First results from an airborne GPS radio occultation system for atmospheric profiling


  • The copyright line for this article was changed on 23 March 2015 after original online publication.


Global Positioning System (GPS) radio occultation (RO) from low Earth-orbiting satellites has increased the quantity of high-vertical resolution atmospheric profiles, especially over oceans, and has significantly improved global weather forecasting. A new system, the Global Navigation Satellite Systems Instrument System for Multistatic and Occultation Sensing (GISMOS), has been developed for RO sounding from aircraft. GISMOS also provides high-vertical resolution profiles that are insensitive to clouds and precipitation, and in addition, provides greater control on the sampling location, useful for targeted regional studies. The feasibility of the system is demonstrated with a flight carried out during development of an Atlantic tropical storm. The data have been evaluated through a comparison with dropsonde data. The new airborne RO system will effectively increase by more than 50% the number of profiles available for studying the evolution of tropical storms during this campaign and could potentially be deployed on commercial aircraft in the future.

1 Introduction

GPS radio occultation uses radio signals to sense the atmosphere as a transmitting GPS satellite sets behind or rises above the horizon relative to a moving receiver (Figure 1). The radio waves undergo refractive bending and a Doppler shift due to variations of refractive index within the atmosphere primarily in the vertical direction. The refractive index in the neutral atmosphere depends on the pressure (hPa), P, temperature (K), T, and water vapor partial pressure (hPa), e,

display math(1)

[Rüeger, 2002]. Thus, information on the structure of the atmosphere can be retrieved from precise measurements of the amplitude and Doppler shift of the radio waves.

Figure 1.

Airborne radio occultation geometry. GPS signals are recorded as a GPS satellite sets, with raypaths successively sampling deeper into the atmosphere or successively shallower as a GPS satellite rises. The tangent points are the points of closest approach of each raypath to the Earth's surface. The consecutive points are indicated by red dots, which drift horizontally away from the aircraft as altitude decreases for a setting satellite. The atmosphere is most dense at the tangent point, so the measured refractive delays are most strongly influenced by the atmospheric properties at this location. Retrievals of atmospheric refractivity are represented as values along these slanted tangent point profiles.

GPS/MET was the first mission to demonstrate that profiles could be retrieved for the Earth's atmosphere using this technique [Kursinski et al., 1996]. Since then, several missions have been launched, most notably the Constellation Observing System for Meteorology Ionosphere and Climate (COSMIC)/Formosa Satellite-3 (FORMOSAT-3) constellation [Anthes et al., 2008], which provides radio occultation (RO) observations on a routine basis. Operationally, spaceborne RO data have improved global numerical weather model forecasts [Healy and Thepaut, 2006; Poli et al., 2008]. However, RO data have only brought limited improvement on regional short-term weather simulations and forecasts of high-impact weather in tropical storms [Kueh et al., 2009]. One reason for this is that for targeted areas, spaceborne RO sampling in time and space is still relatively sparse. For example, in a case study of the 2006 Typhoon Kaemi [Chen et al., 2009], only seven RO profiles were available within the 6000 km × 6000 km model domain and within ± 3 h of the initialization time. The closest profile was more than 600 km from the typhoon center, and the overall impact on the forecast track and rainfall was inconclusive. Other studies have investigated the impact of spaceborne RO in forecasting tropical cyclones [Huang et al., 2005; Kueh et al., 2009] and have shown that localized regions where the moisture increment is high relative to the first guess model have produced significant changes in low-level and midlevel circulations. In a study showing positive results for the 2006 Hurricane Ernesto, assimilation of GPS RO profiles during the 2 days before hurricane genesis resulted in an improvement of the hurricane intensification forecast [Liu et al., 2012]. As in several other cases, the improvement was shown to be a result of assimilating only a few key RO profiles, by chance located nearby and upstream of the storm [Huang et al., 2010; Liu et al., 2012].

While the results of some of these individual cases are promising, the number of cases where an RO profile is serendipitously located near a developing system is very limited. One of the primary motivations for developing an airborne RO system is that the locations of the occultation profiles will occur near the flight track (within 250 km), so that the potential impact of denser sampling of RO profiles near the center of a storm system can be tested. In addition, spaceborne RO observations in the lower troposphere are typically down weighted during assimilation relative to upper troposphere and lower stratosphere observations because of the potentially larger errors attributed to assuming local spherical symmetry in the retrieval process. The airborne system has the potential to provide the data necessary to investigate these errors in detail while simultaneously providing a platform for easily collecting independent collocated measurements.

