One of the largest constraints to the retrieval of accurate ozone profiles
from UV backscatter limb sounding sensors is altitude registration. Two
methods, the Rayleigh scattering attitude sensing (RSAS) and absolute
radiance residual method (ARRM), are able to determine altitude registration
to the accuracy necessary for long-term ozone monitoring. The methods compare
model calculations of radiances to measured radiances and are independent of
onboard tracking devices. RSAS determines absolute altitude errors, but,
because the method is susceptible to aerosol interference, it is limited to
latitudes and time periods with minimal aerosol contamination. ARRM, a new
technique introduced in this paper, can be applied across all seasons and
altitudes. However, it is only appropriate for relative altitude error
estimates. The application of RSAS to Limb Profiler (LP) measurements from
the Ozone Mapping and Profiler Suite (OMPS) on board the Suomi NPP (SNPP)
satellite indicates tangent height (TH) errors greater than 1 km with an absolute accuracy of
Instruments that measure the solar radiation scattered by the Earth's
atmosphere in the limb direction provide a low-cost way of measuring trace
gases, aerosol profiles, and clouds from satellites. The technique can
provide daily full coverage of the sunlit Earth from commonly used polar
sun-synchronous satellites. To meet long-term ozone monitoring needs (3 %
precision between 15 and 50 km) requires the altitude registration of the
radiances to be accurate to within
In this paper we critically examine the performance of two methods of altitude registration that compare measured and simulated radiances. We discuss the inherent strengths and limitations of each method and then assess their performance using data from the Ozone Mapping and Profiler Suite (OMPS) Limb Profiler (LP), launched on board the Suomi NPP (SNPP) satellite on 28 October 2011.
One of these techniques, known as Rayleigh scattering attitude sensing (RSAS),
is relatively insensitive to instrument radiometric errors because
it utilizes measurements at two altitudes (20 and 40 km) where many of the
errors are correlated. However, since the method uses limb radiances
measured at 20 km, it is greatly affected by aerosols and therefore works
best where there is minimal aerosol loading. Under these conditions the
accuracy of the method is limited by the accuracy of the geopotential height (GPH)
data near 3 hPa (
We describe the theoretical basis of these two techniques in Sect. 2 and their application to the OMPS LP instrument in Sect. 3. We present several validations of our uncertainty estimates in Sect. 4 and summarize our findings in Sect. 5.
Most scene-based altitude registration methods applied to limb-scattering
instruments take advantage of the fact that the atmospheric Rayleigh
scattering measured by these instruments varies by 12–14 % km
The RSAS method, described in Sect. 2.1, employs signal ratios in which the DUR effects largely cancel. ARRM, described in Sect. 2.2, uses 295 nm radiances for which ozone absorption screens the DUR signal. The knee method, described in Sect. 2.3, has been used extensively by others (Sioris et al., 2003; Kaiser et al., 2004; Rault, 2005; von Savigny et al., 2005; Taha et al., 2008), but our analysis indicates that it offers no advantages over RSAS and ARRM.
Left panel shows calculated 350 nm radiances as a function of altitude,
normalized at 40.5 km. The GSLS calculation models the OMPS LP field of view
without aerosols. The shape of the curve originates from the competition between
molecular scattering, which increases roughly linearly with pressure, and
attenuation, which becomes important when the Rayleigh optical thicknesses near
the tangent point start to become large. Attenuation causes the slope of 350 nm
radiances to change sharply between 40 and 20 km (right panel), a
This technique is named after the sensor that was flown on the Space Shuttle STS-72
in January 1996 (Janz et al., 1996) to test a concept originally
proposed by Bhartia in 1992. The technique takes advantage of the fact that
the gradient in the log of the LS radiance
If
Figure 2 shows 352 nm sun-normalized radiances from one orbit of OMPS LP when the global aerosol loading is small. The short-scale features at 20.5 and 40.5 km are largely caused by variations in cloud and surface albedo. The 20.5 km curve has sharper features and appears to be shifted toward the South Pole. This is because large Rayleigh attenuation at 20.5 km causes the radiances to have much higher sensitivity to the atmosphere on the satellite side of the tangent point (TP), while the 40.5 km radiances have similar sensitivities to both sides. This effect creates large noise in applying the RSAS technique to orbital data. However, since the noise varies randomly from orbit to orbit, DUR modeling errors are reduced by averaging data from multiple orbits (this is confirmed in daily averages of the sun-normalized radiances where short-scale features are not seen).
