The Swarm mission consists of three polar-orbiting satellites – Swarm A, B, and C . They were launched into near-polar orbits at an initial altitude of about 500 km on 22 November 2013 . Each satellite is equipped with an electric field instrument (EFI) that is mounted on the ram side of the spacecraft . The ion density Ni is derived from the EFI faceplate current assuming that the current is carried by ions hitting the faceplate due to the orbital motion of the spacecraft . However, due to quasi-neutrality, Ni must be equal to the electron density Ne. With the 16 Hz Ne measurements, Swarm observes ionospheric irregularities with sizes of up to 500 m. and provide detailed information on Swarm and the onboard instruments. In this study, we used the 16 Hz Ne measurements to examine topside ionospheric irregularities. The faceplate Ne data are readily available at http://earth.esa.int/swarm (last access: 23 February 2020).

The JULIA radar is a PC-based system for data acquisition. JULIA uses low-power transmitters with a frequency of approximately 50 MHz and Jicamarca's main antenna . Its aim is to record ionospheric irregularities and neutral atmospheric waves at the equatorial region for long periods of time. The pulse width used in the JULIA experiments during the period of study was 25 µs, and the pulse repetition was 160 pulses per second. In addition, 248 range gates that were separated by 3.75 km were sampled – from 0 km to about 930 km – during the period of study. For the identification of 3 m scale ionospheric irregularities, the backscattered 50 MHz JULIA radar echo was used. The radar observations provide the signal-to-noise ratio (SNR), Doppler velocity, and spectral width as a function of height and time. Data collected by the JULIA radar are readily available at http://jro.igp.gob.pe/madrigal/ (last access: 23 February 2020). The website provides JULIA data from 1996 to date. Our analysis was restricted to the years from 2014 to 2018, as this was the period for which JULIA, Swarm, and ionosonde data sets were available.

The equatorial spread-F (ESF) signatures are often recorded by ionosondes installed at the JRO. The ionosonde at the JRO is a Digisonde DPS-4 which records ionograms (altitude versus frequency plots) at 15 min intervals. The Automatic Real-Time Ionogram Scaler with True height (ARTIST) ionogram autoscaling tool, developed at University of Massachusetts Lowell Center for Atmospheric Research (UMLCAR), is used to scale the ionograms, and the outputs are plasma frequency profiles versus altitude . The equatorial spread-F signatures were analysed from the ionograms. The ionosonde data are also available on the Madrigal website, and we used all of the Jicamarca ionograms for the years from 2014 to 2018.

To examine topside ionospheric irregularities, the 16 Hz Ne Swarm faceplate data were used. We followed the same method as , , and to derive the absolute electron density perturbation along Swarm orbital tracks, but we focused mainly on small-scale equatorial plasma structures. We utilized a 2 s running mean filter to determine the mean Ne. The selected running mean is equivalent to a 15 km scale length, given Swarm's velocity of about 7.5 km s-1. The mean Ne was subtracted from the original observations to get the residual, ΔNe, in a similar fashion to , who obtained the residual using total electron content (TEC) data. The standard deviation of the residuals was then calculated at a running window of 2 s to represent the magnitude of the perturbation, std(ΔNe). There is no standard threshold definition of how large stdΔNe must be to identify plasma irregularities . However, the period considered in this study was characterized by low solar activity, and the recorded ionospheric irregularities were very weak. Therefore, a threshold value of stdΔNe=1×1010 m-3, similar to that adopted by , was selected to provide a reasonable irregularity event identification at the small scales and the relatively low Swarm altitudes during the study period.

The JULIA system computes and stores measurements of the zeroth and first lags of the autocorrelation function (ACF) of the signals from two receivers connected to the eastern and western quarters of the main antenna . The total power, Doppler velocity at first moment, and Doppler spectral width of the scattering signals can be determined from these measurements . Of particular interest was the SNR measurements derived by the JULIA system to check plasma plumes for a given evening. In addition, we also used the vertical plasma drift measurements made by the Jicamarca incoherent scatter radar (ISR) to examine the pre-reversal enhancement (PRE) drifts . Field-aligned irregularities in the F region are often observed by the JULIA radar between 18:00 and 06:00 LT . Therefore, to compare the Swarm observations with the JULIA measurements, only swarm satellite passes for the time between 18:00 and 06:00 LT were considered.

