Monitoring the atmospheric boundary layer by GPS radio occultation signals recorded in the open-loop mode



[1] A new type of radio occultation (RO) data, recorded in open-loop (OL) mode from the SAC-C satellite, has been tested for monitoring refractivity in the Atmospheric Boundary Layer (ABL). Previously available RO signals, recorded in phase-locked loop mode were often unusable for sensing the lower troposphere (LT) or resulted in significant inversion errors, especially in the tropics. The OL RO signals allow sensing of the LT and accurate monitoring of the ABL and, especially, its depth. Comparison of RO-inverted refractivity profiles to ECMWF analysis and available radiosondes generally shows good agreement in the depth of the ABL. However, in a number of cases, ECMWF fails to reproduce the top of ABL. Future OL RO signals will provide information about the ABL depth which is an important parameter for weather prediction and climate monitoring.

1. Introduction

[2] The Atmospheric Boundary Layer (ABL) is the lowest layer of the troposphere, directly affected by the Earth's surface [Garratt, 1994]. The main distinctive feature of the ABL is its turbulent structure. In many cases the transition between the mixed ABL and the more stably stratified troposphere above (the interfacial layer) is rather sharp and accompanied by a temperature inversion and, in tropical regions, by a sharp decrease of absolute and relative humidity. Commonly the top of the ABL is sharper over the sea surface than over land. Knowing the properties of the ABL, and in particular its depth (height of its top), is important for understanding transport processes in the troposphere, weather prediction and climate monitoring [Garratt, 1993].

[3] A relatively new space-borne method for remote sensing of refractivity, radio occultation (RO) by use of the Global Positioning System (GPS) and low-Earth orbiting satellites (LEO) [Melbourne et al., 1994; Ware et al., 1996; Kursinski et al., 1997; Rocken et al., 1997; Wickert et al., 2001; Hajj et al., 2004] can contribute significantly to remote sensing of the ABL by complementing traditional instruments such as radars, sodars and lidars. Radio signals traversing the atmosphere are sensitive to the vertical refractivity (N) gradients that are commonly large on top of the ABL. Figure 1 shows an example of temperature T, partial pressure of water vapor Pw and the corresponding refractivity N profile [Thayer, 1974] obtained from a radiosonde (01/24/02, 15.97S, 5.70W). It is seen that the interfacial (inversion) layer, indicated by two horizontal lines, is characterized by significant decrease of Pw and increase of T. Thus the top of ABL can be approximately identified as the break point in the refractivity profile (shown by arrow).

Figure 1.

An example of T, Pw, N radiosonde profiles for a sharp sub-tropical marine ABL. The arrow shows the top of the ABL.

[4] The vertical N-profile is retrieved from GPS RO signals under the assumption of local horizontal homogeneity of N around the ray tangent points (TP). Radio-holographic (RH) retrieval methods [Gorbunov, 2002; Jensen et al., 2003] allow sub-Fresnel resolution, of about 10–100 m. When the N-gradient exceeds a critical value (dN/dz < −157 km−1), known as super-refraction (SR) or ducting, which commonly occurs on top of the marine ABL [von Engeln and Teixeira, 2004], the RO inversion becomes an ill-posed problem and the retrieved N below the SR layer becomes negatively biased [Sokolovskiy, 2003; Ao et al., 2003]. Another, technical problem of RO method is related to the strong perturbations of the RO signals caused by sharp N-gradients. Such signals can cause significant errors or loss of lock in a GPS receiver operating in phase-locked loop (PLL) mode, including the so-called “fly-wheeling” mode [Hajj et al., 2004]. These tracking errors depend on tunable parameters of the tracking loop [Sokolovskiy, 2001; Beyerle et al., 2006].

[5] Von Engeln et al. [2005] suggested determining the depth of the ABL from RO on the basis of the cut-off height defined in the Full Spectrum Inversion (FSI) method [Jensen et al., 2003]. With PLL tracking this truncation is in many cases simply related to the time when the RO receiver declares loss of lock or starts tracking with significant errors due to low signal to noise ratio (SNR). While some correlation of the cut-off height with the top of ABL exists, as explained above, it is strongly dependent on PLL specific tunable parameters and on the RO signal structure affected by the vertical N structure above and below the top of the ABL. For example, it is known that in the summer season, over deserts, the top of the ABL can be as high as 5 km or more [Garratt, 1994], but the N-gradient is weaker than that for the marine ABL. Figure 2 shows a few examples of N-profiles retrieved by the use of FSI from CHAMP RO signals recorded in PLL mode. Panels (A–C) show RO N-profiles over North Africa penetrating well below the top of ABL not captured by ECMWF global analyses. Panels (D–F) show RO N-profiles over the Atlantic Ocean that stop substantially above the top of ABL captured by ECMWF global analyses. Thus the method proposed by von Engeln et al. [2005] is not reliable.

Figure 2.

