The geomagnetic storm analysed occurred on St. Patrick’s day [day of year (DOY) 076–077] in 2015 and was among the most intense storms in the 24th solar cycle (Astafyeva et al. 2015). We reconstruct the ionosphere vertical structure during this period (at a resolution of 15 min) and briefly analyse the variations in reference to the ionosphere disturbance indicator, Dst index. Specifically, the vertical structure is analysed at ionosonde locations for easy comparison with ground ionosonde observations. In-situ densities from the Swarm constellation and radio occultation density profiles are also utilized to assess the reconstructed F-region topside densities. Throughout the analysis, international reference ionosphere (IRI-2016, Bilitza et al. 2017) model densities are also presented as a form of reference. However, it should be noted that the IRI model is a climatological model for average values, so might not be the best candidate for performance evaluation against ionosphere tomography models that include weather.
Storm chronological highlights
A coronal mass ejection (CME) was observed erupting between ~ 00:30 UT and 00:04 UT on DOY 074 and was predicted to encounter the Earth’s magnetosphere on DOY 076. Figure 3 illustrates how the abundant energy injected into the magnetosphere-ionosphere system perturbed the terrestrial geomagnetic field leading to a Dst (Kp index) minimum (maximum) recording of approximately − 223 nT (8). Kp as blue is scaled on the right and Dst as black is scaled on the left. Dashed vertical lines indicate the main events of the storm. Both indices can be obtained from the World Data Center for Geomagnetism, Kyoto http://wdc.kugi.kyoto-u.ac.jp/ (Nose et al. 2015) or https://omniweb.gsfc.nasa.gov/form/dx1.html.
Before DOY 076, the geomagnetic indices Kp and Dst were relatively stable with magnitudes below 4 and 50 nT, respectively. On DOY 076, at 04:45 UT (storm sudden commencement, SSC), the CME hit the Earth, leading to an increase in Dst. On the same day, at ~ 07:30 UT the main phase of the storm commenced and the Dst decreased to a local minimum of ~ − 80 nT. During this period the interplanetary magnetic field (IMF) Bz component was reported to have turned southward (see bottom panel in Fig. 3 and also refer to Cherniak et al. 2015). From about 9:30 UT to 12:20 UT there was a short-lived recovery in the Dst index to ~ − 50 nT; expected to be due to the IMF Bz component turning Northward (Astafyeva et al. 2015). From 12:20 UT the Dst continued with a gradually decreasing trend reaching a global minimum (− 223 nT) at ~ 22:00 UT on DOY 076. The recovery phase continued through the following days with the Kp index ranging below 6.
Analysis of reconstructions at ionosonde locations
In Fig. 4 we have generated images from a time series of vertical profiles corresponding to locations of ionosondes within the high-resolution sub-grid region marked out in blue dashed lines in Fig. 1; Ssessanga et al. 2021 found that a coarse horizontal spacing (5° × 5°) in the outside sub-grid might not reflect a true representation of the ionospheric state and dynamics, particularly at the low latitudes where the ionosphere is expected to have steep gradients. Consequently, results presented here are limited to the region where we expect a coherent and consistent reconstruction. Refer to Table 1 for the locations, four character code names, geographic coordinates and geomagnetic latitude of each ionosonde used in analysis. Ionosondes located in the region of high-resolution are labeled with an “a” superscript.
The images in Fig. 4 are arranged in accordance with increasing ionosonde geomagnetic latitude. The top, middle and bottom panels represent the IRI-2016 model, assimilated CIT, and observations from ionosondes, respectively. Tick labels at the top of each panel show days of the year while those at the bottom show time in UT (the East Asian sector local time (LT) is ~ 9 h ahead of UT). Preferably, IRI and reconstructed densities were converted to frequency (MHz) for easy comparison with ionosonde measurements. The colour scale of the frequency in each panel is given in the far right bottom corner.
In each image plate: for clarity, the altitude range is limited to the region of uttermost interest (F-region ~ 200–600 km). Superimposed is the Dst index (ordinate axis on the right) in order to track the response of the ionosphere during the different storm dynamics. As indicated earlier, dashed vertical lines indicate the main events of the storm. The horizontal dashed line at 300 km is included as a baseline to vividly track the uplift of the plasma density. The time-altitude variation of the peak-density is represented by the purple solid-asterisk line. In the reconstructed, and ionosonde images, white spaces indicate periods when GPS-ground receiver ray links and ionosonde observations did not exist or intersect that particular location. At a particular time-stamp, if there exist any data gaps within the 200–500 km altitude range, the peak density is not determined.