The 350 nm sun-normalized radiances from one orbit of OMPS LP (center slit) taken on 2 February 2012. The blue line shows 40.5 km values, and the green line shows 20.5 km values (divided by 8 to put both curves on a similar scale) versus latitude. Since the global aerosol loading on this day was small, the short-scale features in both curves are largely caused by variations in cloud and surface albedo. The 20.5 km curve has sharper features and appears to be shifted toward the South Pole. This is because large Rayleigh attenuation at 20.5 km causes the radiances to have much higher sensitivity to the atmosphere on the satellite side of the tangent point (TP), while 40.5 km radiances have similar sensitivities to both sides. This effect creates large noise in applying the RSAS technique to orbital data. However, since the noise varies randomly from orbit to orbit, it can be reduced by averaging data from multiple orbits.
The GSLS-modeled ratio of 350 nm limb-scattered radiances at 20.5 km with and without aerosols (left axis) and single-scattering angle (right axis) as a function of latitude. A nominal latitude-independent aerosol extinction profile was used in the calculation for the OMPS LP viewing geometry on 2 February 2012. The strong latitude dependence is caused by an order of magnitude change in aerosol scattering phase function with latitude combined with the attenuation of Rayleigh-scattered radiation by aerosols along the line of sight (LOS). In the Southern Hemisphere, where LP measures aerosols in the backscatter direction, the latter effect dominates and the radiation decreases. The net effect is very sensitive to altitude, variation of aerosol extinction profile along the LOS, and aerosol particle size distribution, and it is therefore difficult to calculate accurately.
Aerosols in the instrument's LOS are a more significant source of error for
the RSAS method. Though the effect of aerosols near 350 nm is small compared
to longer wavelengths, it is difficult to model due to subtle differences
between two large effects: the reduction of Rayleigh scattering by aerosol
extinction and the enhancement of limb radiances by aerosol scattering.
Figure 3 shows the ratio of 350 nm limb-scattered radiances at 20.5 km with
and without aerosols as a function of latitude. The strong latitude
dependence is caused by an order-of-magnitude change in the aerosol
scattering phase function with latitude combined with the attenuation of
Rayleigh-scattered radiation by aerosols along the LOS. In the Southern
Hemisphere, where LP measures backscattered radiation, the latter effect
dominates and the radiation decreases. The net effect is difficult to
calculate accurately since it is very sensitive to aerosol altitude,
variation of the aerosol extinction profile along the LOS, and aerosol
particle size distribution. Model calculations with and without an average
loading of aerosols erroneously attribute TH errors of
Though the effect of aerosols is similar to the cloud effect mentioned earlier, it is not random because aerosols tend to have systematic latitudinal variability at these altitudes we consider. Given this complexity, the RSAS method works best in latitudes and months where the 350 nm aerosol extinction at 20 km is relatively small.
Another potential source of uncertainty in applying the RSAS technique comes
from uncertainty in simulating radiances at 40 km; one needs to have
accurate pressure profiles for altitudes at and above 40 km. If the pressure
profiles are obtained from GPH profiles provided by
meteorological data assimilation systems, a one-to-one relationship exists
between the two errors: a 100 m error in GPH at 3 hPa translates into
We developed ARRM to be applicable over many latitudes and times. In part
this method uses radiances measured by a limb instrument near 295 nm at
Though 295 nm radiances can be very ozone sensitive, this sensitivity drops to less than 0.2 % for a 10 % change in ozone above 65 km because the ozone density at high altitudes is exceedingly low. At this altitude the ozone concentration typically changes by 25 % per kilometer, so model radiance errors will be within 0.5 % provided the OMPS TH is accurate to 1 km.
