Comparative analysis of SBUV/2 and HALOE ozone profiles and trends

Authors


Abstract

[1] The Halogen Occultation Experiment (HALOE) (version 19) ozone vertical profiles are compared with the data obtained by the second generation Solar Backscatter Ultraviolet (SBUV)/2 instruments from the National Oceanic and Atmospheric Administration’s NOAA-11 and NOAA-16 satellites using its version 8 algorithm. We examine the inverse noise-weighted mean percent differences between the HALOE and SBUV/2 ozone profiles at near coincident points and obtain a bias of less than 9% from 40 to 1.5 mb for the NOAA-11 and NOAA-16 satellite data records, with few exceptions. The weighted root mean square (RMS) differences between the HALOE and SBUV/2 observations are generally between 4 and 15% for pressure levels 40–1.5 mb. The RMS differences are larger for the NOAA-16 SBUV/2 data for higher latitudes. We also analyze the time-dependent differences between the two instruments in order to establish whether there is a significant relative drift. The slopes of time series of differences between the HALOE ozone profiles and the NOAA-11 and NOAA-16 SBUV/2 retrievals are, by and large, less than 1% per year and 2% per year, respectively. These results suggest a changing calibration in the NOAA-11 SBUV/2 instrument. Then we investigate the differences between ozone trends determined by the HALOE and SBUV/2 instruments. Trend differences are less than 0.5% per year in almost all latitude bands at pressure levels 10 and 3 mb. There are statistically significant biases of about 0.7% per year at 5 mb from 20°S to 50°N, and the trend differences are less than 1.3% per year at 60°–30°S.

1. Introduction

[2] Investigation of the processes that control the ozone abundance in the atmosphere and a scientifically robust understanding of the ozone change require accurate and consistent long-term observations of the ozone distribution. The presence of ozone is essential for terrestrial life and vital for human well-being in particular because of its role in protecting the Earth from biologically damaging solar ultraviolet radiation and the impact it has on global climate change.

[3] Most ozone resides in the stratosphere. The second-generation Solar Backscatter Ultraviolet instruments (SBUV/2) onboard the National Oceanic and Atmospheric Administration’s NOAA-9, NOAA-11, NOAA-14, NOAA-16, NOAA-17, and NOAA-18 satellites and the Halogen Occultation Experiment (HALOE) on board the Upper Atmosphere Research Satellite (UARS) launched in September 1991 [see Russell et al., 1993] have provided long-term dependable measurements of stratospheric ozone.

[4] HALOE employed the solar occultation technique to make ozone measurements in the mid-infrared (IR) at 9.6 μm. HALOE made 15 sunrise and 15 sunset measurements daily, with vertical resolution equal to about 2.3 km. In our study, we use the third public release data record (v19).

[5] The SBUV and SBUV/2 instruments employ the nadir-viewing backscattered ultraviolet technique [Bhartia et al., 1996] to measure ozone concentration profiles in the atmosphere on a global scale and constitute a 25-year record of ozone retrievals. SBUV/2 employs the ratio of the observed backscattered radiance to the incoming solar spectral irradiance to retrieve the ozone values. Hence the measurements are made on the daylight portion of each orbit, and there are about 14 orbits per day with 26° of separation at the equator. SBUV/2 has better spatial coverage than HALOE and provides daily retrievals on a global basis. These two instruments use quite different measurement techniques, and the vertical resolution is accordingly different. The vertical resolution of the ozone profiles from the SBUV/2 version 8 data record is about 6 km near 3 hPa and degrades to about 8 km near 1 and 10 hPa [Bhartia et al., 2004]. The ozone profiles are reported in about 3.2-km layers, and there is only one piece of independent information between the ozone density peak and the clouds or surface. During our study, we filter the SBUV/2 (v8) ozone profiles with solar zenith angle (SZA) greater than 80°, since the ozone retrieval algorithm has poorer performance at the viewing conditions with higher solar zenith angles. We employ the version v81713 reprocessed data files from the SBUV/2 v8 data set.

[6] The objective of this study is to assess the relative accuracy of the ozone profiles and the ozone trend estimates obtained from the HALOE and SBUV/2 (v8) data sets. Data validation is necessary to develop confidence in various ozone data records that were obtained employing diverse physical principles and to identify data with errors, biases, and physically unrealistic values. The comparison of vertical ozone profiles from these different measuring systems is complicated by the lack of temporal and spatial simultaneity of observations and different vertical resolution of the retrievals.

