GPS radio occultation for climate monitoring and change detection


  • A. K. Steiner,

    1. Wegener Center for Climate and Global Change and Institute for Geophysics, Astrophysics, and Meteorology/Institute of Physics, University of Graz, Graz, Austria
    Search for more papers by this author
  • B. C. Lackner,

    1. Wegener Center for Climate and Global Change and Institute for Geophysics, Astrophysics, and Meteorology/Institute of Physics, University of Graz, Graz, Austria
    Search for more papers by this author
  • F. Ladstädter,

    1. Wegener Center for Climate and Global Change and Institute for Geophysics, Astrophysics, and Meteorology/Institute of Physics, University of Graz, Graz, Austria
    Search for more papers by this author
  • B. Scherllin-Pirscher,

    1. Wegener Center for Climate and Global Change and Institute for Geophysics, Astrophysics, and Meteorology/Institute of Physics, University of Graz, Graz, Austria
    2. Advanced Study Program, National Center for Atmospheric Research, Boulder, Colorado, USA
    3. COSMIC Project Office, University Corporation for Atmospheric Research, Boulder, Colorado, USA
    Search for more papers by this author
  • U. Foelsche,

    1. Wegener Center for Climate and Global Change and Institute for Geophysics, Astrophysics, and Meteorology/Institute of Physics, University of Graz, Graz, Austria
    Search for more papers by this author
  • G. Kirchengast

    1. Wegener Center for Climate and Global Change and Institute for Geophysics, Astrophysics, and Meteorology/Institute of Physics, University of Graz, Graz, Austria
    Search for more papers by this author


[1] Observation of the atmospheric climate and detection of changes require high quality data. Radio Occultation (RO) using Global Positioning System (GPS) signals is based on time measurements with precise atomic clocks. It provides a long-term stable and consistent data record with global coverage and favorable error characteristics. Highest quality and vertical resolution is given in the upper troposphere and lower stratosphere (UTLS). RO data exist from the GPS/Met mission within 1995–1997, and continuous observations are available since 2001. We give a review on studies using RO data for climate monitoring and change detection in the UTLS and discuss RO characteristics and error estimates, climate change indicators, trend detection, and comparison to conventional upper-air data. These studies showed that RO parameters cover the whole UTLS with useful indicators of climate change, being most robust in the tropics. Refractivity is most sensitive in the lower stratosphere (LS) and tropopause region, pressure/geopotential height and temperature over the UTLS region. An emerging climate change signal in the RO record can be detected for geopotential height of pressure levels and for temperature, reflecting warming of the troposphere and cooling of the LS. The results are in agreement with trends in radiosonde and ERA-Interim records. Climate model trends basically agree as well but they show less warming/cooling contrast across the tropical tropopause. (Advanced) Microwave Sounding Unit LS bulk temperature anomalies show significant differences to RO. Overall, the quality of RO climate records is suitable to fulfill the requirements of a global climate change monitoring system.

1. Introduction

[2] Long-term upper air records exist from radiosonde data and from operational satellite data based on the (Advanced) Microwave Sounding Unit ((A)MSU). Intensive efforts have been put into reconciling differences of temperature trends [Karl et al., 2006; Trenberth et al., 2007]. Basic agreement in tropospheric warming and in stratospheric cooling was found for observations, being broadly consistent with climate model trends [Santer et al., 2008]. Despite newly homogenized radiosonde data [McCarthy et al., 2008; Haimberger, 2007; Haimberger et al., 2008; Sherwood et al., 2008] and updated satellite products [Christy et al., 2007] disparities in quantitative trends remain. Structural uncertainties arise due to different choices in data processing and in the climate record construction [Thorne et al., 2005a, 2007; Titchner et al., 2009], since the instruments were designed for short-term meteorological observations and not for long-term climate monitoring purposes [Randel et al., 2009].

[3] Based on the shortcomings of conventional observation methods, requirements for climate observation, processing and support systems were discussed by Trenberth et al. [2002a]. The Global Climate Observing System (GCOS) program defined basic monitoring principles and updated its guidelines for the generation of data sets just recently [GCOS, 2010]. Climate data products based on observations from surface-based, airborne, and satellite-based instruments have to meet the GCOS requirements, which address long-term stability, homogeneity, reproducibility, global coverage, accuracy, resolution in space and time, and description and validation of the products. In this sense a fundamental climate data record (FCDR) involves a series of! instruments with sufficient calibration and quality control for the generation of homogeneous products, which are accurate and stable enough for climate monitoring [GCOS, 2010]. This means that separate data sets from different platforms need to be directly comparable to give reliable long-term records. Therefore, observations must be traceable to standards of the international system of units (SI) [Ohring, 2007]. Further essential requirements are data quality, knowledge of the measurement uncertainty budget and validation of data products [Immler et al., 2010].

[4] Beside the monitoring principles, GCOS [2010] also defines observation requirements for essential climate variables (ECV) such as upper air temperature. For temperature a resolution of 500 km horizontally, 0.5 km vertically in the upper troposphere (UT), 1.5 km vertically in the lower stratosphere (LS), and a root mean square (RMS) accuracy of <0.5 K is defined. Currently available upper-air records generally fail to fulfill the requirements of GCOS [Immler et al., 2010] and there is need for more accurate and more diverse data. According to Goody et al. [2002] any quantity that can be calculated precisely from the output of a climate model, and that can be accurately measured, can be used as climate variable. They point out microwave refractivity derived from Global Positioning System (GPS) Radio Occultation (RO) as promising climate variable.

[5] GPS RO is a new and independent data source with the potential to overcome problems of conventional observations. RO has best data quality in the upper troposphere and lower stratosphere (UTLS, ≈5 km to 35 km altitude) with global coverage, all-weather capability, long-term stability, and consistency. GPS radio signals are influenced by the atmospheric density field. The resulting signal delay can be detected by a receiver on a satellite in Low Earth Orbit (LEO). An occultation event occurs whenever a GPS satellite sets behind (or rises from behind) the horizon. Its signals are then occulted by the Earth's limb from the viewpoint of the receiver. Information on the atmospheric thermal structure is gained from phase measurements and orbit information based on precise timing with atomic clocks. This allows traceability to the international time standard [Leroy et al., 2006a] and ensures long-term measurement stability.

[6] Atmospheric profiles of bending angle, refractivity, pressure, geopotential height, and temperature are retrieved with high accuracy (<1 K) and vertical resolution (0.5 km to 1 km) in the UTLS [Kursinski et al., 1997; Rocken et al., 1997; Steiner et al., 1999]. The error budget can be estimated [e.g., Kursinski et al., 1997; Rieder and Kirchengast, 2001; Steiner and Kirchengast, 2005; Foelsche et al., 2008a; Steiner et al., 2009b; Scherllin-Pirscher et al., 2011a] and data from different RO missions can be combined without inter-calibration and overlap [Hajj et al., 2004; Foelsche et al., 2009b]. Structural uncertainties among RO data from different processing centers are low [Ho et al., 2009a], allowing the provision of RO-based atmospheric fields for climate studies.

