Radio Science

Ionospheric response to the solar flare of 14 July 2000



[1] We present our observations of the global ionospheric response to the X-class solar flare that started the Bastille Day storm on 14 June 2000. The observations were made using the Low-Resolution Airglow and Aurora Spectrograph (LORAAS) instrument on the Advanced Research and Global Observation Satellite (ARGOS). The ARGOS is in a Sun-synchronous orbit at 0230/1430 LT at approximately 840 km altitude. During the daytime the LORAAS observes the 911-Å emission; this emission is produced by radiative recombination of F region O+ and electrons. We simultaneously invert approximately one-half of an orbit of 911-Å limb scans (approximately 30–45 min of observing) using a tomographic inversion technique to produce dayside electron density maps in the orbit plane. We compare our observations of orbits immediately before and after the flare to the orbit in which the flare occurred. We observed a 41% increase in the average 911-Å brightness during the flare orbit compared to the orbits before and after the flare. This corresponds to an overall increase of electron density by ∼20%. This density enhancement has largely decayed by the next orbit. At altitudes near the F region peak, 250–450 km, the density enhancement has decayed to the preflare value. An enhancement of the electron densities at approximately −50° geomagnetic latitude has moved northward in the flare orbit, which may be indicative of enhanced meridional transport postflare. We compared our results to output from the SAMI-2 model, which was run for the flare study presents in the work of Meier et al. [2002].

1. Introduction

[2] Observation and inversion of the emission that results when O+ ions and electrons recombine has been proposed and demonstrated as a means of remotely sensing the electron density in the daytime and nighttime F region ionosphere [Chandra et al., 1975; Tinsley and Bittencourt, 1975; Meier, 1991; Dymond et al., 1996, 1997, 2001a]. Dymond et al. [2001a] demonstrated that the radiative recombination emission at 1356-Å could be used to accurately determine the nighttime electron density; the electron densities they derived were in good agreement with coincident ionosonde measurements. Dymond et al. [2001b] demonstrated that the recombination emissions at 911-Å could be used to accurately infer the daytime F region electron density. The 911-Å emission was not expected to be of any use for ionospheric sensing because it was thought to be too heavily contaminated by molecular and atomic nitrogen emission. Dymond et al. [2001b] showed that the contamination does not affect the accuracy of F region electron densities; they demonstrated accurate electron density determinations during coincident ionosonde overflights.

[3] Dymond and Thomas [1999, 2001] and Kamalabadi et al. [1999] recently demonstrated tomographic inversion algorithms for determining the two-dimensional, altitude versus latitude, distribution of the F region electron density. These algorithms simultaneously invert a set of limb observations of the recombination radiation to produce the electron density.

[4] Previous observations of the electron density enhancements owing to solar flares have been made using ground-based techniques. Thome and Wagner [1971] observed the changes in the F region ionosopheric electron density owing to class 2B flares using incoherent scatter radar. They found that the electron density in the F region increased by roughly 15% and that the increased density decayed out with a time constant of approximately 1 hour at F region heights. Additionally, they observed small electron density decreases on the topside. Mendillo et al. [1974] studied the global response of the ionosphere to the 3B solar flare on 7 August 1972 using total electron content (TEC) measurements derived by measuring the change in the Faraday rotation of signals transmitted by geostationary satellites. They found that the TEC changed by 15–30% and that there was some latitudinal structure to the response. Mendillo and Evans [1974] followed up the work of Mendillo and Evans using the incoherent scatter radar at Millstone Hill, MA. They observed a 60% enhancement of the electron density in the F region at 199 km and 20% at 300 km with the magnitude of the enhancement increasing with altitude above 400 km. The height dependence of the electron density enhancement and its magnitude were shown to be consistent with thermal expansion of the ionosphere owing to heating by the flare. More recent work by Afraimovich [2000] using ground-based TEC measurements derived from global positioning satellite observations demonstrates the use of the global distribution of GPS receivers to study the global response of the ionosphere to solar flares.

[5] A unique opportunity to observe the ionospheric response to a large change in solar X-ray and extreme-ultraviolet emission occurred on 14 July 2000 (Bastille Day storm). During the solar flare the ionizing flux changed dramatically, which resulted in a near-impulsive change in the ionizing radiation that creates and maintains the ionosphere. Yet the geomagnetic impact of the subsequent geomagnetic storm had not yet occurred; thus the ionospheric production and loss mechanisms could be observed directly with little or no interference from dynamical effects associated with the geomagnetic storm.