The comparison of JULIA and Swarm observations was supplemented with ionosonde measurements from the JRO. The ionosonde data analysis was carried out using the SAO Explorer software . To display ionograms, both the raw and processed (SAO) data were loaded into SAO Explorer. In addition, the spread-F index QF, known as the mean spread of the diffusing F-layer trace, was obtained by the ARTIST directly from the ionograms using SAO Explorer , and this was also used in this study to analyse the spread-F signatures. The spread-F index QF is defined as the extent of the diffuse reflection in kilometres averaged over all frequencies where a diffuse echo appeared. For simplicity, the virtual height is used at each frequency to determine the range extent of the reflection. The ARTIST software for data analysis is described by . Spread-F ionograms were similarly studied by using magnetically conjugate ionosondes in South America and by for ionosondes and scintillation receivers at Sanya. In the following section, the results of this study are presented and discussed.

The Swarm satellites regress in longitude by about 22.5∘ between orbital ascending nodes. Therefore, in comparison with JULIA and ionosonde data, the Swarm passes were allowed to be within ±5∘ magnetic longitude of the JRO to make sure that a sufficient amount of Swarm passes could be used for the statistical examination. Both JULIA and ionosonde data during the time when Swarm was within a ±5∘ longitudinal range of the longitude of the ground site were selected. Summary plots, such as those presented in Fig. 4, were generated for all days during the years from 2014 to 2018 for which the data were available. In total, 560 night-time orbits were used for which JULIA, Swarm, and ionosonde data were concurrently available. The outputs of the summary plots could be categorized into four cases considering the presence (or absence) of irregularities. In general, these four cases are as follows: irregularities observed both on the ground and in situ, no irregularities observed on the ground or in situ, irregularities observed only in situ, and irregularities observed only on the ground. For each range–time–intensity plot, the SNR corresponding to the peak height was determined and an event was identified as a significant irregularity when the peak height was ≥400 km. For a peak height less than 400 km, the events were classified as weak irregularities or were not considered to be irregularities. It is important to note that a peak height less than 400 km is a representation of bottom-type, bottom-side, and no equatorial spread-F signatures; therefore, all spread-F altitudes were taken into consideration during the analysis. For the in situ Swarm observations, we considered a threshold of 1×1010 m-3 for std(ΔNe) as a significant irregularity event, whereas std(ΔNe) values less than the threshold were not considered to be irregularities. For the ionosonde measurements, QF values greater than or equal to 20 km were considered to be significant irregularity events. For each category, the percentage occurrence was computed as a ratio of the total number of events in that category to the number of observations. These cases are presented in Fig. a for Swarm and JULIA and in Fig. b for Swarm and ionosonde.