[7] Here we give a review on the suitability of GPS RO data for climate monitoring and change detection. We review and discuss studies that have been performed so far related to climate change detection and give an update on most recent results in the field. We first introduce RO missions and describe RO data characteristics and error estimates in section 2. We give an extensive review on climate change studies carried out, discuss the utility of RO based climatologies for climate change indication and trend detection, and present a comparison to conventional upper air data in section 3. Conclusions and outlook are given in section 4.

2. RO Missions, Data Characteristics, and Error Estimates

2.1. RO Missions

[8] First RO observations of the Earth's atmosphere are available from the U.S. GPS/Meteorology (GPS/Met) proof-of-concept mission launched in 1995. Measurements were made for several periods within the years 1995–1997 [Ware et al., 1996]. GPS/Met successfully demonstrated the applicability of the RO technique for probing the Earth's atmosphere and the provision of accurate data [Ware et al., 1996; Kursinski et al., 1996; Rocken et al., 1997; Steiner et al., 1999]. Continuous RO observations were delivered by the German CHAMP (CHAllenging Minisatellite Payload for geoscientific research) satellite from mid 2001 to 4 October 2008 [Wickert et al., 2004]. The Argentine SAC-C (Satélite de Aplicaciones Científicas) [Hajj et al., 2004] and the U.S./German GRACE (Gravity Recovery and Climate Experiment) mission [Wickert et al., 2005; Beyerle et al., 2005], both launched in 2002, complement and continue CHAMP and are still active. SAC-C and GRACE data can be used to fill the only CHAMP data gap from 3 July to 8 August 2006.

[9] An ongoing RO mission is the Taiwan/U.S. Formosat-3/COSMIC (Formosa Satellite Mission 3/Constellation Observing System for Meteorology, Ionosphere, and Climate; F3C) [Anthes et al., 2000; Rocken et al., 2000; Wu et al., 2005; Anthes et al., 2008], which was launched in 2006 and consists of six receiving satellites. Meanwhile, the F3C satellites are slowly degrading but RO data collection is intended to be continued in order to minimize data gaps until a follow-on constellation mission will be implemented. The European MetOp/GRAS mission [Loiselet et al., 2000] started in 2006. It consists of a series of three satellites launched in sequence to operationally provide RO data until at least the year 2020 [Luntama et al., 2008]. Since recently, also the German TerraSAR-X and Tandem-X satellites provide RO data [Wickert et al., 2009]. Additional small satellites including RO receivers are due for contributing further data. The prospects for long-term RO data availability for climate applications are thus good.

[10] In the following main characteristics of RO data are described and explained in view of their climate relevance.

2.2. Main Characteristics of RO

2.2.1. Geographical Coverage

[11] RO measurements are made in limb sounding geometry. Through movement of the GPS transmitting satellite and mainly the LEO receiving satellite, vertical scanning of the atmosphere is provided. An occultation event is termed setting if the scan is made downward to the Earth's surface or rising if it progresses upwards. The number of occultation events per day depends on the number of transmitting and receiving satellites. The GPS system consists of nominally 24 satellites (practically of about 30 satellites in recent years). The number of occultations increases with the availability of GPS satellites. A single LEO satellite such as CHAMP acquires about 250 occultation events per day with one receiving antenna. The MetOp/GRAS instrument has two rising and two setting dual-frequency channels and can track about 650 occultation events per day. The six-satellite F3C constellation provides up to 2500 events per day. Global coverage is given if the receiving satellite is in a near-polar orbit, which is the case for most RO missions so far. Analyses related to coverage and geographical sampling can be found in the works of, e.g., Pirscher et al. [2007], Anthes et al. [2008], and Foelsche et al. [2008a, 2009a].

2.2.2. Self-Calibration, SI-Traceability, Long-Term Stability, and Reproducibility

[12] GPS satellites transmit dual-frequency signals in the microwave range at wavelengths of 0.19 m (L1 signal) and 0.24 m (L2 signal), which are virtually unaffected by clouds and aerosols, providing all-weather capability. Each GPS satellite carries a precise atomic clock producing the reference frequency by stimulated radiation of rubidium or caesium. Two Pseudo-Random Noise (PRN) codes are modulated onto the signals: the Coarse/Acquisition code (C/A-code) and the Precision code (P-code) together with navigation information and transmitter clock information. The GPS receiver on the LEO satellite is able to replicate the code of a GPS satellite signal and this receiver-generated signal is compared to the received signal.

[13] The basic measurement is the signal phase, which is proportional to the distance measurement between the transmitter and the receiver. Potential clock errors of GPS satellites or of the receiving satellites can be removed by (single- or double-) differencing methods using an additional non-occulted GPS satellite as reference and by relating the measurement to even more stable oscillators (e.g., Cs-133 atomic clocks) on the ground [Hajj et al., 2002]. In case there are ultra-stable clocks aboard LEO satellites, no differencing is needed (“zero-differencing”). The stable clocks aboard the GPS satellites ensure sufficient measurement stability over the occultation event duration of 1 to 2 min. Given the same atmospheric and ionospheric structure, and the same atmospheric path, the timing measurement will be the same whether measured from a U.S. or European or any other satellite, now or in 50 years time [Goody et al., 2002]. Thus, long-term stability and reproducibility is assured.

2.2.3. Resolution, Accuracy, and Precision

[14] In order to separate the phase change resulting from the atmosphere, the kinematic Doppler shift introduced by the relative motion of the GPS and LEO satellites is removed from the raw data based on precise orbit and velocity information (<0.3 m and <0.2 mm/s) [Schreiner et al., 2009]. For microwave refraction, geometric optics can be applied for converting Doppler shift profiles to bending angle profiles assuming local spherical symmetry of the atmosphere. The contribution of the ionosphere can be removed to second order by differencing the dual frequency GPS signals either at phase level, Doppler level, or bending angle level [Vorob'ev and Krasil'nikova, 1994; Ladreiter and Kirchengast, 1996; Steiner et al., 1999; Syndergaard, 2000].

[15] After initialization, bending angle profiles are converted to refractivity profiles via an Abel transform [Fjeldbo et al., 1971]. Density, pressure, and temperature are retrieved by applying the refractivity equation, the hydrostatic equation, and the equation of state. In the lower troposphere, multipath effects, diffraction, and potentially super-refraction become important due to atmospheric humidity causing strong refractivity gradients. Here, wave optics methods are required for the retrieval of bending angle profiles [e.g., Gorbunov, 2002; Gorbunov and Lauritsen, 2004; Jensen et al., 2003, 2004; Sokolovskiy et al., 2007]. Separation of temperature and humidity information in refractivity (“wet-dry ambiguity”) in the moist lower and middle troposphere also requires specific retrieval including external background information [e.g., Kursinski et al., 1997; Healy and Eyre, 2000; Steiner and Kirchengast, 2005].