[7] Nazaryan and McCormick [2005] carried out validation studies of the SBUV/2 version 8 data set and compared the NOAA-11 and NOAA-16 SBUV/2 v8 data with the Stratospheric Aerosol and Gas Experiment II (SAGE II) measurements [Chu et al., 1989; McCormick, 1987], employing its version 6.2 retrieval code. We note that SAGE II also employed the solar occultation technique to make its measurements, but unlike HALOE, the ozone retrievals were made in the UV. The weighted average differences between the SAGE II (v6.2) ozone profiles and the SBUV/2 (v8) data set from the NOAA-11 and NOAA-16 satellites were less than 5 and 10%, respectively, at pressure levels 50–0.7 mb, with some exceptions [see Nazaryan and McCormick, 2005].

[8] In our work, we study the inverse noise-weighted mean and weighted root mean square (RMS) differences between the SBUV/2 (v8) and HALOE (v19) ozone profiles collocated in time and space [SPARC/IOC/GAW, 1998; Nazaryan et al., 2005]. We also investigate the time dependence of the differences between the measurements [SPARC/IOC/GAW, 1998; Cunnold et al., 2000]. Then we examine the differences between the ozone trend estimates obtained from the HALOE (v19) and SBUV/2 (v8) data sets [SPARC/IOC/GAW, 1998]. In our current study, we employ the same method as in the investigation of Nazaryan et al. [2005]. We choose the same parameters of comparison between SBUV/2 and HALOE as in the study of Nazaryan and McCormick [2005] to be able to compare our current results with the previous published studies involving SAGE II, HALOE, and SBUV/2. Nevertheless, it may be possible to make only some qualitative comparisons between the results obtained in those studies. In some cases, one cannot simply use the SAGE II data set as a transfer standard, and a new study is needed to compare the HALOE and SBUV/2 measurements.

[9] We also compare the SBUV/2 and SAGE II data sets and employ the same time periods that we use during the HALOE-SBUV/2 comparisons. The study of the SAGE II and SBUV/2 differences by Nazaryan and McCormick [2005] was for different time periods, and the SBUV/2 data record was also reprocessed since that publication. This new investigation will make it easier to evaluate the results obtained during SAGE II-SBUV/2 and HALOE-SBUV/2 comparison studies.

2. SBUV/2 and HALOE Ozone Mean Profile Comparisons

[10] In this section, we study the bias between the ozone profiles from the HALOE and SBUV/2 data records. We consider individual ozone mixing ratio profiles from those data sets colocated in time and space. We use coincidence criteria of: less than 1° latitude, 12° longitude and 24 hours difference between the match-up ozone profiles for each data set. We consider the NOAA-11 and NOAA-16 SBUV/2 v8 ozone mixing ratio profiles recorded at pressure levels 40 to 1.5 mb (about 22 to 45 km). The high-resolution HALOE ozone retrievals are smoothed vertically using the boxcar average algorithm before comparing them to the SBUV/2 ozone profiles. The smoothing varies with altitude to match the SBUV/2 resolution. In the HALOE data set, the ozone values are reported for about every 0.3 km. We degrade the HALOE ozone profiles by interpolating over the broader altitude intervals to match the pressure levels corresponding to the SBUV/2 retrievals. The SBUV/2 ozone data points are recorded in about 3.2-km layers.

[11] We employ the following formula to calculate the percent differences (PD) between the HALOE and SBUV/2 data pairs:

equation image

[12] Thus for each coincident HALOE and SBUV/2 data pair, we obtain a new data point in time and space, which represents the bias between these experiments.

[13] The individual ozone values for the SBUV/2 and HALOE data records have different measurement precision. In this study, we obtain the uncertainty of each PD data point using the method of propagation of errors [see Leo, 1994]. We employ the following formula to calculate the uncertainty σu of the quantity u = f(s,h), where s and h are the data points having errors σs and σh, respectively:

equation image

[14] We denoted by cov(s,h) the covariance of the quantities s and h. In our case, the covariance term is zero. During our calculations, the quantities s and h denote the coincident data points from SBUV/2 and HALOE, respectively. We obtain the quantity u = f(s,h) ≡ f(SBUV,HALOE), i.e., the percent difference data point, using formula (1).

[15] We examine the bias between the HALOE and SBUV/2 ozone profiles by calculating the weighted zonal averages and RMS of the percent difference data points for each 10° latitude bin using the following formulas:

equation image

where we define xi as the percent difference for the HALOE and SBUV/2 data pair i. The weighting factor wi in equation (2) is the inverse square of the error: wi = 1/σi2. In the statistical literature, the RMS is sometimes referred to as the quadratic mean. Thus we weight each data point in proportion to its error, and hence more importance is given to the data values with smaller uncertainty [see Leo, 1994]. The method of calculating the weighted averages is quite universal, and the above technique was also discussed in more detail by Nazaryan et al. [2005].