Ionospheric irregularities were detected by Swarm and JULIA in about 55.08 % of the cases for Swarm A, 56.89 % of the cases for Swarm C, and 58.62 % of the cases for Swarm B (as seen from Fig. a). No ionospheric irregularities were detected by the Swarm satellites and JULIA in about 27.12 % of the cases for Swarm A, 25.0 % of the cases for Swarm C, and 21.55 % of the cases for Swarm B. In Fig. b, a high percentage occurrence was also observed when there was agreement (irregularities observed by both ionosonde and Swarm and no irregularities observed by the ionosonde and Swarm) between Swarm and ionosonde. The two categories where there was an agreement in Fig. a and b indicate that Swarm satellites, JULIA, and the ionosonde simultaneously observed ionospheric irregularities. also examined the relationship between measurements from JULIA and a DMSP satellite during 110 nights for the years from 1998 to 1999. In comparison with the statistical results presented in Fig. 7 for Swarm and JULIA, the DMSP satellite sampled very few EPBs compared with plumes detected by the JULIA radar. According to , there was a low probability that JULIA and the DMSP satellite would encounter ionospheric irregularities because most plumes could not ascend to altitudes greater than 600 km. There were also some disagreements between the ground-based and space-based observations: JULIA and the ionosonde detected plume structures, whereas Swarm registered no events (as seen from the statistical results in Fig. a and b). For these cases, the Swarm altitudes during the pass were examined. It was observed that the plume structures did not ascended to Swarm altitudes by the time the satellites passed over Jicamarca or that the satellites were simply in a different location. For instances when Swarm registered events while JULIA and ionosonde recorded no signatures, we checked on the longitudinal separation between the satellite passes and the ground site. The longitudinal separations obtained between the Swarm passes and the ground site were often ≈5∘, and the magnitude of the in situ perturbations were relatively low. Ionospheric irregularities tend to be magnetic field aligned ; therefore, Swarm may encounter irregularities of relatively small magnitudes in situ, while JULIA and ionosonde do not identify any events, for a wider longitudinal offset of a pass from the ground site.

showed two cases of Swarm passes over JULIA: one case where Swarm A encountered ionospheric irregularities and JULIA recorded spread F and another case where Swarm B never registered an ionospheric irregularity and JULIA never recorded a spread F. The statistical results shown in Fig.  assert that Swarm B can also detect irregularities and plasma bubbles associated with plumes and spread F, although with more mismatches than for Swarm A and C. Swarm B recorded more mismatches than Swarm A and C due to the progressive temporal and altitudinal separation between Swarm B and Swarm A and C . Swarm B orbits at a higher altitude than Swarm A and C, and it crosses the same region later than Swarm A and C. Generally, a difference in the percentage occurrence in all categories is observed between Swarm A and C (Fig. ), although they orbit at the same altitude above sea level. The large-scale longitudinal bubble structure is sometimes observed with the Swarm A and C satellites ; however, for small-scale irregularities, the 1.5∘ longitudinal separation between the satellites is too large for a significant correlation between them.

Numerous studies have shown that ionospheric irregularities at low latitudes are a post-sunset phenomenon owing to the electrodynamics launched after sunset e.g.. Here, we also compared the local time dependence of the occurrence of plasma plumes observed by the JULIA radar, spread F recorded by the ionosonde, and small-scale ionospheric irregularities encountered by Swarm for different seasons. Figure  shows the percentage occurrence of plasma plumes as a function of local time and height grouped into different seasons: December solstice, June solstice, March equinox, and September equinox.

To obtain the results presented in Fig. , ground-based JULIA SNR data for the years from 2014 to 2018 were used. To eliminate the impact of geomagnetically disturbed conditions on the statistical outcomes, the data were filtered and only those recorded during quiet geomagnetic conditions (Kp ≤3) were taken into account. The JULIA data accumulated for the years from 2014 to 2018 were sufficient for examining the seasonal variation. Therefore, the seasonal dependence of the local time distribution of JULIA observations of ionospheric irregularities was also examined by grouping all of the data into different seasons corresponding to the March equinox (February–March–April), the June solstice (May–June–July), the September equinox (August–September–October), and the December solstice (November–December–January). For each local time–height bin, the percentage occurrence was obtained by dividing the number of observations with SNR>10 dB by the total number of observations e.g..

It is visible from Fig.  that the plasma plumes only occurred at night. Figure  shows the occurrence of irregularities in plasma plumes starting at about 19:00 LT that generally last past midnight. This observation is similar to those of previous studies e.g.. The observed plasma plumes are connected with the non-linear development of the RTI, which is initiated at the bottom of the F region . The highest percentage occurrence takes place at the December solstice and the equinoxes. The lowest percentage occurrence is observed at the June solstice. The daily variations in the vertical plasma drift measured by the ISR were used to better understand the seasonal patterns observed in Fig. . Figure  presents the local time variation of the F-region vertical drift velocity for the years from 2014 to 2018.