[16] The vertical resolution of the processed data set is high with ≈0.5 km in the troposphere to ≈1.5 km in the stratosphere. The horizontal resolution of ≈300 km is comparatively low. However, this ratio of vertical to horizontal resolution is appropriate for weather characterization at synoptic scales and for provision of gridded atmospheric climatologies.

[17] The accuracy of RO measurements was estimated to be better than 1 K between 8 km to 30 km for worst case scenarios [Kursinski et al., 1997; Rieder and Kirchengast, 2001; Steiner and Kirchengast, 2005]. Accuracy depends, e.g., on precise orbit determination, thermal noise of the receiver, residual ionospheric errors, wet-dry ambiguity, and atmospheric multipath effects. Effects due to water vapor become important below about 8 km, initialization and residual ionospheric effects in the stratosphere above about 30 km. Sokolovskiy et al. [2009] introduced an optimal filtering method to reduce the small scale-ionospheric residual effects and to enhance accuracy, being useful in case of weak and/or encrypted L2 signals, and especially for climate applications.

[18] The close proximity of the F3C satellites after launch provided a unique opportunity to estimate an upper-bound on the precision of the RO data from closely collocated occultations (<10 km separation of tangent points and <2 s time separation) with almost parallel occultation planes since the signals from the same GPS satellite were received. The RMS difference of refractivity at 10 km to 20 km altitude was found less than 0.2% [Schreiner et al., 2007], which is approximately twice better than previous estimates obtained from less closely collocated CHAMP and SAC-C occultations [Hajj et al., 2004].

2.3. Error Estimates for RO Profiles and Climatologies

[19] At the Wegener Center we focus on the climatological application of RO data and on the provision and use of RO based atmospheric fields for such applications. The Wegener Center Occultation Processing System (OPS) was developed for best possible exploitation of RO data for climate research with focus on minimizing the influence of ionospheric residuals and of background information in the retrieval. The current RO processing version OPSv5.4 starts with phase and orbit data from the University Corporation of Atmospheric Research/COSMIC Data Analysis and Archive Center (UCAR/CDAAC), Boulder, CO, USA (versions 2007.3200, 2009.2650, 2010.2640). A short overview on OPSv5.4 is given by Steiner et al. [2009b], and a detailed description can be found in the work of Pirscher [2010].

[20] Background information (European Centre for Medium-range Weather Forecasts (ECMWF) short-range forecasting) is included only at the stage of bending angle initialization with statistical optimization at high altitudes. This processing step stabilizes the retrieval in the upper stratosphere and mesosphere and minimizes residual ionospheric errors, which depend on solar activity [Gobiet and Kirchengast, 2004]. Temperature profiles are retrieved assuming dry air conditions (“dry temperature”), which is closely valid in the UTLS down to about 8 km [Steiner et al., 2009a, 2009b]. The Wegener Center OPSv5.4 data, both profiles and climatologies, are available online via

[21] The main contributions to retrieval errors stem from residual errors of the GPS dual-frequency ionospheric correction and from the initialization of profiles at high altitudes using background information. The level of ionization and therefore residual ionospheric errors are correlated with solar activity. The residual error is larger under solar maximum than solar minimum conditions. The change in bending angle due to the 11-year solar cycle was estimated to induce a temperature bias of up to 0.1 K at 20 km and of about 0.2 K at 25 km [Gobiet and Kirchengast, 2004; Rocken et al., 2009]. For comparable solar cycle conditions this bias becomes negligible in >10-year temperature trends below 25 km altitude. An upper bound estimate of 0.04 K/decade (resulting from doubling an empirical bias estimate of 0.2 K over 10 years) is considered conservative, provided that careful quality control excludes those RO profiles under highest ionization conditions for climate applications.

[22] Temperature biases caused by the initialization process are <0.1 K to 0.2 K below 30 km height for non-polar (<60° latitude) large-scale means [Gobiet and Kirchengast, 2004; Gobiet et al., 2007]; and long-term variations are expected to be smaller [Steiner et al., 2007; Ho et al., 2009a]. Recent empirical error analyses, which were performed in reference to ECMWF analysis fields, provided observational error estimates for RO parameter profiles from bending angle to temperature [Scherllin-Pirscher et al., 2011a] confirming and extending former studies [Kuo et al., 2004; Steiner and Kirchengast, 2005; Steiner et al., 2006]. Observational RMS errors were found to be 0.35% to 0.5% in refractivity and 0.7 K to 1 K in temperature within 8 km to 25 km. The errors were found very similar for different RO missions (CHAMP, GRACE, F3C).

[23] Monthly mean atmospheric fields are computed through sampling and averaging of RO profiles. Accurate 10° zonal-mean RO climatologies can already be obtained for single RO satellites like GPS/Met and CHAMP with well defined error characteristics. RO climatologies are affected by statistical observational errors, sampling errors, and systematic errors, yielding a total climatological error. The statistical observational error is based on averaging over hundreds of profiles per zonal band. It can be regarded negligible in large-scale climatologies (<0.01 K to 0.1 K) [Scherllin-Pirscher et al., 2011b].

[24] The sampling error stems from uneven and sparse sampling in space and time. It amounts to <0.3 K in the UTLS with a local time component error of <0.15 K [Pirscher et al., 2007]. It is smallest at low latitudes equatorwards of 40° where atmospheric variability is low and it increases at higher latitudes. The sampling error can be estimated and should be subtracted from raw RO climatologies [Pirscher et al., 2007; Foelsche et al., 2008a]. Remaining residual sampling errors are very small (<0.1 K vertically resolved and <0.03 K for UTLS means in large-scale areas) [Foelsche et al., 2008a, 2009b; Scherllin-Pirscher et al., 2011b]. Generally they are less than 30% of the original sampling error as found by Scherllin-Pirscher et al. [2011b], confirming that the estimate of 50% reduction used by Steiner et al. [2009b] in a detection study was conservative. Residual sampling errors and statistical errors are at a comparable magnitude at low and middle latitudes.

[25] Scherllin-Pirscher et al. [2011b] furthermore give a detailed discussion of systematic error sources and provide systematic error bound estimates. Briefly, initialization errors and residual ionospheric errors are regarded as dominant systematic error contributions in the lower stratosphere. Sustained horizontal gradients and degraded signal tracking and GPS L2 signal quality are dominant in the troposphere. The systematic error generally dominates the total climatological error at low to middle latitudes, where the total climatological error in 10° zonal monthly means is estimated to be smaller than 0.15 K in temperature, increasing toward higher latitudes to 0.6 K in wintertime. Overall the errors of RO climatological fields are small compared to any other UTLS observing systems for thermodynamic atmospheric variables, making these data particularly valuable as reference data set.