A difficulty in applying ARRM is the inaccuracy of GPH data near 0.1 hPa needed to calculate 295 nm radiances at 65 km. GPH uncertainty increases with increased altitude (Schwartz et al., 2008). To reduce the effect of GPH inaccuracies, we developed a variation of a technique that has been used for many years to derive mesospheric temperature profiles from the vertical slope of Rayleigh-scattered radiances measured by ground-based UV lidars (McGee et al., 1991). The technique described in their paper computes temperatures using the relative density differences between successive altitudes where the scattering mechanism is purely Rayleigh. Since we rely on the same region of Rayleigh dominance, we can apply their technique to correct for errors in the GPH assumptions. GPH is related to temperature by assuming hydrostatic balance. The 350 and 295 nm residuals are affected similarly by the errors in the GPH with altitude, so we use the 350 nm residual to correct for the GPH errors at 295 nm. To the extent that stray light is also wavelength independent, this correction will remove stray-light errors.
The residual at wavelength
The 350 nm differential residuals on the right side of Eq. (2) provide
an estimate of the relative error in calculating radiances using
meteorological data between
The TH error estimated using this method is given by
ARRM has two primary drawbacks. ARRM uses 350 nm differential residuals to
correct for GPH errors between 40 and 65 km, so like RSAS this method is
sensitive to errors in GPH profiles near 3 hPa. To the extent these errors
are time-invariant, ARRM works best to monitor changes in the TH error. ARRM
is also sensitive to instrument calibration: a 1 % error in radiance
calibration at 65 km produces
The name of this method is derived from the characteristic knee shape of the limb radiance profiles (Fig. 4). Above the knee the radiances decrease with altitude due to the exponential decrease in Rayleigh scattering and ozone density. Below the knee ozone absorption becomes so large that it essentially blocks most of the Rayleigh-scattered radiation from reaching the satellite, making the radiances insensitive to atmospheric pressure. This characteristic shape allows estimations of altitude registration error in a manner very similar to that of RSAS. An advantage of this method is the ability to use shorter wavelengths that are less sensitive to aerosols.
A disadvantage is the method requires accurate ozone and pressure profiles near and above the knee region. A rough estimate of the ozone profile error caused by a TH error can be determined by simply shifting an ozone profile up and down (Fig. 5). From this analysis, errors in the ozone profile are found to be within 8 % at 40 km from TH errors of 300 m. It is important to note that the shift of the ozone profile does not necessarily translate to the exact error in altitude registration for limb radiances due to the nonlinearity of the inversion, which is especially important below 20 km. Differences of 8 % between the various ozone profile measurements are not unusual, and this will directly translate to uncertainty in the knee method.
The method also has a sensitivity to GPH errors that are similar to RSAS and ARRM. In our view this method provides no compelling advantage over a direct comparison between the limb ozone profile and a truth profile. Indeed, direct ozone comparisons are simpler and more reliable if the altitude registration error is the largest error source, and we use this technique to evaluate the results of RSAS and ARRM in Sect. 4.
In this section we discuss altitude registration errors for the OMPS LP.
Shortly after launch, RSAS analysis indicated a
An observed OMPS LP altitude error of 1 km translates to an along-track pointing error of 250 m for nadir-looking instruments. A positive TH error (the sensor is aimed higher than the indicated geolocation) means the nadir footprint is further south than believed.
Typical ozone profile in the tropics (left panel). By shifting the ozone profile, we can estimate an order and pattern of error in ozone profiles due to TH shift (right panel). Errors in ozone retrievals are within 8 % at 40 km from TH errors of 300 m. Errors are least sensitive at the ozone peak (25 and 30 km) and are more variable below. It is important to note that the shift of the ozone profile does not necessarily translate to the same error in altitude registration for limb radiances due to the nonlinearity of the inversion, which is especially important at lower altitudes (below 20 km).
Slit edge analysis is a method of deriving pointing errors internal to the instrument, much like onboard star trackers. The analysis was performed early in the S-NPP mission and found to be extremely robust; subsequent slit edge analyses indicate that there have been no changes from initial results to date.
The OMPS LP sensor utilizes a two-dimensional charge-coupled device (CCD)
detector to capture spectrally dispersed (along the 740-pixel row dimension)
and vertically distributed (along the 340-pixel column dimension) radiation
(Fig. 6). Three long vertical entrance slits spaced 4.25
An unexpected thermal sensitivity was discovered in the LP instrument soon
after launch (Jaross et al., 2014). Expansion of the LP instrument's
entrance baffle as the sun illuminates it midway through the Northern
Hemisphere causes mirrors in the telescope to rotate slightly, which in turn
moves the limb radiance image on the detector. Since there are separate
mirrors for each entrance slit, the three slit images move independently.
