a function of changing time period (panel a) and changing data source (panel b) as well as NO 2 and H 2 O for changing time period (panels c and d, respectively).
As mentioned earlier, one selection criteria for the BDBP data sources was the goal of high temporal and spatial coverage for ozone and temperature. It can be seen in Fig. 4 that ozone has the highest B-factors for all three shown species, exceeding 80% between ∼10 km and ∼30 km even at a high resolution of 2 • by 5 • . For O 3 , the periods with a late start year have higher B-factors as a result of sparser coverage early in the 10 period (see Fig. 3). B-factors for temperature are even slightly higher than those for O 3 but are not shown here. Above 60 km coverage of ozone decreases rapidly.
In panel (b) of Fig. 4 the effects of the cumulative addition of different data sources on the temporal and spatial coverage of the BDBP are shown. First the effects of adding SAGE II data are shown, followed, in order, by SAGE I, HALOE, POAM II, POAM III and 15 finally ozonesondes. Even though the addition of POAM III data extends the temporal coverage of the BDBP, the B-factors do not change with the addition of the POAM III data because it does not extend the spatial coverage. Clearly, at a resolution of 2 • by 5 • , the inclusion of the ozonesondes is vital to achieve B-factors above 80% in the stratosphere and any coverage at all in the lower troposphere. 20 NO 2 coverage does not reach the 80% mark although it is close in the altitude range ∼15 km to ∼45 km (see Fig. 4c). Only very sparse data are available at altitude levels above ∼45 km and below ∼15 km.
Of the 3 species shown, H 2 O is the only one with almost constant B-factors throughout the stratosphere and lower mesosphere (Fig. 4d), from ∼15 km up to 70 km. While 25 relative humidity from the ozonesonde flights is added to the BDBP in the troposphere, these measurements are excluded from panel (d) of Fig. 4 Table 3).

Comparison with Randel and Wu database
An often used and well established database for ozone trend analyses and calculations of changes in radiative forcing by ozone is that from Randel and Wu (2007), hereafter referred to as R&W. In this section, monthly means of ozone at selected altitudes and 5 latitudes calculated from the BDBP are compared with R&W. As mentioned in Sect. 1, R&W consists of monthly mean zonal mean ozone and is based mainly on SAGE I and SAGE II data, with ozonesonde profiles from Syowa and Resolute providing high latitude coverage. Figure 5 compares monthly mean times series from the BDBP and R&W at the 10 equator and 25 km altitude. The monthly means for the BDBP and for SAGE I+SAGE II were calculated over the same latitude band as in R&W, but with the requirement that there had to be at least 6 values available at the given latitude and altitude for the monthly mean to be valid. For both the BDBP and SAGE I+SAGE II, of the individual values available for the calculation of the monthly means, the highest and lowest 25% 15 were discarded to ensure that extreme values were not included. This rather simple method of rejecting extreme values is sufficient for this exercise; more sophisticated methods are in development for a 3-D (latitude, altitude, time) monthly mean ozone database that will be created from the BDBP. Even though R&W are based on SAGE I and SAGE II only in this part of the atmo- Interactive Discussion agreement between the BDBP and R&W is excellent. It is also clear that R&W does not capture some of the outliers in the BDBP (e.g. November 1984) because R&W is the regression model fit and not the raw monthly mean data. Figure 6 shows another comparison between the BDBP and R&W where ozone anomalies were calculated by subtracting the mean annual cycle from monthly mean 5 time series, for altitudes between 1 km and 70 km and for the latitude zone from 40 • N to 50 • N.
The mean annual cycles (right hand panels in Fig. 6) compare well. The BDBP covers a greater altitude range than R&W providing data between 50 and 70 km and, more importantly, good coverage in the troposphere. With a latitude band of 10 • the 10 number of missing monthly means calculated using the BDBP is small; the gap in the earlier 1980s above ∼33 km is between SAGE I and SAGE II (ozonesondes provide data at lower altitudes). Anomalies (left hand panels in Fig. 6) of the same sign are found during similar periods and at similar altitudes. However, the BDBP anomalies have greater vertical scale, extending down to ∼10 km altitude whereas those of R&W 15 are suppressed below 20 km. The anomalies calculated using the BDBP show greater temporal variability and larger amplitudes than R&W. This is because R&W is based on regression model output and regression models cannot capture all of the variability. Because the regression model used by R&W does not include a basis function to describe the effects of volcanic eruptions on ozone, negative ozone anomalies in the 20 lower stratosphere related to the Pinatubo eruption in the early 1990s are significantly more apparent in the BDBP than in R&W. There is a discontinuity in the BDBP anomalies at ∼5 km altitude. This is an artifact of the monthly mean calculations using the BDBP where satellite data are not available for the monthly means at one altitude but are at the next higher altitude. Data from all of the added satellite instruments (see 25 Sect. 3) have large error values in the troposphere for all measured species due to the tropospheric aerosol loading. It is suggested in various studies not to use these data for quantitative analyses (see for example Kar et al., 2002). For the purposes of this preliminary comparison with R&W, the approach used to calculate the monthly means

