Comparison of Superdarn Peak Electron Density Estimates Based on Elevation Angle Measurements to Ionosonde and Incoherent Scatter Radar Measurements Alexander V

Koustov et al. Earth, Planets and Space (2020) 72:43 https://doi.org/10.1186/s40623-020-01170-w

FULL PAPER Open Access Comparison of SuperDARN peak electron density estimates based on elevation angle measurements to ionosonde and incoherent scatter radar measurements Alexander V. Koustov1* , Sydney Ullrich1, Pavlo V. Ponomarenko1, Robert G. Gillies2, David R. Themens3 and Nozomu Nishitani4

Abstract Measurements of the electron density at the F region peak by the Canadian Advanced Digital Ionosonde (CADI) and the Resolute Incoherent Scatter Radar (RISR) are used to assess the quality of peak electron density estimates made from elevation angle measurements by the Super Dual Auroral Radar Network (SuperDARN) high-frequency radar at Rankin Inlet (RKN). All three instruments monitor the ionosphere near Resolute Bay. The CADI-RKN joint dataset comprises measurements between 2008 and 2017 while RISR-RKN dataset covers about 60 daylong events in 2016. Reasonable agreement between the RKN estimates and measurements by CADI and RISR is shown. Two minor dis- crepancies are discussed: RKN radar daytime peak electron density overestimation by ~ 10% and underestimation by up to 30% in other time sectors. In winter nighttime and dawn, cases were identifed in which the RKN radar signifcantly overestimates the peak electron density. This occurs when the phase in the RKN interferometer measurements is incorrectly shifted by 2π , and this is most signifcant when electron densities are low. Statistical ftting to the joint data sets, split into four time sectors of a day, has been done and parameters of the ft have been determined. These allow slight adjustment of measured real-time RKN values to better refect real peak electron densities in the ionosphere within its feld of view. Keywords: F region electron density, SuperDARN radars, Elevation angles, CADI ionosonde, RISR incoherent scatter radar

Introduction radio waves toward the ground where they can be either Knowledge of the electron density distribution in the detected by a communication receiver or can be back- high-latitude ionosphere is fundamentally important for scattered by the surface and detected by a ground-based various practical applications such as high-frequency radar receiver near their location of transmission. (HF) radio wave communication (Davies 1990; Hun- Decades of research has led to a general understanding sucker 1991; Rawer 2013). At HF, radio waves expe- of the horizontal and vertical distributions of the electron rience signifcant ionospheric refraction resulting in density in the ionosphere, culminating in the develop- strong bending of radio paths and occasionally turning ment of empirical models of the ionosphere such as the International Reference Ionosphere (IRI, Bilitza et al. 2017) and the Empirical Canadian High Arctic Iono- *Correspondence: [email protected] spheric Model (E-CHAIM, Temens et al. 2017, 2019), 1 Department of Physics and Engineering Physics, University which is an improved empirical model for high-latitude of Saskatchewan, Saskatoon, SK, Canada Full list of author information is available at the end of the article regions.

© The Author(s) 2020. This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativeco mmons.org/licen ses/by/4.0/ . Koustov et al. Earth, Planets and Space (2020) 72:43 Page 2 of 14

It is well established that the ionospheric F region has the estimates rely on the occurrence of Pedersen rays the largest electron density and thus afects HF radio (Davies 1990) that reach the top of the ionospheric layer waves in the most signifcant way. Te electron density and return with about the same elevation angle from a at the F layer maximum NmF2 has always been of spe- number of ranges (Ponomarenko et al. 2011). Te algo- cial interest, both experimentally and in the theoretical rithm by Ponomarenko et al. (2011) requires an echoing modeling of the physical processes leading to ionosphere region with Pedersen rays extending at least ~ 200 km. formation (e.g., Kutiev et al. 2013). Despite signifcant Although the possibility of producing peak electron progress in this area, numerous details and trends in the density measurements with the SuperDARN radars has electron density distribution require further investigation been known for years, the method has not been regularly so that forecasting capabilities can be improved. implemented and very limited testing of the method has Te electron density at the F region peak has been been done so far (André et al. 1998; Ponomarenko et al. traditionally studied through ionosonde observations 2011). Part of the problem is the need for reliable echo (Davies 1990; Hunsucker 1991; Rawer 2013). Tis is elevation angle measurements which require difcult because obtaining the maximum electron density value calibration of phased arrays in the HF band (e.g., Pon- from routinely recorded ionograms is a relatively easy omarenko et al. 2015, 2018). Testing has also been limited task provided that the ionogram traces are well defned. by the need for an independent instrument measuring Ionosonde data are usually available with 1 to 15 min res- the electron density distribution in the ionosphere within olution. Over the years, a signifcant body of data, cover- the radar feld of view. Very few SuperDARN radars have ing a wide range of latitudes, has been accumulated (e.g., ionosondes or ISRs positioned at ranges where electron https://www.sws.bom.gov.au/World_Data_Centre). densities are typically obtained. Te electron density distribution with height, along Te SuperDARN radar at Rankin Inlet (RKN) is one with the F layer peak values, is also routinely retrieved of a few radars that has reliable elevation angle data for from incoherent scatter radar (ISR) measurements (Bey- many years of operation (Chisham 2018; Ponomarenko non and Williams 1978; Hunsucker 1991; Rawer 2013). et al. 2011). Continuous operation of the Canadian Modern phased-array ISRs sound the ionosphere along Advanced Digital Ionosonde (CADI) at Resolute Bay several beams nearly simultaneously, allowing them (Jayachandran et al. 2009) and the recent deployment of to build a 3-D distribution of the electron density with an ISR at the same location (to be referred to as RISR) temporal resolutions often as good as 1 min (e.g., Gil- provide an excellent opportunity for testing the quality of lies et al. 2016). One important aspect of both ionosonde electron density estimates from RKN observations near and ISR operations is that the region of measurements the Resolute Bay zenith. is reasonably constrained and known. Tis is necessary Tis study expands the initial work by Ponomarenko for multi-instrument studies, and for the development et al. (2011) by performing a multi-year and two-instru- of ionospheric models, such as the IRI series (Bilitza ment comparison of the F region peak electron density 2018), E-CHAIM and others (e.g., Temens et al. 2017, estimates from RKN observations of ionospheric echoes. 2019). However, for some applications a global coverage is highly desired. Despite the continuously growing array SuperDARN method of F region peak electron of ionosondes and ISRs, their numbers are still limited density estimates from elevation angle and achieving global coverage for instantaneous electron measurements density measurements is still a challenging task. SuperDARN HF radar waves transmitted into the iono- Relatively recently, the Super Dual Auroral Radar Net- sphere experience strong refraction that depends on the work (SuperDARN) radars, which operate within the HF vertical distribution of the electron density. An important band, have been used for measurements of the F region result of the refraction is that radio waves can propagate peak electron density (André et al. 1998; Bland et al. almost perpendicular to the geomagnetic feld lines in 2014; Ponomarenko et al. 2011). To make electron den- extended regions of the high-latitude F layer, often stretch- sity estimates, elevation angles of echo arrival are con- ing from 700 to 1200 km in range. In this “quasi-orthogo- sidered. Both types of SuperDARN echoes, ionospheric nal” radio wave propagation, the presence of feld-aligned scatter (IS) and ground scatter (GS), have been used for decameter irregularities allows for return signals detectable electron density measurements. Identifcation of the ion- by radar. Of special interest are ionospheric signals cor- ospheric region for the electron density estimates is less responding to Pedersen rays. For Pedersen rays, typically certain than with the ionosondes and ISRs. For the work coming from above the F layer maximum, the returned ele- with GS signals, the region of strong radio wave bending vation angles are about the same irrespective of the radar (leading to radio waves turning toward the ground) can range. One important aspect here is that elevation angles be as large as several hundred kilometers. For IS signals, for Pedersen rays and those coming from the maximum Koustov et al. Earth, Planets and Space (2020) 72:43 Page 3 of 14