[16] In our study, we weight the HALOE ozone individual data points using the ozone measurement errors provided in the HALOE v19 data files. Those reported errors include only noise and aerosol effects. We employ the uncertainties reported in the SBUV/2 v8 data record to weight the SBUV/2 ozone data values.

[17] During the winter season, the amplitude of the errors may be larger than during summer in some pressure levels and smaller in others (see Figure 12). They also change from latitude to latitude. The errors are larger during the period immediately following the Mt. Pinatubo eruption in 1991.

[18] Figure 1 shows the mean and RMS percent differences between the HALOE v19 and NOAA-11 SBUV/2 data records. In our study, we use the NOAA-11 SBUV/2 ozone profiles from October 1991 to March 2001. The weighted mean percent differences of the ozone profiles between these experiments at coincident points are less than 9% and often within ±5%, at pressure levels from 40 to 1.5 mb, with few exceptions. In almost all latitude bands, the weighted RMS percent differences are generally between 4 and 15% from 40 to 1.5 mb, and the RMS differences are larger at 50° ± 5°S and 50° ± 5°N.

Figure 1.

Weighted zonal average and weighted zonal RMS percent differences between the HALOE (v19) and NOAA-11 SBUV/2 (v8) ozone data sets at coincident points. The comparisons are for the period October 1991 to March 2001. Note the change of scale at 50°N and 50°S.

[19] The reaction forming the atomic oxygen and ozone are initiated by sunlight, and the simple chemistry dominating the upper stratosphere predicts that these constituents may exhibit a diurnal variation [Brasseur et al., 1999]. The SBUV/2 is a daytime sensor, and HALOE makes the measurements at sunrise and sunset. The diurnal variation of ozone is small for the pressure levels below 1 mb. Planet et al. [1996] studied the differences between the NOAA-11 SBUV/2 and microwave correlative measurements, and their results indicate that the diurnal effect may be of concern at 1 mb and above. The diurnal ozone variation has been estimated to be about 5% at 1 mb [Huang et al., 1997]. We therefore do not consider differences above 1.5 mb.

[20] We study the weighted mean and RMS percent differences between the HALOE v19 and NOAA-16 SBUV/2 v8 ozone data sets from October 2000 to September 2005 (see Figure 2). The weighted average percent difference between the ozone profiles from these instruments is less than 9% from 30 to 1.5 mb (about 24–45 km) with few exceptions. The weighted RMS (HALOE-NOAA 16) percent differences are, in general, less than 15% and greater than 4% in the latitude bands from 35°S to 35°N at pressure levels 40–1.5 mb. The RMS differences are usually between 10% and 25% in the 50° ± 5°S, 40° ± 5°S, 40° ± 5°N, and 50° ± 5°N latitude bands.

Figure 2.

Weighted zonal average and weighted zonal RMS percent differences between the HALOE (v19) and NOAA-16 SBUV/2 (v8) ozone data sets at coincident points. The comparisons are for the period October 2000 to September 2005. Note the change of scale at 50°N and 50°S.

[21] McPeters et al. [2004] reported that the bias between the SBUV/2 version 8 ozone profiles and measurements from the ECC sondes, lidar and microwave retrievals, and the SAGE II and HALOE ozone data sets is less than 10% between 30 and 1 hPa, and the data records frequently agree to within ±5%. Petropavlovskikh et al. [2005] compared the SBUV/2 data (from the Nimbus-7, NOAA-9, NOAA-11, and NOAA-16) with Umkehr measurements from ground-based systems in the northern midlatitudes. They reported anomaly differences between the SBUV and Umkehr data for layers 6, 7, and 8 (about 30, 35, and 40 km) to within ±5%. Our results generally agree with those obtained in the previous studies.

[22] Figure 3 shows that the (SAGE II-NOAA 11) differences generally agree with the HALOE comparison results to within 5%. The (HALOE-NOAA 11) bias is more negative especially above 4 mb. This may be due to higher SAGE II ozone values compared with HALOE. Nazaryan et al. [2005] reported a 5% bias between the SAGE II (v6.2) and HALOE (v19) ozone measurements (with SAGE II values being higher) from 1991 to 2000 over the altitude range 20.5–50.5 km, with some exceptions. The investigation by Morris et al. [2002] show generally small differences between the SAGE II (v6.0) and HALOE (v19) ozone data sets of less than 4% throughout most of the stratosphere from 1991 to 1999.

Figure 3.