[6] The Bastille Day geomagnetic storm was initiated by an X-class solar flare on 14 July at approximately 1000 UT, followed by a coronal mass ejection (CME). Approximately 25–30 min after the flare occurred, high-energy particle precipitation in the polar caps was observed by ARGOS and several other near-Earth orbiting satellites. These particles were generated when the CME shock front overtook the slower solar wind resulting in acceleration of the ambient charged particles to ∼20–30% of the speed of light. The high-energy particle precipitation persisted over both poles for approximately 38 hours when the CME affected Earth's magnetosphere, initiating the very strong geomagnetic disturbance. During this storm the 3-hour ap reached its maximum value of 400. Meier et al. [2002] and Immel et al. [2003] studied the radiative phase of the Bastille Day storm. (We define the radiative phase as the portion of the storm starting with the flare yet before the high-energy particle precipitation had substantially heated the thermosphere, ∼1000–1200 UT on 14 July.) Both studies observed 40–70% increases in the photoelectron impact excited emissions from atomic and molecular nitrogen and atomic oxygen. Meier et al. also calculated the thermospheric heating rates owing to the increased EUV flux and found the heating rates to be consistent with observations. They also calculated the change in electron density resulting from the flare input and predicted increases of 20–30% in the F region and ∼40% in the E and F1 regions.

[7] We have observed the 911-Å emission altitude profiles during the radiative phase of the Bastille Day event using the LORAAS instrument on ARGOS and inverted the altitude profiles using the Dymond and Thomas [2001] algorithm, assuming negligible contamination from the nitrogen emissions. We consider three orbits: the orbit immediately before the solar flare occurred (orbit 5), the orbit in which the flare occurred (orbit 6), and the orbit immediately after the flare occurred (orbit 7). We observed an increase in the average 911-Å brightness during the flare orbit compared to the preflare orbit. We also determined that the electron density increased from the preflare orbit to the flare orbit and that the electron density had largely decayed to preflare values in the postflare orbit. There was, however, an increase in the topside electron density in the postflare orbit, indicating that some of the electrons generated by the flare had been transported to higher altitudes.

2. Data

[8] The data used in this study were acquired by the Low Resolution Airglow and Aurora Spectrograph (LORAAS) [Dymond and McCoy, 1993; McCoy et al., 1992, 1994; Thonnard et al., 1999], which is part of the High Resolution Airglow and Aurora Spectroscopy (HIRAAS) experiment. The LORAAS observes the extreme- and far-ultraviolet airglow in the 80–170 nm passband at approximately 1.7 nm resolution. The LORAAS instrument is orbiting on the U.S. Air Force (USAF) Space Test Program's Advanced Research and Global Observation Satellite (ARGOS or STP 91-1). The ARGOS was launched into a Sun-synchronous (0230/1430 LT) circular orbit at 840 km altitude on 23 February 1999. The LORAAS observed continuously from mid-May 1999 through 8 April 2002, when the ARGOS satellite was turned off.

[9] The LORAAS acquires a spectrum covering the 800–1700 Å passband each second. Ninety spectra are gathered during each limb scan in the normal operating mode; these spectra are “stacked” to form limb images. The ARGOS satellite moves approximately 5.3° of latitude during a limb scan (Figure 1). The LORAAS's field-of-view is mechanically swept from 10° to 27° below the satellite's local horizon by a scan mirror to cover tangent point altitudes of 750 to 100 km. Each limb scan constitutes a vertical sounding of the ionospheric electron density.

Figure 1.

The figure illustrates the observing concept for the HIRAAS experiment. The vehicle is traveling from right to left on the picture. The HIRAAS experiment views aft in the orbit plane. The wedges in the picture indicate the sweeps of the instrument's field-of-view across the limb. A limb scan is acquired every 90 s yielding a limb profile approximately every 5.6° of latitude when near the equator.

