We present electron density profiles derived by inversion of ultraviolet limb radiances observed during November 2000 through April 2001 by the low resolution airglow and aurora spectrograph (LORAAS) instrument on the Advanced Research and Global Observing Satellite (ARGOS). The O+ density profile, which is approximately equal to the electron density profile in the F region ionosphere, was determined by inverting the limb radiance profile of O I 1356 Å emission of atomic oxygen; the 1356 Å emission is produced purely by radiative recombination of O+ ions and electrons at night and has been shown to be a viable means of globally sensing the ionospheric state. Dymond et al. (2001b) showed a preliminary validation of this technique covering a single day of observations. We have performed a more extended validation of the retrieved peak electron density and peak height determined by inversion of nighttime observations of the 1356 Å altitude profiles with coincident ionosonde measurements and climatology.
 The low resolution airglow and aurora spectrograph (LORAAS), flown on the Advanced Research Global Observation Satellite (ARGOS), observed the ionosphere and neutral thermosphere, beginning in mid-May 1999 and continuing observations until early April 2002. LORAAS was an ultraviolet limb scan imager which gathered vertical profiles viewing aft of the ARGOS, covering the 75–750 km altitude range with ∼5 km altitude resolution. Nighttime limb scan observations were inverted to determine the ionospheric electron density as a function of altitude. The retrieval algorithm utilizes ultraviolet spectral lines at 911 and 1356 Å, which are sensitive to emissions from radiative recombination of electrons and oxygen ions [Dymond and Thomas, 2001; Dymond et al., 1997]. Electron densities retrieved by inversion of single limb scans were successfully compared to ionosonde observations of peak density, NmF2, and peak height, hmF2, [Dymond et al., 2001a, 2001b]. Additionally, multiple sequential limb scans were inverted using a two-dimensional algorithm and compared with coincident total electron content (TEC) data from TOPEX satellite passes [Coker et al., 2004] validating the shape of the electron density profile and the determination of the equatorial anomaly peaks. Additional testing is required to validate the density, height, and shape of the retrieved ionospheric profiles over a wider range of local times, seasons, geomagnetic conditions, and solar activity levels.
 This paper presents comparisons of single limb scan ionospheric retrievals with coincident observations of the ionosphere from globally distributed ionosondes and climatology over a period from 1 November 2000 to 30 April 2001. This provides a validation of the limb scan radiance observations during three seasons, from middle fall into spring.
2. LORAAS Instrumentation and Algorithm
 ARGOS is in a Sun-synchronous 832 km, 98° inclination orbit with a descending node at 0230 LT. The LORAAS instrument observes the naturally occurring UV emissions produced by radiative recombination on the Earth's limb. It has a field of view of 2.4° × 0.08° and sweeps out a 2.4° × 17° field of regard during each 90 s scan in the aft direction of the satellite. The observed wavelength range is 800–1700 Å with 19 Å resolution [McCoy et al., 1992, 1994].
 Approximately 90 spectra, with 1-s integration, are gathered per limb scan. During each scan, the tangent altitude ranges from 750 to 75 km. When the motion of the spacecraft is accounted for, each limb scan is separated by approximately 5°–6° in latitude at the equator.
 The 1356 Å spectral line is used during nighttime observations of the ionosphere because it is brighter than the 911 Å emission. The higher brightness yields a more precise determination of the ionospheric parameters, as the signal to noise of the limb data is higher. In the F region, the intensity of the radiative recombination emission is proportional to the integral along the line of sight of the electron density squared [Dymond et al., 1997]. The instrument sensitivity degrades over time as the detector ages, requiring the instrument calibration factor to be estimated and used to scale the observed count rate. For the period in this investigation, a value of 0.17 counts s−1 R−1 was used for the sensitivity coefficient of the instrument as derived from earlier studies [Coker et al., 2004].
 The dependence on density squared combined with long horizontal path lengths through the ionosphere causes the limb scan observations to be most sensitive to the ionosphere in the vicinity of the F region peak. Whether height or density variations, gradients in the ionosphere contribute to neighboring limb scans as either foreground or background emissions. These affect the performance of the one-dimensional retrieval algorithm and need to be considered when comparing with ionosonde data. The algorithm fits a three-parameter Chapman layer to the inverted radiance profile in order to represent the electron density at the location of each limb scan. Generally, a Chapman layer is a reasonable approximation for describing the ionospheric density. However, a more robust parameterization is preferred at low latitudes. The inversion algorithm [Dymond and Thomas, 2001] is based on discrete inverse theory [Menke, 1989] and uses the iterative Levenberg-Marquardt scheme [Press et al., 1992] to seek a maximum likelihood estimate of the ionospheric parameters based on the fit of the model to the data.
