Intercomparison of ground-based microwave remote sensing measurements of stratospheric ozone over the Mendoza region, Argentina with HALOE data

Remote Sensing of Environment 94 (2005) 61 – 82 www.elsevier.com/locate/rse

Intercomparison of ground-based microwave remote sensing measurements of stratospheric ozone over the Mendoza region, Argentina with HALOE data Carlos M. Puliafito*,1, S. Enrique Puliafito1,2 Universidad de Mendoza, Arı´stides Villanueva 773, 5500 Mendoza, Argentina Received 3 March 2004; received in revised form 8 September 2004; accepted 8 September 2004

Abstract The Tropospheric Water Vapour and Stratospheric Ozone (TROPWA) project has measured ground-based stratospheric ozone by means of millimetre wave radiometry tuned at 142 GHz from 1993 to 2000 in Mendoza, Argentina. Additionally, tropospheric water vapour was measured using a 92-GHz radiometer. This paper presents the theoretical error analysis used to characterize the ozone instrument, and a comparative study of the retrieved profiles with the coincident measurements taken with different instruments. To evaluate and validate the retrieved stratospheric ozone profiles, we have used a set of ozone profiles measured with the Halogen Occultation Experiment (HALOE); while the water vapour data was calibrated against a set of 3-year-radiosounding-balloon data taken by the Argentine National Weather Service. This study also includes a comparison of individual ozone profiles measured using a second ground-based millimetre wave radiometer–spectrometer tuned at 276 GHz from the Max-Planck-Institut fqr Aeronomie (MPAE), Germany. During this particular campaign carried out in November 1994, the ground-based measurements were contrasted with two space-born experiments: the Millimetre Wave Atmospheric Sounder (MAS), flown in the NASA-ATLAS 3 mission and the above-mentioned HALOE. From the error analysis and the comparison tests, it follows that between 20 to 40 km the TROPWA instrument is able to retrieve ozone profiles with absolute errors varying from 10% to 20%, relative errors less than 5%, and with a height resolution, calculated as full width at half maximum (FWHM), varying from 5 to 11 km depending on the altitude. The major discrepancies between the different set of profiles are about +8% to
10% (+0.4 to
0.8 ppmv), mainly due to the coarser height resolution of our instrument. D 2004 Elsevier Inc. All rights reserved. Keywords: Atmosphere; Stratospheric ozone; Ground-based; Radiometry; Comparison; Validation

1. Introduction The Institute for Environmental Studies (IEMA), University of Mendoza, performed ground-based measurements of tropospheric water vapour and stratospheric ozone by means of millimetre wave radiometry from 1993 to 2000, according to Table 1 and Fig. 1, from Benegas Station (850 m.a.s.l.), 5 km south of the city of Mendoza, Argentina. Additionally, five other measuring campaigns were per* Corresponding author. Tel.: +54 261 4202017. E-mail address: [email protected] (C.M. Puliafito). 1 Also members of CONICET Argentina. 2 Now at Universidad Tecnolo´gica Nacional. 0034-4257/$ - see front matter D 2004 Elsevier Inc. All rights reserved. doi:10.1016/j.rse.2004.09.007

formed in high locations, i.e., Uspallata (1950 m.a.s.l.), Puente del Inca (2700 m.a.s.l.) and Cristo Redentor (4200 m.a.s.l.)—near Puente del Inca—in the eastern slope of the Argentinean Andes and about 50 to 100 km northwest from the city of Mendoza. Mendoza (338S, 688W, 750 m.a.s.l.) is located in the western semi-arid region of Argentina at the east side of the Andes Range. It has low rainfall of 120–400 mm/year, that is a mean of 230 mm/year which occurs especially during the summer months (November to March). The annual mean wind intensity is 2.6 m/s, with 19% of quiet days. The Andes Range, near the city of Mendoza, reaches an average height of 5000 m with peaks of up to 7000 m. Due to its closeness to the mountains, Zonda winds (from 5–6 to 12 m/s)—

62

C.M. Puliafito, S.E. Puliafito / Remote Sensing of Environment 94 (2005) 61–82

Table 1 Measurement periods and sites HALOE

TROPWA

Profile #

Date

Lat

Long

Profile #

Date

Site

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61

30-oct-93 20-nov-93 21-nov-93 23-mar-94 12-apr-94 14-may-94 23-jun-94 14-jul-94 15-jul-94 16-jul-94 13-aug-94 28-sep-94 26-oct-94 16-nov-94 18-mar-95 19-mar-95 8-apr-95 7-may-95 30-may-95 31-may-95 10-jul-95 18-jul-95 30-aug-95 24-sep-95 27-sep-95 22-oct-95 10-nov-95 11-nov-95 12-mar-96 13-mar-96 14-mar-96 2-apr-96 3-apr-96 30-apr-96 1-may-96 2-may-96 24-may-96 25-may-96 14-jun-96 2-jul-96 3-jul-96 4-jul-96 5-jul-96 12-jul-96 13-jul-96 3-aug-96 4-aug-96 24-aug-96 25-aug-96 26-aug-96 19-sep-96 20-sep-96 5-nov-96 30-jun-97 1-jul-97 14-sep-97 15-sep-97 16-sep-97 17-sep-97 13-oct-97 18-apr-98


34.67
32.31
36.24
32.01
33.08
41.00
35.08
36.46
33.62
36.46
38.68
31.23
32.78
34.13
33.93
29.30
35.01
35.18
35.17
30.77
34.60
32.31
34.11
32.94
33.24
31.05
27.61
31.94
35.73
31.36
26.46
32.83
37.11
30.94
34.91
38.32
37.06
33.18
38.02
40.24
38.11
35.63
32.78
29.56
33.16
38.83
35.19
32.32
36.57
40.17
35.10
40.43
29.82
34.04
30.95
29.32
37.38
33.12
25.45
25.71
34.41


70.36
64.63
67.94
65.04
79.53
68.40
72.31
68.20
71.05
68.20
59.05
76.70
75.86
68.74
66.82
68.29
60.71
65.93
68.03
66.69
75.89
70.92
71.41
61.77
60.39
61.65
70.59
73.56
66.94
68.60
70.22
63.17
63.47
62.07
64.94
67.76
68.28
67.20
77.66
68.13
70.73
73.46
76.32
68.33
71.26
66.58
65.66
67.38
67.57
67.86
59.99
77.24
67.03
65.85
68.77
73.36
75.09
66.34
68.36
74.30
60.91

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

21-nov-93 22-nov-93 23-nov-93 24-nov-93 25-nov-93 26-nov-93 15-dec-93 25-apr-94 1-jul-94 3-aug-94 4-aug-94 5-aug-94 8-nov-94 9-nov-94 10-nov-94 11-nov-94 10-apr-95 12-jun-95 13-jun-95 14-jun-95 15-jun-95 16-jun-95 22-sep-95 26-sep-95 18-oct-95 19-oct-95 20-oct-95 19-mar-96 11-jun-96 23-aug-96 1-oct-96 13-aug-97 15-sep-97 16-sep-97 17-sep-97 8-oct-97 9-oct-97 10-oct-97 5-may-98 7-may-98 8-may-98 11-may-98 12-may-98 29-jul-98 31-jul-98 3-aug-98 4-aug-98 5-aug-98 26-aug-98 28-aug-98 31-aug-98 10-sep-98 28-oct-98 29-oct-98 5-may-99 9-jun-99 7-jul-99 26-jul-99 28-mar-00 29-mar-00

