Comparison of temporal fluctuations in the total electr on content estimates from EISCAT and GPS along the same line of sight

The impact of space weather events on satellitebased technologies (e.g. satellite navigation and precise po­ sitioning) is typically quantified on the basis of the total elec­ tron content (TEC) and temporal fluctuations associated with it. GNSS (global navigation satellite systems) TEC measure­ ments are integrated over a long distance and thus may in­ clude contributions from different regions of the ionised at­ mosphere which may prevent the resolution of the mecha­ nisms ultimately responsible for given observations. The pur­ pose of the experiment presented here was to compare TEC estimates from EISCAT and GPS measurements. The EIS­ CAT measurements were obtained along the same line of sight of a given GPS satellite observed from Tromsø. The present analyses focussed on the comparison of temporal fluctuations in the TEC between aligned GPS and EISCAT measurements. A reasonably good agreement was found be­ tween temporal fluctuations in TEC observed by EISCAT and those observed by a co-located GPS ionospheric mon­ itor along the same line of sight, indicating a contribution from structures at E and F altitudes mainly to the total TEC in the presence of ionisation enhancements possibly caused by particle precipitation in the nighttime sector. The experi­ ment suggests the great potential in the measurements to be performed by the future EISCAT 3D system, limited only in the localised geographic region to be covered.


Introduction
The total electron content (TEC) integrated along the line of sight of a given radio link is associated with group delay and phase advance resulting from the propagation of a ra dio signal throughout the ionised part of Earth's atmosphere (Davies, 1965). With the advent of satellite-based technolo gies, such as satellite telecommunications and satellite-based navigation and precise positioning applications (hereafter, GNSS), our society has become increasingly more reliant on those systems, which may be critically vulnerable to disrup tive space weather events.
The impact of space weather events on satellite-based navigation and telecommunication systems may be subdi vided into two categories: (a) the distortions introduced by large-scale inhomogeneities in the spatial distribution of the plasma density and (b) the disruption introduced by smallscale structures, by means of the phenomenon known as ra dio wave scintillation. The former has been typically anal ysed on the basis of TEC measurements and TEC temporal fluctuations which may be associated with degradation of po sitioning accuracy in the case of satellite navigation applica tions (Mannucci et al., 1998;Schaer, 1999;Jakowski et al., 2002).
TEC measurements typically deduced from observations of GNSS radio signals are utilised to calculate TEC maps which may be at local (http://swaciweb.dlr.de) or global scale (http://aiuws.unibe.ch/ionosphere/, http://iono.jpl.nasa. gov, http://swaciweb.dlr.de). An important aspect connected with this operation is the absolute calibration of TEC Published by Copernicus Publications on behalf of the European Geosciences Union.
observations against offsets introduced for example by clock errors, cycle slips, and phase ambiguity (Ciraolo et al., 2007). A further complication is introduced by the conversion of TEC measurements corresponding to slant ray path into val ues corresponding to vertical ray paths, in order to increase the map coverage (Meggs et al., 2004). Calibrated TEC mea surements are also used in tomographic imaging of the up per ionised atmosphere (Bust and Mitchell, 2008;Yin et al., 2008;Yizengaw et al., 2006a, b). Similarly to the approach used in the case of maps, tomographic images rely on cali brated slant TEC measurements, which are then used to re construct 2-D or 3-D images by assuming a theoretical model for the electron density vertical profile (Bust and Mitchell, 2008, and references therein).
Previous analyses compared TEC estimates obtained by the integration of electron density profiles obtained by the EISCAT (http://www.eiscat.se) receiver in Tromsø with ver ticalised TEC estimates obtained from GPS stations nearly co-located with the radar and the ionosonde (Lilensten and Cander, 2003;Lilensten et al., 2005;Pokhotelov et al., 2011;Stolle et al., 2006). In those cases, the incoherent scatter radar measured electron density profiles in a direction parallel to the local magnetic field line. The electron density profiles were calibrated against two nearly co-located ionosondes, and the TEC estimate was obtained by integrating the elec tron density profiles from 90 km up to 498 km. The results of the comparison showed instances of substantial agree ment between GPS TEC and EISCAT TEC together with cases of disagreement where GPS TEC appeared lower than the EISCAT TEC (Lilensten and Cander, 2003). The origin of discrepancies was hypothesised to rely on factors such as localised particle precipitation enhancing EISCAT TEC estimates, slant GPS satellites' lines of sight cross-cutting areas of particle precipitation resulting in underestimated TEC, and the protonospheric contribution after geomagnetic storms (Lilensten and Cander, 2003;Lunt et al., 1999).
A different analysis showed a comparison between EIS CAT TEC obtained by integration of electron density profiles and verticalised GPS TEC from nearly co-located stations, with equivalent ionospheric pierce points overlapping the area covered by EISCAT (Jakowski et al., 1996). In that case, EISCAT was operated in a scanning mode (CP3) and electron density profiles were measured at different latitudes between 62 and 78 • N during a 30 min north-south scan. GPS TEC estimates as from TEC maps were then compared with EIS CAT TEC within an overlapping region. The EISCAT TEC was obtained by integration of CP3 electron density profiles from about 150 km to about 500 km height and then verti calised at ionospheric pierce points by means of a mapping function (Jakowski et al., 1996). The results indicated larger GPS TEC as compared with EISCAT TEC values, given the geometry considered, which suggested a plasmaspheric con tribution not captured by the radar, yet present on GPS radio signals (Jakowski et al., 1996).
On the basis of these previous analyses, the present exper iment was proposed. Within the framework of the ongoing Marie Curie Initial Training Network TRANSMIT (http:// www.nottingham.ac.uk/transmit), 25 h of measurement time were allocated to the proposed experiment. The measure ments took place from 12 December 2011 to 16 Decem ber 2011 between 15:00 UT and 19:00 UT approximately ev ery day. The UHF EISCAT radar in Tromsø was made to point towards given GPS satellites. The idea was to follow a GPS satellite and to compare temporal fluctuations in EIS CAT TEC estimates (obtained by integration of electron den sity profiles along the line of sight) and in GPS TEC esti mates from the tracked satellite, in order to gather more ev idence in addition to what was shown earlier. This approach avoided indeed any assumption about verticalisation or abso lute calibration of TEC measurements by providing possible insights on the influence of D/E layers, F layer and topside (and the problem of disentangling them on GPS TEC mea surements). The measurements were carried out during quiet magnetic conditions; however this type of experiment re peated in more active conditions might reveal additional de tails. Here, the results of only one day of measurement were reported, namely those from Tromsø on Monday 12 Decem ber 2011. The results suggest the potential advantage of using EISCAT 3D (http://www.eiscat3d.se) measurements for the investigation of physical processes over different spatial and temporal scales as well as for their application to satellitebased technologies exposed to adverse space weather events.