of the F layer difer insignifcantly, on the order of ~ 1°–2°. be a sphere of the radius of 6370 km) and γ is the angle Tis can be shown by ray tracing for typical ionospheric of the magnetic feld line (within the scattering volume) conditions, see for example Greenwald et al. (2017), Figs. 3 with respect to the horizontal plane. Since the index of and 4. refraction for HF radio waves at frequency f0 is related to Figure 1 gives an example of SuperDARN elevation angle the plasma frequency fp as data taken from RKN radar observations on 03 March 2 2016. To plot these data, an instrumental delay of 3 ns was fp = f0 1 − nS, (2) applied. For all other data considered in this study, the delays were computed as described in Ponomarenko et al. the electron density at the F2 layer maximum can be (2015). In Fig. 1, the elevation angles do not change much evaluated from with range gate in many instances, as evidenced by the 2 2 4π mε0f dominating blue and violet color. Tese measurements cor- = p NmF2 2 , (3) respond to Pedersen ray detection. Tere are also high-ele- e vation echoes, colored in brown, at far ranges of the echo where m and e are the electron mass and charge and ε0 is band. Tese are mixed with low elevation (blue) echoes. the permittivity of free space. Te echoes with the lowest elevation angles at the far edge Determination of NmF2 is done by fnding the nearly of echo bands are typical and are expected if the height of constant elevation angles that indicate the occur- the scatter does not change much. Te occurrence of high- rence of Pedersen rays. In the algorithm implemented elevation echoes is sporadic. Tis is a nonphysical efect in the present study, the P2011 technique was used. caused by the radar software fipping the phase diference π Five sequential range gates with about the same eleva- between the main and interferometer arrays by 2 (e.g., tion angles anywhere between range gates 15 and 40 Milan et al. 1997; Ponomarenko et al. 2018). were used to identify occurrences of Pedersen rays. Ponomarenko et al. (2011) suggested estimating the elec- Similar to P2011, we determined the linear ft slope tron density at the F region maximum NmF2 by measuring to the elevation‐range dependence and required that the elevation angle of Pedersen rays. In the Ponomarenko the ftted slope error, δm , did not exceed 0.5 °/range et al. ’s (2011) technique (we will call it “the P2011 tech- gate and the slope was within a doubled error margin, nique”) it is assumed that the ionosphere is spherically |m|≤2 · δm. stratifed and symmetric so that Snell’s law for the rays Te P2011 technique assumes that the elevation launched at an elevation angle (at the radar site) and ◦ angles can be measured for any angle between 0 and reaching the orthogonality condition in the ionosphere ◦ 90 . In reality, however, because the interferometer at the height of hs can be written in the form (Gillies et al. base exceeds the radar wavelength by a factor of three 2009): to fve, the measured elevation angle has an uncertainty RE + hs (e.g., Milan et al. 1997; McDonald et al. 2013; Pon- cos� = nS sinγ , (1) RE omarenko et al. 2015, 2018) in that a single reported value of can correspond to several propagation paths where nS is the refractive index of the ionospheric in the ionosphere. Tis is because the phase angle in the plasma, RE is the Earth’s radius (with Earth assumed to SuperDARN interferometric measurements is reported π between ± while it actually can be diferent by multi- π ples of 2 . SuperDARN radars report only the lowest ◦ ◦ ◦ possible elevations 2π , from 0 to ∼ 40 − 45 (Milan 03 Mar 2016: RKN, beam 5, 10.4 MHz 40 et al. 1997; McDonald et al. 2013; Ponomarenko et al. 35 30 2015, 2018). Tis ambiguity has not been addressed in 30 25 20 the P2011 approach that we implemented in this study. 15 20 10 Although it does not dramatically afect the measure- 5 Range Gate Elevation, deg ion scat ments in a statistical sense because the events are 10 only 06:50 07:20 07:50 08:20 infrequent, it might be critically important for some Solar Local Time (LT) individual events. Fig. 1 An example of elevation angle data for the RKN radar recorded Two aspects of the P2011 method are important to on 03 March 2016. Corrections of the instrumental time delay of keep in mind. When actual elevation angles are larger 3 ns have been applied. The nearly constant elevation angles, used than the SuperDARN limit angle 2π , smaller eleva- for electron density estimates, are colored in blue and violet. Notice presence of gates with high-elevation angles (brown color) mixed tion angles and, consequently, electron densities would with low elevation angles (blue color) at large range gates be reported. Tis scenario might occur on the dayside, Koustov et al. Earth, Planets and Space (2020) 72:43 Page 4 of 14