N-profiles inverted from PLL RO signals compared to ECMWF analyses. Each profile is shifted by 3 km with respect to previous one (horizontal lines correspond to z = 0 for each profile). Arrows show the top of the ABL: 10/06/03, 01:50, 13N, 13E (curve A); 13/07/03, 09:38, 21N, 13E (curve B); 9/07/03, 09:01, 17N, 13E (curve C); 16/06/03, 02:56, 10S, 11W (curve D); 22/06/03, 02:23, 12S, 12W (curve E); 01/07/03, 23:52, 9S, 8W (curve F).

2. Use of the Open-Loop RO Signals for Determining the Depth of ABL

[6] To overcome the problems of PLL tracking, an alternative open-loop (OL) tracking technique has been developed. Fundamental concepts of the OL tracking were suggested by Sokolovskiy [2001]. For this study we used data collected with a flight implementation of OL tracking developed by JPL and tested with the BlackJack receiver on the SAC-C satellite. In OL tracking, a RO signal is recorded for both setting and rising occultations being down-converted (reduced in frequency for low-pass filtering and sampling) by use of a Doppler frequency model calculated without feedback from the RO signal. Theoretically, the accuracy of such a Doppler model, which is based on predicted orbits and a bending angle model, is sufficient to maintain SNR at 80–90% of its maximal level which depends on atmospheric propagation conditions. Then the cut-off point of bending angle profiles determined by RH methods [Gorbunov, 2002] depends on the shadowing of RO signals by the Earth's surface, but not on loss of lock as it often happens in PLL mode. With OL tracking, it is quite natural to identify the top of the ABL by the break point in the retrieved N-profile (see Figure 1).

[7] The six COSMIC (Constellation Observing System for Meteorology Ionosphere and Climate) satellites will track the GPS L1 signal in the troposphere in OL mode. In preparation for COSMIC, JPL tested their OL flight implementation on the SAC-C RO satellite. While an insufficient number of OL RO profiles are presently available for reliable statistical analysis, they clearly demonstrate the improvement of OL over PLL signals in terms of the capability to observe the ABL.

[8] The OL RO data used in this study were collected by JPL in March-May 2005. The SAC-C RO receiver operated in OL mode for about 6 hours on several days by recording L1 C/A GPS signal when the height of the ray TP was lower than ∼10 km. Above that height both L1 and L2 signals were recorded in PLL mode. Below that height the L2 signal was not recorded and ionospheric correction was based on extrapolation. The OL and PLL sampling frequency was 50 Hz. We analyzed 422 rising and setting OL and 2548 setting PLL occultations.

[9] In order to show the difference in the structure of the OL and PLL tropospheric RO signals, for each occultation that passed the post-inversion quality control (QC) [Kuo et al., 2004] we calculated the max ∣dN/dz∣ and the corresponding height, below 10 km. These results are shown as dots in Figure 3. Solid lines show the percentage of retrieved N-profiles penetrating to a given height. Note that the estimates of the maximal N-gradient depend on the smoothing applied to the bending angle, calculated from the phase of the RH-transformed RO signal, and then subjected to the Abel inversion. But since the same smoothing is applied for inverting OL and PLL signals, this does not affect their comparison. Figure 3 demonstrates a very clear difference in the structure of the OL and PLL inverted N-profiles due to the different tracking algorithms in the receiver. The penetration of the OL inversions is significantly better than that of the PLL. It is seen that for PLL data, the heights of the max ∣dN/dz∣ are more or less evenly distributed below 10 km. Apparently, this is related to the fact that PLL RO signals in most cases not only do not penetrate in ABL, but they can terminate at any height in the troposphere. For the PLL data, max ∣dN/dz∣ almost never exceeds 60 km−1. For OL data, there is a noticeable increase of the density of samples with large max ∣dN/dz∣ below 2–3 km. We associate this increase with the top of ABL by validating this with selected retrieved N-profiles. It is important to note that we use the large ∣dN/dz∣ as an indicator of high likelihood of the sharp top of ABL, but the height of max ∣dN/dz∣ does not necessarily correspond to the break point in N(z).

Figure 3.

Heights and magnitudes of the maximal ∣N∣ gradient below 10 km for RO signals recorded in (top) PLL and (bottom) OL modes. Solid lines show percentage of occultations penetrating to a given height. Circled and squared points indicate occultations used in Figures 4 and 5.

[10] Figure 4 shows retrieved N-profiles for 10 OL occultations with large ∣dN/dz∣ below 2–3 km but above 1 km (the corresponding points are marked by circles in the bottom plot of Figure 3). Arrows show the points of max ∣dN/dz∣ (not the break points). Most of these occultations occurred over the tropical ocean. The RO N profiles are compared with the ECMWF global analysis and, where available, with collocated radiosondes. Unlike the tropical PLL occultations, all OL occultations penetrate well below the pronounced top of the ABL. Some of the oceanic OL occultations still do not penetrate to the surface. This is related to the loss of SNR due to the in-real-time RO signal frequency and/or range miss-modeling by the receiver firmware at the bottom of these occultations. This is a technical problem which is expected to be overcome for the COSMIC mission. In some cases, the top of the ABL from RO is in good agreement with that from ECMWF (G,H,I). In other cases, ECMWF is under-resolving the top of the ABL and in some cases completely failing to reproduce the ABL (A,C,J). When radiosonde N-profiles are available, in most cases, they are in better agreement with RO than the ECMWF (E,F,J). This is in spite of the fact that the vertical structures reproduced by RO should be interpreted not as point measurements, but rather as horizontally weighted averages with the scale L ∼ 2equation image [Kursinski et al., 1997], where re is the Earth's radius and H is the vertical scale of the N-structures (for H ∼ 0.1–1 km L ∼ 50–200 km). Thus Figure 4 demonstrates that OL RO retrievals penetrate below the top of the ABL and can be used for determining the horizontally-averaged depth of the ABL associated with the break point in the retrieved N-profile.