Data and methodology
The position of GPS satellite PRN23 was determined in ad vance on the basis of the projection of the ephemeris in the future by using a SP3 file (http://igscb.jpl.nasa.gov/igscb/ data/format/sp3 docu.txt) released the day before each of the days during the measurement campaign. Those positions were determined at 5 min intervals to cover the entire du ration of the measurement. PRN23 was followed between 16:00 UT and 18:00 UT. The radar was pointed towards the same satellite by remaining fixed in a given position (de fined in terms of azimuth and elevation) for 5 min, then re positioning in the new direction in the next interval, and so on. During each position the GPS satellite was moving and traversing the radar line of sight during each 5 min interval. During each 5 min interval the radar was measuring and col lecting backscattered power, which was then converted into electron density profiles by using the typical GUISDAP anal ysis toolbox (http://www.eiscat.com/groups/Documentation/ UserGuides/GUISDAP/).
For the sake of completeness, estimates of calibrated GPS slant TEC were calculated as well on the basis of ray trac ing through reconstructed electron density structures, follow ing the inversion procedure detailed in Mitchell and Spencer (2003) and subsequently refined in Chartier et al. (2012).
The inversion method produced three-dimensional images of electron density. The intersections of a given ray path with the electron density images were used to calculate the slant TEC along the given ray path. Two different calibration methods were utilised. The first method ("daily calibration") assumes slant TEC extracted from the images should match the GPS observations when no local structuring is present. The second method ("weekly calibration") calculates slant TEC along a given ray path for the whole week of mea surements; the average difference between imaged slant TEC and observed slant TEC provides a calibration constant. Both methods provide estimates for the GPS slant TEC accurate within few TECU. However, the point here was to appreci ate and quantify the sensitivity of the two instruments in the presence of ionospheric structures (and their evolution) by means of temporal fluctuations in TEC along the same line of sight, which can provide possible insights on the separate influence of D/E layers, F layer and topside on GPS TEC measurements.

