Comparison of FORMOSAT-3/COSMIC radio occultation measurements with radio tomography



[1] The paper reports a comparison between the continuous tomographic observations of the ionosphere during July–August 2008 and radio occultation (RO) measurements made by the Formosa Satellite 3/Constellation Observing System for Meteorology, Ionosphere, and Climate (FORMOSAT-3/COSMIC or F3/C). The two sets of observations agree well qualitatively and bring out the main features of the low-latitude to midlatitude ionosphere: (1) The electron density enhancements observed at ∼26°N–31°N geographic (∼16°N–21°N magnetic) latitudes during daytime are related to the equatorial ionization anomaly (EIA), and (2) the enhancements observed at night at latitudes poleward of ∼35°N geographic are related to the midlatitude summer nighttime anomaly. The comparison shows that the EIA crest densities are highly underestimated in the RO inversions that use F3/C data. It is also seen that the agreement is better in the midlatitude region, with overall features represented quite well in the RO data, highlighting its usefulness as the two techniques capture similar electron density structure, although there are some magnitude differences.

1. Introduction

[2] Knowledge of the ionospheric electron density (Ne) distribution over a wide geographic area is imperative in the context of satellite communication and navigation, not only in the HF and VHF range but also in the L band, especially for single-frequency Global Positioning System (GPS) users to achieve optimal accuracy. One of the earliest and most widely used methods to probe the ionosphere is by means of an ionosonde that yields vertical electron density profiles up to but not above the altitude of maximum electron density (Nmax). Although incoherent scatter radars (ISRs) have the ability to measure the electron density at altitudes well above the peak, they are rather limited in number. Ionosondes and ISRs are unable to give large spatial coverage, though they can give good temporal resolutions. In this context, a tomographic technique is a viable alternative to obtain the vertical electron density profiles over a large geographic region. The primary data for the tomographic inversion are the line of sight total electron content (TEC) values, estimated along a number of raypaths connecting an orbiting satellite and ground receivers aligned along nearly the same longitude. Ionospheric tomography, first proposed by Austen et al. [1988], has evolved during the past two decades to become an important ionospheric diagnostic tool.

[3] The first preliminary experimental tests by Pryse and Kersley [1992] using TEC measurements from a single satellite pass at two stations in Scandinavia were followed by the Mid-America Computerized Tomography Experiment (MACE), the Russian American Tomography Experiment (RATE), and the experimental campaigns carried out in the United Kingdom [Pryse and Kersley, 1992; Raymund et al., 1993; Pryse et al., 1993]. A review of the results from these experiments is given by Kersley and Pryse [1994], and a comparison of different reconstruction algorithms is made by Raymund [1995]. Subsequently, it was established that these techniques are also useful for studying features such as traveling ionospheric disturbances [Cook and Close, 1995; Markkanen et al., 1995] and the equatorial ionization anomaly (EIA) [Huang et al., 1999; Andreeva et al., 2000; Franke et al., 2003, Thampi et al., 2007]. Apart from this, experimental campaigns were conducted to study the midlatitude ionosphere in Japan with four receiving stations covering the 140°E meridian [Kunitake et al., 1995]. In some recent studies, tomographic imaging of the ionosphere has also been performed using TEC data from closely located GPS receivers [e.g., Yizengaw et al., 2005, 2007; Lee et al., 2008].

[4] Reviews outlining the theoretical development of the method with discussion of its potentials and limitations have been published (see special issues of the International Journal of Imaging Systems and Technology, 5(2), 1994, and Annales Geophysicae, 13(12), 1995). Currently, there are a few operational radio beacon chains for tomography in midlatitude to high-latitude regions. One of them is in Alaska; it is maintained as part of the High frequency Active Auroral Research Program (HAARP), jointly directed by the U.S. Air Force Research Laboratory and Office of Naval Research. Another tomography chain currently operational is the Finnish chain in the 24°E meridian; its receivers are located in the 60°N–69°N sector. The low-latitude networks currently operational are the Low-latitude Ionospheric Tomography Network (LITN) of Taiwan [Yeh et al., 1994; Huang et al., 1999] and the Coherent Radio Beacon Network of India [Thampi et al., 2007]. However, at present there are no active tomography beacon networks in the far-pole midlatitude regions. In this context, the GNU Radio Beacon Receiver (GRBR) network of Japan is the first chain of receivers for the tomographic imaging of the ionosphere over this region.