[10] Limb profiles of the 911-Å emission were produced by summing over the 890–920 Å spectral range and correcting the resulting profile for scattered light. The two primary contributors to the instrumental scattered light are H I 1216 Å (Lyman-α) and O I 1304 Å. Because the Lyman-α line shape has evolved since launch owing to a degradation of the microchannel plate in the detector, the Lyman-α line shape was determined empirically by coadding nightglow spectra at tangent point altitudes above 700 km. The dominant emission at high altitudes at night is Lyman-α with a much weaker emission at 1026 Å (Lyman-β). The ratio of the Lyman-α emission in the 890–920 Å passband to the emission in the line core (1200–1230 Å) was then calculated. This ratio was used to scale the dayglow Lyman-α emission in the line core to determine the scattered component in the 890–920 Å passband. A similar procedure was used to determine the scattered light owing to the O I 1304 Å emission. However, the 1304 Å altitude profile we used was first corrected for scattered Lyman-α, then scaled and subtracted from the 911-Å profile. We assume that the daytime 911-Å emission tends to zero at high altitudes, as shown by model simulations. We then assume that any residual emission at high altitudes, in our case above 720 km, is due to scattered Lyman-α and 1304-Å emissions. The 1304-Å scaling factor was determined by calculating the 911-Å emission at altitudes above 720 km and taking the ratio to the 1304-Å emission above 720 km altitudes. This scalar was then applied to the 1304-Å emission profiles and subtracted from the 911-Å profiles. Counting uncertainties in the 911-Å emission profile are calculated using standard propagation of errors techniques [Bevington, 1969]. Once the scattered light correction was performed, the limb profiles were then converted to radiance units by dividing by the instrument's sensitivity at 911-Å, 0.3 counts s−1 Rayleigh−1. These limb profiles were then fit using the algorithm described in the work of Dymond and Thomas [2001] and the ionospheric parameters were extracted.

3. Data Inversion Approach

[11] The Dymond and Thomas [2001] algorithm parameterizes the electron density profile using the conventional Chapman layer representation [Chamberlain and Hunten, 1987]. The Chapman representation uses three parameters to characterize the ionosphere: the altitude where the density peaks, hmF2; the density at the peak, nmF2; and the O scale height, H, which is one-half the plasma scale height for the form of the Chapman layer given below. The Chapman function for describing the O+ density, equation image(z), is:

equation image

In the algorithm we assume that the electron density equals the O+ density, which is approximately correct below the H+/O+ transition height in the midlatitude F region. This assumption is less accurate in the low-latitude ionosphere. Separate Chapman layers were placed at the tangent point location of each limb scan, approximately every 5.3° degrees of latitude at the equator. These individual profiles were uncoupled, and the parameters for each were varied separately. Since each limb scan contains information from several latitude cells, the Chapman layers for each latitude cell are not necessarily linearly independent, especially in the presence of measurement noise. Thus the solution for the densities can be very “rough” in that large variations of electron density can occur from cell to cell. To reduce the roughness, we impose a physicality constraint by enforcing smoothness in the solution. Smoothness is enforced by applying regularization to the electron density field at each iteration of the solution. The selection of the optimal regularization weight is discussed in greater detail below.

[12] The inversion algorithm uses an iterative approach based on discrete inverse theory [Menke, 1989] to seek the maximum likelihood estimate (minimum of the chi-squared statistic, χ2, see Press et al. [1992] for the definition of χ2) of the ionospheric parameters on the basis of the fit of the model to the data. The forward model used the electron density; the electron density was then squared and the volume emission rate was calculated. Then the volume emission rates were integrated to yield the modeled intensities. These modeled intensities were compared to the observed intensities to evaluate the χ2 [Dymond and Thomas, 2001]. We used the Levenberg-Marquardt scheme [Press et al., 1992] to adjust the parameters and minimize the χ2. The fit to the data was deemed to be good when the χ2 changed by less than 0.1% between iterations. Once the minimum χ2 was found, the covariance matrix was evaluated and used to estimate the uncertainties in the retrieved parameters. The uncertainty of the peak density, nmF2, and hence the whole profile, was adjusted to include the calibration uncertainty (10%) and the uncertainty in the radiative recombination rate coefficients (5%; Melendez-Alvira et al. [1999]). These additional uncertainties were halved and added in quadrature to the uncertainty in the retrieved nmF2 based on counting statistics. (The uncertainties are halved because the retrieved nmF2 is proportional to the square root of the calibration factor and to the square root of the radiative recombination rate coefficient. See Bevington [1969] for details.) The uncertainties in the retrievals are approximately ∼15% for the peak density, ∼10 km for the plasma scale height (∼10%), and ∼20 km for the peak heights (∼15%).

