Interactive comment on “ Refractivity and temperature climate records from multiple radio occultation satellites consistent within 0 . 05 % ” by U .

Thank you also for the comment regarding the small font size in the figure labels. The format of the discussion paper (with its landscape orientation) differs considerably from the format of the final AMT paper. This is particularly unfavorable in the case of fullpage figures in portrait orientation (Fig. 2, Fig 3., Fig. 5 and Fig. 6). When preparing the manuscript we had the final layout in mind. We will make sure that the font size in the final version is sufficiently large.


Introduction and context
Accurate, consistent, long-term data are required for any attempt to detect, understand, and attribute climate variability and change.GNSS (Global Navigation Satellite System) Radio Occultation (RO) data (Kursinski et al., 1997) are meanwhile recognized as a promising source for such a climate record in the upper troposphere and lower stratosphere.The signal phase measurements as function of time are based on precise atomic clocks and can be traced to the international standard for the second (Leroy et al., 2006a).Atmospheric profiles are not derived from absolute values (excess phase) but from phase change profiles (Doppler shift).Therefore, RO (raw) measurements require no external calibration and only short-term measurement stability over the occultation event duration (1-2 min), which is provided by very stable oscillators onboard the GNSS satellites.Potential residual GNSS clock errors and clock errors on the receiving satellites Published by Copernicus Publications on behalf of the European Geosciences Union.

2008
U. Foelsche et al.: Multiple radio occultation satellites consistent within 0.05 % can be corrected by relating the measurements to even more stable oscillators on the ground (Hardy et al., 1994;Hajj et al., 2002).With this "self-calibration", it should be possible to combine data from different sensors and different occultation missions without need for inter-calibration and overlap, provided that the same data processing scheme is used (von Engeln, 2006;Ho et al., 2009b) and spatio-temporal sampling (Foelsche et al., 2008a;Pirscher et al., 2007) is well understood.
During the last few years, RO measurements have been increasingly used by weather centers around the globe, and they show a surprisingly large positive impact on the quality of atmospheric analyses (Cardinali, 2009), partly due to the fact that they are the only satellite data (so far) that can be assimilated without bias correction (Healy, 2008).
The German research satellite CHAMP (Challenging Minisatellite Payload for geoscientific research) has (almost) continuously been recording RO profiles (Wickert et al., 2001(Wickert et al., , 2004) ) from September 2001 to October 2008.RO data from CHAMP, covering more than 7 yr, provide the first opportunity to create RO based multi-year climatologies.The number of RO profiles is sufficient to build monthly and seasonal mean, zonal mean climatologies (Foelsche et al., 2005(Foelsche et al., , 2008a)).The RO receiver on satellite A from the US-German GRACE mission (Gravity Recovery and Climate Experiment) has been continuously activated in May 2006 (Beyerle et al., 2005).
Formosat-3/COSMIC (Formosa Satellite Mission 3/Constellation Observing System for Meteorology, Ionosphere, and Climate; F3C hereafter) is a Taiwan/US RO mission consisting of six receiving satellites (Schreiner et al., 2007;Anthes et al., 2008).In April 2006, all six F3C satellites were originally launched into the same parking orbit with ∼515 km orbit altitude and then sequentially raised to their final orbit altitudes of ∼800 km.At this altitude the precession due to the oblateness of the Earth is smaller than in the parking orbit, leading to an intended deployment of the orbit planes to a final separation of 30 • .Five F3C satellites are in their final orbits (with optimal distribution of RO events in space and local time), but orbit raising for Flight Model 3 (FM-3) has been stopped in July 2007 due to problems with the solar panels (Anthes et al., 2008).
The European operational meteorological satellite MetOp-A was launched into a sun-synchronous orbit in October 2006.One of the eleven scientific instruments is the "GNSS Receiver for Atmospheric Sounding" (GRAS), which performs the first operational RO measurements (Luntama et al., 2008;von Engeln et al., 2009).MetOp-A will be followed by two almost identical satellites, both equipped with GRAS receivers.Together, the MetOp satellites will therefore provide RO measurements with essentially the same instruments over a period of about 15 yr.Since a F3C follow-on mission is currently planned, as are several RO missions of opportunity, a continuous RO data stream can be anticipated at least until 2020.
RO data from these satellites allow for testing the consistency of climatologies from multiple satellites.Hajj et al. (2004), Schreiner et al. (2007), and Ho et al. (2009a) have confirmed the consistency of RO data from different satellites by comparing closely co-located RO profiles.In Foelsche et al. (2009) we have developed an alternative approach, by comparing global climatologies from different satellites.Here we advance those results, inspect more satellites, and focus on the temporal evolution of differences in global monthly mean climatologies, with and without sampling error estimates subtracted, from January 2007 to December 2009.Data and methods are explained in Sect.2, the results on the consistency of RO climatologies are presented and discussed in Sect.3, followed by conclusions in Sect. 4.

