The GPS Segment of the AFRL-SCINDA Global Network and the Challenges of Real-Time TEC Estimation in the Equatorial Ionosphere

Transcription

1 The GPS Segment of the AFRL-SCINDA Global Network and the Challenges of Real-Time TEC Estimation in the Equatorial Ionosphere Charles S. Carrano 1 and Keith M. Groves 2 1) AER/Radex, Inc., 131 Hartwell Avenue, Lexington, MA ) Air Force Research Laboratory, Space Vehicles Directorate, Hanscom AFB, MA BIOGRAPHIES Dr. Carrano leads the Ionospheric Environments and Impacts Group at Atmospheric and Environmental Research, Inc. His research interests include the effects of the ionosphere on GPS and radar. He designed the real-time ionospheric monitoring software currently used by the dual frequency GPS receivers of the AFRL-SCINDA network. He has a PhD in Aerospace Engineering from The Pennsylvania State University and a BS in Mechanical Engineering from Cornell University. Dr. Groves is currently a Program Manager in the Space Vehicles Directorate of the Air Force Research Laboratory where he investigates ionospheric scintillation and its impact on satellite-based communication and navigation systems. He has a PhD in Space Physics from MIT and a BS in Physics from Andrews University. ABSTRACT The estimation of Total Electron Content (TEC) in the equatorial ionosphere using GPS presents a number of challenges due to the presence of strong spatio-temporal density gradients and scintillation of the satellite signals caused by F-region irregularities. In this paper we describe a methodology for real-time calibrated TEC estimation in the presence of scintillation and a highly structured ionosphere. The inter-frequency biases of the GPS satellites are assumed known; we use estimates provided by the Center for Orbit Determination in Europe (CODE). The inter-frequency bias associated with a particular receiver is estimated late at night when the ionosphere is minimally structured, using an iterative approach that minimizes the variance of verticalized TEC measured along the different satellite links. The nightly estimated receiver bias is shown to be insensitive to the assumed centroid height used in the single-layer approximation of the ionosphere. It is also relatively stable on a night to night basis, deviating from its running average most when nighttime gradients in density are largest (commonly associated with geomagnetic activity and/or equatorial spread F). A 14 day running average of the bias is used to minimize the effect of this variability on the calibrated TEC. The effectiveness of the technique is illustrated by comparing the calibrated TEC estimated using two GPS receivers connected to the same antenna. During quiescent ionospheric conditions the difference in TEC estimated with the two receivers is generally less than a couple of TECU, despite their substantially different internal biases. During scintillating conditions, the TEC from the two receivers exhibit substantial differences due to receiver errors in the measurement of pseudorange and phase, unless strict quality control techniques are applied to exclude this data from the analysis. Methods for the automated detection of receiver error due to scintillation are presented and are shown to yield reliable TEC estimates. The GPS segment of the AFRL-SCINDA network currently includes 7 dual-frequency receivers that monitor equatorial scintillation and TEC in the Pacific, South American, Indian, and East Asian sectors. Several additional deployments are planned, particularly in Africa and Asia. The TEC data provided by these sensors are expected to be made publicly available and should make a valuable contribution to space weather monitoring and forecast models, since much of the dynamics of the storm-time ionosphere originates in the equatorial region and other GPS networks tend to sparse in this area. INTRODUCTION The Scintillation Network and Decision Aid (SCINDA) is a network of ground-based receivers that monitor scintillations at the UHF and L-Band frequencies caused by electron density irregularities in the equatorial ionosphere [Groves et al., 1997]. Established by the Air Force Re- ION NTM 2006, January 2006, Monterey, CA 1036

2 Figure 1. Map of current SCINDA ground stations and planned deployments for Blue markers indicate stations currently equipped with a dual-frequency receiver. All stations are expected to be equipped with dual-frequency receivers in the future. The solid and dashed curves shows the approximate locations of the geomagnetic equator and the crests of the Appleton anomaly, respectively. search Laboratory (AFRL) to provide regional specification and short-term forecasts of scintillation to operational users in real-time, the network currently includes 7 dualfrequency GPS receivers capable of measuring Total Electron Content (TEC). A map showing the locations of existing and planned SCINDA ground stations is shown in Figure 1. The SCINDA stations are generally positioned between the ionization crests of the Appleton anomaly, as these locations experience the strongest global levels of scintillation. The standard GPS data delivery services, such as the International GPS Service (IGS), do not provide measurements at data rates sufficient to resolve the temporal dynamics of ionospheric scintillation (10-50 Hz measurements are typically used). Therefore, specialized software has been developed for this purpose. This software is called GPS- SCINDA and it provides measurements of S 4, TEC, and