[13] The selection of the regularization weight was accomplished using the “zero order regularization method” [Press et al., 1992]. Enforcing any form of regularization is equivalent to reducing the number of degrees of freedom allowed in the fit. The zero-order regularization method starts by evaluating the χ2 statistic using a very small regularization weight. This usually produces a rough solution for the electron density but one that yields the smallest χ2. The zero-order regularization method generates a more realistic χ2, reflecting the reduction of the number of degrees of freedom in the fit via regularization, by adding unity to the minimum reduced χ2 with no regularization, to produce a χ2 representative of the regularized solution, χF2. We used a bisection search algorithm to vary the regularization weight until the χ2 for the fit was equal to χF2. At each step in this search process the full Levenberg-Marquardt solution to minimize χ2 was carried out. The bisection search was terminated when the χ2 was equal to χF2 ± 0.1. We have determined that this approach to yields a value for the regularization weight that is nearly equal to that derived using the L curve criterion to determine the optimum regularization weight [Schrimpf and Schreier, 1997].

4. Results

[14] Figure 2 shows the 911-Å emission intensities for the three orbits considered in this work. The orbits 5, 6, and 7 crossed the equator at 75.6°E, 54.8°E, and 29.5°E longitude, respectively. The ARGOS satellite is traveling northward viewing aft (southward) during the daytime. The ARGOS satellite crossed the equator at approximately 8.89, 10.58, and 12.28 UT during orbits, 5, 6, and 7, respectively. Orbit 5 is the preflare orbit that will be used as the baseline for the comparisons of electron densities and 911-Å intensities. As seen in the figure the 911-Å intensity increased by approximately 40%, on average, in orbit 6 compared to orbit 5. This is especially prominent in the F region at altitudes above 200 km where the brightness increased by 40%. This intensity increase corresponds to an average electron density increase of approximately 20%. At altitudes below 200 km the intensity increase is 25%. We note that the 911-Å emission is produced by radiative recombination of O+, which is the dominant ion in the F2 region but is becoming less important in the F1 and E regions; thus the LORAAS data are relatively insensitive to the F1 and E region electron density. Figure 2c, shows that the intensity of the postflare orbit (orbit 7) had decreased to nearly the intensity of the preflare orbit.

Figure 2.

(a–c) The 911 Å altitude versus latitude maps for the preflare, flare, and postflare orbits, respectively. The signal-to-noise ratio in these images is approximately 10.

[15] Figure 3 shows the results of inverting the 911-Å emission intensities using the Dymond and Thomas [2001] tomographic inversion technique. The LORAAS densities are most accurate in the 250–600 km range. Below 250 km, O+ is no longer the dominant ion, and the tomographic inversion algorithm underestimates the electron density. Above 600 km the signal-to-noise ratio in the intensity data is less than unity, which indicates that little information is available to constrain the density. Thus the data used in the inversions were constrained to lie between 200 and 600 km to suppress spurious results due to noise. Figure 3b indicates that the electron density distribution changed considerably during and immediately following the flare. Figure 3c shows that the electron density has decayed to a distribution similar to the preflare distribution but with increased densities above the F2 peak compared to the preflare orbit. This is consistent with the observations of Mendillo and Evans [1974] and is most likely caused by the heating of the ionosphere by the flare. There is, however, evidence of some dynamical effect present in the data. The southern anomaly crest has weakened and perhaps joined with the northern anomaly crest near the dip equator (near 12°N in the images). This may be an indication of meridional wind flow transporting the plasma equatorward. These winds may be produced by intense heating in the polar caps by the high-energy protons produced by the CME shock. This particle deposition was observed by several satellites, including ARGOS. Figure 4 shows the densities produced by the SAMI-2 model [Huba et al., 2000] for the conditions of the flare. The SAMI-2 results were run at the observation longitudes; the time steps closest to the equatorial crossing time of the LORAAS tangent point were selected to make the plots. The adjustments made to the solar flux to account for the flare are described in the work of Meier et al. [2002]. The simulations show a strong density increase during the flare orbit compared to the prefare densities. However, the overall morphologies do not differ as strongly from orbit to orbit as those inferred from the LORAAS data. Figure 4c shows a change in the altitudes of the anomaly crests due to southward flowing meridional winds that originate in the Northern Hemisphere. The LORAAS results are suggestive of a wind field that produces a pile up of ionization near the magnetic equator. The SAMI-2 simulations show the enhancement of the electron density owing to heating by the flare as observed by LORAAS and Mendillo and Evans [1974].

Figure 3.