Weighted zonal average and weighted zonal RMS percent differences between the SAGE II (v6.2) and NOAA-11 SBUV/2 (v8) ozone data sets at coincident points. The comparisons are for the period October 1991 to March 2001.

[23] The most recent major eruption of a volcano was the eruption of Mt. Pinatubo in June 1991. The steady removal of Mt. Pinatubo aerosols is observed in the early 1990s until an apparent steady state is reached after 1997. We also compare the SAGE II, HALOE, and NOAA-11 SBUV/2 data sets from 1997 to 2001 when the effects of the Mt. Pinatubo eruption are minimal. The comparison results for the period 1997–2001 are within less than 2% from the values obtained for the period 1991–2001. At 50° ± 5°S latitude a, smaller bias of about 10% is obtained between HALOE and NOAA-11 SBUV/2 at pressure levels 30–40 mb for the volcanically quiescent period 1997–2001. We obtained a better agreement between SAGE II and NOAA-11 SBUV/2 by about 4% for the period 1997–2001 at pressure levels 30–40 mb in the 50° ± 5°S, 40° ± 5°S, and 50° ± 5°N latitude bands compared with the values for 1991–2001.

[24] Figure 4 shows that the SAGE II-NOAA 16 comparison results for the period 2000–2005 are generally within 5% from the values obtained for the HALOE and NOAA-16 SBUV/2 differences, with some exceptions. The SAGE II-NOAA 16 differences are within about 10° ± 2% from the HALOE-NOAA 16 comparison values at pressure levels 2–1.5 mb at 50° ± 5°N, 40° ± 5°N, and in the southern latitudes. The bias between those comparison results is about 10% at 40–30 mb in the 50° ± 5°S latitude band. To explain those differences, we employ the same method as in the study of Nazaryan et al. [2005] (i.e., almost the same method as in the current study) to compare the HALOE (v19) and SAGE II (v6.2) ozone profiles for the period 2000 to 2005. We obtain differences of about 10% ± 2% between the SAGE II and HALOE measurements at altitudes 42–45 km (about 2–1.5 mb) in the 50° ± 5°S, 40° ± 5°S, 40° ± 5°N, and 50° ± 5°N latitude bands, with SAGE II values being higher. The SAGE II-HALOE differences are about 10% ± 2% from 22 to 29 km (about 40 to 15 mb) in the 50° ± 5°S and 50° ± 5°N latitude bands for the period 2000 to 2005, which is consistent with the results we obtain during the SAGE II-NOAA16 and HALOE-NOAA16 comparisons.

Figure 4.

Weighted zonal average and weighted zonal RMS percent differences between the SAGE II (v6.2) and NOAA-16 SBUV/2 (v8) ozone data sets at coincident points. The comparisons are for the period October 2000 to August 2005.

3. Time Dependence of the Differences

[25] The topic of time dependence of the bias statistic between the HALOE and SBUV/2 measurements is particularly important for estimating long-term changes of the ozone abundance in the atmosphere and employing both data sets for the trend analysis. The investigation of the drift in the differences between the two experiments is essential for assessing possible uncertainties of ozone trend estimates that may have been caused by experimental effects or changing calibration uncertainties.

[26] In this part of the study, we use the same coincidence criteria as in the previous section. We analyze time series of the percent difference data points that we obtained using formula (1) and calculate weighted monthly zonal averages for each 10° latitude bin. The top panel of Figure 5 presents the time series of the monthly zonal averages of the percent difference data points calculated from the HALOE (v19) and NOAA-11 SBUV/2 (v8) coincident retrievals at 25° ± 5°S latitude and 5-mb pressure surface. There is a statistically significant time-dependent bias of about 0.6% per year between the measurements at that location.

Figure 5.

Time series of the percent difference data points calculated from the coincident data from the HALOE (v19) and SBUV/2 (v8) measurements. The circles are the weighted monthly zonal averages of the percent difference data points.

[27] We note that the differences are larger during the months immediately following the Mt. Pinatubo eruption in June 1991. The higher bias may be caused by the impact of heavy aerosol loading on the HALOE or the SBUV/2 ozone profiles. Figure 5 shows that the HALOE values are larger than the NOAA-11 SBUV/2 ozone estimates from 1997 to 2001, though this is not true for all pressure levels and latitudes. The SBUV/2 instrument’s grating drive experienced wavelength-selection problems during that period, and wavelength errors produced increased noise in the SBUV/2 ozone profile data.

[28] We apply the weighted least squares approach [see Neter et al., 1990] to make a linear fit to the monthly averaged difference time series and determine statistically significant slopes.