ROTI (rate of change of TEC), as well as receiver position in real-time using the full temporal resolution available for each parameter. All ionospheric parameters are computed in real-time, thereby enabling the use of receivers with high internal data sampling rates at remote stations with low bandwidth Internet connections. At the same time, the realtime processing avoids the man-hours and cost normally required to post-process large volumes of raw data. Support has been added to GPS-SCINDA for a number of relatively inexpensive and off-the-shelf receiver models, including the Ashtech Z-12, Ashtech uz-cgrs, Ashtech ZXtreme, and the NovAtel dual frequency GISTM, all driven by a common code with common data output formats. The focus of this paper is to describe the method by which slant TEC measured with this system is calibrated in realtime for use by the space weather nowcasting and modeling community. TEC ESTIMATION USING GPS The estimation of Total Electron Content (TEC) using dual frequency GPS receivers is made possible by the dispersive nature of the ionosphere. The ionosphere, being a weakly ionized plasma, imparts a group delay (D) and carrier advance to an RF signal that, to first order, are equal in magnitude and proportional to the total number of electrons encountered along the line of sight (LOS) and inversely proportional to the square of the signal frequency: D = K TEC / f 2, K = m 3 s -1 [1] By measuring the group delay or carrier phase advance imparted by the ionosphere on the two GPS carrier signals, L1 (f 1 = MHz) and L2 (f 2 = MHz), the TEC encountered along the signal propagation path may be inferred. In practice, complications arise due to hardware timing biases and cycle slips, which are breaks in the measured phase. An estimate of TEC along the LOS to each GPS satellite may be obtained in terms of the pseudorange measurements on the L1 and L2 frequencies as follows: where TEC P = A { [ P 2 - P 1 ] - [ B R + B S ] + D P + E P } [2] ION NTM 2006, January 2006, Monterey, CA 1037

3 P1 is the pseudorange on L1 (ns) P2 is the pseudorange on L2 (ns) BR is the receiver differential code bias (ns) BS is the satellite differential code bias (ns) DP is the pseudorange multipath error (ns) EP is the pseudorange measurement noise (ns) The constants A and B have been determined such that the computed TEC has units of TECU (1 TECU = el/m 2 ) and are given by: A = TECU/ns B = TECU/L1 cycle Measuring TEC using the pseudoranges depends on accurately determining the hardware differential (inter-frequency) code biases, BR and BS. Methods for accomplishing this are referred to as TEC calibration and are central to this discussion. Accuracy in measurement of TEC using the pseudoranges alone, however, is limited by the multipath and measurement noise terms which are difficult to model and can exceed those of the phases by an order of magnitude or more. We note that BR is often referred to as the station bias because this contribution to the delay generally depends on the antenna and cable configuration in addition to the receiver hardware itself. An alternative way to estimate the TEC along the LOS to each satellite involves the measurements of the carrier phases on the L1 and L2 frequencies as follows: where TECL = B { [ L1 - (f1/f2) L2 ] - [ N1 - (f1/f2) N2 ] + DL + EL } [3] L1 L2 is the carrier phase on L1 (cycles) is the carrier phase on L2 (cycles) N1 is the integer ambiguity of L1 phase (cycles) N2 is the integer ambiguity of L2 phase (cycles) DL is the phase multipath error (cycles) EL is the phase measurement noise (cycles) Measuring TEC using the carrier phases is generally more precise in that multipath error and measurement noise are smaller and may generally be neglected. The disadvantage is that the integer numbers of accumulated cycles of phase for each frequency, N 1 and N 2, are unknown and change after each cycle slip. Standard practice for estimating TEC using dual frequency GPS combines the strengths of both approaches. The estimate we use is based on the phase formulation [3] with the multipath and noise terms neglected, while the pseudoranges are used to estimate the unknown accumulated phase cycles: TEC = TECR - A ( BR + BS ) [4] Here TECR is referred to as the relative or leveled TEC; it is determined from the phases and a filtered combination of ranges and phases as follows: TECR = DCP + <DPR DCP> ARC [5] where the differential carrier phase, DCP, and differential pseudorange, DPR, are defined (in units of TECU) as follows: DCP = B [ L1 - (f1/f2) L2 ] [6] DPR = A [ P2- P1 ] [7] The notation < >ARC in [5] indicates an average taken over a phase connected arc (between successive cycle slips). The relative total electron content provides an absolute estimate of total electron content prior to calibration by subtraction of the receiver and satellite differential biases. Phase Leveling Figure 2 illustrates the phase leveling process, whereby the more precise but level-ambiguous DCP is leveled to the noisier DPR. A single phase connected arc is shown. Note the wavelike structure in TEC that occurred this night during the time when the satellite's elevation exceeded approximately 40 degrees. This structure is not multipath; instead it is characteristic of a moderately disturbed equatorial ionosphere several hours after sunset where scintillationcausing electron density irregularities have largely decayed away. Ionospheric disturbances such as these are not at all uncommon between the ionization crests of the equatorial anomaly, and care must be taken that this structure does not adversely affect the phase leveling process. The phase leveling procedure must be performed for each phase connected arc, and therefore cycle slips (in either or both frequency carriers) must be identified. Since our software monitors the phase at the maximum data rate provided by the receiver (see Table 1), cycle slips are easily identified as occurring when the DCP/B changes by more than one L2 cycle between samples. At these sample rates, a change in DCP/B of one cycle between samples would be much too fast for the ionosphere to impart even under the most disturbed conditions. ION NTM 2006, January 2006, Monterey, CA 1038