[26] Structural uncertainty, defined as unintentional bias arising from the chosen methodological approaches, was analyzed for four different RO processing centers including Wegener Center. Ho et al. [2009a] found that structural uncertainties are low with <0.03% per 5 years for refractivity trends in large-scale means, corresponding to <0.06 K per 5 years for temperature trends. A follow-on study is currently ongoing and extended to six main RO processing centers worldwide. Preliminary results confirm the findings of Ho et al. [2009a]. Since the different processing centers use different techniques with different systematic error contributions, the small structural uncertainty in trends can be regarded as evidence that the systematic error contributions do not introduce a significant bias in trends (i.e., any possible small residual bias is essentially constant over time).

[27] Altitude and latitude resolved zonal-mean temperature climatologies based on RO data from different F3C satellites show excellent agreement with differences of <0.1 K between 8 km and 35 km altitude, and of <0.04 K in the overall mean [Foelsche et al., 2008c, 2009a]. Pirscher [2010] and Foelsche et al. [2011] carried out a detailed validation of multisatellite climatologies of fundamental atmospheric variables. Figure 1 illustrates the consistency of monthly temperature climatologies (with sampling error subtracted) from different satellites, GPS/Met, CHAMP, SAC-C, GRACE, and F3C-FM1 to F3C-FM6 [cf. Pirscher, 2010, section 2.3]. The deviation of dry temperatures from the satellite mean is shown for the tropics (20°S to 20°N). The agreement at 25 km to 35 km is basically around 0.2 K and at most <0.3 K. At 8 km to 25 km the agreement is better than 0.1 K, and <0.05 K in the global mean. No indications for instrument degradation, instationarities in the RO records or temporal trends in sampling pattern were found [Foelsche et al., 2011], demonstrating that it is possible to combine RO climatologies from different satellites to a single climate record without inter-calibration or correction, provided the same processing system is used.

Figure 1.

Consistency of RO data from different satellites at low latitudes (20°S to 20°N). For each month the deviation of dry temperatures from the all-satellite mean of the month is shown for the (top) 25 km to 35 km and for the (bottom) 8 km to 25 km altitude range. If only a single satellite contributes data in a given month the deviation is by default zero.

3. RO for Climate Change Research

3.1. A Review of RO Based Studies on Climate Change Monitoring

[28] Regarding climate change applications, Yuan et al. [1993] were the first to test the hypothesis that the GPS can be used to detect climate change. They simulated GPS signals in a doubled CO2 model climate and found a decent increase in the signal phase path due to increase in water vapor and reasonable trends of refractivity in the troposphere, indicating the potential of GPS RO to detect climate change. Melbourne et al. [1994] and Ware et al. [1996] pointed out that RO will be well suited for the task of climate monitoring and change detection in fulfilling the requirements of long-term continuity and stability, high vertical resolution and accuracy in the UTLS. The more precise and consistent the measurements, the shorter the time period required to identify climate trends. Melbourne et al. [1994] stated that RO could provide an absolute standard and/or calibration system for other temperature monitoring systems.

[29] Leroy and North [2000] investigated the application of (at that time simulated) COSMIC data for global change research. They showed that climate signal detection can be a useful method for climate model testing. By examining climate trends and anomalies as revealed by RO data, it can be tested whether climate models reproduce those trends and anomalies. Leroy et al. [2006b] performed a related sensitivity analysis using simulated GPS RO data for testing 12 climate models of the Intergovernmental Panel of Climate Change Fourth Assessment Report (IPCC AR4). Simulating trends in logarithmic dry pressure (proportional to geopotential height of pressure levels) they found that features common to all model predictions for the 21st century climate change can be tested with 95% confidence in 7 to 13 years, and that an indicator of climate change in upper air dry pressure is the poleward motion of the midlatitude jet streams.

[30] The capability of GPS RO for climate change monitoring and modeling was also tested based on observing system simulation experiments [Steiner et al., 2001; Stendel, 2006; Foelsche et al., 2008b]. Observational and sampling errors from those results are consistent with recent estimates of these errors from real RO data and those results suggested the errors being sufficiently small for detecting temperature trends in the global atmosphere within 10 to 20 years in most regions of the UTLS. RO-accessible atmospheric parameters (microwave refractivity, pressure/geopotential height, temperature) showed complementary climate change sensitivity in different regions of the UTLS suggesting that the combined information of key RO parameters for UTLS monitoring is of high value for climate studies.

[31] Climate signal detection times and constraints on climate benchmark accuracy requirements were investigated by Leroy et al. [2008]. They showed that detection time increases with measurement uncertainty and that it depends on the correlation time of natural variability and on satellite lifetime. Ringer and Healy [2008] examined the effect of increasing greenhouse gas concentrations on RO bending angle profiles based on climate model simulations. Their estimates of climate signal detection times in UTLS bending angle trends of 10 to 16 years are consistent with Leroy et al. [2008] and Foelsche et al. [2008b].

[32] Based on real data, validation of atmospheric analyses pointed out the value of RO high vertical resolution and accuracy not only for the investigation of the tropopause region [Gobiet et al., 2005; Schmidt et al., 2004, 2005; Borsche et al., 2007] but also for the improvement of atmospheric analyses themselves. Schmidt et al. [2008] inferred tropopause height trends from CHAMP RO data. They found a global increase of the tropopause height of 4 m to 7 m per year in the investigated period 05/2001–12/2007 in that data set.

[33] A first study demonstrating the utility of RO for climate change detection was performed by Steiner et al. [2009b] based on GPS/Met and CHAMP data (1995/1997/2001–2008); an update of the study with the data extended into 2010 and a further advanced study by Lackner et al. [2011] are presented in section 3.3. Schmidt et al. [2010] investigated trends in RO time series for the period 2001–2009 including CHAMP, GRACE, and F3C data. They computed tropopause heights based on bending angles and found a global increase of the tropopause height of 5 m to 9 m per year. Basically consistent with the findings of Steiner et al. [2009b], upper tropospheric warming and lower stratospheric cooling was noted.

[34] Furthermore, a broad range of studies has been undertaken that compared RO data to conventional upper-air data not only for validation purposes but also in view of the long-standing discussions of upper air trends [e.g., Thorne et al., 2011; Bengtsson and Hodges, 2011]. Comparison of RO to radiosonde data was able to reveal quality issues of different radiosonde types [e.g., Kuo et al., 2005; He et al., 2009]. Comparison of RO to (A)MSU satellite data [Schrøder et al., 2003; Ho et al., 2007; Steiner et al., 2007] in terms of bulk atmospheric temperatures was performed also including radiosonde data. The difference of anomalies of the (A)MSU data sets relative to RO showed a statistically significant trend pointing to a gradual divergence of the two data sets over 2001–2009 [Ladstädter et al., 2011].