especially during high solar activity, when the ionospheric electron density can be high (e.g., Temens et al. 2014). Tere is also a limit on the minimum electron density that can be measured with the P2011 technique. Tis is because in a spherically stratifed ionosphere even a zero-elevation ray enters plasma at a non-zero angle with respect to the contours of constant refractive index (electron density). Te above two efects imply that the P2011 technique has upper and lower limits on measurable NmF2 , on the 11 −3 11 −3 order of 8 × 10 m and 0.55 × 10 m , respectively (Ponomarenko et al. 2011). We note that for low RKN ◦ elevation angles (≤ 5 ), statistical fuctuations inherent to π SuperDARN signals can result in an erroneous 2 shift in the interferometer phase angle diference, and the elec- Fig. 2 The feld of view (FoV) of the Rankin Inlet (RKN) SuperDARN tron density inferred with the P2011 method would be radar, large sector area, and RKN beam 5 (beam-like structure) oriented toward Resolute Bay (RB). At RB the CADI ionosonde and signifcantly overestimated. Tis scenario is expected to incoherent scatter radar RISR are stationed. Open circles are range occur on the nightside, especially in winter and generally gate locations for various RISR beams in the imaging mode (51 at the far edges of echo bands. beams) of observations. Colored circles are range gate locations for the RISR beams in the imaging mode. Red crosses are range Geometry of observations and example of joint gate locations for the RISR beam 3 while the radar was operated SuperDARN, CADI and RISR measurements in the world-day mode. Data from the shown colored beams were considered in this study Te P2011 technique was implemented for the RKN radar observations. One of the reasons for this radar selection was that it has been showing stable instrumental delay of the phase between the main and interferometer arrays. to a range resolution of 50 km. Te RISR beams were In addition, over a decade of its operation (since 2007), spread around the Resolute Bay zenith (Fig. 2) to provide the radar has been showing one of the highest echo detailed measurements in a relatively small spatial region. occurrence rates within the network (Ghezelbash, 2013; In this study, a subset of RISR data from 20 beams (large Koustov et al. 2019). Te fact that independent NmF2 colored circles) was considered (for the beam numbers measurements could be derived from the CADI iono- see Table 1). Te selected beams (colored large circles in sonde within the radar feld of view was also an impor- Fig. 2) are oriented toward the RKN radar so that the two tant consideration, although high-resolution ionogram instruments are observing approximately the same iono- scaling was not available at the time of the P2011 study. spheric region. In the imaging mode, the return signals One of the instigating factors for the renewal of interest were integrated over two intervals: 60 s and 300 s. We into testing of the P2011 method is the installation and used the 300-s averaged electron densities. We also used successful operation of the RISR at Resolute Bay (Gillies RISR radar data collected in its “world-day mode” with et al. 2016) that, starting from 2016, has been providing 11 beams (Gillies et al. 2016). Tis mode of RISR opera- high quality and detailed measurements of the electron tions is designed to measure ionospheric parameters over density in the ionosphere for ~ 60 days in a year. a much wider area which is critical for the reliable deri- In the present work, routine observations of all three vation of the plasma fow vectors from the line-of-sight instruments were considered. Figure 2 shows the feld velocity in multiple beams (Gillies et al. 2016). Te radar of view (FoV) of the RKN radar and the orientation of range resolution was the same as in the imaging mode. its beam 5, which (in addition to adjacent beams 4 and For the world-day mode, only beam 3 data were consid- 6) was used in this study. Tis beam crosses the Reso- ered. Te reason is that the beam separation in this mode lute Bay (RB) zenith. Figure 2 also shows the pierce is much larger, while the orientation of beam 3 is almost points (at various heights) of multiple RISR radar beams exactly along the RKN beam 5. Tis allowed us to com- whenever it was operating in its “imaging mode”, with a pare the RKN and RISR data in as close directions and signal being transmitted and received in 51 beams. Te spatial regions as possible. Even so, the spatial overlap radar has an electronically steerable array so that it can between the regions of the RKN and RISR measure- scan through pre-decided beam orientations on a pulse‐ ments is not ideal. It is not only that the RKN data were by‐pulse basis. In the imaging mode, the radar used a obtained in a band of ranges, gates 15–40 (see dashed long pulse with a pulse length of 0.33 ms, corresponding lines in Fig. 2), but also because RISR electron density Koustov et al. Earth, Planets and Space (2020) 72:43 Page 5 of 14

Table 1 RISR and RKN radar beams selected for the comparison Beams RISR Beams RKN

World-day 3 5 Imaging 4, 7, 8, 11, 14, 15, 18, 21, 22, 25, 28, 29, 32, 33, 36, 39, 40, 43, 46, 47 4, 5, 6

profles were treated as being in the vertical direction while the actual data have been collected with tilted beams. Te electron density was inferred from a multi- parameter ft to the measured autocorrelation functions and the measured total power. Te RISR electron densities were routinely calibrated using the RB ionosonde (Temens et al. 2014). Te electron density profles from RISR measurements Fig. 3 The F region peak electron density NmF2 according to Rankin were often not very smooth. To infer the peak electron Inlet (RKN) HF radar elevation measurements in beams 4–6 (blue density, we analyzed each profle individually by consid- squares), CADI ionosonde (dark green open circles) data and RISR ering all the electron density profle points close to the values (red diamonds) averaged over multiple beams (listed in vicinity of the maximum, usually in the range of 200– Table 1). One full day of observations, 3 March 2016, is considered. For this plot, all RKN data with the antennae cross-phase deviation 350 km, and then made a ft to a Chapman-like function ◦ from its maximum possible value Ψadj > 250 (see the text for the that had three free parameters, the scale height, the peak explanation) are shown by pink squares. This type of “anomalous” electron density, and its altitude. We note that the ftting points was excluded from further analysis in this study. The dotted N F provided values that were, typically, only slightly smaller line denotes the m 2 of the instrumental low limit for RKN (less than 10% diferent) than the absolute maximum measurements electron density in the profles. Te RKN radar was mostly operated with 1-min scans and used the standard 300 µs pulse length, providing 45-km resolution. An important feature of this radar is RKN data are shown by squares for every measurement that its operating frequency has often been alternated available, but the data have two diferent colors: blue between ~ 10 and ~ 12 MHz every new scan (especially color for regular points and pink color for anomalous in the last 5 years). Te data from the standard FITACF points. While analyzing the electron density estimates, output (Ponomarenko and Waters 2007; Ponomarenko we realized that some points were well above the gen- et al. 2007) were analyzed. We note that there is some erally expected electron density for the time of the day. uncertainty in SuperDARN echo location on the order Tese were presumably associated with measurements of 100 km (Yeoman et al. 2008). Tis uncertainty is not made when the antennae cross-phase deviation from its Ψadj critical for the present work, as we consider RKN data maximum possible value, (Ponomarenko et al. 2018), Ψadj in many gates (gates 15–40), covering > 500 km in range. was large. We remind the reader that was introduced Ψadj = kd − Ψ Te larger uncertainty is because no location for the echo by Ponomarenko et al. (2018) as , where k is the wave number for radio waves, d is the separa- band (involved in each 1-min electron density estimate) kd was determined. CADI observations are usually taken at tion between the main and interferometer arrays ( is the maximum possible phase diference for observations a resolution of 1 to 5 min, but because of a desire to have Ψ long-term coverage, ionograms were manually scaled for along the radar boresight at zero elevation angle) and is every 30 min. For limited periods, 5-min ionograms were the phase diference between the main and interferome- obtained. ter arrays. For this reason, such measurements have been discarded in this study (apart from Fig. 5 which serves to illustrate such data). Te threshold for data removal was Example of three instrument observations ◦ set at the level of Ψadj ≥ 250 . Such points are shown as Figure 3 presents an example of joint 24-h-long measure- pink-colored squares in Fig. 3. ments of NmF2 with the three systems. RISR 5-min data According to Fig. 3, NmF2 varies signifcantly over are available continuously throughout the day. In these the day, and the NmF2 values from all three instru- diagrams and other fgures, we calculate Solar Local Time ments vary consistently. Te RKN radar shows an after- (LT) by assuming LT UT 6.5 h. CADI ionograms con- = − noon maximum. Tere are several data gaps at dawn sidered for this plot were scaled at a 5-min resolution. (05–07 LT), at near noon and after 20 LT. At ~ 12 LT, Koustov et al. Earth, Planets and Space (2020) 72:43 Page 6 of 14