[5] Another powerful technique used to derive the vertical profiles of atmospheric and ionospheric parameters is the radio occultation (RO) technique [Hajj and Romans, 1998; Schreiner et al., 1999; Hajj et al., 2000; Yunck, 2002]. The RO technique was first applied to the observation of Earth's atmosphere using an experimental satellite called GPS/MET in 1995. Following the successful GPS/MET experiment, similar satellite missions, such as CHAMP, SAC-C, the Gravity Recovery and Climate Experiment (GRACE), and the Ionospheric Occultation Experiment (IOX), were carried out. Many studies have shown occultation results of ionospheric soundings in comparison with ground-based radar measurements since then [e.g., Hajj and Romans, 1998; Schreiner et al., 1999; Hernández-Pajares et al., 2000; Hajj et al., 2000; Jakowski et al., 2002; Tsai and Tsai, 2004]. Following these missions, six microsatellites known collectively as the Formosa Satellite 3/Constellation Observing System for Meteorology, Ionosphere, and Climate (FORMOSAT-3/COSMIC or F3/C) were launched in April 2006, with the principal goal of making the RO measurements; these satellites yield more than 1500 profiles per day.

[6] The F3/C RO measurements were compared previously with ISR measurements and ionosonde-derived foF2 values [e.g., Kelley et al., 2009; Lei et al., 2007]; both are only point measurements. Tomography provides a viable method for comparing the RO measurements at a fixed longitude with more details about the altitude variation, with considerable latitude coverage. Tomography and the RO measurements are both indirect observations of electron density. Careful comparisons, combined with a realistic view of the limitations of these techniques, are essential to understanding the large-scale features of the ionosphere that are revealed by these methods. This paper presents a summary of the results of continuous tomographic observations conducted in the July–August 2008 period, along with comparisons using F3/C RO data and ionosonde data from Kokubunji, Japan (35.7°N, 139.5°E, geographic).

2. Experiment

[7] A new digital receiver, the GNU Radio Beacon Receiver, was developed using the open-source hardware called Universal Software Radio Peripheral along with the open-source software toolkit for the software-defined radio (GNU Radio). The technical details of the receiver have been reported [Yamamoto, 2008]. GRBR receivers were installed at Shionomisaki (33.45°N, 135.8°E), Shigaraki (34.8°N, 136.1°E), and Fukui (36°N, 136°E) in order to obtain simultaneous TEC data through continuous tracking of low Earth orbiting satellites, mainly the OSCAR, Cosmos, RADCAL, DMSPF 15, and F3/C satellites. The reconstruction was performed for a 2-D domain, horizontally from 22°N to 45°N and vertically from 100 km to the satellite altitude. However, because reliability of the structures toward the extremes of the image field of view would have been limited, we discarded data from 3° on either side in latitude and above 650 km in altitude. The typical data duration for the GRBR is 800–1000 s, so the coverage is more than 20° in latitude at satellite altitude (more than 10° in both directions from the receiver location). Hence, the region from 25°N to 42°N is typically covered by a minimum of two receivers or by all three receivers, especially for the pixels above 150 km. However, the region of maximum accuracy is directly above the receiver chain and within a few degrees on either side.