For a receiver within the atmosphere, the RO technique differs from the spaceborne technique [Zuffada et al., 1999] in that the raypath through the neutral atmosphere is not symmetric with respect to the tangent point (point of closest approach to the Earth). The retrieval technique has been adapted to the airborne geometry and several simulation studies have been carried out [Healy et al., 2002; Lesne et al., 2002; Xie et al., 2008]. An exploratory flight system was tested in Japan [Yoshihara et al., 2006]; however, only qualitative conclusions about the performance were drawn. Here we report the first results to provide a quantitative evaluation of the airborne GPS RO measurement technique and illustrate its first use in tropical storms.

2 Measurement Campaign

The Global Navigation Satellite Systems (GNSS) Instrument System for Multistatic and Occultation Sensing (GISMOS) was deployed in August–September 2010 for the Pre-Depression Investigation of Cloud systems in the Tropics (PREDICT) [Montgomery et al., 2012]. The objective of the campaign was to investigate the thermodynamics and circulation within the interior of African easterly waves prior to development of tropical storms. While not all African easterly waves develop into hurricanes, more than 80% of major hurricanes originate as easterly waves [Landsea, 1993]. Few in situ observations have been made of tropical systems within easterly waves during the transition from disorganized convection to tropical depression because development usually happens far from shore, over the Atlantic. This experiment targeted early measurements in the wave axis where the mean flow and wave speed were similar, leading to the isolation of areas where atmospheric moisture potentially could be concentrated [Dunkerton et al., 2008; Wang et al., 2010]. The National Science Foundation (NSF) Gulfstream-V aircraft flew six research flights into the tropical disturbance that eventually developed into Hurricane Karl. On 8 September, as an easterly wave approached the windward islands, it merged with a trough over South America to produce a broad area of low pressure, which then moved westward producing disorganized convection. Late on 13 September the central convection increased and a closed circulation developed, forming a tropical depression around 12:00 UTC on 14 September [Stewart, 2011], several days later than anticipated. We have analyzed data from research flight RF18 from 10:00 to 16:00 UTC on 13 September 2010, the fifth of six research flights into the storm system. In situ pressure and temperature measurements were made at flight level, and dropsondes were released during this research flight in order to sample the thermodynamic profiles and the winds in the interior of the wave as the circulation developed. During the flight, the GISMOS system made continuous observations of GPS signal carrier phase at 5 Hz using conventional geodetic receivers and recorded the GPS raw RF signal at 10 MHz from side-looking antennas [Garrison et al., 2007]. An Applanix POS/AVTM inertial navigation system was used to achieve velocity precision better than 5 mm/s as required for accurate airborne retrievals [Muradyan, 2012; Muradyan et al., 2010; Xie et al., 2008]. GISMOS measurements of the excess GPS signal propagation delay were used to derive profiles of atmospheric refractivity, which is directly related to moisture and temperature (see the supporting information for full details of the method).

3 Results

The aircraft flew at an average altitude of ~ 14 km in a lawn mower pattern over the central region of deep convection (Figure 2). During this time, satellite PRN25 set beyond the horizon at ~11:40 UTC, producing an occultation on the starboard side of the aircraft. The line of sight from the aircraft to the setting GPS satellite at consecutive times is shown in Figure 2 (dotted lines), for the segment of the raypath passing through the atmosphere below the aircraft height. The tangent point is the closest point to the surface of the Earth along the raypath (Figures 1 and 2). The highest tangent point of the occultation is located at the aircraft position and drifts farther from the aircraft as the satellite sets, in this case drifting 327 km at the lowest point tracked. The bold line in the figure connects these tangent points at successively lower height. Since the accumulated refractive bending or delay is greatest where the atmosphere is the densest, the measurement is most sensitive to the refractivity at the tangent point, although it samples the entire path illustrated by the dotted lines. The retrieved refractivity profile is therefore not vertical but slanted along the direction of the tangent point drift. This sampling geometry can be handled in a straightforward manner in data assimilation by assigning the observation to the horizontal location of the tangent point. The locations of the dropsondes released on this flight are also shown in Figure 2 (stars).