Data and methods
At WEGC we have developed a retrieval scheme for climate applications.Background information, which is needed for high altitude initialization in the step from bending angle to refractivity, is introduced in a transparent way (Gobiet and Kirchengast, 2004;Borsche et al., 2006;Gobiet et al., 2007;Foelsche et al., 2008a;Pirscher, 2010).The profile retrieval, termed "Occultation Processing System" (OPS; current version: 5.4), starts from excess phase data for each occultation event, including precise position and velocity information for the transmitting and receiving satellites ("level 1 data" provided by different processing centers).Background information is only introduced at bending angle level, by performing statistical optimization at altitudes above 30 km: the retrieved bending angle profiles and background information are combined using an inverse covariance weighting approach (Rieder and Kirchengast, 2001;Healy, 2001).As background we use short-term forecasts from the European Centre for Medium-Range Weather Forecasts (ECMWF).Ionospheric correction is performed via a linear combination of bending angles (Vorob'ev and Krasil'nikova, 1994).A concise summary description of OPSv5.4 was provided by Steiner et al. (2009), a detailed description by Pirscher (2010).
The same profile retrieval (OPSv5.4)has been applied to the input data from all satellites.For the results shown here we used excess phase and orbit data (level 1 data) provided by UCAR/CDAAC (COSMIC Data Analysis and Archive Center), Boulder, CO, USA, for CHAMP (UCAR version 2009.2650 for the entire record), GRACE-A (version 2007.3200 until May 2009, version 2009.2650from June 2009 until December 2009) and all six F3C satellites (version 2007.3200 until March 2009, version 2009.2650from April 2009until December 2009), respectively.MetOp/GRAS bending angle data are operationally available, but unfortunately no excess phase and orbit data (which are needed for our retrieval).We will not show MetOp/GRAS climatologies before these data become available, but we have received MetOp/GRAS data for a single month, which have been processed in offline-mode (A. v. Engeln, EUMETSAT, personal communication, 2010).With this data set we could at least address the bending angle quality at high altitudes (see Sect. 3.1) and obtained very reasonable results, which are in line with those by von Engeln et al. (2009).
WEGC RO climatologies of the atmospheric parameters bending angle, (microwave) refractivity, dry pressure, dry geopotential height, and dry temperature are obtained by "binning and averaging" with the CLIPS (Climate Processing System, version 1.3) software tool (Pirscher, 2010), using the approach explained by Foelsche et al. (2008a) and(2009).All RO profiles in a predefined geographic domain ("bin") are sampled and averaged (weighted by the cosine of the latitude), using a common (mean-sea-level) altitude grid with regular 200 m spacing.For single-satellite climatologies we use "fundamental" bins with 5 • latitudinal and 60 • longitudinal width, which are averaged in longitude to build zonal mean monthly mean climatologies.Cosine weighting within the fundamental bins is performed, because the latitudinal distribution of RO data of all current missions (with high orbit inclinations) is closer to "equal numbers per latitude" than to "equal numbers per area" (for the latter "no weighting" would be the appropriate approach).With a latitudinal width of 5 • the effect of cosine weighting is, however, minimal.Our basic latitudinal resolution is 10 • : each of the 18 latitude bands (pole to pole) contains two 5 • bands, and the mean profiles for these two bands are averaged, weighted with the surface area.Compared to direct averaging (within 10 • zonal bands), our spatial averaging approach slightly reduces the effect of uneven sampling within the latitude bands.Larger-area aggregation, up to global, also uses weighting by surface area.
Seasonal climatologies are obtained by averaging three monthly mean climatologies, annual climatologies by averaging twelve monthly mean climatologies within a calendar year.In addition we provide error estimates, like the systematic difference between retrieved and co-located reference (the latter from ECMWF analyses), the standard deviation of the retrieved parameter, and the estimated sampling error, caused by undersampling of the "true" spatial and temporal variability (for more details see Foelsche et al., 2008a, Pirscher, 2010).All WEGC OPSv5.4/CLIPSv1.3 climatologies are available online via http://www.globclim.org.