4 Figure 2. Differential pseudorange (black solid), differential carrier phase (blue), elevation (black dashed) and relative TEC (red) measured using an Ashtech uz receiver at Cuiaba, Brazil. We estimate the error in phase leveling rather crudely by measuring the RMS variation of DPR-DCP in each phase connected arc, noting that larger variations are generally indicative of greater multipath and/or thermal noise variations which are problematic for the arc averaging filter. We note that the effect of scintillation is to both shorten the arcs by causing phase slips and also to increase the level of fluctuations within each arc. For this reason we filter the measurements to exclude from the processing all samples for which S 4 exceeds 0.3 or ROTI exceeds 10 TECU/min. Figure 3 shows the estimated error in leveling the phase for one day of data collected at Cuiaba. The increase in leveling error that occurs after 19:00 LT is due to scintillation of the GPS signals at this time. Table 1. Sampling rates used to calculate TEC. Receiver Sampling Rate Ashtech Z-12 Ashtech uz/z-xtreme NovAtel GISM 20 Hz 10 Hz 1 Hz Between cycle slips, 60 second running averages of DCP and DPR are computed to average down the noise and multipath. The software evaluates the arc average in [4] in two ways, first as a running average that includes all data for a real-time estimate of TEC, and then again including only the highest elevation 25% of the arc for a post-processed (refined) estimate. The ROTI is calculated by numerical differentiation of the unaveraged DCP, and then this value is averaged over 60 seconds as well. We use 60 second averages of DPR and DCP in the phase leveling process, although instantaneous values may also be used. When a cycle slip is detected, the event is recorded and all the running averages (both the 60 second averages and the arc averages) are reset. The errors incurred during the leveling process will propagate into the net error in the TEC estimation and are therefore important to estimate. Contributors to the leveling error include multipath (which is largely dependent on the quality of the antenna and the relative obstruction of the viewing environment), the type of averaging filter used in the arc average, and the length of the arc (shorter arcs allow less time for the averaging filter to settle). Arcs shorter than 2 minutes are therefore excluded from the processing. Leveling errors can be increased by ionospheric activity, which acts to increase the frequency of cycle slips and thus to reduce the data available for inclusion in the arc average. Figure 3. Estimated phase leveling error at Cuiaba, calculated as the RMS of the differential pseudorange minus the differential carrier phase within each phase connected arc. Given estimates for the inter-frequency biases the absolute, or calibrated, TEC along each LOS is given in terms of the relative TEC straightforwardly using equation [4]. The challenge in evaluating [4] is in the proper estimation of the receiver and satellite differential biases (DCB). The satellite DCBs have been estimated successfully by several organizations and shown to vary only slowly [Sardon et al., 1997]. We use estimates for the satellite biases provided by the Center for Orbit Determination in Europe (CODE). These biases are available by FTP download as 30 day averages with the mean satellite differential biases removed [Schaer, et al., 1998]. The receiver bias we determine thus contains a contribution from the satellite biases, but this is of no consequence in the calibration since both contributions are removed in the end. Estimating the Receiver Bias There are many techniques available for estimating the hardware inter-frequency bias for a GPS receiver, e.g. Mannucci et al. [1998]. These approaches generally rely on ION NTM 2006, January 2006, Monterey, CA 1039

5 common volume observations provided by intersecting lines of sight to the satellites from multiple GPS receivers on the ground. The SCINDA network is generally too sparse for this approach to be effective (unless it is augmented by an external data source, such as IGS measurements). Our motivation for the bias estimation procedure considered here is that it is suited to stand-alone operation, possibly even on a moving platform. The approach we use to estimate the receiver bias is founded on the principle that the LOS, or slant, TEC is a function of elevation (due to the increased pathlength through the ionosphere) while the receiver bias is not. If the single layer approximation for the ionosphere is valid (which assumes the effect of the ionosphere on the signal can be represented as having been concentrated in a single thin layer at an assumed height) and in the absence of spatio-temporal density