Figure 2.

Aircraft flight path for RF18 over the Caribbean Sea on 13 September 2010 (dash-dotted line), location of dropsonde released on 13 September 2010 at 13:24 UTC (large square), and other dropsondes (stars). For each occultation, the horizontal locations of the tangent points are shown (thin black lines), starting from the highest tangent point adjacent to the flight path and ending at the lowest tangent point furthest from the aircraft. The occultation of PRN25, is shown as a bold black line; for this event, the line of sight from the aircraft to the setting GPS satellite at consecutive times is shown with dotted lines (defined in the text). The hurricane symbol indicates the center of the tropical disturbance at 12:00 UTC before it developed into Hurricane Karl.

Our approach is to directly compare the retrieved refractivity with refractivity derived from the nearest dropsonde profile, in order to address the accuracy of the measurement with as few assumptions as possible. The Doppler shift of the setting satellite PRN25 is shown in Figure 3a, along with the Doppler of PRN14 used to correct for receiver clock errors. The final retrieved refractivity is derived using the method described in the supporting information, and is shown in Figure 3b. It is compared to the refractivity calculated from dropsonde 12 that was released during the same flight at 13:24 UTC at −76.640300°E, 17.642200°N. The dropsonde horizontal location was 25 km from the deepest tangent point of the occultation profile. GISMOS sampled tangent point altitudes in the height interval 14 km to 4 km for PRN25. The difference in refractivity between the dropsonde measurement and the occultation measurement is less than 2%. The agreement with the dropsonde is comparable to the accuracy of COSMIC spaceborne RO observations in the lower troposphere. Chen et al. [2011] estimated COSMIC measurement accuracy using the National Meteorological Center method [Parrish and Derber, 1992], which is based on comparing apparent errors with an estimate of the model errors from lagged forecast differences. The COSMIC refractivity errors were shown to vary approximately linearly from 0.5% at 10 km to about 2.5% at the surface, in summer months and at latitudes south of 45°.

Figure 3.

(a) Excess Doppler shift for the occulting satellite PRN25 (black) and a reference high-elevation angle satellite PRN14 (blue) for recordings made on 13 September 2010 during RF18. The difference in the excess Doppler (red), PRN25 minus PRN14, removes variations due to the receiver clock. (b) Bending angle (gray) derived from the excess Doppler shift. Noise due to turbulence creates large errors in bending angle near the maximum impact parameter. Bending angle values are replaced with simulated values for impact parameters above 1 km below the maximum impact height (black; see supporting information). (c) Retrieved RO refractivity (black) compared to the refractivity calculated from dropsonde 12 released on 13 September 2010 at 13:24 UTC (dashed). (d) RO minus dropsonde percent refractivity difference for the conventional geodetic receiver (gray) and the open loop method (black dashed line).

We use the entire set of dropsondes released on the 13 September 2010 flight to examine a concern that is often raised with RO measurements: that the horizontal resolution is insufficient to be useful for mesoscale systems, especially tropical storms. While the horizontal resolution is certainly limited, several studies have indicated that spaceborne RO profiles assimilated into weather models near tropical cyclones have improved forecasts [Huang et al., 2010; Liu et al., 2012]. In this case we quantify the agreement of the RO profile with the nearest dropsonde (#12), and compare that with the agreement with the overall suite of dropsondes. The average background refractivity profile in the study area was calculated from 11 dropsondes within 520 km of the RO profile (Figure 2) and was used as the reference. Figure 4 shows that both dropsonde 12 and the RO profile have significantly higher refractivity than the background (at zero line), which we attribute to higher moisture in the altitude range from 4 to 8 km. Thus, we demonstrate that the difference between the RO profile and dropsonde 12, which we can consider an estimate of the accuracy, is better than the range of the horizontal refractivity variations that are important in the midtropospheric levels of this developing system. As an additional validation, we include in the comparison the European Center for Medium-Range Weather Forecasting Reanalysis-Interim (ERAI) at 0.75° resolution, which assimilated the dropsonde data that were transmitted via the Global Telecommunication System for operational use during the campaign. ERAI closely agrees with the RO profile and the dropsonde profile in this altitude range where moisture variations are important. The RO profile below 9 km agrees more closely with the ERAI profile than the dropsonde profile, even though the dropsonde was assimilated. We suggest that the dropsonde often shows a greater difference from the ERAI model because of representativeness errors, that is, the dropsonde point measurement is not representative of the atmosphere at the scale of the model grid. This has been described in assimilation tests that compare the impact of dense profile data averaged over 10 km compared to 40 km [Frehlich, 2011; Weissmann and Cardinali, 2007]. RO observations, on the other hand, naturally average over this spatial scale and thus might be expected to agree better with the ERAI analysis.

