The Global Navigation Satellite System (GNSS) Occultation Sounder (GNOS) is one of the new-generation payloads onboard the
Chinese FengYun 3 (FY-3) series of operational meteorological satellites for sounding the Earth's neutral atmosphere and
ionosphere. The GNOS was designed for acquiring setting and rising radio occultation (RO) data by using GNSS signals from
both the Chinese BeiDou System (BDS) and the US Global Positioning System (GPS). An ultra-stable oscillator with 1 s
stability (Allan deviation) at the level of
The radio occultation (RO) technique (Melbourne et al., 1994; Ware et al., 1996) using signals from global navigation satellite systems (GNSSs), in particular from GPS, has been widely used to observe the Earth's atmospheric parameters (e.g., bending angle, refractivity, temperature, pressure, and water vapor) for applications such as numerical weather prediction (NWP; e.g., Healy and Eyre, 2000; Kuo et al., 2000; Healy and Thepaut, 2006; Aparicio and Deblonde, 2008; Cucurull and Derber, 2008; Poli et al., 2008; Huang et al., 2010; Le Marshall et al., 2010; Harnisch et al., 2013) and global climate monitoring (GCM; e.g., Steiner et al., 2001, 2009, 2011, 2013; Schmidt et al., 2005, 2008, 2010; Loescher and Kirchengast, 2008; Ho et al., 2009, 2012; Foelsche et al., 2011a; Lackner et al., 2011).
The RO concept was experimentally tested by the first experimental Global Positioning System/Meteorology (GPS/MET) mission launched in 1995 right after full operational capacity of GPS was achieved (Ware et al., 1996; Kursinski et al., 1996; Kuo et al., 1998). GPS/MET has demonstrated the unique properties of the GPS RO technique, such as high vertical resolution, high accuracy, all-weather capability and global coverage (Ware et al., 1996; Gorbunov et al., 1996; Rocken et al., 1997; Leroy, 1997; Steiner et al., 1999).
The subsequent low Earth Orbit (LEO) satellite missions such as the CHAllenging Minisatellite Payload (CHAMP; Wickert
et al., 2001, 2002), the Constellation Observing System for Meteorology, Ionosphere and Climate (COSMIC; Anthes et al.,
2000, 2008; Schreiner et al., 2007), the Gravity Recovery And Climate Experiment (GRACE; Beyerle et al., 2005; Wickert
et al., 2005), and the Meteorological Operational (MetOp; Edwards and Pawlak, 2000; Luntama et al., 2008) satellites have
further affirmed the long-term stability and remarkable consistency (e.g.,
The development of GNSS such as China's BeiDou Navigation Satellite System (BDS), Russia's GLObal NAvigation Satellite System (GLONASS), and the European Galileo system, has significantly enhanced the availability and capacity of the GPS-like satellites which will make RO even more attractive in the future. These new GNSS navigation satellites together with planned LEO missions will offer many more RO observations. One of these LEO missions is the FengYun 3 series C satellite (FY-3C), carrying China's first GNSS, Occultation Sounder (GNOS). FY-3C was successfully launched on 23 September 2013 (Bai et al., 2014; Liao et al., 2016).
FY-3C GNOS, developed by National Space Science Center/Chinese Academy of Sciences (NSSC/CAS), is the first BDS/GPS compatible sounder and combines a state-of-the-art RO receiver with an ultra-stable oscillator. The future satellites of the Chinese FY-3 series of operational meteorological satellites, the next being FY-3D, scheduled for launch in 2017, will also carry GNOS instruments, similar to the MetOp series of European satellites with its GNSS Receiver for Atmospheric Sounding (GRAS) instruments (Loiselet et al., 2000).