Bending angle quality at high altitudes
We estimate the RO data quality based on bending angle statistics in the mesosphere, at impact heights (impact parameter minus local radius of curvature) between 65 km and 80 km (Pirscher, 2010).Here, the contribution from the neutral atmosphere is almost negligible and measurement noise and ionospheric residuals dominate.For data from the same day we can expect that the residual ionospheric noise is very similar for the different satellites and that differences in measurement noise are dominated by the different qualities of the RO receivers and antennas under considerationas long as the same differencing scheme is used for the removal of clock errors.All CHAMP and F3C excess phase data used in this study have been obtained using "singledifferencing": in addition to the occulted GNSS satellite, a second GNSS satellite is observed to remove clock errors on the receiving satellite (Schreiner et al., 2009).This reference link adds, however, additional ionospheric noise.MetOp is equipped with an ultra stable oscillator.Clock errors are therefore so small that high-quality data can be obtained without performing differencing at all ("zero-differencing").GRACE-A data at CDAAC are also processed using zerodifferencing (W.Schreiner, UCAR Boulder, personal communication, 2009).
Figure 1 shows, for all satellites under consideration, time series of the daily median bending angle bias with respect to the MSIS-E climatology and daily median noise The MSIS-E climatology (Hedin, 1991) is not the "truth", but it serves as a good reference at the heights above 65 km for those very small bending angles.A small negative bias could indeed be expected due to uncorrected ionospheric residuals (Sokolovskiy et al., 2009).The median bending angle values for all satellites in the overlap period are extremely consistent, the mean values for all six F3C satellites (over the available time period) equal −0.annual and semi-annual cycles with respect to MSIS-E.The mean value for MetOp/GRAS (−0.09 µrad) refers only to October 2007, where the semi-annual cycle shows a maximum.The value is therefore also consistent with those from the other satellites.
CHAMP RO measurements started in August 2001, but there have been several changes in the receiver software until March 2002 (Wickert et al., 2004).From this time onwards CHAMP RO data show a noise level (standard deviation) of about 4.7 µrad (Fig. 1, bottom), which is distinctly larger than the noise in F3C and GRACE-A data (from about 2.5 µrad to 2.7 µrad).This reflects the higher receiver quality on the F3C satellites, but also the benefit of zero-differencing in terms of less noise in case of GRACE-A (see Sect. 2), since the RO receiver on GRACE-A is similar to the one on CHAMP (Beyerle et al., 2005).The noise in MetOp/GRAS bending angle data is even smaller (about 1.0 µrad), but this is in part due to the fact that MetOp/GRAS data have also been obtained using zero-differencing.