gradients, then the verticalized TEC from each satellite should be the same [Datta-Barua et al., 2003]. This condition is most closely met just before local sunrise, although there are cases where the ionosphere is highly structured even late at night so that this assumption is invalid, as we shall see. We estimate the receiver bias once per day using data acquired between 03:00-06:00 LT, according to the following algorithm. Given the relative TEC, the satellite bias for each link, and the ionospheric height, h, the verticalized TEC becomes a function solely of the receiver bias: TECV(BR) = [ TECR - A ( BR + BS ) ] / M(ε, h) [8] We compute the total variance, V(BR), of the calculated TECV values summed over bins (i) of local time (at the ionospheric penetration points) taken between 03:00 and 06:00 LT when ionospheric gradients are generally smallest: V(BR) = Σi Variance [TECV(BR)]i [10] The notation [ ]i above indicates that the data is to be taken from the i th local time bin. Only data from satellites above 20 degrees in elevation are included to minimize multipath effects. For a given layer height, h, the variance in [10] is a function only of the assumed receiver bias. We minimize this variance as a function of the assumed receiver bias using Brent's method with parabolic interpolation [Press et al., 1992]. The subdivision and binning of the data in the local time window is intended to decrease contribution of small local time gradients in TEC to the net variance in the window. We found these small temporal gradients to have minimal adverse effects on the bias determination and that minimizing the variance of vertical TEC in 12 minute bins of local time yielded better results than simply minimizing the variance over the entire three-hour window. We also noted that while small temporal gradients do not adversely affect the receiver bias determination, spatial gradients do and we shall explore this later in this paper. In our approach, the best estimate for the receiver bias each night is the value of RB that minimizes [10]. Figure 4 shows the verticalized TEC with the minimum total variance for a night at Cuiaba. Figure 5 shows the total variance as a function of the assumed receiver bias. In our experience, the minimum value is well defined and its minimum depends principally on the amount of structure in the ionosphere during the late-night calibration window. In the above, M(ε,h) is the single layer mapping function of the ionosphere, defined as M(ε, h) = sec { arcsin[(re cos ε)/(re+h)] } [9] where Re is the Earth radius and ε is the elevation. The ionospheric height is unknown in general. This height may be determined using an ionospheric model as is done in Burrell et al. [2006], or held fixed at a value representative of typical conditions. In this work 350 km has been used, except where stated otherwise. We will show later that our estimate for the receiver bias is insensitive to the assumed ionospheric layer height, so that this choice for the assumed height is not critical. Figure 4. Evaluating the total variance of vertical electron content in bins of local time. Each bin is taken to be 12 minutes long. The error bars show twice the variance in each bin. The vertical TEC corresponding to the best estimate of 22.3 TECU for the receiver bias is shown. ION NTM 2006, January 2006, Monterey, CA 1040

6 small during the 03:00-06:00 local time window used to calibrate the TEC. Coloration of the verticalized traces by geomagnetic longitude is a useful way to identify the presence of meridional gradients as a function of local time. As can been seen in the plot, as ionization levels approached their peak values at approximately 14:00 LT, the magnitude of meridional gradients also increased. The rapid fluctuations in TEC that occurred after 19:00 LT are indicative of a turbulent ionosphere, and indeed intense GPS scintillations were encountered this night. Figure 5. Total variance of verticalized TEC between 03:00-06:00 LT as a function of the assumed receiver bias. The best estimate for the receiver bias (22.3 TECU) is as the value for which this curve attains its minimum value (0.42 TECU, in this case). Evaluating the Calibrated TEC If the receiver bias has been reliably determined, the calibrated TEC can be computed from the relative TEC using equation [4]. We find the estimates of the receiver bias to fluctuate unsatisfactorily from night to night, and therefore we use a 14 day running average of the estimated receiver bias to evaluate the calibrated TEC for that day. This approach has been taken to improve the method's resilience to the effects of ionospheric disturbances that occur during the calibration window. Figure 6 shows the variation of the estimated receiver bias at Cuiaba over the 14 day period leading up to October 15, The average of the receiver bias during this period (22.4 TECU) is used to calibrate the TEC on this day. The verticalized calibrated TEC itself is shown in Figure 7. Figure 7. Verticalized TEC versus local time at Cuiaba. The data is colored by the geomagnetic latitude of the ionospheric penetration points. The presence of large meridional gradients implies that the single layer approximation may present an oversimplified view of the ionosphere in this case. Despite this limitation, we can provide a crude estimate of errors incurred in the verticalization process by subtracting estimates of TECV made using assumed layer heights above and below the expected layer height. For example, Figure 8 shows the estimated verticalization error obtained by subtracting estimates of TECV made using assumed ionospheric heights of 400 and 300 km. Figure 6. Daily bias estimates and a 14 day running average of the estimated receiver bias (shown as a dotted line). The error bars show the square root of the total variance of verticalized TEC within the 03:00-06:00 LT window. The verticalized TEC shown in Figure 7 confirms our assumption, at least in this case, that spatial gradients are Figure 8. Estimated verticalization error computed by subtracting TECV estimates using values of 400 km and 300 km for the assumed ionospheric height. ION NTM 2006, January 2006, Monterey, CA 1041