[8] The tomographic inversion is carried out using the algebraic reconstruction technique, with initialization from the International Reference Ionosphere 2007 (IRI-2007) model [Bilitza and Reinisch, 2008]. One of the most important data sources for the IRI electron density is the worldwide network of ionosonde stations that has monitored the ionosphere with varying station density since the 1930s. (Wakkanai (45.4°N, 141.7°E) and Kokubunji (35.7°N, 139.5°E) along the Japanese sector are among the earliest ionosonde observatories.) IRI predictions are most accurate for northern midlatitudes because of the generally high station density in this part of the globe [Bilitza and Reinisch, 2008]. Hence, IRI-2007 is a very reasonable choice as the initial guess for the reconstruction along the northern midlatitude region. The details of the tomographic inversion procedure have been published [Thampi et al., 2007; Thampi and Yamamoto, 2010]. Examples of the TEC data from the GRBR chain over Japan, the first few tomographic images, and validation of these images by comparison with foF2 observations have also been reported [Thampi and Yamamoto, 2010]. However, although IRI-2007 has significantly improved from the previous versions as far as the representation of the topside densities are concerned [Bilitza and Reinisch, 2008], the topside density values may be less accurate than the bottomside values, which in turn could affect the accuracy of the reconstruction of the topside electron densities. In addition to this, since the observation period corresponds to one with extremely low levels of solar activity, the utility of the IRI profiles may be limited. This issue is discussed in detail in section 4.2.

[9] The network has been operational since July 2008. During the observation period under discussion, i.e., July–August 2008, we recorded data for the satellite passes with maximum elevations greater than 40°. Six to eight passes could be obtained every day (at various local times) during this period. If the data from a particular station had glitches or losses of lock lasting more than 1 min, the data were discarded. In this manner, we chose simultaneous data from the three stations for the tomographic inversion. It must be noted that because the polar-orbiting satellites were being tracked, it was not possible to obtain data at exactly the same time on two consecutive days. Using these data sets, we calculated average images for each 2 h interval. For instance, the image representing 1400 local time (LT) is the average of nine images obtained on different days and at different times between 1300 and 1500 LT, and the image representing 2000 LT is the average of eight images obtained between 1900 and 2100 LT. A single IRI run corresponding to the solar minimum and representing the LT interval is used for all the inversions corresponding to that LT interval. All days included in the analysis were magnetically quiet, and the average F10.7 measurements for July and August were 65.7 and 66.3, respectively. Hence, the results are representative of the average northern midlatitude summertime variation during a magnetically quiet solar minimum period.

[10] It should be noted that the observed F3/C electron density profiles were obtained from the standard inversion routine in the COSMIC Data Analysis and Archival Center (CDAAC). The CDAAC ionospheric inversion assumes straight-line propagation of the GPS signal through the ionosphere by neglecting the bending angle. The TECs along the raypaths between the GPS satellite and an F3/C microsatellite are converted to a vertical electron density profile using the Abel transform with an analytical solution to avoid the singularities at the top and lower limits (for details, see Hajj et al. [2000], Syndergaard et al. [2006], Anthes et al. [2008], and references therein).

[11] The ionospheric electron density maps used for comparison, derived from the radio occultation observations of the F3/C satellites, were obtained for the 135°E meridian, and the plasma density structure plot for each time interval was constructed by taking the mean of the F3/C observations made within the same 2 h interval during July–August 2008. The number of daily COSMIC profiles can reach 2000+. Therefore, accumulated data for 60 days can amount to more than 120,000 profiles covering the entire globe. From such profiles, we chose data collected during magnetically quiet times. From these, the profiles that had an Nmax appearing below 200 km or above 500 km were discarded, and if the electron density above 200 km showed a negative value, the entire profile was excluded. The remaining vertical electron density profiles were binned in latitude-longitude (5° intervals) and LT bins (2 h intervals) to produce the 2-D pictures. In other words, one 2-D image of the entire spatial region is made up of different RO profiles, obtained by different occultations, since the entire area could not be covered in a single occultation. When there is more than one observation for any latitude-longitude bin, the mean value is used to represent the value for that bin. In contrast, the tomography gives a 2-D image for each single satellite passage, and the seasonal averages were produced by averaging these 2-D images.

[12] In the case of tomographic reconstruction, the degree of accuracy depends on the initial guess, and the accuracy of reconstruction is not the same over the entire region. The maximum accuracy is obtained for the area containing the highest number of path-pixel intersections with wider angles, with accuracy decreasing toward the edges of the reconstruction [Kunitsyn and Tereshchenko, 2003]. Because the tomography, unlike the RO inversion, does not presume a uniform distribution of electron density in a wide region of the ionosphere, it is less liable to error and reflects a more accurate electron density structure at the specific longitudes where the receivers are located. Therefore, we compare the average electron density structures observed by the tomography and the RO techniques. Additionally, the ionosonde foF2 data from Kokubunji (35.7°N, 139.5°E) are also used for cross comparisons.