Components of the GNOS instrument (setting/rising occultation antenna and RF unit,
left/right; navigation antenna and RF unit, middle in front; tracking and data processing unit,
middle in back)
The GNOS instrument consists of three antennas, three radio frequency (RF) units and a data processor (Fig. 1a), which
uses high-dynamic, high-sensitivity signal acquisition and tracking techniques, in which the navigation antenna has a stable phase. Additionally, the different features of BDS and GPS signals have been taken into account in the GNOS
design. The GNOS can observe the atmosphere and ionosphere, and its detection height range is from Earth's surface to
around 800
Regarding the excess phase processing, a single-difference method removes the LEO satellite clock offset by the difference between the GNSS occultation satellite and its GNSS reference satellite (Wickert et al., 2002). Comparing with the original double-difference method (Ware et al., 1996; Rocken et al., 1997), the single-difference method uses the solved GNSS satellite clock offset estimates instead of further differencing between the GNSS satellites and a GNSS ground station; hence, the single-difference method can minimize the effects of ground data error sources (Hajj et al., 2002; Schreiner et al., 2010). Because single differencing (SD) needs no ground station data, the processing is simpler and easier to realize. Therefore the single-difference approach has become widely used in RO data processing after the switch-off of the GPS “selective availability” (SA) mode as of May 2000 (Hajj et al., 2002), which made GPS clock offset estimation sufficiently reliable.
Even more recently, zero-difference processing was started to be used (Beyerle et al., 2005; Wickert et al., 2005), which can compute excess phase data by applying prior estimated LEO and GNSS clock offsets without need of a reference satellite or ground station. However, it requires that the LEO receiver is equipped with an ultra-stable oscillator that, so far, was only available for the GRACE and MetOp missions (Beyerle et al., 2005; Luntama et al., 2008). The FY-3 GNOS instrument is equipped with such an ultra-stable oscillator as well.
BDS is China's global navigation satellite system designed to provide global coverage starting around 2020, with positioning,
navigation, timing, and short-message communication service capabilities (Li, 2016). So far, BDS can provide good regional
coverage in the Asia-Pacific area with an incomplete constellation, by using two L band frequencies,
B1I
This still growing constellation also provides a practical motivation for zero differencing (ZD) because not all of the FY-3C GNOS BDS RO events can be processed by single differencing, since the incomplete BDS system cannot provide reference satellites for all RO events. On the other hand, the ultra-stable oscillator driving the GNOS receiver makes zero differencing attractive to be potentially used as the method of choice for all BDS RO events. To investigate the feasibility of the zero-difference algorithm for BDS RO data processing, and to evaluate the quality of the retrieved RO data products, we therefore perform in this study a comparative analysis of zero- and single-difference processing for GNOS.
The paper is structured as follows. Section 2 provides a description of our single- and zero-difference excess phase processing. Section 3 presents the FY-3C GNOS data sets and the methods for the inter-comparison analysis. Section 4 presents the statistical analysis results for the various reference data sets. Finally, conclusions are drawn in Sect. 5.
Excess phase is a key variable during the radio occultation data processing, and GNSS satellite and LEO satellite clock errors are main factors affecting the excess phase accuracy. As summarized above, these two clock error components can either be eliminated by double differencing or (for GPS after the SA mode has been deactivated) the GNSS clock errors can be estimated and subtracted, and so single differencing can be applied, or (given an ultra-stable oscillator at the LEO) both clock errors can be estimated and subtracted, and so zero differencing is possible. Recently, because of its higher complexity and degraded accuracy, double differencing has rarely been used. In this section we describe the single- and zero-difference procedures which we used for the FY-3C GNOS excess phase processing.
The GNOS RO excess phase processing determines the total excess phase, which is caused by both the atmosphere and ionosphere, of the GPS L1, L2 and BDS B1, B2 signals as a function of coordinate (GPS) time in the Earth-centered inertial (ECI) true-of-date (TOD) reference frame. The inputs to the processing are GPS, BDS and LEO satellite positions, velocities and clock offsets as a function of coordinate time, LEO satellite attitude information, carrier-phase measurements, antenna-phase center information and Earth orientation information.
Schematic geometry of GNSS radio occultation for single differencing (using link
The outputs of this process include GPS time of the RO event observations, where we adopt the LEO's signal reception time,
GPS L1, L2 and BDS B1, B2 total excess phases, position and velocity of the LEO satellite at signal reception time and
position and velocity of the GNSS satellite at signal transmission time. Hereafter, we will use the term GNSS to refer to
GPS and BDS satellites, as well as
The observed carrier-phase
As it is needed for single-difference processing only, the carrier phase observable
The geometric range
The GNSS satellite orbits (positions and velocities) and the GNSS clock offset estimates
The gravitational delay
The phase wind-up correction term
In the single-difference processing, we use Eq. (
Next, the excess phase of the reference link (which is only an ionospheric excess phase
Finally, the effects of the receiver clock,
The single-difference approach has some advantages over the double-difference approach, as noted in the introduction, and has therefore been widely used in GPS RO data processing. However, it is difficult to find a suitable reference satellite for each RO event to calculate the excess phase using single-difference when the GNSS space segment is still an incomplete constellation, as with the current BDS.