the RISR electron densities are somewhat larger than Comparison of Rankin Inlet NmF2 estimates those measured by CADI. In the dusk/midnight sector and CADI ionosonde measurements of 20–24 LT CADI shows larger electron densities than Te Resolute Bay CADI has been in operation for almost the other two systems. 2 decades (Temens et al. 2017). Te RKN radar data are In Fig. 3, a prolonged electron density minimum available from 2008 onwards, allowing joint radar–iono- is seen between 04 and 07 LT where RISR was capa- sonde measurements that cover the period of 2008–2017. ble of continuously reporting the values, while the Figure 4 presents all of the data on NmF2 collected by RKN radar only measured a few points. Tere were these systems as an occurrence plot on a local time– also very few CADI data points. Te gap in HF radar month plane, separately for each system. For this spe- and CADI ionosonde data indicates that these instru- cifc plot, the RKN measurements in beams 4–6, at gates ments cannot measure electron densities below 15–40, and at all radar operating frequencies were con- 11 −3 NmF2 ∼ 1.0 × 10 m . Tis instrumental limitation is sidered to ensure continuous coverage. known from previous publications (e.g., Davies (1990) Te plots of Fig. 4 are similar to those presented by and Rawer (2013) for ionosondes and Davies (1990) and Fig. 4 of Ponomarenko et al. (2011), but here we cover Ponomarenko et al. (2011) for HF radars). a signifcantly longer time period including the years of Te data presented in Fig. 3 indicate that RKN- the solar cycle 24 maximum (2013–2015). Te similarity based NmF2 can be somewhat larger or smaller than between the plots of Fig. 4a, b is obvious. Indeed, elec- that measured by CADI or RISR, but the agreement is tron density enhancements are centered around local generally reasonable. Figure 3 also indicates that the noon and they are well seen in both plots. On both plots, “anomalous” (pink) RKN-based points occur at the electron density decreases in winter are obvious, espe- time of low electron density in the ionosphere as meas- cially in the midnight sector. For some years, for example ured by RISR, one point is in the dawn sector and a in 2011, the electron density maxima are achieved at the couple of points are in the dusk sector. Reduced over- equinoctial time, consistently in both data sets. all electron densities in the dawn sector and toward the Diferences between the RKN radar and CADI iono- midnight sector are generally expected (Temens et al. sonde NmF2 values are also recognizable in Fig. 4a, b. 2017, Fig. 2). One such diference is the larger nighttime RKN electron

RKN Radar CADI 2018 abCADI Density 2017

NFm 2, 2016 11 -3 10 m 2015 5.0 2014 4.5 4.0 3.5 ar 2013 3.0 Ye 2.5 2012 2.0 1.5 2011 1.0

2010

2009

2008 0 6 12 18 24 6121824 Solar Local Time (LT) Fig. 4 A contour plot of the F region peak electron density versus local time a for the Rankin Inlet (RKN) radar measurements in beams 2–6 and gates 15–40 and b according to CADI measurements at Resolute Bay. Solar local time was counted for the radar location, LT UT 6.5. Data at all = − RKN radar operating frequencies were considered Koustov et al. Earth, Planets and Space (2020) 72:43 Page 7 of 14

densities in most years (e.g., 2011). Te other diference 5 Correlations: RKN<250o o a RKN all is generally larger daytime electron densities according RKN<250 : 0.35 CADI 4 RKN all: 0.04 to CADI (e.g., 2012–2015 data). However, in 2008–2010 3 (years of low solar activity), the near noon RKN electron 2 densities are larger than those given by CADI. Interest- 1 ingly, in 2017 (also a year of low solar activity), the rela- Night 5 Correlations: o tionship reverses. RKN<250 : 0.55 b To assess the data presented in Fig. 4a, b more explic- 4 RKN all: 0.33 3

itly, we present in Fig. 5a–d a series of line plots built for 3 0 ± 3 LT 6 ± 3 LT 2 four time sectors: night ( ), dawn ( ), m 1- day (12 ± 3 LT ) and dusk (18 ± 3 LT ). Here the average 1 1 Dawn N F2 5 Correlations: m is plotted for the entire period under considera- o

2, 10 c 4 RKN<250 : 0.77 tion in each time sector. Te RKN data are given for all m RKN all: 0.78 measurements available (red lines) and for the data with NF 3 cross-phase deviation from its maximum possible value 2 ◦ Ψadj (Ponomarenko et al. 2018) being < 250 . CADI data 1 Day 5 Correlations: are shown by grey lines. Both CADI and RKN curves o d ± 4 RKN<250 : 0.66 were obtained by applying 30-day boxcar sliding flter RKN all: 0.57 (computing median values at each moment). To quantify 3 the agreement between the data trends, the correlation 2 coefcients between the curves are presented in the left- 1 Dusk upper corner of each panel. 2008 2010 2012 2014 2016 2018 Let us frst focus on the daytime observations, Fig. 5c, Universal Time (years) where agreement between the RKN and CADI data Fig. 5 Line plots of the F region peak electron density NmF2 trends is best, as indicated by the correlation coefcients. averaged in four time sectors (a night: 0 ± 3 LT ; b dawn: 6 ± 3 LT ; Reasonable consistency, not only in trends but in values c noon 12 ± 3 LT ; d dusk 18 ± 3 LT ) according to Rankin Inlet radar elevation measurements, red and blue lines. Red lines show NmF2 as well, is seen for some years (2011, 2015–2016). Both estimates based on all radars records while blue lines represent instruments show overall larger electron densities dur- NmF2 estimates for limited data sets with discarded measurements ing years of high solar activity (2012–2014) which can be for which the antennae cross-phase deviation from its maximum ◦ recognized by tracing the envelope of highest electron possible value Ψadj (Ponomarenko et al. 2018) was above value 250 . densities in each year. One can also see that anomalous Radar data at all operating frequencies were considered. Grey lines are CADI-based data on NmF2 , averaged over the same local time points do not signifcantly afect the average electron intervals. Shown by numbers are the correlation coefcients between densities, as the red curves are located only slightly above the radar and CADI curves. Lines for each instrument were obtained the blue curves, with the strongest efect at the equinoc- by applying ± 30-day boxcar sliding flter tial time in 2012 and 2013. Trends for the dusk sector (Fig. 5d) are very similar but show a slightly worse correlation. Te largest RKN NmF2 CADI. Tis shows that the high-phase RKN data have the underestimations are seen in 2012–2014. most signifcant impact when electron densities are low. For the dawn sector, shown in Fig. 5b, the red- and We also see the anti-correlation in 2008–2009, but here blue-colored curves are of the grey line and the trends the elimination of high-phase data gives only a very small divert, with the correlation coefcients between blue and improvement. Also, signifcant RKN electron density grey curves and red and grey curves dropping down fur- underestimations are seen in the summers of 2011–2013. ther to 0.55 and 0.33, respectively. In this case, one can One criticism of the above data trend analysis is that hardly think of there being a linear relationship between the amount of radar and ionosonde data points involved the RKN and CADI measurements. Te solar cycle efect difer signifcantly and the measurements are not simul- is evident in the CADI data, but hardly recognizable in taneous. To further the comparison, we present in the RKN data. Fig. 6a–d scatter plots of the RKN-based NmF2 estimate Electron densities in the midnight sector (Fig. 5a) are versus CADI-based NmF2 for matched moments. Each expected to be low, especially during winter. While CADI plot in Fig. 6a–d is for a corresponding local time sector (grey line) densities behave as expected, RKN observa- of Fig. 5a–d. Here the RKN data at all operating frequen- tions show the opposite, with spikes of electron density cies were considered. in the winter. From 2013 onward, the elimination of high- phase RKN data greatly improves overall agreement with Koustov et al. Earth, Planets and Space (2020) 72:43 Page 8 of 14