[35] Calibration of temperature in the lower stratosphere from microwave measurements using COSMIC RO was investigated by Ho et al. [2009b]. Preliminary results demonstrated that RO data are very useful in identifying inter-satellite offsets among (A)MSU measurements from different satellites. Furthermore, due to its superior vertical resolution and assimilation without bias correction, GPS RO has reduced stratospheric temperature biases and has anchored radiance bias correction in numerical weather prediction (NWP). Its significant impact on weather forecasting [e.g., Healy et al., 2005; Healy and Thépaut, 2006; Aparicio and Deblonde, 2008; Cucurull and Derber, 2008; Cardinali, 2009; Rennie, 2010] has successfully been proven. Meanwhile all main weather centers worldwide operationally assimilate RO observations which also results in improved operational atmospheric analyses and climate re-analyses, the latter being frequently used in climate change studies.

3.2. RO Based Climate Change Indicators

[36] Leroy [1997] discussed geopotential height at constant pressure levels (proportional to tropospheric bulk temperature) as useful RO parameter for monitoring climate change. Vedel and Stendel [2003] pointed to refractivity as an ideal observable for climate detection. Leroy et al. [2006b] proposed to use refractivity as function of geopotential height, arguing that this is the more natural independent vertical coordinate. Schmidt et al. [2010] derived tropopause height trends from RO bending angles, following Lewis [2009]. For fully understanding of their results, temperature was additionally needed to interpret RO bending angle climatologies.

[37] Ladstädter et al. [2009] performed a systematic exploration of climate data sets with an innovative visualization technique (SimVis software) and identified regions and parameters reacting most sensitive to climate change. Subsequent statistical analysis by Lackner et al. [2009] assessed the set of RO parameters depending on season, region, and height domain to deduce the most sensitive and robust indicators in space and time, implying a high signal-to-noise ratio (SNR). Climate change indicators are defined here as climate trends that are statistically significant over (inter-annual) climate variability. Due to the still limited length of continuous RO measurements (starting in fall 2001), an ensemble of runs of global circulation models (GCMs) was used as proxy data. Simulations of three representative GCMs of the IPCC AR4 were used: CCSM3 from the National Centers for Environmental Prediction and the National Center for Atmospheric Research; ECHAM5 from the Max-Planck-Institute for Meteorology in Hamburg; HadCM3 from the Hadley Centre for Climate Prediction and Research of the UK MetOffice. Special Report on Emission Scenarios (SRES) A2 and B1 simulations were used. Based on monthly mean gridded fields for the period 2001–2050, large-scale zonal-mean regions (typical for single satellite RO studies) and also smaller regions were analyzed since RO multisatellite missions will also enable to provide regional climatologies (Table 1). Further data information and a detailed description of the study set up can be found in the works of Lackner [2010] and Lackner et al. [2009]. Refractivity, pressure, and temperature fields were analyzed at geopotential height levels as well as layer gradients of these parameters. The relative refractivity gradient can be regarded as mean bending angle for a layer and the relative pressure gradient as mean layer refractivity (closely proportional to density).

Table 1. IPCC+ Regions Defined and Extended After Christensen et al. [2007]
Zonal Bands
NHMNorthern Hemisphere Midlatitudes30°N–60°N180°W–180°E
SHMSouthern Hemisphere Midlatitudes60°S–30°S180°W–180°E
ALAAlaska, NW Canada60°N–72°N170°W–103°W
CGIEast Canada, Greenland and Iceland50°N–85°N103°W–010°W
WNAWestern North America30°N–60°N130°W–103°W
CNACentral North America10°N–30°N116°W–083°W
ENAEastern North America25°N–50°N085°W–050°W
CAMCentral America10°N–30°N116°W–083°W
SSASouthern South America56°S–20°S076°W–040°W
NEUNorthern Europe48°N–75°N010°W–040°E
SEMSouthern Europe Mediterranean30°N–48°N010°W–040°E
WAFWestern Africa12°S–18°N020°W–022°E
EAFEastern Africa12°S–18°N022°E–052°E
SAFSouthern Africa35°S–12°S010°E–052°E
NASNorthern Asia50°N–70°N040°E–180°E
CASCentral Asia30°N–50°N040°E–075°E
TIBTibetan Plateau30°N–50°N075°E–100°E
EASEastern Asia20°N–50°N100°E–145°E
SASSouthern Asia05°N–30°N065°E–100°E
SEASoutheast Asia11°S–20°N095°E–155°E
NAUNorthern Australia30°S–11°S110°E–155°E
SAUSouthern Australia45°S–30°S110°E–155°E
NPANorthern Pacific00°N–40°N150°E–120°W
SPASouthern Pacific55°S–00°S150°E–080°W
TNETropical Northeast Atlantic00°N–40°N030°W–010°W
MEDMediterranean Basin30°N–45°N005°W–035°E
INDIndian Ocean35°S–17.5°N050°E–100°E

[38] Results are presented in Figure 2 showing RO-accessible climate change indicators for annual means. Refractivity turned out as suitable indicator in the LS and tropopause (TP) region at about 14 km to 24 km in all large-scale regions except the polar caps. Pressure emerges as indicator for the whole UTLS (except over the polar caps). Global and hemispheric temperature means qualify consistently as indicators below about 16 km and above about 21 km. Most robust indicators are found in the tropics for all parameters as well as for layer gradients. When the boundary height levels of the layers used to calculate gradients are properly chosen, the layer gradients appear more sensitive to climate change than single level data, providing additional information. It is beneficial to use the whole set of RO-accessible parameters for climate monitoring in the UTLS since the sensitivity of the parameters to climate change differs with height. Specifically, going from raw to derived parameters, refractivity, pressure and the respective layer gradients are adequate for indicating climate change. In addition, temperature directly represents warming or cooling of the atmosphere and facilitates the interpretation of the results.

Figure 2.

Climate change indicators (i.e., significant trends) based on GCM simulations for the 2001–2050 period shown (from top to bottom) for the RO-accessible parameters refractivity, pressure, temperature as function of geopotential height, and for relative refractivity gradient, relative pressure gradient, and temperature gradient for UT, TP, LS. Annual means are depicted (left to right) for large-scale and zonal means to regional means, the latter as defined in the IPCC AR4 [Christensen et al., 2007] and detailed in Table 1. The significance level of positive trends (red) and negative trends (blue) is indicated by color shading. Trends were analyzed at geopotential height levels (above mean sea level), and corresponding pressure levels are shown for convenience.

3.3. RO Based Climate Trend Detection

[39] Steiner et al. [2009b] investigated the GPS/Met and CHAMP data with respect to their temperature change detection capability in the UTLS. We present an update of this study using the full available record of RO data until the end of 2010. We investigated monthly mean zonal-mean RO climatologies within 1995–2010 from GPS/Met (10/1995, 02/1997), CHAMP (09/2001–09/2008), GRACE (03/2007–12/2010), and F3C (08/2006–12/2010). For GPS/Met sufficient observations of good quality for building climatologies exist only for October 1995 and February 1997. We note that these early GPS/Met months are fully comparable with the later RO data in terms of quality and representativeness. We focus here on RO records for February (from years 1997 and 2002–2010). We test the trend detection capability in the UTLS within 9 km to 25 km (300 hPa to 30 hPa) and the region 50°S to 50°N. Error characteristics are fully accounted for. Observational error, residual sampling error, and a systematic error bound sum up to a total error of about 0.2 K in the temperature climatologies, only in regions of higher atmospheric variability it can reach 0.5 K [cf. Steiner et al., 2009b, Figure 1].