Throughout the selected analysis period, at all stations, the IRI model only shows the general diurnal variation of the F region without detailed features. This is expected since IRI is a climatological model that represents monthly averages of terrestrial plasma densities/frequencies (Bilitza et al. 2017). Consequently, the IRI model performance is usually found lacking when ionosphere underlying driving mechanisms are deviant from the general trend.
Contrary to IRI, reconstructed results reflect changes in the plasma frequency following variations in the Dst index as detailed: the onset of the storm occurred when the East Asian sector was on the day-side. Compared to reconstructed densities on the day before the storm (DOY 075); nearly just after the SSC, at OK426 station that is located geomagnetically in low latitudes, there is a slight noticeable instantaneous uplift (marked with a grey arrow) of the peak densities to altitudes above the 300 km baseline. The uplift reduces in amplitude towards the high latitudes (not noticeable at TO536) and is most probably due to geomagnetic storm induced eastward prompt penetration electric fields (PPEF), which end up modifying the equatorial and low‐latitude ionospheric phenomenology (Maruyama et al. 2004).
Approximately two hours after the SSC, there was a slight increase in F region plasma densities (~ 280–340 km) particularly at stations located near the equatorial anomaly (OK426 and JJ433). In addition, there was a well defined wavy modulation of the peak density altitude at mid-latitude stations (JJ433, TO536 and IC437). The ~ 2-h delay falls within the time frame the traveling atmospheric disturbance (TAD) created at high latitudes during geomagnetic storms, would take to propagate to the mid and low latitudes (for example see the works of Shiokawa et al. 2003 and references therein). Therefore, low latitude ionisation and wavy modulation of the mid-latitude F region could be related to an equatorwards propagating surge and plasma density enhancements in equatorial ionization anomaly (EIA) crests, following day-side PPEF.
By the time the East Asian sector enters the night side (~ 10:30 UT), there is a short-lived recovery in the plasma frequency between 10:30 and 12:30 UT. Thereafter (~ 12:30–24:00 UT), specifically at stations located in the mid-latitudes (JJ433, TO536 and IC437) and towards high latitudes (WK546), the plasma is uplifted to high altitudes (~ 500 km), gradually falls to ~ 320 km, rises again to about ~ 380 km and finally settles to an average altitude of ~ 300 km. Through this whole process, at all stations, the initial daytime plasma enhancements decrease gradually (to values nearly below 4 MHz), following the Dst index which hits a minimum at about 22:00 UT. The decrease in F-region densities accompanied by descent in altitude of the peak densities could be attributed to an equatorward wind surge and the night-side westward PPEF which has opposite effects (reduction in EIA ionisation) in reference to day-side PPEF.
On the day after the storm (DOY 077), at all stations, the plasma density (frequency) remained nearly ~ 60% below the values observed on the day before the storm. Interesting to note through this period, is the total or partial absence of data at all ionosonde stations. From the reconstructions, it is clear that these data gaps were due to a significant decrease in the F region plasma frequency (< 4 MHz) such that the ionosonde could not detect the reflected echoes. This is a well-known/typical negative ionospheric response following a major geomagnetic storm (e.g., Fuller-Rowell et al. 1994; Prölss 1995; Tsagouri et al. 2000). A possible explanation for the intense plasma reduction is that the substorm activity on DOY 076 generated a composition bulge that entered the Asian sector, and persisted through the next day (DOY 077). In fact, Astafyeva et al. (2015) and Nava et al. (2016), analysed the thermospheric column integrated O/N2 ratio changes measured by the GUVI (global ultraviolet imager) instrument onboard the TIMED (thermosphere, ionosphere, mesosphere energetics and dynamics) satellite during the same geomagnetic storm (DOY 076–078, 2015), and found that the Asian sector had significant composition changes that could have led to the negative ionosphere response.
Latitudinal wise, stations towards the high latitudes (WK546) exhibit the highest reduction in plasma density. This enormous reduction in F plasma densities led to an increase in slab thickness, which is seen to increase polewards (extending > 300 km in altitude), and most pronounced during the evening of DOY 076 and the early hours of DOY 077. Indeed, different studies have already shown that the slab thickness is dependent on diurnal, annual, latitudinal and storm-time variations [see Stankov and Warnant, (2009), and references therein]. In the longitudinal perspective, stations TO536 and IC4337 that are almost on the same geomagnetic latitude (the difference is ~ 1°) show that the negative storm effect was greater on the Eastern side.