3. Comparison of Limb Scan and Ionosonde Data
 Ionosonde peak density, NmF2, and peak height, hmF2, data were obtained from the National Geophysical Data Center for four globally distributed stations, Eglin, United States; Ahmedabad, India; Darwin, Australia; and Learmonth, Australia. The magnetic latitudes of the stations (42°N, 16°N, 22°S, and 33°S, respectively) include midlatitudes and low latitudes in the Northern and Southern hemispheres. Limb scans with tangent points within 5° latitude, 15° longitude, and 30 min of the ionosonde observations were selected for comparison.
 Before presenting the comparisons, a brief discussion comparing the observation geometries is in order, as these are a source of error. Ionosondes sample the ionosphere vertically and provide little, if any, information outside the immediate vicinity of the station. In contrast, limb scans from orbiting satellites sample the ionosphere across large horizontal ranges (∼2000 km). The electron density profile retrieved from a limb scan and assigned to the tangent point of the limb scan can be thought of as an average profile over the scanned ionosphere. If the scanned ionosphere is smooth and linear, then the average is an accurate representation of the middle of the scan. If the ionosphere is structured (has nonlinear gradients), then the average is not as accurate a representation of the middle of the scan. If the ionospheric structure is persistent, then systematic errors will bias the comparison between limb scans and ionosonde observations. Additionally, in the presence of large ionospheric gradients, differences in the location and timing of the limb scan and the ionosonde observations will contribute error to the comparison.
Figure 1 compares limb scan and ionosonde peak density observations at the four stations over a 6 month period. Additionally, the climatological trend in peak density is shown from the International Reference Ionosphere (IRI) model [Bilitza, 2001]. For Eglin (42°N magnetic latitude), the limb scan densities follow the ionosonde densities and the climatology quite well. The average difference between limb scan and ionosonde peak density is −0.2 × 105 cm−3 with a sigma of 1.1 × 105 cm−3. One notable deviation from the climatology is observed near the end of March 2001 and is associated with a period of high solar activity and major geomagnetic storms. For Ahmedabad (16°N magnetic latitude), limb scan densities follow the climatological trend for the 6 month period with perhaps twice the daily variation than the midlatitude Eglin data. This is attributed to increased variability in the low-latitude ionosphere. Again there is a departure from climatology near the end of March, associated with increased solar and magnetic activity. The ionosonde data record is not complete for this station and departs somewhat from the limb scan data and climatology starting in mid-February to the end of the period. The average difference between limb scan and ionosonde densities is 1.9 × 105 cm−3 with a sigma of 2.2 × 105 cm−3.
 For Darwin (22°S magnetic latitude), the limb scan densities follow the climatology and the ionosonde data but tend to report lower average densities. The average difference between limb scan and ionosonde peak density at Darwin is −2.6 × 105 cm−3 with a sigma of 2.2 × 105 cm−3. Density enhancements associated with increased solar and magnetic activity in late March are seen in both the ionosonde and limb scan data. For Learmonth (33°S magnetic latitude), the limb scan densities follow the climatology and the ionosonde data but tend to report lower average densities. Upon closer inspection of the raw spectral data, there is evidence that most of the scans near Learmonth are affected by auroral particle noise as the spacecraft approaches the high-latitude region. This increases the distortion of the spectral information and results in an underestimation of the 1356 Å radiance and the resulting peak density. For this reason, Learmonth is ignored in the following discussions pertaining to average density differences; however, Learmonth is included in discussions pertaining to the peak height, since the particle noise has little effect on the determination of the height of the layer.
 The average density differences and sigmas for all four stations are summarized in Table 1. The low latitudes (Ahmedabad and Darwin) have consistently larger sigmas than the midlatitudes (Eglin and Learmonth). This is attributed to increased density structure in the low latitudes compared with midlatitudes. The average differences are consistent with those expected from the one-dimensional (1-D) algorithm. Dymond and Thomas  showed that the 1-D algorithm underestimates the peak density at the crest of the dominant tropical arc by about 30%, overestimates the peak density by about 15% at the crest of the lesser tropical arc, overestimates the peak density by about 40% immediately poleward of the arcs, and accurately estimates the peak density at midlatitudes far from the tropical arcs. Climatology suggests that for much of the period under investigation, the dominant arc has a peak density at 0230 local time near 10 × 105 cm−3, whereas the lesser arc has a peak density near 7 × 105 cm−3. As seen in Figure 1b, the ionosonde peak densities suggest that Ahmedebad is situated either near a lesser arc or poleward of the arc. This implies an overestimation of the peak density by the 1-D algorithm and is consistent with the observed 38% overestimation. Similarly, as seen in Figure 1d, the ionosonde peak densities suggest that Darwin is situated beneath a dominant arc. This implies underestimation of the peak density and is consistent with the observed 29% underestimation. Eglin is situated at midlatitudes far from the tropical arcs, where limb scans accurately estimate peak density.