[13] It must be mentioned here that several factors limit comparison of the individual electron density profiles obtained by means of these two techniques. First, even within a 500 × 500 km (lat-long) grid, we were able to access only one F3/C RO point per day. So the number of points available for such a comparison with tomography was small. Second, tomography corresponds to one 2-D plane (fixed at a longitude) and inversion results in actual vertical profiles. In contrast, if we consider individual RO profiles, they vary in latitude and longitude during the occultation event, and in some cases this latitude variation is more than a few degrees. Hence, in this paper, we limit the comparison to the average electron density variations observed by these two techniques.

3. Results

[14] Figure 1 shows the average temporal evolution of the ionosphere over the 135°E–136°E longitude sector during the period July–August 2008, as obtained from tomography. The time marked in each plot is the central hour of the 2 h interval, and the spatial resolution of the images is 0.5° (latitude) × 25 km (altitude). Following the absence of significant ionization at night, the electron density increases in the image at sunrise (0600 LT), with maximum Ne ≈ 2 × 1011 m−3. The electron density continues to increase with time, and latitudinal gradients become clear at 1000 LT, with higher density in the south. This electron density gradient is associated with the development of EIA crests at lower latitudes [see, e.g., Hanson and Moffett, 1966; Balan and Bailey, 1995]. The density and its gradient increase with time and reach maximum at 1400 LT, with peak Ne ≈ 6 × 1011 m−3 in the 25°N–27°N region. The enhanced density in the low-latitude region can be seen in the images up to 1800 LT. The EIA-related density enhancement disappears after about 1800 LT. Afterward, as shown by the image for 2000 LT, a shallow buildup of electron density occurs at latitudes centered around 35°N. As time progresses, the electron density in the lower-latitude end continues to decrease, while the density near the center and higher latitudes remains more or less steady, as seen in the images for 2000 and 2200 LT. The latitudinal gradient reverses at 2200 LT, with the northern latitudes showing higher densities compared to southern latitudes. Comparing all the images in Figure 1, we find that the electron density at latitudes beyond ∼36°N reaches diurnal maximum at around 2000–2200 LT. That is, in this latitude region the premidnight (2000–2200 LT) electron density is higher than the daytime density, which is the midlatitude summer nighttime anomaly (MSNA) [Thampi et al., 2009; H. Liu et al., 2010; Lin et al., 2010].

Figure 1.

The mean temporal evolution during July–August 2008 from tomography.

[15] Figure 2 shows the temporal evolution of the ionosphere at 135°E longitude obtained from F3/C RO observations during July–August 2008. As mentioned in section 2, to generate such images, the vertical profiles obtained using RO are binned in a latitude-longitude interval of 500 km. The time marked in each plot is the central hour of the 2 h interval. Although the trends mirror those in the tomographic images (Figure 1), the absolute magnitudes of electron density differ. The maximum differences in magnitude are ∼30%–35%, with the tomographic estimates being higher than those obtained by occultation; higher differences are found at latitudes below 30°N. The major difference is at 1400 LT. Tomographic reconstruction shows a considerable increase in EIA from 1200 LT to 1400 LT, while in the F3/C map the EIA remains more or less constant during this 2 h interval, with the enhancement occurring at 1600 LT and 1800 LT.

Figure 2.

The mean temporal evolution during July–August 2008 from FORMOSAT-3/COSMIC radio occultation measurements.

[16] Figure 3 shows a comparison of the LT variation of NmF2 (maximum electron density) from tomography with that from the Kokubunji ionosonde, along with the average NmF2 values obtained from RO data. The star symbols indicate the mean values obtained from tomography, and error bars show the standard deviations, representing the day-to-day variation. The Kokubunji data shown for comparison are the average values of July and August 2008. The individual observations also compared well [Thampi and Yamamoto, 2010]. It must be remembered that the Kokubunji latitude is close to the center of the reconstruction, where maximum accuracy is expected. The RO values are extracted from the map shown in Figure 2. It can be seen that the agreement is very good. This shows that the average seasonal behavior of the midlatitude F region ionosphere is represented reasonably well in the RO inversions.