EISCAT calibration
Typically, EISCAT electron density profiles are calibrated against closest available ionosondes by means of the calibration script CALIB NE within GUISDAP (Lehti nen and Huuskonen, 1996) (http://www.eiscat.com/groups/ Documentation/UserGuides/) in view of similar geometries between the two instruments (e.g. vertical or field-aligned di rections) allowing for common fields of view on average. In the case of the present experiment, the calibration of elec tron density profiles along slant lines of sight by means of CALIB NE was purely indicative, owing to different fields of view between instruments. A refined calibration could have been based on the use of possible ionosonde measurements centred at the equivalent ionospheric pierce point coordinates for the radar positions. However, in this case one limitation would have been the availability of such instruments across the whole interval of directions the radar pointed to, and a second limitation would have been introduced by the slant projection of vertical profiles. That type of calibration was not attempted for the analysis presented here, and the stan dard CALIB NE toolbox was applied to EISCAT measure ments by allowing the minimum elevation angle to be 60 • . This choice was not entirely appropriate as the minimum el evation angle should be limited to 75 • . However owing to the particular geometry used in this experiment, CALIB NE was used to remove possible outliers from the data and increase the confidence in the slant electron density profiles and their accuracy. An example of the output of CALIB NE for the measurements collected during this experiment is shown in Fig. 1 where the actual proportionality between EISCAT and ionosonde data is indicated together with the suggestion for the calibration constant to be used for obtaining calibrated EISCAT electron density profiles. The analysis to follow was entirely based on calibrated electron density profiles accord www.ann-geophys.net/31/745/2013/ ing to what is described above. Another aspect to consider is that ionosonde data from the F region are usually missing when the E layer is dense (as in the case of sporadic E layers).

Time alignment
Co-located with the EISCAT radar was a Novatel GSV4004 ionospheric monitor capable of measuring TEC and rate of change of TEC at 1 min intervals together with 50 Hz signal level and phase (Van Dierendonck et al., 1993). The compar ison of the rate of change of TEC between GPS and EISCAT relied on the integration of EISCAT slant electron density profiles. EISCAT electron density profiles could be obtained at different integration times during each radar position. The maximum integration time was approximately 5 min, corre sponding to the duration of the measurements in one pre cise position. The maximum integration time corresponded to the minimum error in the electron density profiles possi ble in the case of the experiment considered here. Figure 2 shows a representative case of electron density profiles, the error associated with them, and the estimate of the electron temperature profiles for an integration time corresponding to 5 min (Fig. 2a), 60 s (Fig. 2b) and 150 s (Fig. 2c). The error on the electron density profiles increased with decreasing in tegration time, especially at higher ranges. For the sake of the comparison between rate of changes of GPS TEC and EISCAT TEC, an integration time of 150 s was chosen as the most appropriate compromise between temporal averag ing and confidence in the measurements. All the calculations shown hereafter refer to EISCAT electron density profiles ob tained by means of 150 s integration time.

Data processing
The tracked satellite (PRN23) provided TEC measurements, which were compared with the radar's one. In view of the error connected with electron density profiles at 150 s integration time, the EISCAT TEC was calculated by inte grating the electron density profiles from 70 km up to 500 km in range. Because the interest was in the comparison be tween rates of changes of TEC (estimates of GPS slant TEC were calculated in two different ways), the observation times needed to be as close as possible. While the GPS TEC val ues were supplied every 15 s, the radar sampling rate corre sponded, instead, to the integration time used to retrieve the TEC at 150 s. It has to be considered that, to follow the satel lite, the radar antenna moved every 5 min and, because the movement takes a few seconds, the actual measurement time interval could be smaller than 150 s. Furthermore, the two sampling times were not aligned. There were about 10 GPS samples in each radar measure ment interval (a rate of 15 s in about a 150 s interval) which needed to be reduced and aligned with the radar data accord ing to a communal reference time. The GPS samples were then averaged in that interval. Each average value was then referenced to a time corresponding to the centre of the radar measurement interval (Fig. 3). Details of the error analysis are given in Appendices A and B. Figure 2c shows measurements from EISCAT (Tromsø) on 12 December 2011, in terms of electron density profiles, the error associated with them, and the electron temperature pro files. Figure 2d shows the minimum and maximum altitudes corresponding to EISCAT ranges utilised for the TEC inte gration while following PRN23. Figure 4a shows the TEC obtained by integration of the electron density profiles in Fig. 2c, integrating from 70 km altitude to the maximum al titude in Fig. 2d. In addition, Fig. 4a contains estimates of the GPS slant TEC along the same line of sight calibrated according to the "daily" and "weekly" methods described in Sect. 2 (in this particular case, the "daily" method provided a better calibration against the "weekly" method). Figure 4b shows temporal fluctuations in TEC as observed by both the radar and the GPS monitor over a time interval of 150 s. In Fig. 4b, different lower bounds for the TEC integration were used and compared with the GPS observations. Figure 4c shows the contribution to temporal fluctuations in TEC from different ionospheric layers (i.e. D/E, F1 and F2 nominally in terms of altitude intervals) as compared to the GPS obser vations. Figure 5 refers to PRN23 tracked between 17:00 UT and 18:00 UT on 16 December 2011.