[40] Standard and multiple linear regression [von Storch and Zwiers, 1999] was applied to the temperature time series by taking into account the individual total climatological error for each month; for a detailed description see the auxiliary material of Steiner et al. [2009b]. Multiple linear regression included Quasi-Biennial Oscillation (QBO) and El Niño-Southern Oscillation (ENSO). QBO was represented by the monthly QBO indices of zonally averaged winds at 50 hPa and at 30 hPa over the equator. ENSO was represented by the monthly Nino 3.4 (N3.4) Sea Surface Temperature (SST) index. N3.4 and QBO indices are provided by the U.S. Climate Prediction Center (CPC) and available online at

[41] Figure 3 shows RO temperature anomalies for the full RO record 1995, 1997, 09/2001–12/2010, where the QBO pattern is clearly observable in the tropical LS (Figure 3, top), but fairly weak in the February time sequence (Figure 3, bottom). In our context the QBO pattern becomes diminished since we moreover average the LS over a large domain of 20°S to 20°N and of 100 hPa to 30 hPa. We nevertheless account for the QBO. A maximum correlation between QBO 50 hPa wind indices and RO temperature was found at a lag of 0 months [Lackner et al., 2011] in the tropical 10°S–10°N zonal band. QBO 30 hPa wind indices were used at a lag of 5 months due to downward phase propagation at a rate of about 1 km per month [Wallace et al., 1993] as seen in fully resolved RO data (Figure 3, top). However, the multiple regression results were found effectively insensitive to lagging of the QBO 50 hPa wind index from 0 to 3 months. A maximum correlation between N3.4 SST index and RO temperature anomalies in the UTLS was found at a lag of 4 months (temperature lagging ENSO) [Lackner et al., 2011], consistent with the results of Trenberth et al. [2002b] and Free and Seidel [2009].

Figure 3.

(top) Temperature anomalies for the RO record (10/1995, 02/1997, 09/2001–07/2010) show the QBO pattern in the tropical LS between 10°N to 10°S. (bottom) The February only time sequence.

[42] We tested whether the observed trend exceeds inter-annual variability in the study period. We also tested whether the trend exceeds long-term natural variability as estimated from pre-industrial control runs of the GCMs used (see section 3.2). Significance was assessed using a Student's t-test. Regression results are shown in Figure 4. A significant cooling trend (−2.64 K ± 0.20 K per 10 years) was detected relative to inter-annual variability (95% significance level) and to natural variability (99% significance level) in the tropical LS (100 hPa to 30 hPa) in February for the period 1997–2010 (Figure 4b). Inter-annual variability in the LS is explained mainly by an ENSO-related signal while the QBO signal (Figure 4a) vanishes due to the large averaging area. In the tropical UT (300 hPa to 200 hPa; Figure 4d), the trend (0.47 K ± 0.11 K per 10 years) is not significant since atmospheric variability is large due to a strong ENSO signal (Figure 4c). For the October series (1995, 2001–2010), trends in the LS (−1.31 K ± 0.19 K per 10 years) are not yet significant over variability though they show a clear cooling tendency, and trends in the UT (−0.04 K ± 0.09 K per 10 years) are not yet significant at all. We note that using February and October time series indicates no overall climatological trend. Steiner et al. [2009b] was rather a first demonstration of atmospheric change detection, showing the utility of RO for UTLS climate trend analysis.

Figure 4.

RO temperature anomalies (crosses) and trends (solid lines) accounting for ENSO and QBO are shown for February 1997/2002–2010 for the tropical (b) LS and (d) UT. Light gray shading indicates periods with RO data availability. Trends were calculated with multiple linear regression (light blue) including (a) monthly QBO indices of 30 hPa (lag 0; light colored) and 50 hPa winds (lag 5; dark colored) and (c) monthly N3.4 SST indices with a four month lag (black-highlighted bars). The N3.4 index exceeds 0.4 K for El Niño or −0.4 K for La Niña (dark gray). Regression model values are indicated by light blue diamonds. Trends from standard linear regression are shown in Figures 4b and 4d for reference (black).

[43] In a further study the whole record of RO-accessible parameters refractivity, geopotential height, and temperature was used to aim at detection of a forced climate signal by applying an optimal fingerprinting technique [Lackner et al., 2011]. The method can be considered as generalized multivariate regression y = Xa + u, where y contains the observed RO trends, X the forced ensemble GCM trends, a the scaling factors, and u the internal variability [Hegerl et al., 2007]. A climate trend is detected if its likelihood of occurrence due to internal variability alone is small. All calculations are performed in a dimension-reduced Empirical Orthogonal Function (EOF) space. Pre-industrial control runs of three global climate models (see section 3.2) were employed to estimate natural climate variability u. The expected forced climate change signal X was estimated from an ensemble mean of the three GCM climate simulations for the 20th century concatenated with SRES A2 and B1 simulations beyond 1999 (for a detailed description, see Lackner et al. [2011]).

[44] Figure 5 shows that the rebuilt RO trend patterns are consistent with the projections of the forced GCMs for temperature (Figure 5, left) and for geopotential height (Figure 5, right) for the inspected intermittent (1995/1997/2001–2010) (Figure 5, top) and continuous (2001–2010) periods (Figure 5, bottom). When only large-scale aspects of the temperature pattern are resolved, i.e., a small number of EOFs is used for the rebuilt pattern, 90% confidence is not yet achieved. When resolving smaller-scale aspects, i.e., a larger number of EOFs is used to rebuild the temperature pattern, temperature trends in the tropical LS are stronger than GCM trends. This allows for climate change detection at a 95% confidence level.

Figure 5.

(left) Temperature and (right) geopotential height trend patterns for (left panels) RO anomalies and (right panels) fGCM ensemble mean anomalies. (top) Rebuilt trend patterns are shown for the intermittent RO period (10/1995, 02/1997, 09/2001–07/2010) using 8 EOFs, which retain 96% and 99% of the total variability for temperature and geopotential height, respectively. (bottom) Rebuilt trend patterns for the continuous period (09/2001–07/2010) using 5 EOFs retain 88% of the total variability for temperature and 97% for geopotential height.

[45] For geopotential height of pressure levels an emerging climate change signal at a 90% confidence level was detected for both the intermittent and the continuous period, for the latter so far in a broad 50°S to 50°N band only. These UTLS geopotential height changes reflect an overall tropospheric warming. The detection is also consistent with expected detection times estimated from climate model simulations by Leroy et al. [2006b] and Ringer and Healy [2008], and from an observation system simulation experiment by Foelsche et al. [2008b].