Zero differencing also will likely produce lower-noise excess phase data than single differencing, from applying the estimated LEO clock offsets and avoiding the use of a reference link (being an additional error source). It can be employed if the LEO receiver is equipped with an ultra-stable oscillator such as in the case of the GNOS instrument.
In the zero differencing (ZD) approach we employ Eq. (
As mentioned, the single-difference approach involves a GNSS reference satellite, which normally has high signal-to-noise ratio (SNR) and high-phase measurement accuracy. In order to use the specific reference satellite most likely to have the best signal quality and lowest ionospheric influence, our FY-3C GNOS receiver software chooses the GNSS satellite with highest elevation angle seen by the navigation (zenith) antenna. For reasons of robustness and for ensuring best consistency, here we only use BDS reference satellites for BDS occultations (likewise GPS reference satellites only for GPS occultations).
The largest gain and half-power beam width of GNOS's POD antenna is 5
Statistics of FY-3C GNOS BDS carrier-phase SDs (blue for B1 signal carrier phase; red for B2 signal carrier phase) as a function of elevation angle, calculated by using positioning channel measurements.
Figure 3 illustrates the GNOS in-orbit testing results of the BDS B1 and B2 carrier-phase observation error SD, as
a function of elevation angle. As can be clearly seen, both the B1 and B2 carrier-phase measurement errors decrease with
increasing elevation angle. At elevation angles larger than 10
Histogram of the maximum elevation angle of the BDS reference satellites, with the statistics based on the 13 564 BDS RO events that occurred over October–December 2013.
Applying this 10
To evaluate the performance of the zero- and single-difference methods, we have conducted a comparison analysis of the retrieved FY-3C GNOS BDS RO bending angle and refractivity data for the selected 92 days from 1 October to 31 December 2013, retrieved by either including the single-difference or zero-difference method in the excess phase processing.
Geographical and local time distribution of the GNOS BDS RO events that have proper BDS reference satellites for single-difference processing (red from the BDS-GEO satellites; blue from BDS-IGSO; green from BDS-MEO; numbers in parentheses denote the associated number of events during October–December 2013). Distributions are shown as a function of latitude and longitude
In our data processing of bending angle and refractivity, a quality control algorithm has been used (which for single differencing reduced the profile data set by about 2 % and for zero differencing by less than 1 %). The processing statistics we obtain show that, after quality control, the number of RO events obtained by zero differencing is higher by about 13 % than the one obtained by single differencing, which we find is due to some ineffective reference BDS satellite links during the single-difference processing. The geographic and local time distribution of the RO events that also have proper BDS reference satellites for single-difference processing is shown in Fig. 5.
Mean difference (bias) and standard deviation (SD) statistics of GNOS-derived refractivity (
Figure 5a shows that the geographic distribution of events well reflects the different BDS orbit types. BDS-GEO RO events mainly distribute in the Southern and Northern Hemisphere high-latitude zones along the longitude sector of the Chinese region. The number of BDS-IGSO RO events is highest, almost equal to the number of GEO and MEO RO events together. The BDS-IGSO RO event coverage forms a quasi-global “8” shape, with the larger oval over the South American, Pacific, and Atlantic Ocean areas, and the somewhat smaller oval over Southeast Asia, Northwest Australia, Pacific, and Indian Ocean areas. Similar to the typical distribution of GPS RO events (e.g., Pirscher et al., 2007; Anthes et al., 2008), the BDS-MEO RO events show essentially global coverage, with more RO events in the mid- and high-latitude zones and less at low latitudes.
Figure 5b and c show the distribution of the RO events in a complementary way with focus on local time, again reflecting
well the different BDS orbit types and their impact on RO event locations in space and time. It can be seen that the
BDS-GEO RO events occur during all 24 h of the day, while the BDS-IGSO and BDS-MEO RO events distribute mainly in the
09:00–11:00 and 21:00–23:00 LT ranges (best seen in Fig. 5c). In particular, at low and middle latitudes, equatorward of
about 50 to 60
The distribution of the GNOS BDS RO events processed by using zero differencing (not separately shown) is very similar to Fig. 5, though with slightly more RO events (2623 BDS-GEO, 4820 BDS-IGSO and 2863 BDS-MEO) passing the quality control.