7 It is worth noting that no assumption was made regarding the minimum value of TEC that occurs, in this case (Figure 7), at approximately 05:00 LT. Instead, the assumption that the gradients of TEC vanish leads to an automatic determination of the minimum value of TEC, which consists of contributions from both the ionosphere and the plasmasphere. Errors in the estimation of the bias determination propagate directly into errors in the computed TEC, however, and it is possible to obtain a negative TEC estimate using this method. In this work we have purposely left such defects in the data unaltered to illustrate the behavior of the method. In an operational context stricter methods of quality control (lower thresholds on the maximum allowable S4 and ROTI, for example) would be employed to prevent such nonphysical occurrences. The error in the final calibrated (slant) TEC has contributions from several sources, including the phase leveling process and the receiver bias determination (which implicitly includes errors due to an incorrectly assumed ionospheric height). Since the total density at night is relatively low, errors in choosing the incorrect ionospheric height do not change the verticalized TEC appreciably. This situation leads one to suspect that the estimated receiver bias would be only weakly dependent on the height, and this turns out to be the case. We found the estimated receiver bias to be relatively insensitive to the assumed layer height, varying at most by a couple of TECU for all plausible values of the layer height. Figure 9 shows the result of computing BR twice, using the assumed heights 450 km and 250 km and then subtracting. The maximum difference over the course of the year was 2.5 TECU; typically it was less than 1-2 TECU. Since the receiver bias is insensitive to the assumed layer height, we estimate the total error in the calibrated TEC as a weighted average of the errors in the phase leveling and bias determination processes alone. Figure 9. Sensitivity of the estimated receiver bias to the assumed ionospheric layer height. A 100 km change in the height usually yields less than 2 TECU change in the bias. Validation of the Bias Estimation As evidence that our methodology leads to the correct determination of the receiver bias, we performed this analysis for two Ashtech Z-12 GPS receivers that were connected to the same choke-ring antenna using a cable splitter. Since the two receivers measured the same ionosphere and experienced the same multipath environment, equation [2] implies that the difference between the differential pseudorange to the same satellite measured by the two receivers should equal the difference between the two receiver biases plus the thermal noise (and also possibly receiver errors due to scintillation). Figure 10 shows the DPR measured by the two receivers with a relative receiver bias of approximately TECU. This relative bias was confirmed to be the same for all the other PRNs as well to within +/- 0.1 TECU. Figure 10. Differential pseudoranges for PRN 23 at Ascension Island measured by two different GPS receivers connected to the same antenna on 03/16/2002. The increase in fluctuations after 05:00 is due to low-elevation multipath. These two receivers were operated at Ascension Island for roughly one week while connected to the same antenna. Figure 11 shows the relative bias between the two receivers calculated nightly (using data between 03:00-06:00) by subtracting the DPR from each receiver using only segments with low multipath. Also shown is the relative bias determined by minimizing the variance in TECV late at night. The relative receiver biases estimated using these two independent methodologies generally agree to within a couple of TECU. The largest difference in the two methods for estimating the relative receiver biases occurred on the last night during which the strongest GPS scintillations were observed. While this close agreement suggests that the receiver bias estimation was accurate, some caution is warranted since it is possible that systematic errors in the estimation of the bias for each receiver could possibly cancel. ION NTM 2006, January 2006, Monterey, CA 1042