Puente del Puente del Puente del Puente del Puente del Puente del Benegas Benegas Benegas Benegas Benegas Benegas Puente del Puente del Puente del Puente del Benegas Uspallata Uspallata Uspallata Uspallata Uspallata Uspallata Uspallata Benegas Benegas Benegas Benegas Benegas Benegas Benegas Benegas Benegas Benegas Benegas Benegas Benegas Benegas Benegas Benegas Benegas Benegas Benegas Benegas Benegas Benegas Benegas Benegas Benegas Benegas Benegas Benegas Benegas Benegas Benegas Benegas Benegas Benegas Benegas Benegas

Inca Inca Inca Inca Inca Inca

Inca Inca Inca Inca

C.M. Puliafito, S.E. Puliafito / Remote Sensing of Environment 94 (2005) 61–82

63

Table 1 (continued) HALOE

TROPWA

Profile #

Date

Lat

Long

62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83

19-apr-98 21-apr-98 22-apr-98 14-may-98 15-may-98 16-may-98 25-jul-98 26-jul-98 10-sep-98 11-sep-98 19-jun-99 20-jun-99 21-jun-99 29-jun-99 30-jun-99 1-jul-99 10-aug-99 12-aug-99 22-feb-00 8-apr-00 7-apr-00 10-apr-00


28.59
35.25
38.97
39.83
36.57
32.50
38.59
35.01
38.37
32.33
38.33
35.86
33.03
30.74
33.94
36.75
33.41
40.68
31.36
34.08
40.20
39.05


63.44
61.56
64.17
66.14
65.48
64.57
62.97
61.77
63.88
66.13
66.48
68.98
71.62
65.14
67.85
70.44
72.66
71.71
60.97
68.07
66.26
62.85

similar to Ffhn or Chinook winds—prevail in the highest layers most of the year. The Tropospheric Water Vapour and Stratospheric Ozone (TROPWA) Project has at present 3 mm wave radiometers working at different frequencies: 21.8, 22.2, 31.5, 92, and 142 GHz for water vapour, water vapour continuum and ozone, respectively. From these data, it is possible to retrieve stratospheric ozone profiles from approximately 20 to 40 km altitude with an altitude resolution ranging from 5 to 11 km, depending on the altitude and with a minimal temporal resolution of 1 h. Additionally to our radiometric measurements, we have used a data set of ozone profiles from the Halogen Occultation Experiment (HALOE) onboard the Upper Atmosphere Research Satellite (UARS) for comparison purposes. The HALOE Instrument uses a solar occultation technique and has been taking data almost continuously since 1991. It measures vertical profiles of O3, HCl, HF, CH4, H2O, NO, NO2, aerosol extinction and temperature versus pressure. By using solar occultation, HALOE

Profile #

Date

Site

measures the attenuation of the sun’s radiation through all the average middle atmosphere altitudes and views the Earth as the sun rises and sets relative to the satellite. Using different radiometers, the sunlight is measured in several infrared wavebands between 2.45 and 10.0 Am. Ozone is measured in the wavelength range between 9.2 and 10.4 Am. The satellite orbit determines the time and location of each profile. While orbiting, HALOE views as an average, 15 sunrises and 15 sunsets every 24 h. During orbiting, HALOE scans from 80 S to 80 N in latitude approximately every 30 days. A complete description of this instrument and operation is given by Russell et al. (1993). The HALOE error analysis study is presented by Bru¨hl et al. (1996). In the present paper, we also compare profiles from the Millimetre Wave Atmospheric Sounder (MAS) which flew from November 2 to 4, 1994, on the ATLAS III Space Shuttle Missions. The ATLAS-MAS instrument uses a limb sounding technique in atmospheric emission. The ozone profiles have been obtained from radiometric measurements at 184.7 GHz at high spectral resolution (200 kHz), allowing

Fig. 1. Selected data from TROPWA (black squares) and HALOE (gray circles) from October 1993 through March 2000. Sixty daily profiles were obtained from TROPWA and 83 profiles were selected from HALOE, according to Table 1.

64

C.M. Puliafito, S.E. Puliafito / Remote Sensing of Environment 94 (2005) 61–82

retrievals of ozone in the upper part of the middle atmosphere. The instrument description is given by Croskey et al. (1992), and the ozone measurements are presented by Hartmann et al. (1996) and Bevilacqua et al. (1996). Furthermore, during this same measurement period (November 1994) another ground-based radiometer–spectrometer from the former Max-Planck-Institut fqr Aeronomie (MPAE), Germany, measured ozone at 276.9 GHz together with the TROPWA experiment in Puente del Inca. The 276-GHz MPAE radiometer can also measure ClO at 278.6 GHz and N2O at 276.3 GHz. This instrument has a spectrometer with 15 filters ranging from 100 to 2 MHz bandwidth. A detailed technical description of this instrument has been given in Loidl and Ro¨mer (1990).

2. Materials and methods 2.1. Instrumentation Remote sensing of ozone at millimetre wavelength is an adequate technique to measure vertical ozone distribution at the stratosphere and the mesosphere. The main advantage over other techniques is given by the fact that measurements can be taken continuously day and night because the technique does not require any external source of radiation. In addition, it can measure under nearly all weather conditions. Ozone profiles are difficult to obtain only if the tropospheric water vapour content is large or in the presence of rain. A radiometer is a heterodyne radio frequency receiver, which is tuned at the frequency of a rotational (or vibrational) transition of a trace gas. The molecular structure of these constituents results in frequency-dependent absorption and emission characteristics that can be used to uniquely identify each type. The rotational lines of ozone and water vapour are well separated from one another in the millimetre wave range. The 142-GHz spectral line is centred in an atmospheric window of high transparency limited by the strong absorption lines of oxygen at 118 GHz and water vapour at 183 GHz. Thus, the broad wings of the oxygen and water vapour spectral lines do not affect the measurement. The radiation emitted from different altitudes suffers a pressure broadening proportional to the pressure in that altitude. Thus, by measuring the spectral line width of the radiation received at ground level it is possible to derive the altitude from where the radiation is coming from. If the radiation is emitted from more than one altitude, you will observe basically the sum of all the contributions from different altitudes, the task of the inversion is to decompose this sum into the single contributions. These effects are considered in the radiative transfer equation, e.g., Chandrasekhar (1960). The brightness temperature measurement is then very sensitive to the ozone concentration at different altitudes. Finally, if we know the temperature and pressure profile of the atmosphere, it will be possible to retrieve the

vertical ozone distribution through mathematical inverse algorithms. The measured radiation for an upward looking groundbased radiometer expressed in terms of brightness temperature is given by the radiative transfer equation being S B the ground level and S A the farthest atmospheric upper layer: R sB
að f; sÞds TB ð f Þ ¼ TA ð f Þe sA R sB Z sB
að f; sVÞdsV þ T ðsÞyð f ; sÞað f ; sÞe sA ds ð1Þ sA x where yð f ; sÞ ¼ ex
1 , and x=hf/kT,a( f,s)=a(T(s), P(s), VMR, f, spectral parameters): absorption coefficient [dB/ km], P(s): pressure [mbar] , T(s): absolute temperature [K], VMR: volume mixing ratio [ppm: parts per million], f: frequency [Hz], s: position along the line of sight [m]. If the temperature and the pressure profile of the atmosphere are known, it will be possible to determine the volume-mixing ratio of the trace gas from the measured brightness temperature as a function of the frequency (T B( f)). This procedure is known as brightness temperature inversion. In order to convert the output of the instrument (pure voltages or counts) into brightness temperature, the instrument is calibrated against known brightness temperatures, i.e., a hot load (T BH=292 K) and a cold load (T BC=74.64 K). This calibration process is also used to reduce errors introduced by gain drifts in the amplifier’s chain within the instrument. Each measuring cycle is then divided into three steps: hot load–atmosphere–cold load. The radiometric sensitivity or radiometric resolution DT is defined as the smallest change in the received T B that can be detected by the radiometric output and can be written as:

TSYS DT ¼ pffiffiffiffiffiffiffiffi Df s

ð2Þ

where, DT is the temperature resolution [K], T SYS is the radiometer noise temperature [K], Df is the radiometer bandwidth [Hz], and s is the integration time [s]. Eq. (2) is completely valid if the bandwidth is sufficiently small compared to the centre frequency, which is our case, e.g., Ulaby et al. (1981). The output signal coming from the 142-GHz ozone total power radiometer is fed into the backend, i.e., a filter bank spectrometer, which analyses the spectral components of the intermediate frequency band (IF). The IF was selected to 3.7 GHz and has a bandwidth of F600 MHz. The filter bank spectrometer has nine channels with different bandwidths, and it was recently extended to 19 channels. The data are integrated and recorded in a personal computer, generally, in files of 10 min long. Hence, the real integration time corresponding to the ozone measurement is 200 s. Table 2 indicates the bandwidth, the offset frequency, the measured noise temperature and the resulting radiometric resolution DT for each channel of the spectrometer.