Results and discussion
The normal ionisation decay between 16:00 UT and 16:30 UT (Fig. 2c) was followed by a sporadic E (Es) layer which formed at approximately 16:20 UT and lasted through out the measurement interval (Kirkwood and Nilsson, 2000;Nygrén et al., 1984). Typically, about 2 TECU of integrated ionisation could be associated with that Es layer (Fig. 4a). At 17:10 UT an enhancement in the ionisation at higher ranges showed three distinct peaks (i.e. at 17:15 UT, at 17:30 UT and at 17:50 UT), with the Es layer still present underneath (Fig. 2c). The integrated ionisation corresponding to those ionisation enhancements corresponded to about 5-7 TECU (Fig. 4a). The ionisation enhancement might have originated from plasma patches transported over the line of sight (Foster et al., 2005;Moen et al., 2006). Such a pattern can be observed as well through the elec tron density integrated along the path (i.e. TEC) as mea sured from EISCAT (Fig. 4a), where enhancements in TEC can be associated with enhanced ionisation structures appear ing in Fig. 2c. The integration ranges utilised in the calcula tion of TEC as observed from the radar (Fig. 2d) included altitudes typical for both E and F regions. Consequently, temporal fluctuations in slant EISCAT TEC along the radar line of sight in correspondence with the ionisation enhance ments (compare Figs. 2c and 4b) seemed to be correlated with temporal fluctuations in the TEC observed from PRN23 (Fig. 4b). Different contributions are shown in Fig. 4b: tem poral fluctuations in TEC after integrating electron density profiles from 70 km in altitude (blue line), from 150 km in altitude (red line), and from 200 km in altitude (green line). TEC fluctuations obtained when integrating from 200 km in altitude (Fig. 4b) seemed to be similar to the fluctuations obtained when integrating from 70 km or 150 km in alti tude, suggesting the bulk of ionisation causing TEC enhance ments was located at F layer altitudes. In order to verify such an aspect, the contributions to TEC fluctuations from radar observations were isolated from different ionospheric layers (Fig. 4c): i.e. from altitude intervals nominally asso ciated with D/E (70-150 km), F1 (150-200 km) and F2 lay ers (200 km to maximum range). The largest contribution ap peared to result from the F2 layer (Fig. 4c), while the whole TEC fluctuations observed from PRN23 would include the contributions from each different layer with its own phase (different phases might stem, for example, from layers mix ing along the line of sight as observed in Swartz et al. (2009) for example). The comparison between Fig. 4b and c sug gests that the bulk of TEC fluctuations originated in the F region with nothing attributed to the topside or to a redistri bution of plasma in the plasmasphere in this specific case.
The differences in TEC fluctuations observed from the radar and PRN23, within the time period of Fig. 4b and c, are within ±0.5 TECU over 150 s. This may reflect different sensitivities to different regions, but further studies would be required.
The experiment described here was intended to provide possible clues on the real structuring along the line of sight as compared with integrated GPS measurements. Satellite data can indeed be used for 3-D imaging on either regional or global scales, while present EISCAT profiling is limited to narrow regions and to specific temporal windows with dif ferent sensing geometries (see for example a comparative study in Meggs et al., 2005). The modelling of structured features such as those described here might equip tomogra phy reconstruction algorithms with higher resolution, which is needed to infer the physics of the observed phenomena. On the other hand, from a purely applicative point of view, the measurements performed by both GPS and EISCAT showed a substantial agreement which could illustrate the great po tential behind the measurements to be performed by the fu ture EISCAT 3D instrument (Aikio et al., 2012;Johansson et al., 2010). Of course, the results reported are dependent on the accuracy of the EISCAT calibration using ionosondes; there is scope for improvement especially under Es layers such as those reported in this experiment.
In view of the experiment described here, EISCAT 3D (see, for example, http://www.eiscat3d.se) appears to pro vide additional details on spatial and temporal distribu tion of plasma density structures. On the other hand, EIS CAT 3D measurements could be used to provide maps of TEC (as suggested in Lilensten and Cander (2003), Lilensten et al. (2005), and Jakowski et al. (2002) for example) as well as of rate of change of TEC with higher spatial and temporal resolution than what is available from standard TEC maps at present. The only limitation for EISCAT 3D would be the re gion covered (northern Scandinavia). However, in that region Ann. Geophys., 31, 745-753, 2013 www.ann-geophys.net/31/745/2013/ � the measurements would be very relevant for precise satellite navigation and positioning used there in applications such as aviation, land surveying, property management, offshore drilling, amongst others. Finally, additional experiments are needed in order to gather more evidence on possible layer-mixing mechanisms (of the type detected in Swartz et al., 2009, for example).