These image motions cause mis-registration of both the vertical pointing and
center wavelength of each pixel. Vertical pointing changes are detected most
clearly by observing the location (detector column) of the lower slit edge,
which has a sharp signal gradient. Figure 7 contains plots of the average
edge locations in the vertical (altitude) dimension along the orbit. These
pointing shifts are very repeatable (ranging only
OMPS LP CCD high-gain Earth-viewing radiance images for the three slits
(east/center/west). The east and west slit images are separated in longitude by
2.25
Since the same slit edge analysis can be applied to pre-launch test data, it
is possible to obtain the pixel line-of-sight shift relative to its
calibrated value in the spacecraft reference frame. There is no evidence of
image distortion, so this shift is the same for all detector pixels within a
slit image. The edge analysis indicates the three slit edges shifted by the
equivalent of 570, 470, and 950 m (east, center, and west slits, respectively) at the
middle of an orbit relative to pre-launch measurements. A mean sensor
temperature decrease exceeding 25
After the application of the slit edge determined corrections, analysis of RSAS and ARRM results indicated remaining TH errors. These remaining errors are presented in the next two sections.
We use the Gauss–Seidel limb-scattering (GSLS) radiative transfer code
described by Loughman et al. (2015) to estimate 350 nm radiances. Since the
40 / 20 km radiance ratio is not sensitive to polarization effects, we use the
faster scalar code rather than the full vector one to calculate DUR in our
model. The calculations assume a pure Rayleigh atmosphere bounded by a
Lambertian reflecting surface at 1013.25 hPa. The reflectivity of this
surface is calculated using limb measurements at 40 km. However, both
measurements and calculations show that the ratio of 40 / 20 km radiances is
not affected by reflectivity or surface pressure and that there is no discernible
cloud effect. Since NO
Slit edge results for the three slits (green: east slit; red: center slit; blue: west slit) plotted against time since southern terminator crossing. A 1-pixel shift corresponds to a 965 m TH shift. The offsets are stable from the southern terminator to the midlatitude Northern Hemisphere, where the exposure to the sun increases thermal effects. The event number, which is an index of LP measurements through each orbit, is shown at the top.
We estimate pressure and temperature vs. altitude at the LP measurement
locations and time from the Modern-Era Retrospective Analysis for
Research and Application (MERRA) data (GEOS-5 FP_IT Np)
from the Global Modeling and Assimilation Office (GMAO) at NASA
Goddard Space Flight Center (GSFC). The data are provided as GPH
at 42 pressures from the surface to 0.1 hPa, on a
0.5
As discussed in Sect. 2.1, RSAS results are affected by aerosols near 20 km. Aerosol profiles derived from the Optical Spectrograph and InfraRed Imaging System (OSIRIS) data (Llewellyn et al, 2004; Bourassa et al., 2007) indicate that tropical aerosols reached a minimum value (during the OMPS lifetime) just before the eruption of the Kelud volcano in Indonesia on 14 February 2014 (Fig. 8). Radiative transfer calculations using OSIRIS-derived aerosol profiles indicate that the aerosol-caused errors in the results shown are less than 100 m. We have therefore chosen to use equatorial RSAS data before the eruption to represent our best estimate of altitude registration errors (listed in Table 1).
Although we determined the best RSAS data point at the time just before the
Kelud eruption, we investigated other locations and time periods to estimate
the method's accuracy. The southern polar region is known to have relatively
minimal aerosol loading, especially during onset of the ozone hole in
October, but the extreme viewing angles at the South Pole make the LS
radiances difficult to model. The RSAS-derived TH errors in the southern
polar region are greater than from the pre-Kelud time period results between
0 and 200 m for all slits. This range of RSAS results is used to estimate an
absolute accuracy of
RSAS results at the Equator before the Kelud eruption in February 2014. The time period had a minimum aerosol loading (during OMPS lifetime) and was chosen using OSIRIS measurements (Fig. 8).