Figure 3.

LT variation of electron density at Kokubunji, Japan, from tomography and F3/C measurements along with ionosonde observations. The error bars indicate standard deviations.

[17] Having considered the overall comparison, we now concentrate on two specific latitudes for a detailed comparison: 28°N, which represents the region where prominent daytime EIA-associated density enhancement is observed, and 40°N, which corresponds to the region where a prominent MSNA is observed. However, the tomographic inversion would be slightly less accurate at these latitudes than it is at 35°N since 28°N and 40°N are both outside the receiver locations. Even though these regions have path-pixel intersections from all three receivers, this degradation in accuracy may arise when the three receivers yield parallel projections that provide no additional information. It must be mentioned here that previous investigations using tomography have also considered regions 5°–6° outside the receiver latitudes to be accurate enough for studying the large-scale structures [see, e.g., Materassi et al., 2003; Kunitsyn and Tereshchenko, 2003].

[18] Figure 4 shows the Nmax variations corresponding to two latitudes: 28°N and 40°N. At 28°N, NmF2 starts to increase from ∼0600 LT up to 1800 LT. Thereafter, the electron density begins to decrease. The F3/C measurements agree fairly well with the tomographically reconstructed values, except for those recorded at 1400 LT. Another difference is that the rate of decrease of electron density from 2000 to 2200 LT is slightly higher in RO than in tomography. At 40°N, the diurnal maximum occurs at 2000–2200 LT, and a decrease is also seen at 1200 LT; both observations are characteristic of the MSNA [Thampi et al., 2009; H. Liu et al., 2010; Lin et al., 2010].

Figure 4.

LT variation of maximum electron density (Nmax) at 28°N and 40°N from tomography and F3/C measurements. The error bars indicate standard deviations.

[19] Figure 5 shows the LT variation of Ne at different altitudes at 28°N and 40°N, extracted from tomographic reconstructions, along with F3/C RO comparisons. These values are basically extracted from Figures 1 and 2 but facilitate a more quantitative comparison. The star symbols indicate the mean values obtained from tomography, and the error bars represent the standard deviations. At 28°N, the electron density starts to increase from ∼0600 LT, and the diurnal maximum occurs at 1400 LT except at 250 km altitude and below. This discrepancy around 1400 LT is discussed in detail in section 4.3. After 1400 LT, the electron density begins to decrease. F3/C data also show this pattern, although maximum electron density occurs at 1800 LT. Apart from this difference in the time of the diurnal maximum, the diurnal trends agree reasonably well in the altitude range of 250–400 km. At 200 km, the difference between tomography and F3/C is seen from 1200 LT, and the maximum difference in terms of magnitude is seen around 1800 LT. At 500 km, the differences between these two methods are large at all local times, as tomographically reconstructed values are always much higher than the RO values. An important point to be noted here is that, for tomography, the initial guess is taken from the IRI-2007 model, which is seen to constrain an overestimate of the topside (section 4.2). Meanwhile, RO may underestimate the electron densities at the F region and above because it assumes spherical symmetry and first-order estimation of the electron density at the apex altitude while inverting each profile [Tsai and Tsai, 2004; J.-Y. Liu et al., 2010b]. These issues are discussed in detail in section 4.

Figure 5.

LT variation of electron density at different altitudes at 28°N and 40°N from tomography along with F3/C measurements. The error bars indicate standard deviations.

[20] At 40°N, the electron density at 200 km derived from both techniques matches very well. Although the electron density starts to increase at sunrise, it shows a decreasing trend from 1000 to 1200 LT above 250 km. In the F3/C RO data this minimum occurs near 1000 LT instead of 1200 LT, as seen by the tomography. In both types of data the electron density begins to increase in the afternoon hours, and diurnal maximum occurs around 1800–2200 LT for altitudes above 250 km. The increase in electron density from 2000 to 2200 LT is greater in tomography than in the F3/C data, but both measurements show that densities at 2000–2200 LT are greater than the daytime values. At 500 km, the difference between these two methods is large at all local times.