8 which should vary smoothly since the effects of the (nonscintillating) ionosphere are removed and the dominant contribution is from the satellite motion [Hofmann-Wellenhof et al., 2001]. Here we take the simpler approach and exclude data from the TEC processing whenever S4>0.3 or ROTI>10.0 TECU/min are detected. Figure 11. Difference between the receiver biases of two GPS receivers connected to the same antenna estimated by subtracting the DPR values from the same satellite (diamonds) and by subtracting the biases estimated by minimizing TECV late at night (crosses). Effects of Scintillation on TEC Estimation Ionospheric scintillation affects GPS receivers in several ways. Intermittent signal fades and enhancements result in errors decoding the GPS data messages and also in estimating the ranges [Carrano et al., 2005]. Phase fluctuations stress the ability of the receiver to maintain lock and cause cycle slips that may prevent the receiver from using the phase to refine its range measurements. As an illustration of how scintillation can directly affect the processing of TEC, Figure 12 shows the effects of scintillation on the differential pseudorange. The C/No ratio for PRN 28 shown here experienced intermittent fades exceeding 25 db. Also shown in Figure 12 are the DPR values measured by the same two receivers (that shared a common antenna) as shown in Figure 10. Note that the two receivers respond differently (by measuring different values for the DPR) to the scintillating signal even though it is caused by the same ionospheric irregularities. This suggests that each receiver is making false measurements of DPR during the deep signal fades, and these errors translate directly to the computed TEC by corrupting the phase leveling process. Nevertheless, by excluding data for which S4 exceeds 0.3 or ROTI exceeds 10.0 TECU/min from the processing, the resulting calibrated TEC from the two receivers is virtually the same (Figure 13). In general, one must detect scintillation and exclude this data from the processing of TEC in order to avoid making false measurements of the ionosphere. In principle, one might do this by detrending the ionosphere-free residual ρ =[ P1 - (f2/f1) 2 P2 ] f1 2 /(f1 2 -f2 2 ) [10] Figure 12. Carrier to noise ratio (top) and differential pseudorange (bottom) for PRN 28 measured at Ascension Island on 03/16/2002 by two different GPS receivers connected to the same antenna (red and blue time-series). The large fluctuations in DPR between 20:15-21:30 LT are the result of scintillation. Figure 13. Verticalized TEC measured by two different GPS receivers connected to the same antenna at Ascension Island on 03/16/2002. Note the development of the Appleton ionization trough to the north and crest to the south after dusk (18:00 LT). Effects of Ionospheric Disturbances on the Receiver Bias Estimation An important consideration in estimating the receiver bias by this method is whether there are systematic errors in doing so that depend on the state of the ionosphere. Not surprisingly, the bias determination is more successful when the ionosphere is quiet and smoothly varying rather than when it is disturbed due to equatorial spread-f (ESF) irregularities, or highly structured due to magnetically active conditions in the magnetosphere. ION NTM 2006, January 2006, Monterey, CA 1043

9 Scintillation Activity Figures show the variation of the daily estimated receiver bias (performed once each night using data between 03:00 and 06:00 LT) for Antofagasta, Cuiaba, Bahrain, and Kwajalein Atoll. Also shown in Figures is the level of scintillation activity measured by the GPS 0-12 hours after local sunset at each station. The diagonal colored streaks running from upper left to lower right represent low intensity and systematic signal fluctuations due to multipath rather than ionospheric scintillation. The vertical streaks of color superimposed on this multipath background indicate scintillation activity on a nightly basis. Note that the SCINDA stations also monitor scintillations at the UHF frequency, but only the GPS scintillations are shown here. Since we have no independent measurement of the receiver biases for these stations, it can be difficult to distinguish a true change in the receiver bias from an outlier in the bias estimation process. Nevertheless, our study at Ascension Island (discussed earlier) suggests that the actual receiver bias probably varies by roughly 1 TECU over a week's time. Therefore, we expect that deviations in the bias larger than this from a one or two week trend are probably the result of poor bias estimation. Our aim here is to attempt to explain when the bias estimation process might be expected to perform poorly, both generally and also by considering a few examples in detail. As can be seen by comparing the estimated bias and scintillation activity (especially for Antofagasta and Cuiaba, Figures 14 and 15, respectively) the nightly variation of the estimated bias tends to be larger during the times of year when scintillation activity is prevalent. This effect might be expected to increase during solar maximum years, and the correlation may actually be better with UHF scintillation than GPS scintillation, although this has not been verified. Essentially no GPS scintillation activity was measured in 2005 at Bahrain, which lies to the north of the northern Appleton ionization crest. Thus, the variation of the estimated bias at this station is relatively consistent throughout the year. That the average variation at this station is typically larger than at the others may be due to the station's geographic location on the northern slope of the ionization crest where nightly density gradients tend to be large even in the absence of scintillation. It is interesting to note that while GPS scintillation activity was present at Kwajalein, it was generally weaker than at the South American stations, and the variations in the estimated receiver bias are correspondingly smaller. Figure 14. Daily estimated receiver bias (top) and scintillation activity (bottom) at Antofagasta, Chile in The error bars for the biases show the square root of the total variance in TECV between 03:00-06:00 LT. Two outliers in the bias estimation are labeled with dates for later discussion. Figure 15. Daily estimated receiver bias (top) and scintillation activity (bottom) at Cuiaba, Brazil in The error bars for the biases show the square root of the total variance in TECV between 03:00-06:00 LT. Two outliers in the bias estimation are labeled with dates for later discussion. ION NTM 2006, January 2006, Monterey, CA 1044