[29] Figure 6 shows the slopes of the difference time series. The error bars are twice the standard errors of the slope estimates (95% confidence interval). We observe statistically significant slopes of differences of less than 0.9% per year between the HALOE v19 and NOAA-11 SBUV/2 measurements in all latitude bands at pressure levels from 15 to 1.5 mb, with few exceptions at 15 mb. The drift in the bias is less than 1.2% per year at 15 mb from 10°S to 40°N. The Mt. Pinatubo eruption occurred at the beginning of the time series under consideration, and it may affect the slope calculations for the lower altitudes with heavy aerosol loading [SPARC/IOC/GAW, 1998]. Therefore the slopes of differences below 15-mb pressure level (about 28.4 km) are not included in our analysis in Figure 6. Although not shown in this paper, we calculate the slopes of time series of the differences from 40 mb to 20 mb (about 22 km to 26.5 km) for the volcanically quiescent period 1997 to 2001. Our results indicate statistically insignificant slopes of differences less than 2.3% per year at pressure levels 40 to 20 mb in all latitude bands and often less than 1% per year in the northern latitudes.

Figure 6.

Slopes of time series of differences between the HALOE (v19) and NOAA-11 SBUV/2 (v8) ozone data sets at coincident points for the period October 1991 to March 2001.

[30] We also investigate the drift in the bias between the SAGE II and SBUV/2 experiments for the period 1991 to 2001 [see also Nazaryan and McCormick, 2005]. Figure 7 shows statistically significant slopes of differences of less than 0.9% per year at pressure levels 7 mb to 4 mb in the latitude bands 60°–10°S and 20°–40°N. We also observe a statistically significant drift in the bias of less than 0.9% per year at 1.5 mb in the latitude bands from 50°S to 50°N. We also note that the sign of the percent difference slopes varies at pressure levels 5 mb to 1.5 mb (about 35.6 and 44.6 km) for the HALOE-NOAA 11 and SAGE II-NOAA 11 difference time series. Morris et al. [2002] and Nazaryan et al. [2005] suggested that there is essentially no statistically significant drift in the bias between the HALOE and SAGE II ozone profiles at those altitudes. On the basis of our investigation, we suspect that the statistically significant slopes illustrated in Figures 6 and 7 may be due to calibration problems in the NOAA-11 SBUV/2 (v8) data set.

Figure 7.

Slopes of time series of differences between the SAGE II (v6.2) and NOAA-11 SBUV/2 (v8) ozone data sets at coincident points for the period October 1991 to March 2001. The slopes of (HALOE-NOAA-11) differences (thick solid line) are duplicated without error bars for comparison.

[31] The slopes of percent difference time series for the SBUV/2 experiment on the NOAA-16 satellite are shown in Figure 8. Our results indicate that the slopes are, in general, less than 2% per year in all latitude bands from 50°S to 60°N at pressure levels 40–1.5 mb. The slopes of differences are less than 4% per year from 30 mb to 1.5 mb at 55° ± 5°S, and the slope is about 7% per year at 40 mb. We observe statistically significant drift in the bias less than 2.5% per year from 3 mb to 1.5 mb in the latitudes between 40°S and 40°N. There are also statistically significant negative slopes less than 1.5% per year from 15 mb to 7 mb at 5° ± 5°S, 35° ± 5°N, and 45° ± 5°N. Hence we note that the ozone values reported by the SBUV/2 instrument on the NOAA-16 satellite are increasing with time relative to the HALOE ozone measurements at those pressure levels and latitudes.

Figure 8.

Slopes of time series of differences between the HALOE (v19) and NOAA-16 SBUV/2 (v8) ozone data sets at coincident points for the period October 2000 to September 2005. Note the change of scale at 55° ± 5°S and 55° ± 5°N.

[32] Time series of monthly zonal averages of the percent difference data points calculated from the coincident HALOE (v19) and NOAA-16 SBUV/2 (v8) data at 35° ± 5°N latitude and 1.5-mb pressure surface are depicted in the lower panel of Figure 5. The trend analysis of this time series revealed a statistically significant time-dependent bias of about −1.7% per year (see Figure 8).

[33] Petropavlovskikh et al. [2005] reported slopes of (SBUV-Umkehr) anomaly differences between the Nimbus-7, NOAA-9, NOAA-11, and NOAA-16 combined data and retrievals from Umkehr ground-based systems in the northern midlatitudes of −1.4%, −0.7%, and 0.1% per decade for layers 6, 7, and 8 (about 30, 35, and 40 km), respectively.