Consistency of refractivity climatologies
In Foelsche et al. (2009) we adopted the approach to look at systematic differences between seasonal zonal mean climatologies from different satellites.This is a rigorous test of the consistency, since these differences contain the sampling errors of both climatologies (F3C satellites in their final constellation, e.g., never simultaneously sample the same region of the atmosphere).We showed that the predominant part of such differences arises indeed from different sampling of the atmosphere.Figure 2 illustrates this effect based on deviations of monthly mean refractivities from the multisatellite mean (i.e.averaging the climatologies of all satellites involved) in October 2007.It is most pronounced at high latitudes, where high atmospheric variability leads to values exceeding 2 % (due to the exponential decrease of refractivity with altitude we show relative errors).
The sampling error (undersampling in space and time) can, however, be quantitatively estimated when an adequate representation of the "true" spatial and temporal variability of the atmosphere is available and the information on the locations and times of RO events is used (Foelsche et al., 2008a;2009).As a proxy for the "true" evolution we use ECMWF analyses, whose four time layers per day are sufficient to sample the diurnal cycle and subdiurnal variations up to the semidiurnal cycle.We estimate the sampling error by comparing climatologies derived from vertical ECMWF profiles at the RO times and locations with climatologies derived from the complete 4D ECMWF field.This allows computing "double-differences", where we subtract the estimated sampling error fields from both single-satellite climatologies before we then difference the sampling-error-corrected climatologies.Even though these double differences contain the errors in the estimation of the sampling errors for both climatologies the results are striking (Fig. 3): relative refractivity differences are now smaller than 0.25 % almost everywhere in the considered domain between 8 km and 35 km altitude.This very good agreement shows furthermore    that the estimation of the sampling error is quite accurate, i.e.ECWMF analysis fields capture the atmospheric variability reasonably well.

CHAMP-Satellite Mean
Figure 4 shows the temporal evolution of global mean refractivity deviations over January 2007 to December 2009.Even the differences without sampling error subtraction (top panel) only occasionally exceed 0.15 % for individual months.Sampling errors largely cancel when averaging over several months and mean values for the entire available time period range from −0.03 % to +0.02 %.After subtraction of the estimated sampling errors from the individual climatologies (bottom panel) the data consistency is even more convincing: values for individual months are smaller than 0.05 % in almost any case, mean values for the entire time period are 0.00 % for each satellite.F3C/FM6 data are not available for all months.In June and until mid-July 2009 there are only very few F3C/FM-6 data, which explains slightly larger deviations during these months.
In Fig. 5 we analyze the altitude dependence of the refractivity deviations by inspecting five altitude layers of 5 km width in the upper troposphere and lower stratosphere (UTLS).Up to 30 km altitude all records are very consistent with temporal mean values between −0.01 % and +0.01 % (0.00 % in most cases).Above 30 km, CHAMP RO data differ from the multi-satellite mean by −0.03 %.The most likely reason for this behavior is the high-altitude initialization: CHAMP bending angles at high altitudes are not biased toward the other satellite records (Fig. 1), but CHAMP data show a higher noise level.This leads to a stronger weighting of the background than in the RO data from the other satellites, and a potential bias in the background will already become visible at lower altitudes (see, e.g.Gobiet and Kirchengast, 2004;Steiner and Kirchengast, 2005;Gobiet et al., 2007, for details on initialization behavior).