Figure 4.

N-profiles inverted from OL RO signals (corresponding to the circled points in Figure 3) compared to ECMWF analyses and radiosondes. Each profile is shifted by 3 km with respect to previous one (horizontal lines correspond to z = 0 for each profile). Arrows show the points of max ∣dN/dz∣: 25/04/05, 15:59, 40S, 96W (curve A); 26/04/05, 11:17, 26N, 28W (curve B); 27/04/05, 12:00, 22N, 28W (curve C); 29/04/05, 15:29, 22S, 82W (curve D); 02/05/05, 12:31, 18S, 53W (curve E); 02/05/05, 14:21, 23S, 61W (curve F); 03/05/05, 14:51, 23N, 56W (curve G); 04/05/05, 12:54, 27S, 154E (curve H); 04/05/05, 15:47, 26S, 77W (curve I); 15/05/05, 13:49, 21S, 70W (curve J). Radiosondes: 20.5S, 54.7W (curves E and F); 27.4S, 153.1E (curve H); 23.4S, 70.4W (curve J).

[11] Figure 5 shows retrieved N-profiles for 4 OL occultations with large ∣dN/dz∣ found below 1 km (the corresponding samples are marked by squares in the bottom plot of Figure 3). Arrows show the points of max ∣dN/dz∣. It is remarkable that all 4 occultations occur over the Tasman sea. This region is well known to the radio-wave propagation community due to strong anomalous VHF propagation effects related to elevated ducts below 1 km [Milnes and Unwin, 1950]. These ducts are associated with Foehn winds blowing from the Australian deserts. The RO N-profiles are compared to the ECMWF global analysis and the 12hr Antarctic Mesoscale Prediction System (AMPS) MM5 (Fifth Generation Penn State/NCAR Mesoscale Model) forecast [Powers et al., 2003]. Generally, ECMWF analysis reproduces N-gradients smaller than those present in the RO retrieved N-profiles below 1 km, while in cases A, C, and D the N-gradients are better reproduced by the AMPS MM5 forecast. Thus, with OL tracking, RO can be used for monitoring radio wave propagation conditions over the sea (except for surface ducts). Although RO cannot retrieve negative N-gradients larger than the critical value shown above, a large retrieved gradient at sufficiently low height can be used as a warning of possible ducting.

Figure 5.

N-profiles inverted from OL RO signals (corresponding to squared points in Figure 3) compared to ECMWF analyses and AMPS forecasts. Each profile is shifted by 3 km with respect to previous one (horizontal lines correspond to z = 0 for each profile). Arrows show the points of max ∣dN/dz∣: 20/04/05, 12:35, 50S, 154E (curve A); 25/04/05, 12:58, 36S, 152E (curve B); 27/04/05, 12:45, 37S, 154E (curve C); 03/05/05, 12:06, 41S, 158E (curve D).

3. Conclusions

[12] Refractivity profiles inverted from the OL RO signals penetrate below the sharp top of the marine tropical ABL. In most cases, the limited OL data set from SAC-C OL shows good correlation between the top height of the ABL (associated with the break point in the RO N-profiles) and that determined from ECMWF analysis and radiosondes. In some cases, the ECMWF analysis fails to properly describe the ABL. In those cases the RO inverted profiles are generally closer to radiosondes than to ECMWF. Thus, OL RO signals provide valuable information about the ABL which is important for understanding transfer processes in the lower troposphere, and for weather forecasting and climate monitoring. The most reliable characterization of the ABL obtained from RO is the depth traced by the break point in the refractivity profile. The refractivity inversion errors below the top of ABL (theoretically studied by Sokolovskiy [2003, 2004], Ao et al. [2003], and Beyerle et al. [2006]) shall be validated in the future, when larger ensembles of the OL RO data are available. Large N-gradients (∼−100 km−1) found below ∼1 km in Tasman Sea indicate high probability of ducting. This information may prove useful for monitoring the radio communication environment over the ocean. Six COSMIC satellites, launched on April 14, 2006, will provide ∼2500 setting and rising occultations, recorded in OL mode, daily, thus allowing for continuous, all-weather global monitoring of the ABL.


[13] This study was supported by the National Science Foundation (NSF) as part of the development of the COSMIC Data Analysis and Archiving Center (CDAAC) at UCAR under the Cooperative Agreement ATM-9732665. The authors are grateful to the Jet Propulsion Laboratory (JPL) team for implementing and testing COSMIC flight software on the SAC-C satellite and providing OL RO data.