C.M. Puliafito, S.E. Puliafito / Remote Sensing of Environment 94 (2005) 61–82 Table 2 Filterbank characteristics Channel number

Offset freq. [MHz]

Bandwidth [MHz]

Typical noise temperature at 200-s int. time

DT [K]

0 1 2 3 4 5 6 7 8


350
250
140
60 0 60 140 250 350

100 100 40 40 2 40 40 100 100

14 000 15 800 19 800 20 700 22 400 18 400 19 500 18 300 21 200

0.06 0.06 0.13 0.13 0.20 0.12 0.13 0.08 0.09

Ozone line center frequency at 142.175040 GHz.

Another important error source for the profile retrieval, is the so-called baseline structure, which appears as an impedance mismatch between the antenna and mixer, in the front-end of the radiometer, creating a standing wave of several 100 MHz period. The elliptical mirror located at the radiometer input, operated by a step motor, sets the three antenna positions (hot load–atmosphere–cold load). Additionally, this step motor is moved back and forth in its axis direction reducing the baseline by changing the wavelength configuration between the mirror and the horn antenna. The mirror is moved several times within one measurement period thus, the mean value of the wave structure is highly reduced. Since the brightness temperature spectrum is expected to be symmetric with respect to the line centre, further baseline removal, when necessary, is performed via software, by

65

reducing the slope of the measured spectrum. Normally, the two most external channels are used to calculate this slope, since at their offset frequencies (
350 and +350 MHz), they provide the least ozone information coming mainly from the lowest altitudes. Finally to reduce the uncertainties, we fold the spectrum about the line centre, using only half of the spectrum, so only that part of the reflection wave which is symmetric, will affect the data. These remaining uncertainties are included as measurement errors. Fig. 2 shows a measured and corrected spectrum: The raw data (black line with diamonds) as it is measured from ground level is depicted at the top of the figure. The gray line represents the symmetrized spectrum (for baseline correction), whereas the remaining spectrum at the bottom of the figure shows the stratospheric contribution (dark line with gray circles), after the implementation of the tropospheric correction. An additional description of the 142 GHz radiometer–spectrometer hardware has been presented in Puliafito et al. (1998). When a ground-based measurement of a spectral line, such as ozone, is carried out, the radiometer not only receives the brightness temperature information from the spectral line, but also from a tropospheric continuum contribution. The tropospheric water vapour represents one of the most important constituents considered in the tropospheric contribution due to its high opacity, producing an important attenuation. This attenuation is almost frequency independent for the considered 1.2 GHz bandwidth. Thus, a tropospheric correction to the measurements must be achieved, i.e., by using the information either from the farthest channel from the ozone line centre, or in our project also from the 92-GHz water

Fig. 2. Data processing to obtain the stratospheric ozone spectrum at 142 GHz. At the top of the figure, the raw data (black line with diamonds) is depicted as it was measured from ground on November 21, 1993. The gray line represents the symmetrized spectrum after baseline removal. The spectrum at the bottom of the figure shows the stratospheric contribution (dark line with gray circles), after the tropospheric correction was implemented.

66

C.M. Puliafito, S.E. Puliafito / Remote Sensing of Environment 94 (2005) 61–82

vapour radiometer. For the retrieval process, it is then necessary to calculate the tropospheric contribution. If the atmosphere is considered as a two-layer medium, the troposphere and the stratosphere, the measured brightness temperature, Eq. (1), at ground level can be written as: TB

MES

¼ TB

TR

þ TB

ð3Þ

ST

whereT B_MES: brightness temperature measured at ground level, T B_TR: tropospheric contribution, mainly due to water vapour, T B_ST: stratospheric part, mainly due to ozone TB

MES

¼ Teff ð1
vÞ þ TB

O3 v

ð4Þ

where v is the atmospheric transmission, and Teffi270 K, is the effective or mean atmospheric temperature in the troposphere. To calculate the unknown atmospheric transmission v in Eq. (4), we use the information from the farthest filter channel (
350 MHz). As it has been previously mentioned, this channel has the least ozone information. Therefore, its brightness temperature T B0 is equivalent to the tropospheric brightness temperature. TB

TR

¼ TB0 þ TB

ST0 cTB0

TB0 ¼ Teff ð1
vÞ v¼1
TB

O3

¼

TB0 Teff TB
Teff ð1
vÞ v

ð5Þ

where T B0 is the brightness temperature for channel 0 (
350 MHz), T B_ST0 is the stratospheric ozone contri-

bution for channel 0. This small contribution is calculated using an ozone model profile (a priori). T B are the uncorrected brightness temperatures, T B_O3 is the calculated stratospheric brightness temperature, for any channel. Integrating the water vapour density q(z) (g/cm3) over the troposphere yields to the so-called total water vapour content TWV or integrated water vapour IWV (mm) or precipitable water vapour (see also Peter & Ka¨mpfer, 1992): Z HT IWV ¼ 10
3 qð zÞdz ð6Þ 0

We study the tropospheric transmission, and the effective temperature, using 3 years of daily radiosounding launched by the Argentine National Weather Service at Mendoza Airport (from 1993 to 1995). Using the radiosounding profiles, we calculated climatologic monthly tropospheric profiles of pressure, temperature and water vapour density. Further, we estimated v, Teff, IWV, and the expected brightness temperature which should be measured at 92 GHz and at channel 0 (
350 MHz offset frequency) of the 142 GHz radiometer. Fig. 3 shows the tropospheric transmission (dark dots) and the calculated brightness temperature for 142 GHz using 3 years of daily radiosounding profiles (white circles), as a function of the day of the year. In Fig. 4, it is also shown, the regression line between the brightness temperature for channel 0 of the 142-GHz radiometer and integrated water vapour using the radiosounding profiles. Figs. 5 and 6 show measured 92 and 142 GHz data, respectively, compared with their

Fig. 3. Tropospheric transmission (dark dots) and brightness temperature for 142 GHz calculated using 3 years (1993–1993) of daily radiosounding profiles (white circles), as a function of the day of the year. Radiosoundig profiles were obtained around noon time, approximately on a daily basis, by the Argentine National Weather Service at Mendoza Airport.

C.M. Puliafito, S.E. Puliafito / Remote Sensing of Environment 94 (2005) 61–82

67

Fig. 4. Regression line between the brightness temperature for channel 0 (
350 MHz offset frequency) of the 142-GHz radiometer and integrated water vapour using the radiosounding profiles (same data as in Fig. 3). This regression line is used to calculate the IWV from the radiometric measurement. Similar calculations are done for the 92-GHz data, and to obtain the atmospheric transmission, from either radiometer, the 142 GHz (channel 0) or the 92 GHz.

calculated brightness temperature using radiosounding profiles. The differences between the calculated and measured data for coincident days from 1993 to 1995 give a mean difference of 19% and a standard deviation of

23%. These variations can be explained since radiosounding and radiometric data were not taken at the same position and hence did not necessarily capture the same volume of air.