[21] When we compare the tomographically reconstructed values of electron density with the RO-derived values, we can see that the difference between them maximizes around 1400 LT at 28°N. To see this effect at different latitudes, we compared the latitude variations obtained from tomography and F3/C. Figure 6a shows a comparison of densities at 300 km altitude corresponding to 0800 LT. The ionosphere is more or less “flat” at this time, before the development of EIA, and the electron densities derived by means of these two methods agree fairly well at all latitudes. However, at 1400 LT, the differences are quite large in the low-latitude sector (Figure 6b). If we compare the tomographically derived latitudinal variation with the RO data, we can see that the EIA is much underestimated in the RO data, and the difference is significant even if we consider that tomographic reconstructions of the low-latitude region are less accurate than those obtained for the region directly over the receiver locations. Hence, we can surmise that the EIA is not represented well in the RO data.

Figure 6.

Latitude variation of electron density at 300 km altitude at 0800 LT and 1400 LT from tomography and F3/C measurements.

4. Discussion

[22] Although the comparison of ionospheric densities measured by these two techniques, satellite radio occultation and ground-based radio tomography, reveals similar latitudinal gradients, there are certain differences. Since both techniques involve mathematical inversions, we need to consider the inherent assumptions of each and their limitations. Section 4.1 gives a brief account of the assumptions inherent in the RO inversions, RO validations using observations, and the simulation studies needed to understand the accuracy of RO inversion and possible sources of error. Similarly, for tomography, the accuracy of reconstruction depends on the initial guess used in the iterative algorithm. We have used the IRI model as the initial guess, which may have some limitations in the present deep solar minimum. To understand these potential limitations, we compare the IRI profiles with Jicamarca ISR measurements (section 4.2). A discussion of the observed differences between the RO inversion and tomography is provided in section 4.3.

4.1. Accuracy of RO Inversion

[23] The major assumptions and approximations used in the CDAAC RO inversion that may affect the accuracy of the inversion are (1) straight-line GPS signal propagation through the ionosphere, (2) spherical symmetry of the electron densities along the signal raypath, and (3) first-order estimate of the electron density at the apex altitude for inverting each profile, which is made to avoid the integral of Abel transform extending to infinity. Of these three, the one most likely to be a source of error is the assumption of spherical symmetry [Schreiner et al., 1999; Syndergaard et al., 2006; Lei et al., 2007]. As a result, RO-inverted profiles suffer from large errors (or even artificial structures) in the electron densities in the E region or below 250 km altitude at middle and low latitudes [Kelley et al., 2009; J.-Y. Liu et al., 2010a, 2010b].

[24] The F3/C electron density profiles were previously validated in several different ways. A comparison with Millstone Hill ISR (42.6°N, 71.5°W) showed that the height of the F2 peak and the shape of the profiles showed good agreement, while the peak electron densities in the F3/C RO data were larger than the ISR measurements by about 15% [Lei et al., 2007]. Similar comparisons were performed with Jicamarca ISR measurements. It was found that even though both measurements agreed fairly well, the agreement was not as good as with Millstone Hill. This discrepancy was attributed to the large horizontal gradients in the ionosphere that are ignored in the retrieval of electron density profiles from occultation measurements [Lei et al., 2007]. The comparison of F3/C-retrieved peak electron density with the densities measured by 31 globally distributed ionosondes under the condition that the latitude-longitude differences between the F3/C occultation and the ionosondes were less than 2° yielded a strong correlation, with a correlation coefficient of 0.85 [Lei et al., 2007]. It was shown that the reconstructed values of foF2 agree at the 13% RMS level, with a marginally significant mean difference. However, some inaccuracies in the F3/C RO profiles have been reported recently. The F3/C occultation-based profiles were also compared with the Arecibo ISR measurements, and it was found that in several cases the RO profiles gave lower density values compared to the ISR values and that the altitude determination yielded underestimated values below 300 km and overestimated values above 300 km [Kelley et al., 2009].