Before the validation against the GNOS-independent reference data from the European Centre for Medium-Range Weather Forecasting (ECMWF) and radiosondes, we did a cross-check of the quality of the BDS RO events based on a limited ensemble of co-located profiles undertaken between GNOS BDS and GPS RO events.
Figure 6 shows the results of our inter-comparison of retrieved refractivity profiles from zero differencing and
single differencing for BDS against the single-differencing results for GPS (zero-differencing GPS data were not
available). A reasonably high consistency of the BDS- and GPS-derived RO profiles is found: the BDS refractivities
both from zero and single differencing appear essentially unbiased against the GPS refractivities within 5 to
25
Mean difference (bias) and standard deviation (SD) statistics of the GNOS bending-angle profiles retrieved by using excess phases from the single-difference processing
Mean difference (bias) and standard deviation (SD) statistics of the GNOS refractivity profiles retrieved by using excess phases from the single-difference processing
For producing the statistical validation analysis results compared to the independent reference data, we calculated the
fractional error of the retrieved bending-angle (
As reference data we used analysis data from the European Centre for Medium-Range Weather Forecasts (ECMWF) as well as radiosonde data obtained from the global radiosonde archive of the National Oceanic and Atmospheric Administration–National Centers for Environmental Information (NOAA-NCEI).
The ECMWF analysis data were used as 6-hourly fields (00:00, 06:00, 12:00, 18:00 UTC time layers every day) at
a horizontal resolution of about 300
The radiosonde profiles had about 0.5 to 3
The target domain for the comparative statistical analysis is from 5 to 35
Figure 7 shows the statistics of the GNOS BDS RO bending-angle results for the different BDS subsystems (GEO, IGSO, MEO)
and the full BDS (total), for both single differencing (Fig. 7a) and zero differencing (Fig. 7b). The bias and SD profiles
have been calculated from the large ensembles of these event data sets, based on the fractional difference profiles
according to Eq. (
In line with expectations, the biases and SDs are slightly smaller for the zero differencing than for the
single differencing (though even smaller SDs might be expected from avoiding the reference link computation; e.g.,
Schreiner et al., 2010), but in general they are very similar. Both cases show a small negative bias of around
Several aspects of small visible differences (e.g., specific difference of GEO results from the other results, increasing
SD differences below 10
Overall the results confirm the high quality of the GNOS retrievals, in line with recent results by Liao et al. (2016), and a robust zero-difference processing being a viable alternative to the single-difference processing. The results also indicate that the BDS retrievals can achieve quality comparable to what is well established for GPS retrievals. Thanks to the diversity of BDS orbits, we can also demonstrate RO retrievals from occultations with GNSS transmitters not in medium Earth orbit (MEO). The results clearly indicate that the GNSS transmitters in GEO and IGSO can also provide quality comparable to that of MEO.
Mean difference (bias) and standard deviation (SD) statistics of the GNOS refractivity profiles retrieved by using either zero differencing (red) or single differencing (blue), with collocated radiosonde refractivity profiles used as reference (
Figure 8 shows the statistics of the GNOS BDS RO refractivity results, again for the different BDS subsystems (GEO, IGSO,
MEO) and the full BDS (total), for both single differencing (Fig. 8a) and zero differencing (Fig. 8b). The bias and SD
profiles have been calculated from these large BDS event ensembles based on the fractional refractivity difference
profiles according to Eq. (
Similar to the bending-angle results (Fig.
As for the bending angles, aspects of small visible differences of refractivity, such as more deviation of the GEO results, will also be explored by detailed excess phase processing and retrieval error analyses as part of follow-up work.
Overall the refractivity results confirm the messages summarized in Sect. 4.1 based on the bending-angle results. That is, they underline the high quality of the GNOS BDS retrievals as being (nearly) comparable to GPS retrievals, the robustness of both the zero- and single-difference processing and the reliable retrieval quality of RO events with GNSS transmitters on MEO satellites as well as on GEO and IGSO satellites.