[46] Overall the findings show an emerging trend signal in the RO climate record, which is expected to increase further in significance as the record grows over the coming years. Lackner et al. [2011] concluded that the small natural changes during the period (assessed in separate studies) suggest that the detected change is mainly caused by anthropogenic influence on climate. In this sense the RO record has provided the first formal detection of a climate change signal in the free atmosphere, since other existing upper air records did not have sufficiently favorable error characteristics.

3.4. Comparison of RO and Conventional Upper-Air Data

[47] Standard linear regression trend results are compared to trends in the radiosonde temperature record, which is the only independent UTLS profile data set available. We used radiosonde data from the Hadley Centre/MetOffice, UK (HadAT2) [Thorne et al., 2005b; Titchner et al., 2009] and newly homogenized records from the University of Vienna, Austria (RAOBCORE [Haimberger, 2007] and RICH [Haimberger et al., 2008], versions v1.5, L. Haimberger, personal communications, 2011). We also compared to an ensemble of 14/16-year running trends within 2001–2020 extracted from the GCM runs (see section 3.2). We used standard linear regression for the computation of trends in all data sets. Figure 6 shows tropical temperature profile trends for February 1997, 2002–2010 between 15°S to 15°N. Agreement between radiosonde data and RO data is given within uncertainty estimates. Climate model simulations basically agree as well but they show less warming/cooling contrast across the tropical tropopause and less cooling in the LS [cf. Steiner et al., 2009b].

Figure 6.

Tropical (15°S to 15°N) temperature profile trends at 6 pressure levels from 300 hPa to 30 hPa shown for February 1997/2002–2010. RO data (black) are compared to HadAT2 (green), RICH (blue), and RAOBCORE (cyan) radiosonde data, and to the mean (red) and individual 10 year trends (gray) of a multimodel multiple realizations data set of IPCC AR4 climate model runs within 2001–2020. Model error bars denote one standard deviation of the trends ensemble, and data error bars denote the error of the trend.

[48] Checking sensitivities, we investigated trend patterns (using standard linear regression) in 10°-zonal bands as presented in Figure 7. RO trend patterns are compared to the equivalent trend patterns in forced GCMs, the ERA-Interim reanalysis (ERA-INT) [Simmons et al., 2007] and the available gridded monthly radiosonde records RAOBCORE, RICH, and HadAT. Trends based on the February series are shown for the intermittent period including GPS/Met (Figure 7a, top) and for the continuous RO period (Figure 7a, bottom). They are compared to trends based on all available RO months intermittently (Figure 7b, top) and continuously (Figure 7b, bottom). Comparison of the intermittent February series shows overall consistent trend patterns between the data sets. Disregarding GPS/Met and inspecting only the continuous February series reveals qualitative agreement in trend patterns between the data sets in the LS. Also the trend patterns of the whole time series (intermittent and continuous) are similar above 150 hPa and exhibit comparable amplitudes.

Figure 7.

Temperature trend patterns (left to right) for RO, GCM, ERA-Interim, RAOBCORE, RICH, and HadAT data. Shown are the (a) February only trends of the intermittent RO period (1997/2002–2010, including GPS/Met) (Figure 7a, top) and of the continuous RO period 2002–2010 (Figure 7a, bottom), as well as the (b) full trends of the intermittent RO period (10/1995, 02/1997, 09/2001–07/2010) (Figure 7b, top) and of the continuous period (09/2001–07/2010) (Figure 7b, bottom).

[49] Below 150 hPa the comparison data sets differ from RO and among themselves. ERA-Interim shows a characteristic warming/cooling feature above/below 150 hPa, which is a peculiarity of this reanalysis [Poli et al., 2010]. GCM trends overall show a continuous warming in the UT and a cooling in the LS with smaller trend amplitudes than the observational data sets. The similarity of the overall LS cooling within 20°S to 20°N from the continuous period and the intermittent period of the February time series (Figure 7a) and the whole time series (Figure 7b) lends confidence that the RO trend results of Figure 4 are credible. Further confirmation is provided by the consistency of detection results between the long and short periods by Lackner et al. [2011], though with less significant but similar (i.e., robust) signals for the shorter period.

[50] A comparison of monthly mean LS temperatures (TLS) of CHAMP RO data with (A)MSU was carried out by Steiner et al. [2007]. They found significant anomaly differences in the tropical region, also confirmed by Ho et al. [2007], who then demonstrated calibration of (A)MSU data using RO [Ho et al., 2009b]. Known error sources (errors in RO data, TLS computation procedure) were found insufficient to account for the differences [see also Steiner et al., 2009a]. Ladstädter et al. [2011] updated and extended the study with most recent data sets. TLS equivalent anomalies for RO were computed by applying the radiative transfer model RTTOV-9_3 [Saunders et al., 2010a, 2010b]. Comparison was performed to the real (A)MSU records from University of Alabama in Huntsville (UAHv5.4) and Remote Sensing Systems (RSSv3.3) [Mears and Wentz, 2009]. Furthermore, Ladstädter et al. [2011] compared RO to a newly reconstructed (A)MSU record (STARv2.0) provided by the National Environmental Satellite, Data, and Information Service (NESDIS) Center for Satellite Applications and Research (STAR) [Zou et al., 2009; Zou and Wang, 2010]. Finally, TLS equivalents of radiosonde data were provided for RAOBCOREv1.5 and RICHv1.5 by L. Haimberger (University of Vienna, Austria, personal communications, 2011).

[51] The spatial sampling characteristics were taken into account for both radiosonde and RO data in a consistent way. Considering the sampling error for radiosondes is important in regions with sparse station coverage, and generally because the inhomogeneous distribution of radiosonde stations does not fully capture the atmospheric variability [Ladstädter et al., 2011]. Considering the sampling error reduces the variability in the radiosonde temperature anomalies by up to 50% in the inspected regions. This reduces the standard error of the trend differences and thus provides a more accurate estimate of the data set deviations.

[52] Figure 8 shows 2001–2010 monthly mean time series of (A)MSU TLS anomalies and anomaly differences relative to RO for the tropics (20°S to 20°N) and for the near-global region (70°S to 70°N). Computation of anomaly differences removes the climatological variability common to both data sets and reveals differences due to measurement and processing. For TLS anomalies the agreement in intraannual variability of UAH, RSS, STAR, and radiosonde data with RO is very good. Comparing radiosondes to RO shows very good agreement globally and in all considered regions (tropics, Northern Hemisphere extratropics, and Southern Hemisphere extratropics), with the minor exception of RICH in the tropics and Northern Hemisphere extratropics. The anomaly difference trend values are summarized in Table 2.

Figure 8.