Figure 4.

RO refractivity profile (bold black line) deviation relative to the environmental mean, defined as the mean refractivity profile for 11 dropsondes released 13 September 2010 (the zero reference). Despite the averaging inherent in the RO measurement, the refractivity local to the tangent point (bold) is significantly closer to the nearby dropsonde (thin black line) than to the environmental mean profile at 0%. The RO profile is also close to the ERA-Interim (ERAI) reanalysis profile at the closest grid point to the lowest tangent point.

Finally, we have also compared the level of agreement between the dropsonde and RO profile with the expected range of variations empirically observed to be important for cyclone development. Specifically, Komaromi [2013] estimated the variation of the daily mean dropsonde profile relative to the background mean for developing versus nondeveloping cases from the PREDICT campaign and found that, on average, the nondeveloping cases had a significant dry humidity anomaly present between 300 hPa and 950 hPa, with a maximum mixing ratio anomaly of −1 g/kg at 550 hPa. This corresponds to a refractivity anomaly of −3% at 550 hPa, which is of the order that can be distinguished by the airborne RO method. In addition, within 150 km of the center of circulation for the genesis events, the moisture between 300 hPa to 950 hPa increased by ~ 1 g/kg, creating an expected change in refractivity on the order of 2–3.5%. While the RO profile has relatively poor horizontal resolution compared to dropsondes, it does give an average refractivity observation near the tangent point that captures the scale of spatial and temporal anomalies shown by Komaromi [2013] to be important and is thus expected to be useful for measuring the general thermodynamic properties in the interior of tropical storm systems, despite its inability to resolve smaller-scale convective features.

In the lowest part of the atmosphere, strong refractivity gradients can lead to complex signal propagation effects causing signal fading and rapid carrier phase fluctuations. The signal variations can exceed the operational range of conventional GPS receiver phase-locked tracking loops and eventually cause the receivers to lose lock. Loss of lock in this profile occurred at approximately 11:50 UTC when the signal was 3.9° below the horizon for PRN25 (corresponding to a tangent height of approximately 4 km above the surface). Loss of lock by the GISMOS geodetic receivers occurred above 5 km for all other occultations from this flight (Figure 2) and the other flights into the pre-Karl system. For that reason, GISMOS was designed with a raw RF sampling GPS recording system (GRS) where open loop tracking could mitigate this effect. Open loop tracking was developed to replace the traditional feedback loop with a method that tracks the signal using an a priori estimate of Doppler shift, as described in Beyerle et al. [2006], rather than determining the Doppler shift from the signal itself. The method has been implemented in a software receiver and adapted to the signal dynamics of the airborne system [Lulich et al., 2010; Wang et al., 2013]. We applied this open loop tracking to the raw 10 MHz RF data recorded by the GRS in postprocessing mode. The result for the same PRN25 occultation is shown in Figure 3c (dashed line) and extends the profile an additional 2 km further toward the surface. The open loop RO profile also shows similar agreement with the coincident dropsonde measurement. The high-moisture environment of the subtropics is a challenging environment for the conventional GPS receivers, and a preliminary inventory of the remainder of the data set shows that recordings down to ~ 6 km occurred for less than one profile per flight or a total of 21 profiles for 26 flights during the entire PREDICT campaign. In comparison, the preliminary analysis of the GRS data from one flight using the open loop method indicates that excess phase was recovered down to a height corresponding to 2 km for both rising and setting satellites for more than seven occultations per flight. The further analysis of the complete GISMOS GRS data set in the future will therefore provide an optimal data set for both a robust statistical study as well as an exceptional data set for assimilation. This proof of concept presents the potential for developing an operational system based on an upgraded GPS receiver that could be deployed on commercial aircraft, which are already equipped with inertial navigation systems and external GPS antennas. We have compared open loop tracking signals from the top and side-looking antennas (not shown) and find that refractivity results differ by less than 0.5%. This illustrates that the inertial navigation system navigation solution is accurate enough when transferred to the location of the side or top antenna [see also Muradyan et al., 2010]. Although the GRS data from the top antenna terminated at 4 km altitude rather than 2 km for the side antenna, it illustrates that data from a simpler installation using the top antenna for occultations on commercial aircraft would still be worthwhile.