stronger for the nighttime. For daytime, the RKN NmF2 a b values tend to be close to those measured by CADI. Overall, however, one can conclude that there is a reasonable agreement between the two instruments, par- ticularly for the daytime. Tis judgment is based on good data clustering not far away from the bisector of perfect agreement. In addition, the linear fts to the data clouds in Fig. 6a–d are 0.49, 0.55, 0.82 and 0.68, respectively. Te c d correlation coefcients are not high ranging from 0.4 to 0.6.

Comparison of Rankin Inlet NmF2 estimates and RISR ISR values RKN and RISR plots of joint electron density measurements, similar to that of Fig. 6a–d, were produced for about 60 events in 2016 for which RISR was operational for at least 2 h. Te plots were built from the world-day Fig. 6 Scatter plots of the Rankin Inlet NmF2 estimates versus CADI-based NmF2 for matched moments and four local time and imaging modes of RISR operation separately, Fig. 7. sectors: a night 0 ± 3 LT ; b dawn: 6 ± 3 LT ; c noon 12 ± 3 LT ; d dusk Since RISR data were given only for ~ 60 days and with 18 ± 3 LT . Radar data in 2008–2017 at all operating frequencies were a 5-min temporal resolution, the number of joint meas- considered. The total number of available points n is shown at the urements is not as large as for the RKN-CADI compari- bottom of each panel (yellow). In all panels, the solid lines are the son of Fig. 6, especially after the data were split onto four lines of best linear ft time sectors; however, the statistics are sufciently large to infer trends. Scatter plots of Fig. 7 show a general agreement Te scatter plots in Fig. 6a–d, for nearly coincident between the instruments although the degree of agree- measurements (a 1-min RKN measurement being < 1 min ment varies. One obvious feature in Fig. 7 is smaller RKN apart from a CADI measurement), show that the RKN- electron densities at night and dusk (Fig. 7a, d, e, h) as based NmF2 values are smaller than those measured evidenced by the cloud of points located below the bisec- by CADI, with the exception of daytime. Te efect is tor of perfect agreement. Te points for the daytime and

a bcd

ef gh

Fig. 7 Scatter plots of the F region peak electron density NmF2 inferred from Rankin Inlet (RKN) radar measurements (beams 4–6) versus NmF2 inferred from RISR measurements. For RKN, data at all operating frequencies were considered. a–h are for RISR observations in the world-day and imaging modes, respectively, with averaging over the beams as described in the text. The local time sectors are the same as for those for Fig. 6. The total number of available points is shown at the bottom of each panel (yellow). In all panels, the solid lines are the lines of the best linear ft Koustov et al. Earth, Planets and Space (2020) 72:43 Page 9 of 14

dawn sectors (Fig. 7b, c, f, g) are spread signifcantly so quantile–quantile (Q–Q) plots of the RKN and CADI/ that the data clouds are of a circular shape. Tey are, RISR data. We adopted an approach similar to that of however, centered on the bisector of perfect agreement Tindale and Chapman (2017). Te data for each of the with the daytime data being slightly shifted to higher instruments were frst ranked according to reported RKN values. Te slopes of the best ft line in the case of values of the electron density and then the quantiles Fig. 7a–d are 0.42, 2.4, 0.82 and 0.67, respectively. were selected between 5 and 95% of the total number For the case of Fig. 7e–h, the slopes are − 1.36, 1.46, of points with a step of 5%. For each quantile level, the 0.86 and 0.56, respectively. Te slopes for daytime and electron density value inferred from the RKN measure- dusk (the last two numbers for each data set) are con- ments was paired with a similarly computed value from sistent with the slopes for the RKN-CADI comparisons another instrument, CADI or RISR. All the point pairs of Fig. 6. Tey support our judgment on the reason- were then placed on a 2-D plane with the RKN values able agreement between the instruments. Slopes for the being plotted along the Y axis, Fig. 8, forming the Q–Q nighttime and dawn data difer greatly from expectations. plots. We selected the same data sets as those consid- Te reasons for this result are discussed below. ered in Figs. 6 and 7 and placed them in a similar fash- Te prominent feature that stands out in Fig. 7 is the ion to aid in comparison. occurrence of points located well above the bisector of Te Q–Q plots of Fig. 8c, g, k show that the daytime perfect agreement whenever the RISR electron densi- electron densities, corresponding to various quantiles, 11 −3 ties are low, below ~ NmF2 ∼ 1 × 10 m (Fig. 7a, e are collocated with the bisector of the perfect agree- and especially Fig. 7b, f). Tese are in abundance for the ment for both the RKN-CADI and the RKN-RISR data nighttime and dawn plots. Tis efect can also be identi- sets (in both modes of the RISR operation). Tis indi- fed in Fig. 5a where nighttime RKN and CADI data are cates that the data distributions for independent instru- compared. We remind the reader that the RKN data of ments are almost the same. Data of Fig. 8 in other time Fig. 7 include only measurements with the cross-phase sectors show departures of the points from the bisector deviation from its maximum possible value Ψadj (Pon- of perfect agreement. ◦ omarenko et al. 2018) being below 250 . One can con- For the dusk observations, Fig. 8d, h, l, the points are clude that this limitation is insufcient to remove all somewhat of the bisector but align reasonably with 11 −3 erroneous RKN measurements with the currently imple- it for NmF2 = (1 − 3) × 10 m . For larger CADI mented electron density estimation algorithm. or RISR densities, the RKN values are systematically Although it is not readily apparent, such anomalous smaller. Tis indicates that the data distributions devi- RKN points are present in the RKN-CADI data set of ate in the tails of the distributions, with the RKN values Fig. 6, but their total number is small in the large body of being systematically smaller. Tis efect had been men- measurements. Another important issue here is that both tioned earlier and can be recognized in the scatter plots ionosondes and HF radars are unable to detect echo sig- of Figs. 6, 7. nals at low electron densities, contrary to ISRs. Te RKN-CADI dawn plot of Fig. 8b is similar to Data of Fig. 7 indicate that the anomalous points are that of the dusk plot of Fig. 8c except for the deviations more frequent for the imaging mode of RISR operation. from the bisector at large electron densities are slightly We believe that this is purely owing to the relatively small stronger. For nighttime, Fig. 8a, the deviations of the number of data points. Aside from this feature, the scat- RKN and CADI distributions are evident at just about ter plots of Fig. 7 for the two modes of RISR measure- any electron density. ments (top and bottom rows) look similar. Generally, Te RKN-RISR data for the dawn and nighttime, similar plots for the single (Fig. 7a) and multiple beam Fig. 8f, i, g, show another type of diference. Te points (Fig. 7b) RISR measurements indicate that SuperDARN here are located well above the bisector of perfect data, based on the analysis of echo signals from spatial agreement, indicating the presence of higher-value regions of ~ 200 km size, is a reasonable representation of records in the RKN data sets. Tis is consistent with the the electron density in a broad area around the Resolute scatter plots of Fig. 7. Bay zenith most of the time. We also performed the Q–Q analysis by considering all the RKN data on the electron density in 2008–2017 Assessment of RKN electron density estimates and all the CADI electron density data that are avail- with quantile analysis able to us. In this analysis, matching in time was not required. We found that the Q–Q plots for various time To alternatively assess the quality of the NmF2 estimates from RKN measurements, we performed sectors looked very much similar to those shown in the additional analysis by producing the so-called Fig. 8a–d indicating the robustness of the results for the joint RKN-CADI data set. Te Q–Q plots for the entire Koustov et al. Earth, Planets and Space (2020) 72:43 Page 10 of 14