(a–c) The observed electron densities for the preflare orbit, flare orbit, and post flare orbit, respectively. The density enhancement at 55°S in panel (a) is observed to slowly move northward in subsequent orbits. In the postflare orbit, the southern equatorial anomaly in panel (a) has weakened and perhaps merged with the northern anomaly crest. The dip equator is near 12°N in these figures. The isopleths Figures 345 are equally spaced at 2 × 105 cm−3.

Figure 4.

Panels (a), (b), and (c) show the SAMI-2 electron densities for the preflare orbit, flare orbit, and post flare orbit, respectively.

[16] The electron density distribution observed by the LORAAS differs from that predicted by SAMI-2. This is likely caused by dynamics, probably the wind field, that were improperly modeled by SAMI-2. Figure 5 shows the electron density distributions reconstructed on a geomagnetically quiet day, 4 July 2000, when LORAAS observed the same longitudes as shown in Figures 2 and 3. The electron density distributions show better agreement with SAMI-2, which indicates the electric fields and winds driving the SAMI-2 model are a better representation of the physical conditions on a quiet day. The ionosphere on 14 July 2000 had already been disturbed by a geomagnetic storm on 13 July and had not yet fully recovered.

Figure 5.

Panels (a)–(c) show ionospheric reconstructions from 4 July 2000. This is a day with no geomagnetic activity in which the ARGOS observed the same longitudes as it did on 14 July. The electron density distributions more closely agree with the SAMI-2 morphologies in Figure 4.

[17] Figures 6a and 6b show the fractional change of the electron density for orbits 6 and 7 with respect to orbit 5. The ∼20% increase in the electron density near the F2 peak is evident in Figure 6a. The electron density in the F1 and F2 regions, shown in Figure 6b, has indeed decreased to preflare values by orbit 7. The observed decay of the electron density back to preflare values is consistent with the decay time observed by Thome and Wagner [1971]. The high-altitude density enhancement is consistent with Mendillo and Evans [1974]. However, the LORAAS measurements show that this enhancement is of a global scale, which could not be ascertained from the Mendillo and Evans results.

Figure 6.

Panel (a) shows the observed fractional change of the electron density from the flare orbit compared to the preflare orbit ((orbit 6 − orbit 5)/(orbit 5)). Panel (b) shows the observed fractional change of the electron density from the flare orbit compared to the preflare orbit ((orbit 7 − orbit 5)/(orbit 5)). Panel c shows the fractional change of the electron density of the flare orbit to the preflare orbit predicted by SAMI-2. Panel (d) shows the fractional change of the electron density change of the postflare orbit to the preflare orbit predicted by SAMI-2. The dashed contours indicate areas where the electron density is lower than it was in the preflare orbit. The isopleths are equally spaced at 0.02.

[18] Figures 5c and 5d show SAMI-2 results of the fractional change in the electron densities produced by the flare compared to the densities under nominal solar input. Figure 5c shows the main increase in the electron density to be in the F1 region, where the ionization is dominated by X-rays. The electron density was predicted to increase by 20–30% in the F1 region and near the F2 peak. This is not consistent with the LORAAS results, which show a decrease in the density in the F1 region; however, we note that the LORAAS results are not very accurate in the F1 region as LORAAS observes O+, which is not the dominant ion at F1 heights. The LORAAS densities are higher than the SAMI-2 densities above 300 km, perhaps indicating that the solar EUV fluxes in the SAMI-2 model need to be further increased to better match the flare heating rates. Additionally, the SAMI-2 results show (Figure 5d) the F region density decaying back to preflare values. This is consistent with the LORAAS results in Figure 5b.

5. Concluding Remarks

[19] We have presented the results of our study of the global ionospheric response to an X-class solar flare. We find that the observed behavior of the electron density agrees well with previous ground-based measurements that did not have the spatial coverage of the LORAAS measurements. The observed electron density increases are qualitatively consistent with predictions of the ionospheric response made using the SAMI-2 model [Huba et al., 2000]. We also observed enhanced electron density above the F2 peak, which is consistent with the observations of Mendillo and Evans [1974]. The electron densities above the F2 peak are underestimated by the SAMI-2 model, perhaps indicating that the solar EUV fluxes used in SAMI-2 need to be further adjusted. The observed decay of the electron density to the preflare values is consistent with previous observations of flare induced density enhancements as observed by Thome and Wagner [1971]. This aspect of the ionospheric response to the flare is well captured by SAMI-2.


[20] This work was supported by the Defense Meteorological Satellite Program (DMSP) and by the Office of Naval Research.