Fig. 5. Measured 92-GHz data (black circles: 10-min integration time) compared with calculated brightness temperature (gray squares) using radiosounding profiles as a function of the day of the year, between 1993 and 1995 (similar to Fig. 3, but for 92 GHz). The long columns of black circles show particular days of large water vapour variability, due to incoming weather fronts. The difference between measured and calculated data for coincident days and time gives a mean value of 19% with a standard deviation of 23%.

68

C.M. Puliafito, S.E. Puliafito / Remote Sensing of Environment 94 (2005) 61–82

2.2. Data analysis We will briefly describe this mathematical method, which is used to retrieve mixing ratio profiles of a given atmospheric gas. The general remote sensing equation can be written in schematic matrix form, according to Rodgers (1976), as y ¼ F ð x; bÞ þ ey

ð7Þ

where y is the measurement vector, F(x,b) is the forward model, b is a set of parameters used in the forward model (e.g., line strength, collisional broadening and atmospheric temperature) which must be estimated, E y is the measurement error with covariance S e, and x is the profile to be inferred from the measurement. The retrieved profile xˆ can be generally expressed as (Puliafito et al., 1995):     xˆ ¼ R y; bˆ ; c; e ¼ T x; bˆ ; c; e where R represents the retrieval algorithm, bˆ is our best guess of the forward parameters, and c represents any other parameter used in the inverse model, for example a priori profiles, etc. It can also be written as a transfer function T, relating the retrieved profile with the true profile x. The problem of retrieving constituent mixing ratio profiles from microwave spectral line measurements is, in general, nonlinear. Therefore, it is necessary to use nonlinear techniques in these retrieval problems. In such techniques, the weighting functions or kernels are first calculated based on a priori profiles and then iterated in the solution algorithm (Puliafito

et al. (1995)). Eq. (7) can be linearized about the a priori value of x, x a, and the corresponding estimate of y from the forward model, as y ¼ yˆ ðxa Þ þ K ð x
xa Þ

ð8Þ

where yˆ (x a)=F(x a,b). K is the measurement kernel or weighting function defined as K¼

BF ð x; bÞ Bx

ð9Þ

The optimal estimation inversion techniques have been described in Rodgers (1976, 1990):  
1 xˆ ¼ xˆ a þ SX K T KSX K T þ Se ð y
K xˆ a Þ

ð10Þ

where Sx is the covariance matrix of the a priori profile xˆ a and Se is the covariance matrix of the measurement error. Several other retrieval methods have been compared in Puliafito et al. (1995), including the error analysis of each method. 2.3. Error analysis An important question in every retrieval technique is to characterize the uncertainties due to different error sources, and how they affect the retrieved profile. Several ways of qualifying these uncertainties have been proposed. However, in this section we will follow the definitions presented by Rodgers (1990), which we have already used for other cases in the abovementioned paper and will be briefly presented here.

Fig. 6. Measurements done for channel 0 (
350 MHz offset frequency) of the 142-GHz data (black squares: 10-min integration time) compared with calculated brightness temperature (gray circles) using the radiosounding profiles as a function of the day of the year (1993–1995).

C.M. Puliafito, S.E. Puliafito / Remote Sensing of Environment 94 (2005) 61–82

The total error covariance in the inversion process can be expressed as the sum of all different contributions, that is, STOT ¼ SN þ SS þ SM

ð11Þ

where SN is the null-space error, SM is the measurement error and SS is the forward model error. SN, the null-space error, characterizes the spatial resolution in the retrieval process mainly due to instrument geometry, mathematical calculations, and already used a priori information. It is defined as: SN ¼ ðA
IÞSX ðA
IÞT

69

other hand, the SN, and SS are responsible for the absolute estimation of the retrieved information. If we compare both retrieved profiles only the relative error should be considered. However, if a profile from different experiments or techniques is inter-compared (validated), then it is necessary to use the absolute errors by convolving the averaging kernels of the lowest-resolution method with the high-resolution profile. Finally, in a retrieved height profile, the error bars at each altitude are then given by the square root of the diagonal elements of STOT (Eq. (11)).

ð12Þ 2.4. Error estimation

where A is the so-called averaging kernel matrix that considers the sensitivity of the retrieval to the true profile, I is the identity matrix and SX is the estimated covariance of the unknown profile x  
1 BT Bˆx ¼ ¼ Sx K T KSX K T þ Se K ð13Þ A¼ Bx Bx The covariance matrix for the measurement error can be calculated using the expression: SM ¼ DSe DT

ð14Þ

where Se is the covariance matrix describing the brightness temperature (measurement) error, and D is called the contribution function, which defines the sensitivity of the retrieved profile, xˆ , to the measurements or to the noise, y, and can be expressed as:  
1 BT Bˆx ¼ ¼ Sx K T KSx K T þ Se D¼ ð15Þ Be By Additionally, uncertainties in determining v and Teff in Eq. (6) due to meteorological variability (Figs. 3–6) produce an uncertainty in the calculated brightness temperature T B_O3, hence it can be treated as a measurement error SM (Eq. (14)), and therefore, it can be included in the Se covariance matrix. Another source of measurement uncertainty is the baseline residual due to the standing waves, as it has already been explained above. In this way, Se includes the measurement error itself, the baseline residual, and the uncertainty produced by the tropospheric attenuation (see also Nedoluha et al., 1995). Generally, for ground-based measurements, the higher the water vapour concentration is, the larger the difficulty or uncertainty in retrieving an ozone profile. The forward model error covariance is given by: SS ¼ DKb Sb KbT DT

ð16Þ

D is again the contribution function, and K b considers the sensitivity of the measurements to the forward model parameter b (spectral parameters). Sb is the forward model parameter error covariance. The variability of the retrieved profile from measurement errors (SM) contributes to the relative errors. On the

Fig. 7 illustrates the averaging kernels (matrix A, Eq. (13)) versus height. This picture indicates the degree of accuracy when retrieving a perturbation for a given height. Each curve represents a row in the A matrix and indicates the altitude resolution. Table 3 shows this resolution measured as full width at half maximum (FWHM) of the kernel. For example, the line indicated with triangles, corresponds to a perturbation or a spike of 1 ppmv and 9 km FWHM, at 27 km. These lines demonstrate how a perturbation in one height affects all other heights in the inversion process. Rodgers and Connors (2003) present a detailed discussion of the use of the averaging kernels and how this information may be used to determine the influence of the selected a priori profile into the retrieval. Furthermore, this paper also presents the necessary methodology for comparing retrieval coming from different instruments, especially with different height resolution. This validation process is also clearly presented in Nedoluha et al. (1997). Following these excellent papers, it can be seen how the retrieved profile xˆ is related to the true profile x and the a priori profile x a by  ð17Þ xˆ
xa ¼ A x
xa Þ þ Dey where A are the averaging kernels, as presented in Eq. (13). Further arrangements may lead to:  ð18Þ xˆ
x ¼ ðA
IÞ x
xa Þ þ Dey where the first term (A
I)(x
x a) can be defined as smoothing error. Eq. (17) can be rewritten to determine the effect of the a priori on the retrieval process: xˆ ¼ Ax þ ðI
AÞxa

ð19Þ

Here we see that if A=I (I: identity matrix), the retrieved profile will recover the true profile; otherwise, a proportion of the a priori will be passed to the retrieved profile through (I
A)x a. Fig. 8 shows the influence of the a priori profile for the TROPWA configuration. It can be seen that from 20 to 40 km, the contribution of the selected a priori profile is less than 30%, dependence is much higher in all other