Night Dawn Day Dusk 6

-3 abc d m 1 1 4

2 2 (RKN), 10

m 2764 2110 4048 3909 NF 0 246 246246 246

11 -3 NmF2 (CADI), 10 m 6 efg h 4 -3 m 11 2 336 242 564 687 6 ijkl F2 (RKN), 10 m

N 4

2 215 280 769 653

0 246 246246 246

11 -3 NmF2 (RISR), 10 m

Fig. 8 Quantile–quantile (Q–Q) plots of the NmF2 nearly simultaneous measurements by the RKN radar and the CADI ionosonde or the RISR radar for various time sectors and experiments. a–d are for the RKN-CADI joint data set, e–h are for the RKN-RISR 1 beam joint data set and i–l are for the RKN-RISR 20 beams joint data set. The total number of points for each data set is shown at the bottom of the plots

RKN data set and the entire RISR data set showed dif- Figure 9 compares NmF2 inferred at 12 MHz versus ferences similar to those shown in Fig. 8i–l. NmF2 inferred at 10 MHz for measurements with a time separation of 1 min. Only daytime and nighttime plots are Does the radar frequency of the SuperDARN presented because they refect the extremes that we iden- measurements afect the electron density tifed in the plots for all time sectors. For the daytime, the estimates? consistency between the data is evident. Similar results As we have already mentioned, the RKN radar has often are seen in plots for dusk and dawn (not presented here). been operating in a dual frequency mode, switching the Data for nighttime (Fig. 9a, b) show noticeable difer- frequency of transmission between 10 and 12 MHz every ences. Here the electron densities inferred from 12 MHz other scan (every other minute). Although the NmF2 measurements are slightly larger than those inferred estimates with the RKN radar should be, ideally, inde- from 10 MHz measurements. Similar, but weaker, incon- pendent of the operating frequency, some diferences sistencies are seen for nighttime winter observations in might be expected. One reason is that the bands of ech- the dawn and dusk sectors (data are not presented here). oes involved in electron density estimates do not typically We note that these are the periods with generally lowest coincide. Te other potential factor is the diference in electron densities and strongest north–south electron the scattering height. For this reason, here we investi- density gradients in terms of season and local time. gate the efect of radar operating frequency on the NmF2 estimates. Koustov et al. Earth, Planets and Space (2020) 72:43 Page 11 of 14

Dusk ab 6

5 -3 Number m of events 11 4 60 40 30 25 15 3 10 5 1 0 cd 2 2 (CADI), 10 m

NF 1

1 2 3 4 5 6 11 -3 NFm 2 (RKN), 10 m

Fig. 10 Scatter plot of the F region peak electron density NmF2 inferred from Rankin Inlet (RKN) radar measurements (beams 4–6) versus NmF2 inferred from CADI measurements. All dusk sector Fig. 9 Contour plots of the F region peak electron density NmF2(12) observations in 2008–2017 were considered. The white line is the inferred from Rankin Inlet (RKN) radar measurements (beams 4–6) linear ft line to the data, as described in the text at 12 MHz NmF2(10) inferred from RKN measurements at 10 MHz. a, b Are for nighttime observations in winter and spring, respectively. c, d Are for daytime observations in winter and spring, respectively. All records in 2008–2017 have been considered. The nighttime sectors (not presented here), are given in Table 2 along and daytime sectors were selected as 0 ± 3 LT and 12 ± 3 LT , with the correlation coefcients of the scatter plots and respectively. The total number of available points n is shown at the upper-right corner of each panel. In all panels, the solid lines are the the total number of points involved. Te slopes of the lines of the best linear ft lines are all above 1.0 signifying a general tendency for the RKN measurements to underestimate the peak electron density in the ionosphere, as measured by CADI. On the improvements of electron density estimates Te negative intercepts indicate that RKN does not accu- from solely RKN data rately measure very low densities, as we have previously Te data of Figs. 6 and 7 show that RKN-based electron discussed. Application of these empirical relationships to densities difer from those measured by the CADI iono- individual measurements should provide a better quality N F2 sonde and the RISR radar, both of which are well estab- estimate of m in the RB zenith. lished instruments used to characterize the electron Similar ftting has been done for the joint RKN- density distribution in the ionosphere. To use only RKN RISR data. We were seeking the dependence of a type N F2(RISR) = a · N F2(RKN) + b radar for NmF2 estimation, the HF measurements have m m . Te ft line slopes to be adjusted to align with the established observations were found to be strongly afected by the erroneous of the F layer maximum. Here we propose a simplifed RKN estimates at low ionospheric electron densities (as ∼ 1 × 1011 m−3 recipe for such an adjustment. measured by RISR) below . On our sec- Figure 10 illustrates the approach. We show here the ond attempt, we excluded all the points with such low scatter plot of the RKN NmF2 versus CADI NmF2 , for RISR electron densities. Te coefcients of these sec- the dusk observations. Tis is a redrawn Fig. 6d. Similar ond-attempt ft lines are given in Table 3 along with the plots have been considered for all other panels of Figs. 6 correlation coefcients of the scatter plots and the total and 7. To characterize the plot in Fig. 10, we conducted a linear Table 2 Coefcients of a linear ft ft by minimizing the standard deviations of all the point NmF2(CADI) = a · NmF2(RKN) + b to the scatter plots distances from the linear ft line in the direction perpen- of RKN electron density versus CADI (Fig. 6) dicular to the ft line. Tis method of ftting was done a b ( × 1011 m−3) Correlation Number because both instrument measurement techniques have coefcient of points some uncertainties. Our goal was to establish the rela- Night 1.92 0.93 0.39 2764 tionship of the type NmF2(CADI) = a · NmF2(RKN) + b . − Dawn 1.56 0.83 0.45 2110 Such a relationship allows one to “calibrate” the RKN − Day 1.11 0.35 0.59 4048 measurements, for the dusk sector. Te coefcients of − Dusk 1.42 0.53 0.47 3909 the ft lines, for Fig. 10 and similar plots for other time − Koustov et al. Earth, Planets and Space (2020) 72:43 Page 12 of 14