On the third day after the storm (DOY 078), except for OK426, all stations maintain a slow (< 6 MHz) gradual recovery towards the normal plasma frequency values observed during the quiet period. Nava et al. (2016) analysed the same storm and illustrated that the recovery process lasted more than 7 days. At OK426, reconstructions indicate that the low-latitudes had an undulating peak, accompanied by reinforced ionisation covering altitudes ~ 200–420 km. Surprisingly, these effects do not extend to mid-latitudes (see reconstructions at JJ433, TO536 and IC437). Nava et al. (2016) also generated regional TEC maps corresponding to periods, before, during and after the St. Patrick’s day storm. Over the Asian sector, on the day after the storm, ionization was confined at low latitudes near the equator, and it almost disappeared at middle latitudes. This characteristic is typical of westward zonal electric field penetration during disturbance dynamo (DD) when EIA is inhibited. Nonetheless, this ambiguous or peculiar behaviour may be related to other post-storm effects that might require further investigation (also see Kutiev et al. 2006).
GPS and ionosonde: The significance of complementing GPS with ionosonde data is well illustrated at most stations; we can reconstruct the full extent of F region plasma structure beyond the bottomside limitations of the ionosondes. Moreover, as noted earlier, at points where the ionosondes failed to record any echoes due to the low plasma electron content, our technique was still able to provide reasonable plasma content estimations. This is a crucial result particularly for applications that utilize transionospheric HF (high frequency) signals.
Comparison with Swarm densities
In-situ densities offer a good test of accuracy since they represent a specific point in space (ionosphere) at a particular time. The Swarm constellation consists of three identical polar-orbiting satellites that fly at two different altitudes ≤ 460 km [Alpha (A) and Charlie (C)] and ≤ 530 km [Bravo (B)]. Each satellite has a Langmuir Probe that facilitates the measurement of in-situ densities. Part of the Swarm density results presented here are already presented in Ssessanga et al. 2021 (as preliminary results) to showcase the performance of the technique. We will add more analysis plots to cover the morning and evening sectors of the analysed period. The data used in the analysis are Level 1b electron density measurements at a 2 Hz rate, accessible at ftp://swarm-diss.eo.esa.int. For a further review on Swarm data see, for example, Olsen et al. (2013).
In Fig. 5, time profiles of Swarm in-situ densities (red) are plotted alongside densities from our reconstructions (blue) and IRI model (black). The top, middle and bottom rows of subplots correspond to satellites A, B and C, respectively. Each row of subplots covers three DOYs (076, 077 and 078) during which St. Patrick’s day storm effects were most evident. In each subplot, the geographic traces of the Swarm satellite during that particular period are marked red on the map in the upper right corner. Gaps in the traces indicate periods when either in-situ measurements or density reconstructions were not available for analysis. To match our low time resolution (15 min) reconstructions with in-situ measurements, the ionosphere was assumed to remain stationary under a period of 10 min. Then, all in-situ measurements within this window were mapped onto the nearest grid altitude plane within 10 km to the satellite orbital altitude. The vertical dashed lines in each subplot indicate the start (green) and end (magenta) of grouped densities, in which the time data gaps are less than 1-h. The corresponding Swarm satellite time coverage of each grouped densities is indicated on the far right below the map. The x-axis in each subplot is not linear but rather readjusted depending on the available grouped densities within the day. The middle subplots corresponding to Swarm A and B are essentially a re-plot from Ssessanga et al. (2021), except we added more observation data points.
In all subplots, the agreement between the reconstructed and in-situ densities is generally good. IRI on the other side exhibits a poor estimation of the densities. Also noticeable is that due to differences in orbital altitudes, there is a slight difference in the level of electron density distributions observed by satellites (A, C) and B. The difference between in-situ and reconstructed densities is on average less than 0.2 × 1012 el/m3. This result offers confidence in the reconstructed topside F-region density structure, which is not attainable while using ground-based instruments such as ionosondes. A point of concern could arise from the poor density reconstructions at the awake of profiles in time range 9:00–11:00 UT (corresponding to satellites A and C on DOY 076 and 077). During this period Swarm A/C traces fall within or near the equatorial latitudes. Ssessanga et al. 2021, ascribed this to a poor grid specification over these latitudes. That is to say, the equatorial ionization anomaly region exhibits steep latitudinal densities gradients, yet, the currently utilized grid assumes a coarse horizontal spacing (5° × 5°) over this region [refer to Fig. 1 and Saito et al. (2017; 2019)], hence reconstructions would not reflect the true ionospheric state and dynamics.