[34] Figure 9 shows the drift in the bias between the NOAA-16 SBUV/2 (v8) and SAGE II (v6.2) ozone profiles from October 2000 to August 2005. The statistically insignificant slopes of differences are generally less than 1% per year at most latitudes and pressure levels from 40 to 1.5 mb with few exceptions: (1) 10°S–40°N at 2 mb, where the statistically significant drifts are less than 2.2% per year; (2) 55° ± 5°N at 3–1.5 and 20–15 mb, where the statistically significant slopes are less than 2.7 % per year and 1.8% per year, respectively; and (3) 25° ± 5°S and 45° ± 5°N at 10 and 15 mb, respectively, where the statistically significant drifts are less than about 1.1% per year.

Figure 9.

Slopes of time series of differences between the SAGE II (v19) and NOAA-16 SBUV/2 (v8) ozone data sets at coincident points for the period October 2000 to August 2005. The slopes of (HALOE-NOAA-16) differences (thick solid line) are duplicated without error bars for comparison. Note the change of scale at 55° ± 5°S.

[35] Compared with the results obtained by Nazaryan and McCormick [2005] for the period 2000–2003, we obtain smaller drift in the bias between the SAGE II and NOAA-16 SBUV/2 retrievals. We note that we employ the same method of calculations, the longer time period 2000–2005, and the reprocessed NOAA-16 SBUV/2 data set.

4. SBUV/2 and HALOE Ozone Trend Comparisons

[36] In this section, we compare the ozone trend estimates obtained from the HALOE v19 data set and the SBUV/2 v8 data record from the NOAA-11 satellite. According to Newchurch et al. [2000], the altitudinal and latitudinal structures of the SAGE I/II and SBUV(/2) trends are similar in the upper stratosphere, with SAGE trends being considerably more negative at almost all latitudes. Cunnold et al. [2000] reported that SBUV/2 (v6) measurements have an upward trend with respect to SAGE II (v5.96) retrievals of about 0.7% ± 0.3% per year over the period from 1989 to 1994. In our investigation, we use longer data sets and a newer data record from SBUV/2, employing the version 8 retrieval algorithm that was optimized to provide an unbiased estimate of ozone trends. We study the long-term changes in ozone determined from the HALOE (v19) and NOAA-11 SBUV/2 experiments for the period October 1991 to March 2001, when data are available from both instruments. We consider ozone-mixing ratio profiles and calculate inverse noise-weighted monthly zonal means for each 10° latitude bin. Then we fit those monthly average time series to a model including linear, annual, semi-annual, quasi-biennial oscillation (QBO), solar cycle, and autoregressive noise terms using the expression:

equation image

We employ the weighted least squares approach to fit the data [see Neter et al., 1990; Nazaryan et al., 2005]. In some cases, the HALOE retrievals at a given latitude may be separated by a time interval larger than 30 days, and those months are not included in the calculations and treated as data gaps as shown in Figure 10.

Figure 10.

HALOE (v19) and NOAA-11 SBUV/2 (v8) ozone time series from October 1991 to March 2001. The diamonds are the weighted monthly zonal averages for the 40° ± 5°N latitude bin and 15-mb pressure level. The solid line is the fitted model.

[37] The QBO component in the ozone time series observed near the equator is associated with the quasi-biennial oscillation in zonal wind and temperature in the tropical stratosphere. The QBO signature in ozone is also detected in the extratropics, and those anomalies are approximately out of phase with the tropical signal [see SPARC/IOC/GAW, 1998]. In order to model the QBO in the ozone time series, we apply the principal component analysis (PCA) to the Singapore zonal winds [see Naujokat, 1986]. The first two principal components, PC1 and PC2 of the PCA, account for 95.5% of the zonal wind variability at the following pressure levels: 70, 50, 40, 30, 20, 15, and 10 hpa [Wallace et al., 1993]. In our model (3), the QBO is represented in the following way:

equation image

where we included the factor A1 + A2cos(2πt) + A3sin(2πt) to model the seasonal synchronization of the QBO component in the midlatitudes [Randel and Wu, 1996], and the variable t represents years.

[38] We model the observed solar cycle variation of ozone by employing the normalized F10.7 cm radio flux as a proxy for the coronal forcing factors which vary in agreement with the 11-year solar cycle [see SPARC/IOC/GAW, 1998].

[39] In the model [see formula (3)] the noise term N(t) = YtYt − 1 is autocorrelated. In our study, we assume that N(t) is an autoregressive series of order one. Hence we use the AR(1) model [see Montgomery and Peck, 1982; Neter et al., 1990] and apply the Prais-Winsten method to correct for the autocorrelation [Prais and Winsten, 1954]. The procedure is described in more detail by Nazaryan et al. [2005].