Time series of OSIRIS aerosol extinction profiles above the tropopause (dashed line). The large aerosol extinction coefficients in 2012 are due to the June 2011 Nabro eruption in Eritrea. The aerosols at 20 km reached a minimum value (during OMPS lifetime) just before the eruption of the Kelud volcano on 14 February 2014.
We calculated the radiances at 295 nm with the same radiative transfer code and profile inputs used for RSAS. The ARRM results presented here were normalized to RSAS results at the Equator before the Kelud eruption. The inter-slit differences have consequently been zeroed in all ARRM figures shown. We note that ARRM and RSAS estimates of the inter-slit TH differences agree to within 50 m at this normalization point.
Time-dependent plots show negative pointing error trends of approximately
200 m over the
Figure 10 summarizes the ARRM time series over a range of latitudes and
accounts for the two adjustments. At southern low to mid-latitudes
variations are
Daily averaged TH errors from ARRM analysis in bins of five events for
the center slit over a year. Colors indicate a different year.
Daily averaged time-dependent plots of TH errors from ARRM analysis
for the three slits. Approximate latitudes of the 10-event band averages are
noted. Values are normalized (by
ARRM is designed to accommodate stray-light errors that are independent of wavelength. That is, no additional TH errors are mis-assigned when stray light at 65 km is the same at 295 and 350 nm. The ground characterization of Limb sensor stray light indicates a small wavelength dependence (Jaross et al., 2014), but this is removed in ground processing. Our subsequent comparisons with RTM predictions indicate that residual stray-light errors at 65 km have a daily mean bias that translates to less than 100 m in TH. This sensitivity may be sufficient to explain the divergence of the west slit results in the northernmost panel of Fig. 10. This separation is seen more clearly in Fig. 11, which contains a mean of the ARRM results plotted as a function of time through each orbit. There are no geophysical errors that would give rise to inter-slit differences, but the west slit does have a known stray-light problem at high northern latitudes where direct solar illumination is near to the field of view (Jaross et al., 2014).
Average (over the
The more puzzling feature of Fig. 11 is the apparent 300–400 m variation in pointing through the course of each orbit. We lack indications of either geophysical or instrument errors to explain this result and must accept the possibility of a true pointing change. One potential explanation involves thermally induced flexing of the spacecraft that would affect the limb pointing but not the location of the slit image.
If the variation seen in Fig. 11 is attributed entirely to pointing errors, it implies that the slit edge analysis (see Fig. 7) underestimates the real changes by as much as 300 m.
In this section we utilize uncertainties in the parameters used to derive TH to indirectly validate the results shown in Sect. 3. Section 4.1 focuses on the validation of 3 hPa GPH information from MERRA. In Sect. 4.2 we consider the sensitivity to DUR modeling errors. Finally, in Sect. 4.3 we compare the ozone mixing ratio at 3 hPa derived from the OMPS LP and the Microwave Limb Sounder (MLS). For the validation studies in this section, the OMPS LP TH has been corrected with the RSAS errors listed in Table 1.
Errors in the assumed GPH profile directly translate into TH errors. For
ARRM, our application of the McGee technique depends upon the accuracy of
the GPH at 3 hPA (
The MLS–MERRA differences provide an estimate of the errors caused by the use of MERRA GPH in our radiative transfer calculations. Fortunately, although the 3 hPA GPH varies over 4 km along an orbit, a comparison of daily averaged values from MLS and MERRA show differences that are usually less than 200 m (Fig. 12).
Daily 5
These differences do not directly explain the orbital dependence of TH
errors shown in Fig. 11 but do provide an estimate of the magnitude of
errors caused by the use of MERRA GPH in our radiative transfer
calculations. There is better ozone agreement at the poles, but this may be
due to the reliance on climatology where there are scant measurements. If
this error were attributed solely to the limb model and only at one
altitude, the resulting TH error would be less than
There is no evidence of either seasonal dependence or north–south bias in the comparison, meaning that it is not clear how these GPH errors influence the ARRM results seen in Fig. 10. However, in Sect. 4.3 we discuss some suggestive but inconclusive results that help untangle GPH errors from TH errors.