Conclusions
Measurements of GPS rates of change of TEC were com pared with EISCAT observations from Tromsø. The mea surement campaign took place from 12 December 2011 until 16 December 2011 and was based on instruments operated by the EISCAT Scientific Association and the University of Bath. The EISCAT UHF radar in Tromsø was pointed to wards a single GPS satellite all the time. The given PRN was followed during its orbit at steps of 5 min intervals over which the GPS satellite was crossing the radar line of sight. In order to increase the radar accuracy, an integration time of approximately 150 s was chosen as the best compromise be tween accuracy, temporal and spatial resolutions. GPS mea surements were collected by means of standard GSV iono spheric monitors capable of outputting TEC values as well as signal components at 50 Hz sampling rate.
The agreement between the two types of instruments ap peared evident in a case of isolated Es layer and ionisa tion enhancement possibly due to plasma patches transported over the line of sight.
The potential benefit from future EISCAT 3D measure ments appeared to be indicated by the simple experiment de scribed here. The benefit could be associated with an accurate tool for the refinement of 3-D tomography imaging based on satellite data as well as for the higher spatial and temporal resolution for equivalent maps of TEC and rate of change of TEC, which would prove very useful for applications based on precise satellite navigation and positioning.
The repetition of the present experiment during more active conditions could provide more details on the level of structuring, possible layer-mixing processes, the origin of scintillation-induced signal fluctuations within the weak scattering approximation, and the associated modelling.

Temporal TEC fluctuations
The tracked satellites (PRN 23) provided TEC measurements which were compared with the radar's. In view of the er ror connected with electron density profiles at 150 s integra tion time, the EISCAT TEC was calculated by integrating the electron density profiles from 70 km up to 500 km. Be cause the interest was in the comparison between rates of changes of TEC (no absolute GPS slant TEC calibration was www.ann-geophys.net/31/745/2013/ attempted here), the observation times needed to be as close as possible. While the GPS TEC values were supplied ev ery 15 s, the radar sample rate corresponded, instead, to the integration time used to retrieve the TEC over 150 s. It has to be considered that, to follow the satellite, the radar an tenna moved every 5 min and, because the movement takes a few seconds, the actual measurement time interval could be slightly smaller than 150 s.
Furthermore, the two sample times were not aligned. There were about 10 GPS samples in each radar measure ment interval (a rate of 15 s in about 150 s interval) which need to be reduced and aligned with the radar data according to a communal reference time (Fig. 3). The reference time t ref was then set as the central point within the radar inte gration time. The GPS samples, in the same reference time, were therefore calculated by averaging the total GPS samples contained within the radar integration time interval (Fig. 3). Assuming, for example, L individual GPS samples within 150 s (radar integration time), where k is index for L elements in window centred about t ref .
Within the same reference time, t ref , the GPS and radar mea surements can be easily compared in terms of TEC variation. Thus, the time variation of the radar TEC is calculated as where �t is the time step and is about 150 s. Similarly, the TEC variation in time retrieved by the GPS is calculated as and, because of the same reference time t ref , the time interval �t is the same as the radar one.