Consistency of dry temperature climatologies
Figure 6 shows the dry temperature deviations from the multi-satellite mean.The respective sampling errors are again subtracted.Between 10 km and 25 km altitude, values for individual months hardly exceed 0.05 K (only 0.02 % in relative terms, given temperatures are near 250 K or lower above about 8 km in the UTLS).Temporal mean values for GRACE-A and F3C do not exceed 0.02 K (0.01 %) up to an altitude of 30 km.CHAMP shows small, but systematic differences from the satellite mean by −0.03 K (10 km to 15 km and 15 km to 20 km), −0.05 K (20 km to 25 km), −0.08 K (25 km to 30 km), and −0.11K (30 km to 35 km).This increasing difference with altitude is consistent with the explanation in Sect.3.2 above: the hydrostatic integration in the retrieval step from refractivity (density) to pressure leads to a downward propagation of high altitude initalization errors, propagating further into temperature (Gobiet and Kirchengast, 2004;Steiner and Kirchengast, 2005).A detailed Fig. 5. Global mean monthly mean single-satellite refractivity deviations from the multi-satellite mean, after subtraction of the respective sampling errors, for five different altitude layers in the UTLS: 30 km to 35 km, 25 km to 30 km, 20 km to 25 km, 15 km to 20 km, and 10 km to 15 km (top to bottom; same layout per panel as in Fig. 4).Fig. 6.Same layout as in Fig. 5 (see that caption for explanation) but here showing the dry temperature deviations for the five altitude layers considered.
U. Foelsche et al.: Multiple radio occultation satellites consistent within 0.05 % discussion of errors in atmospheric RO profiles is given by Scherllin- Pirscher et al. (2011).All satellites show somewhat higher temporal variability in the 30 km to 35 km domain.

Conclusions
We analyzed the consistency of radio occultation (RO) climatologies derived from different satellites, by applying the same profiles retrieval scheme (OPSv5.4) to excess phase and orbit data provided by UCAR/CDAAC, Boulder, CO, USA, as well as the same climatology preparation scheme (CLIPSv1.3).Refractivity climatologies from all satellites agree very well up to 30 km altitude, even more when the estimated respective sampling errors are subtracted.In the UTLS core region between 10 km and 30 km altitude, global mean values of relative refractivity deviations from the multisatellite mean for individual months are smaller than 0.05 % in almost any case, time-averaged values for the entire time period are 0.00 % for each satellite.Consistency of temperature climatologies is correspondingly high, generally within 0.05 K (0.02 % relative).
It should, however, be noted that this consistency is not a proof of absolute accuracy, since there is a possibility of small common systematic errors in all RO records (e.g.due to residual structural uncertainty as analyzed by Ho et al. (2009) or due to uncorrected ionospheric residuals).We found no indications for instrument degradation, temporal inhomogeneities in the RO records, or temporal trends in sampling patterns.We conclude that if applying the same processing scheme for all data, refractivity and dry temperature records from individual satellites with similar bending angle noise can be safely combined up to 30 km altitude (refractivity also up to 35 km) to a consistent single climate record.Long-term climate records of this quality will be of substantial value for climate monitoring in the upper troposphere and lower stratosphere, underpinning the importance of long-term collection of radio occultation data for climate applications.
Figure 1.Daily median bending angle bias with respect to the MSIS-E climatology (top) and daily median bending angle noise (bottom) for nine different satellites (listed left; showing also for each satellite the time-averaged mean value from its full period of observations).

Fig. 1 .
Fig. 1.Daily median bending angle bias with respect to the MSIS-E climatology (top) and daily median bending angle noise (bottom) for nine different satellites (listed left; showing also for each satellite the time-averaged mean value from its full period of observations).

Figure 2 .
Figure2.Monthly mean relative refractivity differences with respect to the multi-satellite 2 mean for October 2007 for seven satellites (name in panel titles, for this month there are no 3 F3C/FM6 data available).4

Fig. 2 .CHAMPFigure 3 .
Fig. 2.Monthly mean relative refractivity differences with respect to the multi-satellite mean for October 2007 for seven satellites (name in panel titles, for this month there are no F3C/FM6 data available).

Fig. 3 .
Fig.3.Monthly mean relative refractivity deviations from the multi-satellite mean for October 2007 for the same satellites as in Fig.2but with the respective sampling errors subtracted.

Fig. 4 .
Fig. 4. Temporal evolution of global mean monthly mean single-satellite refractivity deviations from the multi-satellite mean (averaged over the entire domain from 10 km to 30 km altitude) from January 2007 to December 2009.Deviations with sampling errors included (top) and subtracted (bottom), respectively, are shown for eight satellites (listed left; showing also for each satellite the time-averaged mean value from its full period of observations).