[40] In Figure 10, we present (as an example) the time series of ozone monthly zonal weighted averages from the HALOE and SBUV/2 data records for the 40° ± 5°N latitude band and 15-mb pressure level. We note that the NOAA-11 satellite is not sun-synchronous, and the orbit drifts relative to the terminator. Thus the terminator-crossing location moves to lower latitudes, decreasing coverage and resulting in a data gap from April 1995 to June 1997, as depicted in Figure 10.

[41] The ozone trends (percent-per-year) obtained from the HALOE v19 data set and the SBUV/2 v8 data record from the NOAA-11 satellite for the period 1991 to 2001 are shown in Figure 11. The error bars indicate 95% confidence limits.

Figure 11.

Ozone trends as a function of latitude for the HALOE (v19) (triangles) and NOAA-11 SBUV/2 (v8) (squares) data sets for the period October 1991 to March 2001. The model is fitted to the weighted monthly zonal averages.

[42] The differences between the trends are statistically insignificant and less than 0.5% per year in almost all latitude bands at pressure levels 10 and 3 mb. At 5 mb, there are statistically significant differences of about 0.7% per year from 20°S to 50°N, and the bias is less than 1.3% per year from 60° to 30°S. Figure 6 shows positive statistically significant drift in the bias of about 0.3–0.7% per year between the HALOE and SBUV/2 measurements at 5 mb between 50°S and 40°N. Hence the NOAA-11 SBUV/2 ozone values are decreasing relative to the HALOE values with time at those latitudes and pressure levels. Figure 11 shows that the SBUV/2 ozone trend estimates are more negative than HALOE ozone trends at all latitude bands at 5 mb.

[43] Figure 11 shows that for the majority of latitude bands, the sign of the HALOE trend is of opposite sign to the SBUV/2 trend at pressure levels 10 and 1.5 mb. We note that there is a much better agreement between the SAGE II and HALOE ozone trend estimates [see Nazaryan et al., 2005].

[44] This study reveals serious potential drift problems with the NOAA-11 SBUV/2 instrument, but when included with other measurements in the SBUV and SBUV/2 instrument series, the derived ozone trends are more reasonable. Rosenfield et al. [2005] compared ozone profile trends from the SBUV version 8 data set with trends modeled by the Goddard Space Flight Center (GSFC) zonally averaged coupled model for the 1979–1997 time period. In the southern hemispheric (SH) midlatitudes, the most negative annual upper stratospheric SBUV trend was reported to be −7.0% ± 2.0% per decade, compared with the modeled maximum negative midlatitude trend of −5.8% ± 0.3% per decade in SH. The most negative upper stratospheric SBUV trend in the northern hemispheric (NH) midlatitudes was calculated to be −4.7% ± 1.3% per decade and the modeled trend was −5.2% ± 0.3% per decade at that location.

[45] In order to show the differences that may be introduced into the trends if one does not take into account the uncertainties of the measurements, we also calculate the nonweighted monthly zonal averages and apply the same procedure to obtain the ozone trends for SBUV/2 and HALOE. The trend estimates for the nonweighted case are generally within 0.1–0.2% per year from the trends computed for the weighted monthly zonal averages. We obtain different trends for the “weighted” and “nonweighted” cases because in one case, we use the weighted averages, and in the other case, we use the arithmetic means to compute the time series of monthly zonal means. Figure 12 shows the percent differences between the weighted and nonweighted monthly zonal means of ozone data points from SBUV/2 and HALOE for the 20° ± 5°N latitude bin at 10 mb (about 31 km). The average of weighted and nonweighted data pairs is used as the reference in computing the percent differences. We note that the features shown on these examples are not true for all pressure levels and latitudes. Thus the effects of the error weighting may be larger during the winter season (as seen in Figure 12) in some pressure levels and smaller in others. They also change from latitude to latitude as we mentioned above. Figure 12 shows larger differences for HALOE during the period immediately following the Mt. Pinatubo eruption in 1991.

Figure 12.

Percent differences between the weighted and non-weighted monthly zonal averages of ozone data point from the HALOE and SBUV/2 experiments for the 20° ± 5°N latitude bin and 10 mb pressure level.