70

C.M. Puliafito, S.E. Puliafito / Remote Sensing of Environment 94 (2005) 61–82

Fig. 7. Averaging kernels of the retrieval process versus height. Each curve represents a row in the A matrix, and gives an indication of the altitude resolution. Table 3 shows this resolution measured as full width at half maximum (FWHM) of the kernel.

altitudes. At present, a new spectrometer with narrower filters closer to the line centre is being tested, thus allowing retrieving better information at higher altitudes. As expressed in Rodgers and Connor’s paper, Eq. (17) is used to compare measurements (in our case TROPWA data) with other measurements or model calculation with much higher resolution and then much less dependence on the a priori (in our case, HALOE data). The high-resolution case can be considered as bideal caseQ. Then, Eq. (17) can be written as (see Eq. (4) referred paper): xs ¼ xa þ A ð xh
xa Þ

ð20Þ

where x s is the smoothed version of the highest resolution profile. However, when analysing relative variations of measurements given by the same instrument, this error is not considered, because it is constant for those measurements. Fig. 9 shows the contribution functions (matrix D, Eq. (15)), which indicates how much a 1-K noise in a given channel of the spectrometer influences the inversion process. For example, the black line with bQ represents the weight that 1 K additive noise has in the retrieved profile for the 60-MHz channel. The overall error calculation (Eq. (11)) gives us the final covariance, that is, the foreseen uncertainties for each

height, including all factors that can affect the retrieval procedure (errors in the inversion process, errors due to the a priori profiles, errors due to noise in the information, etc.). Summarising, the error sources can be divided in two main categories: the statistical errors and the systematic errors, and can be calculated as shown above. For our instrument, the statistical sources are the residual variability in the tropospheric attenuation and the measurement noise due to the integration time. These sources give an uncertainty of 3–7% for the tropospheric attenuation, while less than 2% for the measurement noise. Systematic error arises from: sideband suppression and residual baseline structures, leading a 3–5% uncertainty; calibration uncertainties reach 1–2%. These uncertainties yield into relative errors of about 8–11% (0.25–1.1 ppm) between 20 and 40 km. Null space error depends on the A matrix and on variances of the a priori profiles used and may vary between 10% and 30% from 20 to 40 km. These values increase for higher altitudes. By integrating longer, it is possible to reduce measurement noise variances (Se), up to a certain point where no longer integration does improve the spectrum, especially since the forward model uncertainties, the tropospheric uncertainties and the baseline structure cannot be removed by integration time. As the averaging kernels and the

Table 3 Height resolution calculated from the averaging kernels, measured as full width at half maximum (FWHM) Peak FWHM

17 km 5 km

20 km 6 km

23 km 8 km

26 km 9 km

29 km 10 km

32 km 12 km

35 km 12 km

38 km 12 km

C.M. Puliafito, S.E. Puliafito / Remote Sensing of Environment 94 (2005) 61–82

71

Fig. 8. Influence of the a priori profile in the retrieval of stratospheric ozone, calculated using the averaging kernels. Between 20 and 40 km, the contribution from the a priori profile is less than 30%, at all other altitudes the dependence is much higher.

contribution matrices are function of both covariances Sx and Se, reducing Se alone does not necessarily reduce SM, since Sx also contributes to D (Eq. (15)). However, reducing Se will indirectly improve A (Eq. (13)) and thus it may obtain some improvement in the height resolution.

2.5. Simulation tests In order to verify the retrieval behaviour and its capability, a set of simulation tests have been used. As an example of this simulation test, we present the results

Fig. 9. Contribution functions which indicate the sensitivity of each channel to noise in the retrieval process. For example, the black line with bQ represents the weight that 1 K additive noise in the 60 MHz offset frequency channel has over the retrieved profile. The gray line with diamonds represent the influence of the 350-MHz channel; the gray line with gray circles corresponds to the 250-MHz offset frequency; and the gray line with triangles represents the influence from the centre channel.

72

C.M. Puliafito, S.E. Puliafito / Remote Sensing of Environment 94 (2005) 61–82

for 12 spectra using several ozone monthly models taken from Cospar International Reference of the Atmosphere (CIRA) by Keating et al. (1990), as original (brealQ) profiles. These model profiles were used to simulate synthetic spectra or bbrightness temperature measurementsQ, using Eq. (1). Moreover, extra Gaussian noise with a standard deviation from 0.15 to 0.75 K has been added to the synthetic spectra lines according to the uncertainty of each filter bank channel. The tropospheric and baseline uncertainties are considered included in the simulated Gaussian noise. This corresponds to 200 s of integration time in agreement with the radiometric sensitivity given by Eq. (2) and Table 2. The next step to retrieve the generated synthetic spectra is to select the a priori information, such as temperature, pressure, ozone, water vapour profiles, etc. e.g. Remsberg et al. (1990). As more information is gained from the middle atmosphere, it is possible to select a breasonableQ set of profiles as a priori information, together with the uncertainties provided by the different instruments. The degree of coincidence between the original (brealQ) and the retrieved profiles will depend on the noise level of the spectrum, the frequency resolution of the filter bank, and the lower or greater knowledge of the a priori temperature and pressure profiles, which are characterized by the Sx and Se matrices. In these simulation tests, we have used as a priori information the same data set as above (forward model) but for a different month. In general, covariances matrices Sx and Se are not diagonal. Sx of the a priori set of profiles ranges for the main diagonal from 20% to 30% below 20 km, 30–40% up to 50 km, and less than 20% above 50 km.

Non-diagonal elements decay to less than 1% after 10–15 km from the main diagonal. For the Se matrix, we used a diagonal matrix using square values of DT shown in Table 2. Once these profiles have been selected, together with the Se and Sx matrices, we have performed the retrieval process using Eq. (6). We have computed the variances of the differences between the btrueQ profiles and the retrieved profiles as: VSIM ¼ r2 fxˆ
xg ¼ VN þ VM

ð21Þ

Since we know the btrueQ profiles, the variance V SIM represents the sum of the null space error V N plus the measurement error V M of the simulation test. (We use V N and V M to distinguish this simulated calculation from the btheoreticalQ calculation of SN and SM). These variances can be compared with the SN+SM of the theoretical error analysis, using the same Se and Sx matrices. Note that the Ss is not considered since both the btrueQ spectrum and the retrieval process use the same forward model. The result of this comparison process for 12 monthly profiles is illustrated in Fig. 10. This figure shows the dispersion V SIM as a percentage of the mean true profiles, (black dark line) varying around 10% of the mean ozone profiles (0.2 to 0.8 ppmv) up to 35 km and higher for higher altitudes. In the same figure, we show also the theoretical calculations for SN, SM and the sum SN+SM. The null space error SN is represented by the thin line with gray circles, and reaches about 10% up to 35 km, corresponding to the variances of the used profiles. Note that the calculated SM (dark thin line with gray triangles) is about

Fig. 10. Result of the simulation test: VSIM (black dark line), shows the dispersion from the differences between btrueQ and retrieved profiles, and represent the uncertainties, due to null space error and measurement error. SM (dark thin line with gray triangles) is the theoretical measurement error. SN is the calculated null space error (thin line with gray circles). All values are expressed as a percentage of the mean btrueQ ozone profile.

C.M. Puliafito, S.E. Puliafito / Remote Sensing of Environment 94 (2005) 61–82

73

2%, and corresponds mainly to the simulated measurement error. This measurement error has been probably underestimated in the simulation test compared to the real measurements, since as discussed above there are other sources of variabilities besides the measurement error itself. Adding noise to the measurement would have increased the V M value, and hence would have displaced the curves to higher uncertainties, but would not change the overall performance of the retrieval. Nevertheless, these values are consistent with the theoretical error analysis performed above, which helps to understand the behaviour of the inversion process.