Table 3 Coefcients of a linear ft We also found a slight dependence of RKN density N F2(RISR) a N F2(RKN) b m = · m + to scatter plot of RISR measurements on radar frequency; electron densities electron density versus RKN electron density for 1 beam at 12 MHz are slightly larger, on average, than those at (frst 4 lines) and 20 beam (last 4 lines) RISR measurements 10 MHz (Fig. 9) for winter nighttime conditions but (Fig. 7) not for the daytime conditions, Fig. 9. We expect that a b( × 1011 m−3) Correlation Number 12 MHz echoes would come at somewhat larger ranges as coefcient of points compared to those at 10 MHz, statistically speaking. Te Night, 1 beam 2.38 1.84 0.23 304 typical latitudinal diference would be ~ 100 km. How- − Dawn, 1 beam 0.53 0.62 0.28 175 ever, one can fnd cases with spatial separation between Day, 1 beam 0.96 0.13 0.49 563 10 and 12 MHz echo detection zones up to 5 range gates, − Dusk, 1 beam 1.56 0.64 0.44 680 i.e., 2° of latitude. An important feature of the ionosphere − Night, 20 beams 0.44 2.84 0.14 111 equatorward of Resolute Bay is that the maximum elec- − − Dawn, 20 beams 0.60 0.44 0.15 142 tron density can actually decrease toward lower latitudes − × 11 −3 Day, 20 beams 1.09 0.41 0.56 743 by (0.1 0.2) 10 m over two degrees of latitude − Dusk, 20 beams 1.40 0.32 0.47 546 at nighttime (Temens et al. 2017). Although the efect − seems to be weak, it can explain partially the reason for the diference in the peak electron density inferred from 10 and 12 MHz measurements. number of points involved in each plot. Te slopes of the Our comparison for the nightside and morning hours lines are close to or above 1.0 for daytime and dusk obser- showed occasional strong disagreements with CADI and, vations, indicating a general tendency for the RKN elec- especially, RISR. One interesting efect is the generally tron densities to underestimate the peak electron density smaller electron densities in RKN measurements during in the ionosphere, equatorward of the RB location. We nighttime (Figs. 6, 7). Several efects very likely contrib- note that the correlation coefcients are very low (< 0.3) ute to this. Te frst one is the fact that the RKN electron for the nighttime and dawn observations, such that the density derivation procedure considers elevation angles ft lines here do not have practical signifcance and imply- for the Pedersen rays propagating at the top of the F layer, ing that improvement of the SuperDARN-based electron i.e., slightly above the height of the electron density peak. density estimates is needed for the cases of low F region Tus, the real maximum electron densities at the F layer peak electron densities. peak are slightly larger than the reported ones. Te sec- ond potentially contributing factor is an assumption of Discussion and conclusions the fxed height of the scatter at 250 km (Eq. 1). Since the An important general conclusion from the comparisons nighttime heights of the F layer can be as high as 400 km, the real values of the peak electron density can be larger of the RKN-based electron densities NmF2 and those measured by CADI and RISR is that the RKN estimates than those inferred by assuming a 250-km height, by a are reasonable most of the time, as indicated by good factor of up to ~ 1.15, which one can estimate from equa- clustering of the data along the expected bisector line on tions of Ponomarenko et al. (2011). Te impact of the the plots of Figs. 6 and 7. Te data showed better agree- fxed height assumption for the daytime electron density ment for daytime conditions and signifcant diferences estimates is not that important since the real heights of at nighttime. the daytime scatter are much closer to 250 km. For daytime conditions, an interesting very minor We hypothesize that the lower RKN electron densities efect in the reported data is the RKN electron density are also due to the averaging efect in HF radar measure- overestimation during daytime (Figs. 6, 7). Tis can be ments. While CADI and RISR are detecting signals from explained by the fact that under enhanced electron den- localized regions with the strongest electron density, the sity, the RKN echo bands shift to lower latitudes so that elevation angles in RKN measurements are “averaged” the RKN electron density estimates are related to some- over at least fve radar gates (225 km) so that any local- what lower latitudes than the Resolute Bay location for ized electron density enhancement is smoothed out. Te CADI and, to some extent, than the RISR latitudes of smoothing efect is expected to be stronger for iono- measurements. Tis increase of the electron density spheric conditions with higher patchiness and poorer with decreasing latitude at lower polar cap latitudes is propagation conditions. Out of all time sectors, this expected at daytime (Temens et al. 2017). It can be as is very likely to occur at winter nighttime because here 11 −3 large as (0.2 − 0.5) × 10 m over two degrees of lati- the ionosphere is depleted and highly patched. It is not tude (potentially possible separation between the RKN a surprise then that the RKN under-estimation efect is and CADI echo detection zones). stronger at winter nighttime. Koustov et al. Earth, Planets and Space (2020) 72:43 Page 13 of 14

Te data presented show occasional dramatic RKN seasonal and solar cycle efects. Straight application electron density overestimation at low real ionospheric of the SuperDARN electron densities for the analysis electron densities (according to RISR); these measure- of individual events is less certain as local anomalies ments were classifed as “anomalous points” (Figs. 3, 5, 7). do occur, and they have to be investigated on a case- π We indicated that this happens due to the incorrect 2 by-case basis. wrapping of the phase angle in interferometric measurements. Te number of these points is not usually signifcant, as we learned from the RKN data set analysis. For example, we estimated that these points make up ~ 6% of Abbreviations our data for February 2016. In terms of the time sector, CADI: Canadian Advanced Digital Ionosonde; e-CHAIM: Empirical Canadian these points were mostly seen at the nighttime and dawn High Arctic Ionospheric Model; FoV: Field of view; GS: Ground scatter; HF: High frequency; IMF: Interplanetary Magnetic Field; IRI: International Reference and in the afternoon in 2016 (example in Fig. 3). For other Ionosphere; IS: Ionospheric scatter; ISR: Incoherent scatter radar; LOS: Line- years of observations, the number of such points and of-sight; LT: Local time; MHz: Megahertz; RB: Resolute Bay; RKN: Rankin Inlet; their preferential occurrence time varied. Figure 5 shows SuperDARN: Super Dual Auroral Radar Network; RISR: Resolute Incoherent that these points have the most impact on nighttime data, Scatter Radar-Canada. and especially when density is low. Te issue requires fur- Acknowledgements ther investigation. Figure 7 indicates clearly that despite Continuous funding of the SuperDARN radars by National Scientifc Agencies of Australia, Canada, China, France, Italy, Japan, Norway, South Africa, the our eforts, quite a few erroneous measurements are still United Kingdom, and the United States of America is appreciated. The current in the database implying that simply removing all meas- research would be impossible without ongoing support from the Canada urements with the antennae cross-phase deviation from Foundation for Innovation, Canadian Space Agency’s Geospace Observatory ◦ (GO Canada) continuation initiative and NSERC Discovery grant to AVK. AVK Ψadj > 250 its maximum possible value does not fully acknowledges funding from the Institute for Space-Earth Environmental resolve the problem. We are planning to address the issue Research (ISEE), Nagoya University, while he visited Nagoya University and in the future. started work on the paper. The University of Calgary RISR radar is funded by the Canada Foundation for Innovation and is a partnership with the US We can summarize the results of this study as follows: National Science Foundation and SRI International. CHAIN ionosonde operation was supported by the Canada Foundation for Innovation, the Canadian 1. Te RKN radar estimates of the electron density at Space Agency, and the New Brunswick Innovation Foundation. Scientifc analysis of CADI and Rankin inlet radar data is funded by NSERC (Canada). the F region peak in the RB area are reasonably con- Critical comments by two reviewers are appreciated. sistent, most of the time but not always, with measurements by the CADI ionosonde measuring the Authors’ contributions AVK drafted the manuscript. SU performed extensive analysis of the data pro- electron density in the zenith of RB and the RISR ISR vided by PVP (electron density measurements), RG (RISR electron density data) measuring electron density in much broader area and DT (CADI ionosonde data). NN contributed to the radar data analysis. All mostly southward of RB. authors read and approved the fnal manuscript.