Table 1. Limb Scan and Ionosonde Peak Density Differences at Four Stations Over a 6 Month Period
−0.2 × 105
1.1 × 105
1.9 × 105
2.2 × 105
−2.6 × 105
2.2 × 105
−1.7 × 105
1.3 × 105
Figure 2 compares limb scan and ionosonde peak height observations at the four stations over the 6 month period. Additionally, the climatological trend in peak height is shown using IRI. The limb scan heights follow the climatology and ionosonde heights quite well. In particular, note the oscillations in November and December data for Darwin and Learmonth. The limb scan heights track these systematic variations observed by the ionosondes. The average height differences and sigmas for all four stations are summarized in Table 2. The average differences are remarkably within ±8 km for all four stations, and the sigmas range from 29 to 44 km, with evidence for better performance at the lower latitudes. This stands in contrast to the density comparisons, where sigmas increased at low latitudes. This may be attributed to the fact that the height gradients are not as dramatic as density gradients in the low latitudes and that the emission is brighter in the low latitudes, providing better discrimination of the peak emission height late at night. Although the data are not presented here, similar results are observed for the limb scan estimates of topside scale height. The brighter emission at low latitudes allows for a more precise estimation of the topside scale height.
Table 2. Limb Scan and Ionosonde Peak Height Differences at Four Stations Over a 6 Month Period
Figures 3a and 3b display the limb scan data versus the ionosonde data on scatterplots for peak density and peak height, respectively. Three distinct populations can be observed in the peak density comparison, depending on location. Limb scans accurately estimate peak density at midlatitudes near Eglin far from the tropical arcs, overestimate peak density near Ahmedebad, and underestimate peak density near Darwin and the dominant tropical arc. Note that there are a few outliers for the Ahmedebad population, where the ionosonde peak densities are much higher, indicating the presence of a stronger tropical arc and resulting in less overestimation. Similar outliers for the Darwin population are observed, resulting in less underestimation.
 The peak height comparisons do not show evidence of divergent populations; therefore the data are presented as one population (see Figure 3b). Using all four stations provided 603 samples over the 6 month period, resulting in a mean difference between limb scan and ionosonde peak heights of −1 km and a standard deviation of 39 km. This compares favorably with peak height RMS error estimates of ∼45 km reported for GPS radio occultations when compared with ionosondes [Jakowski et al., 2002, 2004]. Assuming that the ionosonde peak height error is on the order of 20 km at night, the limb scans estimate peak heights with a precision approaching 33 km, or 10%. The 1-D algorithm produces reasonably accurate estimates of peak height.
 One-dimensional reconstructions of nighttime ionospheric profiles were obtained from UV limb scans on the ARGOS satellite. Comparisons of limb scan ionospheric retrievals with climatology and coincident ionosonde data from November 2000 through April 2001 demonstrated that the limb scans produced expected results. Peak densities and heights followed climatological trends and daily variations over the 6 month period. Peak heights were estimated with reasonable accuracy (10%) at a variety of latitudes, with slightly better performance at low latitudes. Peak densities were estimated with reasonable accuracy at midlatitudes (20–30%), where the ionospheric gradients were minimal, but suffered from consistent underestimation and overestimation at low latitudes, as predicted by earlier simulations of the 1-D algorithm [Dymond and Thomas, 2001]. These systematic errors arise directly from a limitation of the algorithm, which assumes that the ionosphere is spherically symmetric. Density gradients associated with the tropical arcs violate this assumption, causing the algorithm to underestimate the density of the dominant tropical arc and to overestimate the density at latitudes adjacent to the tropical arc. This study represents a validation of nighttime, 1356 Å limb scan–retrieved ionospheric profiles using the 1-D algorithm over three seasons, at midlatitudes and low latitudes in the Northern and Southern hemispheres. A two-dimensional algorithm has been developed [Dymond and Thomas, 2001] and is being tested [Coker et al., 2004], which takes advantage of multiple limb scans in the orbit plane of the satellite to overcome the limitations of the 1-D algorithm.
 This work was supported by the Office of Naval Research.