4 Conclusions

The GNSS Instrument System for Multistatic and Occultation Sensing (GISMOS) has been developed for airborne high-vertical resolution atmospheric profiling of regional meteorological targets such as developing tropical storms, where assimilation of observations in the upstream environment has the potential to improve forecast accuracy. In August and September 2010, GISMOS was deployed on the National Science Foundation Gulfstream-V aircraft to make atmospheric observations in the southeastern Atlantic and Caribbean during the Pre-Depression Investigation of Cloud systems in the Tropics (PREDICT), with concurrent sampling by dropsondes. Despite the relatively large-horizontal averaging length of the limb-sounding occultations, the retrieved refractivity values at the tangent points agreed to better than 2% with the measurements from the closest dropsonde. This is sufficient to distinguish the order of magnitude variations in refractivity observed by dropsondes between developing and nondeveloping cases quantified by other researchers during the PREDICT campaign, and the temporal changes in refractivity associated with moisture changes in the last 72 h prior to genesis in the 150 km region around the circulation center, as demonstrated by Komaromi [2013]. GISMOS has the capability for 10 MHz sampling of the RF data, which when combined with a software receiver that performs open loop tracking, significantly extends its measurement capability in the lower troposphere as low as 2 km, for both rising and setting satellites. The repeatability of the measurements, shown by the agreement between the independent measurements made by conventional geodetic measurement system and the GRS, is better than 0.5% refractivity. The open loop method has the ability to recover on the order of 1–2 profiles for each hour of flight time, and thus would be a valuable method to develop for real-time operational use. The method is less accurate than spaceborne radio occultation, as a consequence of the turbulent aircraft motion. However, the aircraft deployments can provide many more profiles in the storm region or a region of interest than are possible with the current combination of spaceborne GPS RO missions, and thus, they provide a unique data set for testing the impact of assimilating a dense suite of occultation measurements. In combination with the dropsondes, airborne RO increases by 50% the number of profiles that will be available for studying the development of storm systems within tropical waves during the PREDICT campaign. The technique has the potential to provide hundreds of profiles per day if implemented on commercial aircraft for use in operational data assimilation in numerical weather models. Airborne RO could thus be an important addition to the atmospheric observation system for tropical storms.


Support for this work was provided by HIAPER UCAR subcontract S05-39696 from NSF, NSF grant SGER-0802887, and NSF grant AGS 1015904. Partial support for B.M. was provided by the Ross Fellowship. P.M. was supported by the Schlumberger Faculty for the Future Fellowship. F.G.N. was supported by a Capes/Fulbright Graduate Student Fellowship (1834/07-0) and a NASA Earth System Science Research Fellowship (NNX11AL50H). We would also like to thank the following: A. Johnson who assisted with the GISMOS data collection; J. Jensen, J. Meitin, A. Cooper, A. Schanot, R. Sherman, and the NCAR-EOL staff for logistical support; PREDICT principal investigato M. Montgomery and C. Davis; M. Bell, D. Raymond, and C. Lopez who assisted with GISMOS operation; J. Dunion, J. Cordeira, K. Griffin, and the PREDICT/NASA/NOAA forecast teams; and K. Young, J. Wang, and the NCAR-EOL dropsonde team. We would like to acknowledge the continued support and interest of NSF program officers, in particular J. Fein and E. DeWeaver, B. Smull, and J. Huning. We acknowledge the infrastructure and expertise provided by E. Calais at the Geodesy Laboratory at Purdue (now at Ecole Normale Superieur, Paris) during the early development of the airborne RO system.

The Editor thanks two anonymous reviewers for their assistance in evaluating this paper.