10 Figure 16. Daily estimated receiver bias (top) and scintillation activity (bottom) at Bahrain, Saudi Arabia in The error bars for the biases show the square root of the total variance in TECV between 03:00-06:00 LT. Essentially no GPS scintillation occurred at Bahrain in In general, outliers in the estimated receiver bias correspond to nights that have significant ionospheric gradients present during the calibration window. These late-night gradients are frequently associated with geomagnetic activity and equatorial spread F. When substantial gradients exist during the local time window used to estimate the biases, errors occur because the assumption of small gradients at night is violated. The bias determination method presented here cannot always distinguish between contributions to the total variance in TECV from a poorly estimated receiver bias from those due to true ionospheric gradients. In fact, it is often possible to minimize the TECV variance to artificially low levels when there is substantial ionospheric structure (effectively enforcing the small gradients assumption even when it is not valid) by selecting an incorrect value for the receiver bias. For this reason, we recognize that ionospheric gradients late at night are a major source of error for this receiver bias estimation method, and this is why we use a 14 day average to produce the final calibrated TEC. To help substantiate this claim, we consider four outliers in the bias determination processing (labeled 02/07, 06/13, 01/17, and 07/12 in Figures 14 and 15) in some detail. Figures 18 and 19 show the estimated vertical TEC at Antofagasta and Cuiaba for the day prior to and the day of the poor bias determination (thus the calibration window corresponding to the outlier lies in the right-hand plot for each of these four examples). Note the turbulent structure in the TEC during the 03:00-06:00 LT calibration window in each case. GPS scintillations were exhibited in all four cases. We note that the 06/13 and 07/12 cases are examples of intense scintillation during the off-season, which is relatively uncommon and were likely due to increased magnetic activity in the magnetosphere (a sudden onset of Kp reaching 7 + was measured on 06/12 and a sudden onset of Kp reaching 6 + occurred on 07/10). Response to Magnetic Storms Figure 17. Daily estimated receiver bias (top) and scintillation activity (bottom) at Kwajalein Atoll, Marshall Islands in The error bars for the biases show the square root of the total variance in TECV between 03:00-06:00 LT. Next we examine the effects of geomagnetic storms on the bias determination by considering a specific example. On August 24, 2005 the onset of a severe geomagnetic storm began when the Bz component of the IMF turned southward (-48 nt) at ~09:00 UT. The 1-hour DST index began a precipitous decline at about 10:00 UT, reaching a minimum value of -219 nt at approximately 11:00 UT. The 3-hour Kp index attained a maximum value of 9- while the Ap index reached 300 nt. Figure 20 shows the effect of the storm on vertical TEC measured at Antofagasta, Cuiaba, Bahrain, and Kwajalein along with the 1-hour DST index. ION NTM 2006, January 2006, Monterey, CA 1045

11 modulation of the daily TEC profile. Note that an intense meridional gradient of the same sign was established at both Antofagasta and Cuiaba the day after the storm onset despite the fact that these two stations lie on opposite sides of the magnetic equator. This suggests that the meridional structure of TEC was modified dramatically on August 25. Nevertheless, since the TEC gradients at each the stations were sufficiently small during the local time windows used for the receiver bias determination, no anomalous bias estimates were caused by this particular storm. Figure 18. Estimated vertical TEC for two periods at Antofagasta. A highly structured ionosphere during the TEC calibration window between 03:00-06:00 on 2/7 and 6/13 caused outliers in the bias determination. In both cases GPS scintillations were observed earlier in the evening. Figure 19. Estimated vertical TEC for two periods at Cuiaba. A highly structured ionosphere during the TEC calibration window between 03:00-06:00 on 1/17 and 7/11 caused outliers in the bias determination. In both cases GPS scintillations were observed earlier in the evening. All four of these stations show marked effects in TEC due to the storm and, as noted by other researchers [Basu et al., 2001] the manifestation of these effects depends primarily on the local time of maximum magnetic perturbation. The minimum value of DST occurred just after local sunrise (~07:00 LT) at Antofagasta and Cuiaba, near peak daylight hours (~14:00 LT) at Bahrain, and about four hours after sunset (~22:00 LT) at Kwajalein. The two South American stations observed an anomalous TEC peak just after UT noon during the daytime ramp-up in TEC. A similar peak was observed at Bahrain, which experienced the largest TEC