If we follow the traces of Swarm A and C, we observe that on DOY 076 when the storm commences (~ 10:00–12:00 UT) the topside densities range \(0.5 \sim 2\times {10}^{12}\mathrm{ el}/{\mathrm{m}}^{3}\). On the day that follows, during the same period, when the negative storm effects are dominant, we observe that the densities remain nearly below \(0.25\times {10}^{12}\mathrm{el}/{\mathrm{m}}^{3}\) (~ 50% of the densities on DOY 076). The reduction in plasma densities is also observed at Swarm B altitudes and is maintained throughout the late hours of DOY 077. This result is consistent with the density reductions that were observed in the middle and bottom panels of Fig. 5.
On DOY 078, when the F-region topside densities start returning to normal, the reconstructed densities relatively track well the in-situ measurements, but with a better performance at Swarm B altitudes. The discrepancy in performance may be related to the altitude level; Swarm (A and C) orbit at a lower altitude (≤ 460 km) than B, and at such altitudes the plasma densities are more likely to be influenced by different nonlinear dynamical forces during the storm period, hence making the reconstruction acute. Certainly, the local time (\(\sim \mathrm{UT}+9\)) difference in Swarm B and A/C observations (different ionisation levels) might also influence the results. Nevertheless, the reconstructed densities still outperform the IRI model estimations.
Comparison with radio occultation (RO)
In Fig. 6, a set of RO electron density profiles (red) from COSMIC (constellation observing system for meteorology, climate, and ionosphere; orbital altitude ~ 800 km) and GRACE (gravity recovery and climate experiment; orbital altitude ~ 490 km) constellations are plotted together with profiles from assimilated tomography (blue) and IRI (black) during the analysed period. RO data are accessible in level 2 format at https://cdaac-www.cosmic.ucar.edu/cdaac/. Comparison is limited to reconstructions away from the low latitudes and above the 200 km altitude mark, where RO density profiles are expected to have the best accuracy. That is to say, RO density profiles are an inversion of RO total electron content (ROTEC) using the so-called inverse Abel transform that assumes spherical symmetry in the ionosphere: For the most part, profiles from Abel transform have good accuracy, with the exception of the E-region (where rays have asymmetric contributions from the F region portions of the rays), and low latitudes (where large density gradients exist, Garcia-Fernandez et al. 2003; Wu et al. 2009; Yue et al. 2010). Rather than comparing the RO density profiles to a specific vertical profile within the grid, the comparison is performed at the location of tangent points (which contribute the most density along the RO ray path) shown on the maps in the top right corner of each subplot. Purple and green represent GRACE and COSMIC, respectively. Gaps in the profiles indicate instances when assimilated tomography reconstructions were not available. Fortunately, during the analysed period the GRACE constellation covered the ~ 200–400 km range, whereas COSMIC mostly covered the topside ~ 400–700 km. This snip view gives us a chance to nearly observe and analyse the capabilities of the assimilated tomography in specifying the electron density field both in time and space (horizontally and vertically).
On the DOY 075 (top left corner subplot) before the storm, both assimilated tomography and IRI have a good estimation of the topside structure. However, on days that follow, assimilated tomography and RO profiles show a dramatic decrease in electron densities consequent to the geomagnetic storm. By contrast, the IRI model maintains a high-density output, with the largest deviation from the truth in the 200–400 km altitude range. This result is important because it gives a sample of what might be expected in applications that use models to correct for ionospheric effects during severe conditions.
Despite the variability, reconstructed profiles on average adequately track the RO densities at all height ranges. Surprisingly, a combination of ionosonde densities and ground-based GNSS TEC is adequate to reconstruct a reliable topside structure (~ 400–700 km). Nonetheless, the noisy structure of some of the profiles motivates our suggestion in future to integrate RO data (both TEC and electron density profiles) in the analysis to further constrain the vertical structure.
Results presented in this study can be placed in context with previous performance analyses of the tomography technique during the early stages of development. In this sense, our work is a complement to studies by Seemala et al. 2014; Saito et al. 2017, 2019; Ssessanga et al. 2021, who have already analysed the tomography reconstructions during the quiet period and found the technique to have good fidelity. Therefore, the technique seems fairly robust in handling different ionospheric conditions.