Comparison of monthly mean lower stratospheric temperatures (TLS) for (A)MSU, radiosonde, and RO records for September 2001 to December 2010. Shown are (left) de-seasonalized TLS anomalies relative to 2002–2010, and (right) the anomaly differences and their trends for (top) the tropics (20°S to 20°N) and for (bottom) the near-global region (70°S to 70°N).

[53] Significant differences between (A)MSU and RO TLS temperature anomaly time series were found (global trend of −0.20 K to −0.22 K per 10 years), with the STARv2.0 record showing the largest trend difference. Overall the trend differences are smaller compared to Steiner et al. [2007], since the time series is longer and the negative trend tendency of the (A)MSU record ceased to continue in 2006. The conclusion of Steiner et al. [2007] remains still valid, also based on the extended data set used here, that further understanding and reconciliation of anomaly differences is needed.

Table 2. Trends of Anomaly Differences for the Period September 2001 to December 2010a
Data SetsTrend (K/10 years)Detrended Standard Deviation (K)
  • a

    The ± value defines the 95% confidence intervals for the trends. Trend values which are significantly different from 0 at the 95% level are marked by a double asterisk.

20°S to 20°N
RSS–RO−0.215 ± 0.058**0.08
UAH–RO−0.216 ± 0.061**0.09
STAR–RO−0.242 ± 0.071**0.10
RICH–RO−0.129 ± 0.089**0.13
RAOBCORE–RO−0.012 ± 0.0830.12
70°S to 70°N
RSS–RO−0.206 ± 0.043**0.06
UAH–RO−0.198 ± 0.042**0.06
STAR–RO−0.220 ± 0.045**0.07
RICH–RO−0.013 ± 0.0650.10
RAOBCORE–RO+0.080 ± 0.061**0.09
30°N to 70°N
RSS–RO−0.252 ± 0.045**0.06
UAH–RO−0.256 ± 0.051**0.07
STAR–RO−0.286 ± 0.054**0.08
RICH–RO−0.125 ± 0.063**0.09
RAOBCORE–RO+0.014 ± 0.0610.09
70°S to 30°S
RSS–RO−0.159 ± 0.052**0.08
UAH–RO−0.118 ± 0.055**0.08
STAR–RO−0.124 ± 0.056**0.08
RICH–RO+0.034 ± 0.1390.20
RAOBCORE–RO−0.006 ± 0.1390.20

4. Conclusions and Outlook

[54] The Global Climate Observing System (GCOS) program defined basic monitoring principles and guidelines for the generation of fundamental climate data records and essential climate variable products. Observations from surface-based, airborne, and satellite-based instruments should fulfill the GCOS requirements, which comprise long-term stability, homogeneity, reproducibility, global coverage, accuracy, resolution, description and validation of the products. Since conventional upper-air climate records show shortcomings in meeting the GCOS requirements there is need for more accurate and more diverse data.

[55] GPS RO is an independent climate record of high quality for monitoring the Earth's atmosphere. We reviewed the characteristics of RO, which are advantageous for climate studies. RO provides global observations in virtually all-weather conditions with high accuracy and resolution of 0.5 km to 1.5 km in the UTLS. Precise time measurements guarantee SI-traceability, long-term stability, and reproducibility. Data from different RO missions can be combined without inter-calibration and overlap. Intercomparison of RO data from different processing centers confirmed the consistency of RO data products. Structural uncertainty was found to be low with <0.03% per 5 years for refractivity trends in large-scale means, corresponding to <0.06 K per 5 years for temperature trends. Intensive work is currently ongoing to provide structural uncertainty estimates for all RO products involving six main RO processing centers worldwide.

[56] Based on retrieved profiles of bending angle, refractivity, pressure, geopotential height, and temperature, accurate atmospheric climatologies can be constructed and error estimates for these RO products can be provided. The observational RMS error for single RO profiles is estimated to be 0.7 K to 1 K at 8 km to 25 km altitude for current RO missions. It becomes negligible in climatologies when averaging hundreds of profiles. Systematic errors in the UTLS mainly arise from initialization and ionospheric residuals, and an upper bound of about 0.2 K below 25 km in the LS was estimated. At the Wegener Center we provide climatology products together with respective error estimates. The sampling error due to uneven and sparse sampling in space and time is <0.3 K in the UTLS and can be subtracted from RO climatologies, the residual is <0.1 K for vertically resolved data and <0.03 K for UTLS means. The total error budget including observational error, sampling error, and a systematic error bound sums up to within 0.15 K in temperature at low to mid latitudes.

[57] A review of studies related to climate change monitoring and detection based on GPS RO was given as well as an update on the status in the field. An overview on climate applications was presented from climate change indicator studies via trend detection to upper-air record comparisons. The results showed that the set of RO-accessible parameters covers the whole UTLS with useful climate change indicators. We found the RO data capable to start detecting significant UTLS climate trends in geopotential height and temperature, consistent with expected detection times. Trend results from the current RO record are basically consistent with model data and with recent radiosonde records; RO trends are tentatively more pronounced and show more warming/cooling contrast over the tropical UT/LS. Significant temperature anomaly differences were found comparing to (A)MSU data, which is investigated in ongoing work.

[58] We showed that climate change signals can be detected for the still short record if regions are adequately chosen, i.e., averaging over large zonal bands and investigating regions with low atmospheric variability such as the tropics, where earlier detection is possible. However, of main importance was the demonstration of the accuracy, stability, homogeneity, and high quality of the RO record for climate signal detection purposes. Furthermore, it is necessary to have several independent measurements of the same atmospheric variable in order to estimate overall structural uncertainties. The performance of the RO record underpins its capability to meet GCOS requirements on climate monitoring and to become a climate benchmark record in a future global climate observing system.


[59] The authors acknowledge UCAR/CDAAC (Boulder, CO, USA) for RO excess phase and orbit data, and ECMWF (Reading, UK) for atmospheric analysis, reanalysis, and forecast data. We thank UAH (Huntsville, AL, USA), RSS (Santa Rosa, CA, USA), and NOAA/NESDIS (Camp Springs, MD, USA) for AMSU data, Hadley Centre/MetOffice (Exeter, UK) for HadAT2 data, and L. Haimberger (University of Vienna, Austria) for RAOBCORE and RICH data. We acknowledge the modeling groups, the PCMDI and WRCP's Working Group on Coupled Modeling (WGCM) for access to CMIP3 multimodel data as well as the Climate Prediction Center (U.S.) for ENSO and QBO index data. J. Fritzer (Wegener Center) is thanked for his efforts in OPS system development and operations. The work was funded by the Austrian Science Fund (FWF) grants P21642-N21 and P22293-N21 and regarding OPS development by ESA/ESTEC Noordwijk and FFG/ALR Austria. B. Scherllin-Pirscher was funded by the National Science Foundation under cooperative agreement AGS-0918398/CSA, AGS-0939962. The National Center for Atmospheric Research is supported by the National Science Foundation.