[25] Recent simulations of the RO inversion show that the RO-inverted profiles for the low-latitude EIA regions are significantly affected by the greater electron density of the EIA crests, resulting in large errors in the measurement of electron densities below 250 km altitude [J.-Y. Liu et al., 2010b]. These simulations also show that artificial plasma caves are created underneath the EIA crests and that the RO inversion reconstructs less distinct EIA crests with underestimation of the electron density. J.-Y. Liu et al. [2010b, Figure 2] show that the electron density of RO inversion around 20°N–30°N geographic latitude has a negative error (underestimation). Although the error in the topside is relatively small, the underestimation is still around 20%–30%.

4.2. Performance of IRI During the Deep Solar Minimum

[26] As mentioned in section 2, the IRI-2007 model is used to obtain the initial guess required for the iterative inversion algorithm used for tomographic reconstruction. It is now well known that the minimum of solar cycle 23/24 was quite unusual since it had a record number of days without sunspots [Livingston and Penn, 2009]. There is no representative data set included in IRI from a comparable solar minimum. Hence, it is quite reasonable to expect that IRI may be unable to predict the ionospheric conditions correctly during the years 2008 and 2009. In a recent study, Lühr and Xiong [2010] compared the electron density predictions of the IRI-2007 model with in situ measurements of the CHAMP and GRACE satellites for the years 2000–2009 and found that from 2005 onward the overestimation of the electron density by the model has been progressively increasing. In this context, to understand the accuracy of the IRI predictions during our period of observation, we compared the model outputs with the ISR measurements from Jicamarca. For this comparison, the IRI model predictions for the Jicamarca ISR location were generated for July 2008 and compared with the ISR profiles. Figure 7 shows this comparison. It can be seen that the IRI overestimates the densities at all local times. This overestimation is greater in the topside compared to the bottomside ionosphere, especially during daytime. For instance, at 1200 LT and 1600 LT, the differences are small up to ∼270 km. The topside has significantly high values in the model predictions even during these times. At 2100 LT, the profile is completely overestimated by the model.

Figure 7.

Comparison of IRI-2007 profiles with Jicamarca incoherent scatter radar measurements. Both correspond to July 2008.

[27] Nonetheless, it must be mentioned here that the largest differences between modeled and measured Ne appear at equatorial and low latitudes compared to the midlatitude region [e.g., Lühr and Xiong, 2010]. This could be due to the fact that the midlatitude ionosphere is better represented (with a longer and denser database) in the model than is the equatorial region. Hence, the discrepancies shown here using Jicamarca data are representative of the maximum deviations, and we expect the performance of the IRI to be better in the midlatitude region, especially in the bottomside. However, for the topside, we can expect a significant overestimation (similar to the case discussed in section 3) even at midlatitudes because the database assimilated into the model is mostly from ground-based ionosonde measurements, which provide information only about the bottomside.

4.3. Comparison of RO Inversion With Tomography

[28] When we compare the RO inversion with tomographic reconstruction, certain differences are seen at all altitudes and latitudes. In general, the electron density estimated by tomography is systematically higher than that estimated by RO, which could be due to the fact that the IRI overestimates the density values of this deep solar minimum, and hence the tomographic inversion produces overestimated results everywhere. Nonetheless, while examining the differences, two features can be recognized: (1) the differences are greater in the low-latitude region than in the midlatitude region, and (2) the differences are greater in the topside than in the bottomside (near the F2 peak). The first feature indicates that a major contributor to the difference could be the assumption of spherical symmetry that is inherent in RO inversion. It must be mentioned here that with tomography the maximum accuracy is expected to be seen from the F layer peak downward, where the maximum separation in path angles occurs, and directly over the receiver locations. The second feature is due to the overestimation of the topside by the IRI, as is evident in the comparison with ISR measurement.