increase due to the storm. Kwajalein experienced intense GPS scintillations due to the storm, but not much Figure 20. Estimated vertical TEC at Antofagasta, Cuiaba, Bahrain, and Kwajalein during a magnetic storm that began on 08/24/05. Also shown is the hourly DST index (nt). CONCLUSIONS In this work we present a methodology used to provide calibrated TEC from the dual-frequency receivers of the AFRL-SCINDA network. The method uses the satellite inter-frequency biases provided by CODE and estimates the receiver bias by minimizing the total variance of verticalized TEC late at night. Given that these receivers operate in the equatorial zone where ionospheric scintillation is a frequent occurrence, methods for detecting scintillation and excluding this data from the TEC processing are discussed. The nightly estimated receiver bias is shown to be relative- ION NTM 2006, January 2006, Monterey, CA 1046

12 ly stable, exhibiting a relatively slow tend (~ 5 TEC/month) that may depend on seasonal conditions. The largest deviations from the trend generally occur when density gradients are large at night (commonly associated with geomagnetic activity and/or equatorial spread F). This observation may be readily explained since the assumption on which the bias determination technique depends, namely that density gradients are small during the local time window used for calibration, is violated. Filtering of the data according to the S4 measured by the GPS is necessary, but may not be sufficient to ensure accurate bias estimation since a disturbed ionosphere does not always generate GPS scintillations (a filter on S4 measured at UHF might be more successful). A 14 day running average of the estimated receiver bias is used to generate the final calibrated TEC in real-time. Future efforts will involve improving techniques used to detect outliers (evenings for which the estimated receiver bias should be excluded from this average). Tighter restrictions on the maximum allowable ROTI for the data to be included in the processing may help to reduce the occurrence of outliers. These restrictions will reduce the data available for bias estimation, of course, and may result in extended periods where a bias determination cannot be made. In this work no attempt was made to account for the C/A-P code bias that affects TEC estimation for receivers that evaluate DPR using the C/A code on L1 and the P code on L2. For the data presented in this work the receivers at Antofagasta, Christmas Island, Bahrain, and Cuiaba should have been corrected. The receivers at Ascension Island and Kwajalein do not require this correction as they operate using the P code on L1 and L2 directly to compute the DPR. Finally, a validation study should be performed using an independently obtained measure of calibrated TEC. This validation could also be used to adjust the weighting coefficients used in our error estimates. Once these shortcomings have been addressed, and adjustments to the algorithms for operational quality controls are in place, it is expected that the TEC data from the SCINDA stations will be made publicly available. ACKNOWLEDGEMENTS The authors would like to thank Patricia Doherty of Boston College for her participation in several helpful discussions on this topic, and Chris Bridgwood for preparing the seasonal plots of GPS scintillation activity. The work at AER, Inc. was supported by AFRL contract F REFERENCES Basu, Su., Sa. Basu, C. Valladares, H.-C. Yeh, S.-Y. Su, E. MacKenzie, P. Sultan, J. Aarons, F. Rich, P. Doherty, K. Groves, and T. Bullett, Ionospheric effects of major magnetic storms during the International Space Weather Period of Sept. and Oct. 1999: GPS observations, VHF/UHF scintillations, and in situ density structures at middle and equatorial latitudes, J. Geophys. Res., 106, , Burrell, A., N. A. Bonito, and C. S. Carrano, Total electron content (TEC) processing from GPS observations to facilitate ionospheric modeling, Proceedings of the American Meteorological Society, submitted Carrano, C. S., K. M. Groves and J. M. Griffin, Empirical Characterization and Modeling of GPS Positioning Errors Due to Ionospheric Scintillation, Proceedings of the Ionospheric Effects Symposium, Alexandria, Virginia, Datta-Barua, S., P. H. Doherty, T. Dehel, J. A. Klobuchar, Ionospheric Scintillation Effects on Single and Dual Frequency GPS Positioning, Proceedings of the Institute of Navigation GPS/GNSS Meeting, Portland, Oregon, Groves, K., Basu, S., Weber, E., Smitham, M., Kuenzler, H., Valladares, C., Sheehan, R., MacKenzie, E., Secan, J., Ning, P., McNeil, W., Moonan, D., and Kendra, M., Equatorial scintillation and systems support, Radio Sci., 32, 2047, Hofmann-Wellenhof, H. Lichtenegger, and J. Collins, GPS Theory and Practice, Springer-Verlag, New York, Mannucci, A., B. Wilson, D. Yuan, C. Ho, U. Lindqwister, and T. Runge, A global mapping technique for GPS derived ionospheric total electron content measurements, Radio Sci., 33 (3), , Press, W H., S. A. Teukolsky, W. T. Vetterling, B. P. Flannery, Numerical Recipes in FORTRAN, the Art of Scientific Computing, Cambridge U. Press, Cambridge, Sardon, E. and N. Zarraoa, Estimation of total electron content using GPS data: How stable are the differential satellite and receiver instrumental biases? Radio Sci., 32, , Schaer, S. and W. Gurtner, IONEX: The ionosphere map exchange format version 1. Proceedings of the IGS AC Workshop, IGS, Darmstadt, Germany, ION NTM 2006, January 2006, Monterey, CA 1047