The RSAS method and ARRM applied two strategies to minimize DUR modeling uncertainties. The RSAS method employs signal ratios in which the DUR effects largely cancel, and the ARRM uses 295 nm radiances for which ozone absorption screens the DUR signal. To ascertain the success of these strategies, we estimate the DUR modeling error by comparing LP measurements and modeled 350 nm radiances at 3 hPa. Model errors of the 350 nm radiances translate directly (Eqs. 1 and 3) into false estimates in the TH errors.
The OMPS Nadir instrument makes nearly simultaneous measurements from a
smaller field of view (50
The radiance comparison, shown in Fig. 13, suggests model or calibration
errors of 2–3 % on average, plus structures caused by the limb and nadir
scene mismatch. If this error were attributed solely to the limb-modeled DUR
effect, the resulting TH error would be less than
We have estimated the DUR modeling error by comparing 350 nm measured
and modeled radiances at 3 hPa. The radiances are modeled using an independent,
nearly simultaneous measure of surface reflectivity derived from the OMPS Nadir
instrument at 340 nm. The 50
Daily 5
At 3 hPa (
The time series of daily ARRM TH error (blue) and zonal mean ozone
differences (%) between OMPS LP and Aura MLS (red) at 50
The ARRM has displayed the ability to track any drifts or sudden changes of 100 m (Sect. 3.3), and time series of TH error derived from the ARRM track very closely to the time series of the LP–MLS 3 hPa ozone differences (Fig. 15). Both ARRM and LP–MLS ozone comparisons depend upon accurate TH and MERRA information, and in the same way. So, while these results suggest some confidence in the ARRM technique, we cannot assign the correlation shown in Fig. 15 to only a TH error or a MERRA error. It is important to note that MLS ozone profiles are reported as volume mixing ratio on a vertical pressure grid, while the LP algorithm retrieves ozone as number density on an altitude grid. Thus, in order to compare LP and MLS ozone retrievals, we had to convert ozone number densities to mixing ratios using MERRA temperature and GPH profiles. This conversion inevitably introduces MERRA GPH and temperature errors into the ozone comparisons. Therefore ozone differences between LP and MLS ozone retrievals depend not only on the LP TH error but also on errors in MERRA GPH as well as on errors in the retrieval algorithms and instrumental sampling (geophysical noise). Furthermore, analysis of LP and MLS ozone retrievals indicates a large daily ozone variability at 3 hPa that ranges from 2 % in the tropics to 20 % at high latitudes with the seasonal maximum during austral winters (results are not shown here); this analysis provides a sense of the geophysical ozone variability that is present. In consideration of all of the above factors, we remain cautious in making definite conclusions regarding applying the ARRM results; further analysis and comparisons with independent ozone observations (like SAGE III) are needed to confirm the results.
Accurate altitude registration is key to the success of using limb-scattered radiances to retrieve atmospheric trace gases. We have described two scene-based techniques that together provide highly precise and accurate estimates of the tangent height. These altitude registration techniques are inexpensive and more comprehensive than external sources of altitude information, such as star trackers mounted on the spacecraft. In fact the star trackers on the SNPP spacecraft failed to detect the 1–1.5 km tangent height error that we derived by applying the RSAS method (see Table 1). This error, if attributed to incorrect spacecraft pointing, means the nadir-viewing instruments on SNPP were locating scenes 340 m too far south. Initial pointing errors of even greater magnitude have been observed (Wolfe et al., 2013).
The RSAS and ARRM techniques are complementary. We developed the latter because changes in tangent height errors are rather more important than static errors. Figure 9 suggests the relative accuracy of ARRM is sufficient to detect 100 m changes in pointing in the absence of GPH errors. The accuracy may not be so small when GPH errors are included, especially for time periods less than 1 year. Our expectation is that multi-year trends in GPH error are small, a position that is supported by the lack of observed OMPS TH trends after accounting for distinct pointing shifts. RSAS too can be used to monitor pointing changes, but its sensitivity to stratospheric aerosols means the results may be influenced by geophysical processes such as volcanic eruptions.
The data set is available here:
The authors gratefully acknowledge the assistance of NASA's limb processing team in providing the data used in this paper. We would also like to thank Dave Flittner, Ernest Nyaku, and Didier Rault, who helped with the development of and updates to the RT model. Finally, we would like to acknowledge the role Didier played in laying the groundwork for the OMPS limb retrieval algorithm. Edited by: C. von Savigny Reviewed by: two anonymous referees