[46] In Figure 13, we show the SAGE II (v6.2) and SBUV/2 (v8) trend estimates for the period 1991 to 2001 to compare them with the results depicted in Figure 11 and illustrate the differences between the SBUV/2, SAGE II, and HALOE ozone trends [Nazaryan and McCormick, 2005]. The results produced for the SAGE II and SBUV/2 ozone trends are generally in agreement with our findings in Figure 11, with few exceptions described below. We obtain small differences of within 0.5% per year between SAGE II and SBUV /2 at 10 and 3 mb in all latitude bands. Comparison of the results depicted in Figures 11 and 13 shows that the bias between the SAGE II and SBUV /2 trends is smaller than between HALOE and SBUV /2 at 50 ± 5°S at 1.5 mb. There are statistically significant trend differences of less than 1% per year between SAGE II and SBUV /2 from 30°S to 30°N at 1.5 mb. The better agreement between HALOE and SBUV /2 near the tropics at 1.5 mb (about 44.6 km) may be explained by the difference of about 0.9% per year between SAGE II and HALOE near the tropics at 45 km reported by Nazaryan et al. [2005].

Figure 13.

Ozone trends as a function of latitude for the SAGE II (v6.2) (circles) and NOAA-11 SBUV/2 (v8) (squares) data sets from October 1991 to March 2001. The model is fitted to the weighted monthly zonal averages. Ozone trends for HALOE (v19) are duplicated as a thick solid line for comparison. The HALOE error bars are not shown.

5. Summary

[47] In our work, we studied the HALOE (v19) and SBUV /2 (v8) ozone profiles collocated in time and space. During our analysis, we calculated the time series of the percent differences between the data sets and obtained inverse noise-weighted average and root mean square differences between the HALOE (v19) ozone data record and the SBUV /2 retrievals from the NOAA-11 and NOAA-16 satellites.

[48] The weighted average differences between the ozone profiles from the HALOE (v19) data set and the SBUV /2 (v8) measurements from the NOAA-11 and NOAA-16 satellites were generally less than 9% at pressure levels from 40 to 1.5 mb. The weighted RMS percent differences between the HALOE and NOAA-11 SBUV/2 ozone profiles were usually between 4 and 15% from 40 to 1.5 mb, and the differences were larger at 50° ± 5°S and 50° ± 5°N. We also showed that the weighted RMS differences between the HALOE (v19) and SBUV/2 NOAA-16 (v8) ozone profiles were between 4 and 15% from 40 to 1.5 mb in the latitude bands from 35°S to 35°N.

[49] The study of the time dependence of the differences between the HALOE (v19) and NOAA-11 SBUV/2 (v8) ozone data showed that the slopes of time series of differences were, in general, less than 1% per year from 15 to 1.5 mb. Our calculations revealed a statistically significant drift in the bias between those instruments in some latitudes and pressure levels. The trend analysis of the difference time series of the collocated data pairs from the HALOE (v19) and NOAA-16 data sets demonstrated that the slopes of percent differences were less than 2% per year from 40 to 1.5 mb in the latitude bands from 50°S to 60°N with few exceptions. The slopes were larger at 55° ± 5°S.

[50] In our work, we also obtained trend estimates for the HALOE (v19) and NOAA-11 SBUV/2 (v8) ozone data records. We employed the weighted least squares approach to fit the ozone model and determined the ozone long-term changes (percent-per-year) as a function of latitude and altitude. Our results showed statistically insignificant differences of less than 0.5% per year between the HALOE and SBUV/2 trends in almost all latitude bands at 10 and 3 mb. We obtained statistically significant biases of about 0.7% per year at 5 mb from 20°S to 50°N, and the difference was less than 1.3% per year from 60° to 30°S.

[51] The space-based solar occultation instruments have provided dependable long-term data records on a global scale of important trace gasses, and they serve as a reliable source for validation of other instruments. We obtained a good agreement while considering the mean differences between the HALOE (v19) and SBUV/2 (v8) ozone profiles, though the statistically significant slopes obtained in the HALOE -SBUV/2 time series may be an indication of calibration problems in the NOAA-11 SBUV/2 data record. The magnitude of the bias between the ozone trends from the HALOE and SBUV/2 data sets varied significantly among different pressure levels. The discrepancy between the ozone trends calculated at some pressure levels and latitude bands was much larger than the 0.5% per year differences obtained at 10 and 3 mb.

Acknowledgments

[52] We thank L. Flynn of NOAA NESDIS for valuable discussions and J. Anderson of Hampton University for providing the ozone model used for the trend analysis. We also thank Barbara Naujokat of the Freie Universitaet Berlin for providing the Singapore zonal winds. We wish to thank the Dominion Astrophysical Observatory of the National Research Council of Canada’s Herzberg Institute of Astrophysics for providing the solar flux data. We acknowledge the support from the NOAA’s Educational Partnership Program, contract number NA17AE1625.

Ancillary