From Fig. , the PRE of the vertical plasma drifts can be seen around the sunset hours (between 17:00 and 20:00 LT) before its reversal. Figure  shows the highest PRE peak during the December solstice and equinox seasons, whereas the PRE peak is the lowest at the June solstice. Similar observations were made by . Comparing the results presented in Fig.  with the local time distributions presented in Fig. , it follows that a high occurrence of post-sunset topside spread F is associated with enhancements of the PRE peak. A high PRE implies significant E×B vertical drifts that boost the rate of RTI growth . The PRE moves the ionospheric F layer to higher altitudes where there is less interaction between ions and neutrals. The decreased interaction between ions and neutrals leads to increased RTI .

Figure  also shows a relatively high occurrence of irregularities after midnight, especially at the solstices and the September equinox. Similar observations were made by and . The extension of the occurrence of plumes post-midnight may be due to the late reversal time and small post-reversal electric fields . Despite the use of low-power transmitters, the JULIA radar can also detect weak post-midnight irregularities, particularly during the solstice seasons (as shown in Fig. ). However, the detected post-midnight ionospheric irregularities often exist at much lower altitudes than those presented by . Using Jicamarca radar measurements, reported that the ionospheric irregularities that occur after midnight are typically well-formed structures that can be connected to the disturbed dynamo.

Figure shows the QF indices derived from ionosonde observations as a function of local time and month. To obtain the results presented in Fig. 10, ground-based ionosonde data for the years from 2014 to 2018 were used. These data were also filtered, and only those recorded during quiet geomagnetic conditions (Kp ≤3) were considered.

Considering that equatorial ionospheric irregularities are night-time phenomena, we only present QF data from 18:00 to 06:00 LT in Fig. . To generate Fig. , the QF data were averaged over 0.1 h local time bins for each month (y axis). Figure  shows that the QF index was high in the post-sunset period with peak values occurring between about 20:00 and 00:00 LT at the December solstice and the equinoxes. The high QF values at the equinoxes and the December solstice are most likely due to the RTI, which is usually triggered at the bottom side of a rising equatorial F layer. The rate of growth of the RTI depends on the meridional wind and the eastward electric field PRE determined by the longitudinal gradient of the flux-tube-integrated conductivity . At the June solstice, the QF values were small after sunset, as seen in Fig. . Generally, Fig.  shows moderate QF values lasting until local midnight or longer (similar to the trend presented in Fig. ). The low post-sunset QF values at the June solstice can be attributed to the small PRE, which can also be seen in Fig. c.

Figure  shows the QLat–LT distributions of std(ΔNe) for the Swarm satellites for the years from 2014 to 2018. Recall that the Swarm passes were allowed to be within ±5∘ magnetic longitude. The Ne data collected for the 5 years were also grouped into different seasons similar to those presented in Fig. . The results presented in Fig.  were also generated considering only the geomagnetically quiet conditions (Kp ≤3). The std(ΔNe) was then calculated in bins of 1∘×0.1 h resolution in QLat and local time. The occurrence rate of ionospheric irregularities does not always correspond to the highest amplitude of ionospheric irregularities . Therefore, we presented the calculated std(ΔNe) per bin as a function of QLat and local time, as seen in Fig. . From Fig. , high std(ΔNe) values frequently occurred between about ±10 and ±20∘ QLat, i.e. at the approximate location of the EIA belts. The distribution of the std(ΔNe), as seen in Fig. , is essentially symmetrical about the quasi-dipole equator. The symmetrical distribution about the magnetic equator has also been observed in earlier studies e.g.; this confirms that equatorial ionospheric irregularities usually extend along the magnetic field lines in the northern and southern directions, and they are concentrated at the EIA belts . The std(ΔNe) attained a maximum between 20:00 and 22:00 LT. A decrease was detected after 22:00 until 06:00 LT. The local time distribution of std(ΔNe) for Swarm A, B, and C is the same as that of the quiet-time F-region echoes presented in Fig.  and the QF distribution presented in Fig. . The distribution of std(ΔNe) shown in Fig.  has peak values at local times and QLat ranges where the RTI is expected. In terms of seasons, as observed from Fig. , high values of std(ΔNe) were seen at the equinoxes and the December solstice, whereas the lowest values were detected at the June solstice. This is similar to the seasonal dependence of quiet-time F-region echoes presented in Fig. . In Fig. , Swarm hardly encountered post-midnight irregularities while orbiting over South America during all seasons. In Fig. , it is observed that the post-midnight plumes often existed at lower altitudes at the solstices and the September equinox. Therefore, the low post-midnight ionospheric irregularity observations by Swarm may be because the plumes failed to reach Swarm altitudes.