Table 4 Daily profiles used for the bimonthly averages (for profiles #, see Table 1)

3. Results

September–October

3.1. Ground-based stratospheric ozone measurements from 1993 to 2000

November–December

As a result of this project, ozone radiometric measurements were obtained from 1993 to 2000. From this database, 1820—10 min—brightness temperature spectra were selected, which in turn generated 303—2-h integration time—ozone profiles, distributed in 60 days. Table 1 and Fig. 1 summarize the temporal distribution of these ozone profiles. Fig. 11 shows the above resulting stratospheric ozone profiles retrieved during this period. Additionally, bimonthly average profiles have been calculated using all available profiles for that particular period, for the 8-year time span as expressed in Table 4. Fig. 12(a) shows these averaged ozone profiles, while Fig. 12(b) illustrates the expected mean uncertainties in the ozone profiles versus the atmospheric height, reaching a relative accuracy from 7% to 11% as a function of the altitude and the month of the year according to the above-described error analysis.

Period

TROPWA Profiles #

HALOE Profiles #

March–April

8, 17, 28, 59, 60

May–June

18, 19, 20, 21, 22, 39, 40, 41, 42, 43, 55, 56 9, 10, 11, 12, 30, 32, 44, 45, 46, 47, 48, 49, 50, 51, 57, 58

15, 16, 17, 29, 30, 31, 32, 33, 34, 61, 62, 63, 64, 80, 81, 82, 83 18, 19, 20, 35, 36, 37, 38, 39, 54, 65, 66, 67 8, 9, 10, 11, 21, 22, 23, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 55, 68, 69, 77, 78, 79 1, 12, 13, 24, 25, 26, 51, 52, 56, 57, 58, 59, 60, 70, 71 2, 3, 14, 27, 28, 53

July–August

23, 24, 25, 26, 27, 31, 33, 34, 35, 36, 37, 38, 53, 54 1, 2, 3, 4, 5, 6, 7, 13, 14, 15, 16

Measurements performed at high locations had very low water vapour content, and lower baseline residual, which results in higher quality of the measurements. Measurements performed with the 92-GHz radiometer have given values of total water vapour content (or integrated water vapour IWV) in Puente del Inca (2700 m.a.s.l.) from 2 to 16 mm for clear sky conditions during November (late spring). In Uspallata, the water vapour content oscillated between 6 and 12 mm for clear sky conditions during June and September (winter and early spring). For Benegas, the water vapour content oscillates between 10 and 60 mm for clear sky conditions, whereas in winter time IWV varies between 10 and 30 mm. During large occurrence of Zonda winds, these values are even lower. Occasionally, retrieved ozone profiles have shown a big temporal dynamic variability, especially in the stratospheric

Fig. 11. Plot of the retrieved stratospheric ozone profiles from the TROPWA experiment. This graphic was created using 60 daily profiles, corresponding to the dates and sites shown in Table 1 and Fig. 1.

74

C.M. Puliafito, S.E. Puliafito / Remote Sensing of Environment 94 (2005) 61–82

Fig. 12. (a) Bimonthly ozone profiles retrieved from the TROPWA experiment between 1993 and 2000. Averaged were obtained on the basis of the available profiles in each period, according to Tables 1 and 4. (b) Mean relative uncertainties of the bimonthly profiles. March–April profile: gray line with black diamonds; May–June profile: gray line with plus signs (+); July–August profile: black line and black triangles; September–October profile: gray line with gray squares and November–December profile: black line and asterisks (*).

region between 20 and 35 km height. Figs. 13 and 14 show the retrieved ozone values for the 23.5-km layer measured at Puente del Inca in November 1994 and in August at Benegas, 1996. The 25–35-km height layer belong to the highest values in the ozone concentration (given in parts per million), which at the same time correspond to the most sensible layers in our retrieval process. These observed variations obviously exceed the error estimation considered for those layers. A larger discussion on these data was reported in Puliafito et al. (2002).

3.2. HALOE stratospheric ozone measurements over Mendoza from 1993 to 2000 In order to compare our results, we selected 83 stratospheric ozone profiles measured by the HALOE instrument (obtained at the NASA-HALOE web site: http://haloedata. larc.nasa.gov/home.html) over Mendoza region (338S, 68 W), within the period 1993–2000. As coincidence criterion, the ozone HALOE profiles were chosen not more than F58 away from Mendoza latitude and at most F108 in longitude.

C.M. Puliafito, S.E. Puliafito / Remote Sensing of Environment 94 (2005) 61–82

75

Fig. 13. Large ozone variations measured during November 1994 in Puente del Inca, corresponding to the (a) 23.5-km and (b) 30.5-km layers. Gray circles show 2-h running mean of 10-min integration time data. The continuous lines show a superimposed damped sinus wave between days 8 and 11 of November. The value of the sinus wave is represented in the legend.

The average profiles resulted in a mean latitude of 34.37 S and a mean longitude of 67.68 W with a mean deviation of 3.58 in latitude and 4.58 in longitude. HALOE measurements for Mendoza region are obtained, in average, around once per month (considering sunset and sunrise profiles). Fig. 15 shows the selected 83 HALOE ozone profiles for Mendoza region, which corresponds to the data shown in Table 1 and Fig. 1. In the comparison process, the different geographical and temporal view of the selected profiles must be taken into account. Additionally, the HALOE ozone profiles have a higher vertical resolution than the TROPWA measurements. As stated above in the discussion of the averaging kernels, and Eq. (20), retrievals from ground-

based radiometry, as TROPWA instruments, use the pressure broadening information to derive the altitude dependence of ozone emission. Therefore, they have a coarser vertical resolution compared to HALOE profiles. For this reason, to be able to compare equivalent resolution, HALOE ozone profiles have been averaged by convolving with the averaging kernels of the lower resolution TROPWA instrument. Comparative studies among different instruments have used similar coincidence criterion and also used the averaging kernels for comparing results, e.g., Nedoluha et al. (1997, 1998). In order to evaluate the degree of coincidence between HALOE and TROPWA ozone profiles, two types of comparisons were performed. First, we compared bimonthly

76

C.M. Puliafito, S.E. Puliafito / Remote Sensing of Environment 94 (2005) 61–82

Fig. 14. Ozone variations for different altitude layers, measured on August 23, 1994 from Benegas, during a Zonda wind event. Retrieval is calculated using 2-h running mean spectra from 10-min integration time individual spectrum.

average profiles, mainly taken from Benegas, and secondly, we compared a set of individual profiles for November 1993, November 1994 and June 1995. The discussion of the comparison among individual profiles will be shown in the Section 3.3 in association with other instruments measurements. Both bimonthly mean ozone profiles, i.e., from HALOE and TROPWA instruments, were compared for

the period 1993–2000 for the area above Mendoza. The averaged bi-monthly HALOE profiles are shown in Fig. 16 corresponding to each selected bimonthly period. In each of these figures, four curves are shown: First, the averaged 1 km layer HALOE profile for the given period, shown with asterisks (*). This curve can be considered as the high-resolution profile; secondly, the convolved HALOE profile with the TROPWA averaging kernels,

Fig. 15. Plot of the selected 83 stratospheric ozone profiles from the HALOE experiment, corresponding to the dates and sites shown in Table 1 and Fig. 1. As coincidence criterion, the ozone HALOE profiles were chosen not more than F58 away from Mendoza latitude and at most F108 in longitude.