Figure 9 shows the single- and zero-difference results for refractivity statistics, bias and SD profiles, against collocated radiosonde profiles and only for the whole set of BDS RO events, since the number of collocations is more limited. The number of RO events entering into the statistics is also strongly height-dependent in this case and is therefore shown as (maximum) number in the legend in addition to as height profiles in the side panel (Fig. 9, right).
Given the smaller ensemble size of about 50 to 200 events (depending on height), and the less strict collocation and thus
somewhat higher representativeness error than for the ECMWF data extracted at the mean RO event location, these
refractivity results are expected to exhibit somewhat more deviations than those in Fig. 8. As Fig. 9 shows, the bias is
nevertheless still fairly small, near
In summary, the comparison to this entirely independent radiosonde data set underpins the finding that both zero and single differencing are robust and that the GNOS BDS retrievals can perform similarly to GPS retrievals. The latter were established to compare well to quality radiosondes (Anthes, 2011; Ladstaedter et al., 2015).
In this study we have introduced our single- and zero-difference excess phase processing of BeiDou System (BDS) RO data of the FY-3C GNOS mission and evaluated the quality of atmospheric profiles derived from this single- and zero-difference processing.
The single differencing can correct the receiver clock offset, and thus it has lower requirements on the receiver clock stability. However, it requires a proper reference GNSS satellite and will induce some of this reference satellite's positioning and carrier-phase measurement errors into the RO processing. The advantage of the zero-difference algorithm is its independence from reference satellites, but it requires a receiver clock of very high quality (ultra-stable oscillator such as available for GNOS) to obtain a highly accurate receiver clock offset estimate, which nevertheless can leave some residual errors after the clock offset correction.
Because BDS is still a regional navigation system, we found that about 20 % of the GNOS BDS RO events do not
have proper reference satellites for single differencing, providing another argument for a zero-difference alternative. We
performed a comparative analysis of the zero-difference and single-difference excess phase processing chains for the FY-3C
GNOS BDS RO observations, in which independent reanalysis data from ECMWF and collocated high-quality data from
radiosondes have been used as reference for evaluating the retrieved bending-angle and refractivity profiles over the
upper troposphere and lower stratosphere (UTLS, 5 to 35
The results showed that the GNOS BDS RO profiles derived by using both the zero-difference and single-difference
algorithms exhibit very good consistency with the ECMWF and radiosonde data. The zero-difference method appeared to
perform slightly better than the single-difference method, especially visible at stratospheric altitudes (above
15
Compared to ECMWF data, the average UTLS bending-angle bias was found to be near
Overall these results indicate the high quality of the GNOS BDS retrievals, and that robust zero-difference processing is a viable alternative to single-difference processing. The results also indicate that the BDS retrievals can achieve quality comparable to the established GPS retrievals. Based on the diversity of BDS orbits, we also demonstrated for the first time RO retrievals from occultations with GNSS transmitters not in MEO. We also found that the GNSS transmitters in GEO and in IGSO provide quality comparable to the ones on MEO satellites.
Currently, the GRAS onboard the European meteorological satellite series MetOp and the GNOS occultation receiver onboard
the Chinese meteorological satellite series FY-3 are the only two RO instruments for long-term operational observations that
include an ultra-stable crystal oscillator featuring a very high-quality Allan deviation at the 10
The software code used for this study is not in the public domain and cannot be distributed. To access the relevant result files of this study, please contact the corresponding author.
The authors declare that they have no conflict of interest.
This article is part of the special issue “Observing Atmosphere and Climate with Occultation Techniques – Results from the OPAC-IROWG 2016 Workshop”. It is a result of the International Workshop on Occultations for Probing Atmosphere and Climate, Leibnitz, Austria, 8–14 September 2016.
This research was supported by the National Natural Science Foundation of China (grant nos. 41775034, 41505030, 41405039, 41405040
and 41606206), the Strategic Priority Research Program of Chinese Academy of Sciences (grant no. XDA15012300) and by the FengYun 3 (FY-3) Global Navigation Satellite System Occultation Sounder (GNOS) development and
manufacture project led by NSSC/CAS. The research at WEGC was supported by the Austrian Aeronautics and Space Agency of
the Austrian Research Promotion Agency (FFG-ALR) under projects OPSCLIMPROP (grant no. 840070) and OPSCLIMTRACE (grant
no. 844395). The ECMWF (Reading, UK) is thanked for access to their archived analysis and forecast data (available at