For comparison, Fig.  shows the quasi-dipole versus local time distribution of ionospheric irregularities based on the Swarm Ionospheric Bubble Index (IBI), which is a standard Level 2 product of the Swarm mission . The IBI provides information on the climatology of ionospheric irregularities and the level of magnetic field disturbance by taking both the electron density and magnetic field measurements into account . It is important to note that the IBI is has a value of 1, 0, or -1 for bubble detected, not detected, or undetermined respectively. Therefore, to generate the results in Fig. , the data sets were first grouped into different seasons corresponding to the March equinox, the June solstice, the September equinox, and the December solstice. For each season, the data were then binned into 1∘×0.1 h quasi-dipole latitude–local time bins. For each quasi-dipole latitude–local time bin, the percentage occurrence was obtained by dividing the number of observations with an IBI of 1 by the total number of observations. The binned percentage occurrence of an IBI of 1 shown in Fig. 12 has similar seasonal characteristics as the in situ irregularities shown in Fig. 11. The percentage occurrence of an IBI of 1 ranges from 0 % to 20 % which is relatively low. The low percentage occurrence of ionospheric irregularities derived from the IBI was also observed by . This may be because the magnitude of ionospheric irregularities must be large enough to cause magnetic field fluctuations . The latitudinal profile of std(ΔNe) and the percentage occurrence of an IBI of 1 have peaks near the anomaly crests (about ±15∘ QLat). The diamagnetic effect in fluctuations of Ne are believed to be the cause of an IBI of 1, and this occurs at the anomaly crests .

The Bragg condition for the JULIA radar implies that the coherent spread echoes are from density variations at about a 3 m wavelength . The Bragg condition for backscatter means that a radar can only observe structures in the refractive index with a size close to the half radar wavelength . The Swarm electron density measurements used in this study are limited by a sampling rate of 16 Hz and an orbital velocity of about 7.5 km s-1 to wavelengths of about 500 m and longer respectively. The good correlation between the Swarm measurements at these wavelengths and the spread echoes from Swarm altitudes suggest that the irregularities seen by Swarm occur over a spectrum of different wavelengths, at least from about 500 m down to the radar wavelength of 3 m. A non-linear decay of unstable waves could explain this. In addition, we expect that the radar signal could be affected by scintillations which are particularly known from one-way signal propagation such as in Global Navigation Satellite System (GNSS) and VHF satellite beacons . For the radar, this could be relevant for echoes where the Bragg reflection occurs at high altitudes, above Swarm. Fresnel theory shows that wavelengths at the Fresnel scale of (2λd) are most relevant for causing scintillations. Here, λ is the wavelength (3 m for the JULIA radar), and d is the distance from the perturbation to the receiver (445–510 km for Swarm). This gives a Fresnel scale between about 1.6 and 1.7 km. The Swarm measurements generally indicate that irregularities at such scales are present near the paths of Bragg-reflected radar signals. Therefore, we suggest that spread-F signals may, at times, be a result of both the Bragg backscattering at the highest altitude and scintillations of the radio waves to and from the scatter region.