C.M. Puliafito, S.E. Puliafito / Remote Sensing of Environment 94 (2005) 61–82

77

Fig. 16. Comparison of the HALOE and TROPWA profiles, figures (a) through (e). Four curves are shown in each figure: (1) 1-km layer average HALOE profile, (line with asterisks, *). This curve can be considered as the high-resolution profile. (2) Convolved HALOE profile with the TROPWA averaging kernels, (line with gray squares). This curve can be considered as smoothed version of the HALOE profile. (3) The a priori profile used in the retrieval (line with gray circles). (4) The TROPWA bimonthly average profiles. The comparison should be done between curves (2) and (4). (a) Period March–April, (b) period May–June, (c) period July–August, (d) period September–October, and (e) period November–December.

(line with gray squares); thirdly, the a priori profile used in the retrieval (line with gray circles). And finally, as fourth curve, the TROPWA bimonthly average profiles. Fig. 17 depicts the relative differences between the smoothed (convolved) HALOE profile and the TROPWA averages. Comparing these measurements, a difference between +0.4 and
0.8 ppmv (about +8% to
10%) can be observed. The major difference occurs for the maximum profile value, i.e., around 30–35 km height, which shows systematic lower values for the TROPWA

profiles. These differences arise from the coarser height resolution of the TROPWA averaging kernel and the influence of the selected a priori profiles used in the retrievals. 3.3. 276 GHz ground-based radiometric ozone measurements and ATLAS III-MAS during November 1994 From November 5 to 11, 1994, a measuring campaign was carried out in Puente del Inca, in the Argentinean

78

C.M. Puliafito, S.E. Puliafito / Remote Sensing of Environment 94 (2005) 61–82

Fig. 17. Relative differences between the TROPWA bimonthly averages and the smoothed (convolved) HALOE profiles. March–April profile: gray line with asterisks, *; May–June profile: gray line with gray squares; July–August profile: gray line and gray triangles; September–October profile: black line with black diamonds; and November–December profile: gray line and plus signs +.

Andes. One of the objectives of this campaign was to measure stratospheric ozone with different instruments and from a high location in the mountains to reduce the influence of tropospheric water vapour in the obtained spectra. During this campaign, two ground-based radiometer–spectrometers participated, TROPWA experiment and an additional radiometer from Max-Planck Institut fqr Aeronomie, Katlenburg-Lindau (MPAE) tuned at 276.9 for the ozone spectral line. The two ozone measurements were planned to be compared with the data coming from the Millimetre Wave Atmospheric Sounder (MAS) experiment of the NASA-ATLAS III Mission occurred from November 2 to 4, 1994. ATLAS III-MAS measurements were planned to last about a week long, but unfortunately the instrument suffered a malfunction on the third day of the ATLAS III mission, permitting coverage from approximately 358S to 758N. Consequently, MAS ozone data for Mendoza region was only available on November 4. Even though the groundbased campaign was planned from 5 to 11 November, due to technical problems on TROPWA experiment, ozone measurements were only available from November 8 to 11. The MPAE uncooled 276 GHz radiometer measured ozone on November 6 and experimentally ClO and N2O for the other days of the campaign. For comparison, the profiles obtained from the MPAE radiometer and the MAS instrument were not convolved with the averaging kernel of TROPWA, since the MPAE radiometer has similar resolution to TROPWA, and MAS has a resolution

of 3 km. Although MAS resolution is smaller than TROPWA, convolving this profile with the TROPWA averaging kernels produces almost no differences in the range of 20–40 km. Fig. 18 illustrates the coincident measurements from MAS, TROPWA and MPAE experiments that occurred during November 4 to 11, 1994. We have selected as coincident measurement criterion those profiles that were close to 338S, 688W (Mendoza coordinates) about F58 in latitude and at most F108 in longitude. In this picture, it is possible to see one of the MAS ozone measurements on November 4 at 328S, 648W, which reached a peak of about 8 ppmv between 32 and 34 km. The other MAS coincident measurement was found to be at 298S, 718W with an approximate maximum of 9 ppmv at 32 km height. The closest ozone HALOE measurement was found at 348S, 688W for November 16, with a maximum of 9.4 ppmv around 31 km, and on November 14 at 268S, 668W. TROPWA measurements carried out from November 8 to 11 had a maximum of around 35 km and oscillate from 7.7 to 8.9 ppmv. MPAE ground-based ozone profile achieved on November 6 has a maximum slightly displaced above the average of all compared profiles, and it is located about 34 km with peak amplitude of around 8.4 ppmv. Fig. 19 shows the difference of each profile with respect to the mean of all profiles. In order to compare measurements with different height resolution, HALOE profile was convolved. All other profiles were averaged in 5-km layers.

C.M. Puliafito, S.E. Puliafito / Remote Sensing of Environment 94 (2005) 61–82

79

Fig. 18. Illustration of the coincident measurements from several instruments during November 4 to 16, 1994. (1) There are two MAS ozone measurements: on November 4 at 328S, 648W, which has peak of about 8 ppmv between 32 and 34 km; and at 298 S, 718 W with an approximate maximum of 9 ppmv at 32 km height (green lines). (2) Two HALOE measurement at 348S, 688W for November 16, with a maximum of 9.4 ppmv around 31 km, and on Nov. 14 at 268S, 668W. (red lines). (3) An MPAE ground-based ozone profile achieved on November 6, has a maximum slightly displaced above the average of all compared profiles, and is located about 34 km, with a peak amplitude of around 8.4 ppmv (orange lines). (4) Four TROPWA profiles from November 8 to 11, having maxima around 35 km and oscillating from 7.7 to 8.9 ppmv (thin blue lines).

It can be observed that from 20 to 40 km, all profiles have a discrepancy to the mean value of about F15% or about F1.3 ppmv.

In the next two figures, we will show two other individual cases to illustrate the comparison study. Fig. 20 depicts a HALOE profile for November 20, 1993 for 328S

Fig. 19. Relative difference of each profile in Fig. 18 with respect to the mean of all profiles. In order to compare measurements with different height resolution, HALOE profiles were convolved; all other profiles were averaged in 5 km layers. HALOE profiles: red lines; TROPWA profiles: blue lines; MAS profiles: green lines; MPAE: orange line.

80

C.M. Puliafito, S.E. Puliafito / Remote Sensing of Environment 94 (2005) 61–82

Fig. 20. TROPWA and HALOE ozone profiles for November 20 and 21, 1993. The HALOE data corresponds to 328 S and 648 W (dark lines), shown as high resolution and convolved profile (dark line and black circles). Several TROPWA profiles (thin lines) obtained during November 21, 1993. These coincidence was very close in time and space.

and 648W (high resolution and convolved profile) and several TROPWA profiles obtained during November 21, 1993. Fig. 21 shows several measurements over Mendoza between June 13 and 15, 1995 from TROPWA and

HALOE. In this figure, a high resolution and a convolved profile for HALOE have been drawn. Based on the error analysis of the TROPWA retrieval process and the comparison with other instruments, it can

Fig. 21. TROPWA and HALOE ozone profiles over Mendoza between June 13 and 15 1995. A HALOE profile for May 30 is added since it was measured very close to Mendoza coordinates. The dark line with small diamonds shows the convolved HALOE June 95 profile. TROPWA profiles are represented with thin lines.

C.M. Puliafito, S.E. Puliafito / Remote Sensing of Environment 94 (2005) 61–82

be concluded that the TROPWA instrument can retrieve profiles satisfactorily within F10% above 20 km and below 40 km.