[29] The maximum electron density (Nmax) values obtained from the tomographic reconstructions were compared with the values derived from the ionosonde located at Kokubunji (near the center of the plane of reconstruction); the correlation was found to be ∼0.93, and the significance was >95% [Thampi and Yamamoto, 2010], which corresponds to the maximum accuracy. However, even if the accuracy degrades, the differences are still significant, and the observations suggest that the EIA is not represented in the F3/C RO inversions. The magnetic equator at this longitude is at about 9°N, where we expect the electron densities to show the EIA trough at 1400 LT, and this feature is completely absent from the latitude variation obtained with RO. This indicates that the observed differences are closely linked to the inaccuracy in the RO inversion. The fact that the differences are smaller at 0800 LT, before the EIA develops, also suggests this possibility. However, since we do not have any ground-based measurements over the areas in the low-latitude region at this longitude, we cannot provide any quantitative estimate of the accuracy of the tomographically obtained values over the low-latitude region, and hence we cannot provide any quantitative estimate of how much the EIA is underestimated by RO.

[30] The fact that the differences are small at altitudes near the F2 peak in the midlatitude region, especially in the comparison of NmF2 with the ionosonde at Kokubunji (Figure 3), indicates that both techniques achieve maximum accuracy over this region. For the topside, the tomography technique has more limitations than does RO. The path-angle separation declines in the topside, and the IRI model also tends to overestimate the topside ionosphere; both factors result in a degradation of accuracy. As mentioned just above, with tomography, the maximum accuracy is expected from the F layer peak downward, where the maximum separation in path angles occurs, and directly over the receiver locations. Hence, the fact that differences between these two inversions increase (even in the 30°N–39°N region, as shown in Figures 1 and 2) when significant gradients are present indicates the limitation associated with the spherical symmetry assumption of RO inversion.

[31] The comparison with tomography also suggests that the EIA is not well represented in F3/C RO data. This could be due to the following: (1) the assumption of a horizontally symmetric stratification of the ionosphere, inherent in the RO inversion, which is not valid for the EIA region, or (2) the first-order estimate of the electron density at the apex altitude, made to avoid the integral of Abel transform extending to infinity. Meanwhile, the observation validation and simulation studies have shown that the strength of the EIA crests is underestimated by the RO observations [cf. Coker et al., 2009; J.-Y. Liu et al., 2010b], a conclusion that agrees with the comparison carried out here. The disagreements between the RO data and the tomographic data grow larger as the latitudes descend. The northern EIA crest is normally located near 20°N–25°N geographic latitude (∼11°N–16°N magnetic latitude) [see, e.g., Lin et al., 2007; Tulasi Ram et al., 2009], and the disagreement is largest around there.

5. Conclusion

[32] The tomographic observations of the ionosphere during July–August 2008 are compared with the FORMOSAT-3/COSMIC (F3/C) radio occultation measurements. The two sets of observations agree well qualitatively and bring out the main features of the low-latitude to midlatitude ionosphere. The enhancement seen in the 26°N–32°N latitude region during daytime is found to be related to the equatorial ionization anomaly (EIA), and that observed over latitudes northward of ∼35° at night is related to the midlatitude summer nighttime anomaly, which is actually a phase reversal of the diurnal cycle of electron density at midlatitudes [Thampi et al., 2009; H. Liu et al., 2010; Lin et al., 2010]. The comparison with tomography shows that the general morphology of the F region ionosphere would not be significantly affected by the RO inversion error, except for the EIA feature. The EIA crest densities are found to be underestimated in the F3/C data. The differences could be due to the fact that horizontally symmetric stratification of the ionosphere, inherent in the RO inversion, is not valid for the EIA region. For the topside, the tomography technique has more limitations than does RO. The path-angle separation declines in the topside, and the IRI model also tends to overestimate the topside ionosphere, and both these features result in the degradation of accuracy. Keeping these differences in mind, we can conclude that the F region electron density structures and morphologies observed by RO and the averaged images obtained by tomography are reliable over the midlatitude region.


[33] The work of S.T. and H.L. is supported by the Japan Society for the Promotion of Science (JSPS) foundation. S.T. also thanks the Indian Space Research Organization. The authors thank NICT, Japan, for the ionosonde data, UCAR CDAAC (, TACC of Taiwan CWB for the F3/C data, and the NSPO operational team for the F3/C triband beacon experiment. We thank the Jicamarca Radio Observatory for providing the ISR data. We also thank the reviewers for their pertinent comments.