Printer-friendly Version
Interactive Discussion is adequate. However, for a more accurate monthly mean database, weighting of the measurements used in each mean with their errors should be incorporated.

Summary and Outlook
The material presented above outlines the construction of a new global database of trace gases and aerosols from multiple sources of high vertical resolution measure-5 ments. The first version of this database, referred to as BDBP version 1.0, includes measurements from several solar occultation satellite instruments (SAGE I and II, POAM II and III, HALOE) and from ozonesonde flights from over 130 stations globally, covering the period 1962 to 2006. It is planned to update the database annually to include the newest ozonesonde data, add historical data sources that have not yet 10 been included (e.g. ILAS and GOMOS, and more ground-based measurements, for example lidar data), add measurements from new satellite-based instruments currently in development, and possibly newer versions of the data already in the database (e.g. from the application of improved satellite retrieval algorithms). The internal construction of the database allows for considerable flexibility:

15
-Where measurements are available at high temporal resolution, e.g. during an intensive ozonesonde campaign, the resolution of the original data is maintained. -Measurement sets can include any number of individual measurements taken by the same instrument at that latitude, longitude, altitude and time and this can vary between data sources or within a data source.

20
-The database has a single common format so that utilities to extract subsets of data from the BDBP are easy to write and quick to execute. -The database consist of three different grids but the structure of the measurement sets is the same for all three grids which simplifies data handling. This flexibility allows for multiple applications: -Because temporal means are not calculated, trace gas profiles at a specific location and time, if available, can be extracted for studies requiring point source data.

10
-Case studies with data from only one data source, or a combination of data sources, can be made.
-Since a longitude value is stored with each measurement set, analyses requiring longitudinal disaggregation can be undertaken.
-The temporal and spatial coverage of the BDBP is sufficient to provide data for 15 statistically significant trend analyses.
-It is possible to bin the data from the BDBP in several different spatial and temporal resolutions, as required for analyses (e.g. monthly means, seasonal means, yearly means, etc.), since data in the BDBP are stored with the information of the exact measurement time, latitude and longitude.

20
-The combination of data sources in the BDBP provides a robust basis for climatology calculations which can be used to validate chemistry-climate models.
-Since the BDBP covers a long time period, climatologies calculated from the BDBP can be used to provide boundary conditions for model simulations. Examples of some of the products available through the BDBP are shown in Fig. 7. The evolution of the ozone hole every spring is clearly visible and the values from the 6 different data sources show no significant offsets or drifts in comparison to each other. The NO 2 time series plotted in panel (b) of Fig. 7 shows decreases in NO 2 in 1991 and 1992 resulting from the Mt. Pinatubo eruption. The H 2 O time series plotted in panel 5 (c) of Fig. 7 shows good agreement between the HALOE and SAGE II data sources and the derived monthly means are in good agreement with Fig. 3a of Rosenlof et al. (2001). In contrast to the high water vapour values seen in the 1991-1994 SAGE II data in Fig. 3a of Rosenlof et al. (2001), the SAGE II data screening implemented here removes most of the H 2 O data points in this period where measurements were strongly 10 affected by the eruption of Mt. Pinatubo.

ACPD
This version of the BDBP does not consider problems of inhomogeneities between the different data sources. While comprehensive screening was applied to the different data sources, systematic offsets and drifts between the data sources are likely. Applications using the BDBP, e.g. the calculation of a monthly mean 3-D ozone database, 15 will need to consider removal of these offsets and drifts.
An advantage of combining measurements from multiple sources is the improved temporal and spatial coverage achieved. It was shown that only with the combination of the different data sources a high index of coverage (B-factor) could be achieved. Especially the ozonesonde data contribute considerably to the coverage of the tropo-20 sphere and the lower stratosphere.