4. Conclusions From 1993 to 2000, the Institute for Environmental Studies (IEMA) of the University of Mendoza carried out ground-based stratospheric ozone measurements by means of millimetre wave radiometry. In this project also tropospheric water vapour was measured to determine the tropospheric attenuation affecting the ozone measurements. For this purpose, 3 years of radiosounding over Mendoza was used to study the water vapour variability. These data were also used to calibrate the 92-GHz radiometer and the channel 0 at
350 MHz offset frequency of the 142-GHz radiometer. These two frequencies were used to calculate the atmospheric transmission. To evaluate and validate the stratospheric ozone profiles retrieved from the TROPWA measurements, a theoretical error analysis is presented followed by a comparative study using the HALOE ozone profiles. This study includes a comparison of individual profiles and of seasonal averages from 1993 to 2000. We selected 60 daily profiles from TROPWA and 83 profiles from HALOE over Mendoza, with a coincidence criterion of F58 in latitude and at most F108 in longitude. The average profiles resulted in a mean latitude of 34.37 S and a mean longitude of 67.68 W with a mean deviation of 3.58 in latitude and 4.58 in longitude. These coincident HALOE measurements for Mendoza region correspond approximately to one profile per month. Given the coarser height resolution of the retrieved ozone profiles from microwave ground-based instrument, compared to those obtained from the solar occultation technique, the HALOE profiles were averaged by convolving them with the averaging kernels of the TROPWA retrieval process. This error analysis and the comparison tests allowed us to evaluate and qualify the retrieval of our instrument. It can be seen that between 20 and 40 km the TROPWA instrument is able to retrieve ozone profiles with absolute errors varying from 10% to 20%, and relative errors less than 5%, with a height resolution (FWHM) that varies from 5 to 11 km depending on the altitude. The seasonal variations show consistent patterns but the TROPWA measurements have a systematic lower peak value of about 0.5 to 0.7 ppmv. The major discrepancies between both set of profiles occur in the period of May–June with relative differences varying around +8% to
10% (+0.4 to
0.8 ppmv). This difference is partially due to the coarser height resolution of our instrument. Additionally, during November 4 to 11, 1994, the comparison study also included a second ground-based millimetre wave radiometer–spectrometer from MPAE, which was added to the TROPWA instrument. These ground-based meas-

81

urements were compared with those coming from the space-born experiments, ATLAS 3-MAS and HALOE. The profiles obtained from the MPAE radiometer and the MAS instrument were not convolved with the TROPWA averaging kernel, since the height resolution is not substantially different for the altitude range of 20–40 km. The results of the inter-comparison among the four instruments show that TROPWA measurements agree well with these coincident observations, especially between 20 and 40 km where the discrepancies among them do not exceed F10% or around 1 ppmv, while for heights below 20 km the discrepancies are larger than 10%.

Acknowledgements The authors wish to thank the authorities of the University of Mendoza and the Max-Planck Institut fqr Aeronomie for their support to this research activity. We also thank the HALOE Project and the MAS Project for the public availability of the data. The Argentine National Weather Service is here acknowledged for kindly permitting the use of the radiosounding data.

References Bevilacqua, R. M., Kriebel, D. L., Pauls, T. A., Aellig, C. P., Siskind, D. E., Daehler, M., et al. (1996). MAS measurements of the latitudinal distribution of water vapor and ozone in the mesosphere and lower thermosphere. Geophysical Research Letters, 23, 2317 – 2320. Brqhl, C., Drayson, S. R., Russell III, J. M., Crutzen, P. J., McInerney, J., Purcell, P. N., et al. (1996). HALOE ozone channel validation. Journal of Geophysical Research, 101, 10217 – 10240. Chandrasekhar, S. (1960). Radiative transfer. New York7 Dover Publications. Croskey, C. L., K7mpfer, N., Bevilacqua, R. M., Hartmann, G. K., Kqnzi, K. F., Schwartz, P. R., et al. (1992). The millimeter-wave atmospheric sounder (MAS): A shuttle based remote sensing experiment. IEEE Transactions on Microwave Theory and Techniques, 40, 1090 – 1100. Hartmann, G. K., Bevilaqua, R. M., Schwartz, P. R., K7mpfer, N., Kqnzi, K. F., Aellig, C. P., et al. (1996). Measurements of O3 H2O and CIO in the middle atmosphere using the millimeter-wave atmospheric sounder (MAS). Geophysical Research Letters, 23, 2313 – 2316. Keating, G. M., Pitts, M., & Young, D. (1990). Ozone reference models for the middle atmosphere. Advances in Space Research, 10, 12317 – 12355. Loidl, A., Rfmer, C., 1990. b278 GHz (ClO) Radiometer-FrontendQ. Technical report, Max-Planck Institut fqr Aeronomie, Germany, MPAE-T-70-90-36. NASA-HALOE web site: http://haloedata.larc.nasa.gov/home.html Nedoluha, G. E., Bevilacqua, R. M., Gomez, R. M., Siskind, D. E., Hicks, B. C., Russell III, J. M., et al. (1998). Increases in middle atmospheric water vapor as observed by the halogen occultation experiment and the ground-based water vapor millimeter-wave spectrometer from 1991 to 1997. Journal of Geophysical Research, 103, 3531 – 3543. Nedoluha, G., Bevilacqua, R. M., Gomez, R. M., Thacker, D. L., Waltman, W. B., & Pauls, T. (1995). Ground-based measurements of water vapor in the middle atmosphere. Journal of Geophysical Research, 100(D2), 2927 – 2939.

82

C.M. Puliafito, S.E. Puliafito / Remote Sensing of Environment 94 (2005) 61–82

Nedoluha, G. E., Bevilacqua, R. M., Gomez, R. M., Waltman, W. B., Hicks, B. C., Thacker, D. L., et al. (1997). A comparative study of mesospheric water vapor measurements from the ground-based water vapor millimeter-wavespectrometer and space-based instruments. Journal of Geophysical Research, 102, 16647 – 16661. Peter, R., & K7mpfer, N. (1992). Radiometric determination of water vapor and liquid water and its validation with other techniques. Journal of Geophysical Research, 97(D16), 18173 – 18183. Puliafito, S. E., Bevilacqua, R., Olivero, J., & Degenhardt, W. (1995). Retrieval error comparison for several inversion techniques used in limb-scanning millimeter wave spectroscopy. Journal of Geophysical Research, 100, 14257 – 14268. Puliafito, C., Puliafito, E., Hartmann, G., & Quero, J. L. (1998). Determination of stratospheric ozone profiles and tropospheric water vapor content by means of microwave radiometry. Proceedings of the VIII Simpo´sio Brasileiro de Microondas e O´ptica (SBMO 98) (pp. 274 – 278). Brasil7 Joinville-SC. Puliafito, C. M., S. Puliafito, S. E., & Hartmann, G. K. H. (2002). Observations of large stratospheric ozone variations over Men-

doza, Argentina. Atmospheric Chemistry and Physics Discussion, 2, 507 – 523. Remsberg, E., Russell III, J. M., & Wu, C. Y. (1990). An interim reference model for the variability of the middle atmosphere water vapor distributionQ. Advances in Space Research, 10, (6)51 – (6)64. Rodgers, C. D. (1976). Retrieval of atmospheric temperature and composition from remote measurements of thermal radiationQ. Reviews of Geophysics, 14, 609 – 624. Rodgers, C. D. (1990). Characterization and error analyses of profiles retrieved from remote sounding measurements. Journal of Geophysical Research, 95, 5587 – 5595. Rodgers, C. D., & Connor, B. (2003). Intercomparison of remote sounding instrumentsQ. Journal of Geophysical Research, 108, (D3), 4116, ACH13, 1–14. Russell, J. M., Gordley, L. L., Park, J. H., Drayson, S. R., Hesketh, W. D., Cicerone, R. J., et al. (1993). The halogen occultation experiment. Journal of Geophysical Research, 98, 10777 – 10797. Ulaby, F. T., Moore, R. K., & Fung, A. K. (1981). Microwave remote sensing. Active and passive. Artech House Inc.