2. Te RKN peak electron density estimates are in Funding better agreement with CADI and RISR for day- A.V.K. was supported by NSERC Discovery (Canada) Grant (RGPIN 03741-2016) time and for electron densities in the range of and International Joint Research Program of the Institute for Space-Earth 11 −3 Environmental Research, Nagoya University. S.U. was funded by NSERC CREATE NmF2 ∼ (1 − 3) × 10 m at dusk. For daytime, (Canada) program (RS346079). P.V.P. was funded by NSERC Discovery (Canada) there is a very minor overestimation efect. For other Grant (RGPIN1774-2002). N.N. was supported by JSPS KAKENHI (Japan) Grants time sectors, there is a tendency for the RKN elec- 16H06286 and 18KK0099. RG was funded by the Canadian Foundation for Innovation project number 21425. D.R. Themens is supported by Defence tron density to be underestimated. Research and Development Canada contract number W7714-186507/001/SS. 3. Tere is a very subtle tendency for the NmF2 inferred from 12 MHz measurements to be larger than that Availability of data and materials SuperDARN data can be obtained from https://superdarn.ca. RISR data are inferred from 10 MHz measurements during night- available at Madrigal database http://madrigal.phys.ucalgary.ca or http://data. time, away from summer. phys.ucalgary.ca. CADI data are available for download at http://chain.physi 4. Te RKN peak electron density estimates for night- cs.unb.ca. time and dawn conditions when the electron densi- 11 −3 Ethics approval and consent to participate ties are below NmF2 ∼ 1 × 10 m can often be Not applicable. erroneous due to 2π shifts of the phase in interfero- Competing interests metric measurements. In the present study, to dimin- The authors declare that they have no competing interests. ish the role of this efect, some potentially afected RKN data were discarded. Author details 1 Department of Physics and Engineering Physics, University of Saskatchewan, 5. Te results imply that SuperDARN dayside and dusk- Saskatoon, SK, Canada. 2 Department of Physics and Astronomy, University side electron density estimates are justifable for their of Calgary, Calgary, AB, Canada. 3 Department of Physics, University of New use in statistical studies, such as those focused on Koustov et al. Earth, Planets and Space (2020) 72:43 Page 14 of 14

Brunswick, Fredericton, NB, Canada. 4 Institute for Space‑Earth Environmental S, Parisi M, Torta JM (2013) Solar activity impact on the Earth’s upper Research, Nagoya University, Nagoya, Aichi, Japan. atmosphere. J Space Weather Space Clim 3:A06. https://doi.org/10.1051/ SWSC/2013028 Received: 29 August 2019 Accepted: 21 March 2020 McDonald AJ, Whittington J, de Larquier S, Custovic E, Kane TA, Devlin JC (2013) Elevation angle-of-arrival determination for a standard and a modifed SuperDARN HF radar layout. Radio Sci 48:709–721. https://doi. org/10.1002/2013RS005157 Milan SE, Jones TB, Robinson TR, Thomas EC, Yeoman TK (1997) Interferometric References evidence for the observation of ground backscatter originating behind André D, Sofko GJ, Baker K, MacDougall J (1998) SuperDARN interferometry: the CUTLASS coherent HF radars. Ann Geophys 15:29–39. https://doi. meteor echoes and electron densities from ground scatter. J Geophys org/10.1007/s00585-997-0029-y Res 103:7003–7015. https://doi.org/10.1029/97JA02923 Ponomarenko PV, Waters CL (2007) Spectral width of SuperDARN echoes: Beynon WJG, Williams PJS (1978) Incoherent scatter of radio waves measurement, use and physical interpretation. Ann Geophys 24:115–128. from the ionosphere. Rep Prog Phys 41:909. https://doi. https://doi.org/10.5194/angeo-24-115-2006 org/10.1088/0034-4885/41/6/003 Ponomarenko PV, Waters CL, Menk FW (2007) Factors determining spectral Bilitza D (2018) IRI the international standard for the ionosphere. Adv Radio Sci width of HF echoes from high latitudes. Ann Geophys 25:675–687. https 16:1–11. https://doi.org/10.5194/ars-16-1-2018 ://doi.org/10.5194/angeo-25-675-2007 Bilitza D, Altadiil D, Truhlik V, Shubin V, Galkin I, Reinisch B, Huang X (2017) Ponomarenko PV, Koustov AV, St-Maurice J-P, Wiid J (2011) Monitoring the International reference ionosphere 2016: from ionospheric climate to F-region peak electron density using HF backscatter interferometry. Geo- real-time weather predictions. Space Weather 15:418–429. https://doi. phys Res Lett 38:L21102. https://doi.org/10.1029/2011GL049675 org/10.1002/2016SW001593 Ponomarenko P, Nishitani N, Oinats AV, Tsuya T, St-Maurice J-P (2015) Applica- Bland EC, McDonald AJ, Larquier S, Devlin JC (2014) Determination of iono- tion of ground scatter returns for calibration of HF interferometry data. spheric parameters in real time using SuperDARN HF radars. J Geophys Earth Planets Space 67:138. https://doi.org/10.1186/s40623-015-0310-3 Res Space Phys 119:5830–5846. https://doi.org/10.1002/2014JA020076 Ponomarenko P, St-Maurice J-P, McWilliams KA (2018) Calibrating HF radar Chisham G (2018) Calibrating SuperDARN interferometers using meteor back- elevation angle measurements using E layer backscatter echoes. Radio scatter. Radio Sci 53:761–774. https://doi.org/10.1029/2017RS006492 Sci 53:1438–1449. https://doi.org/10.1029/2018RS006638 Davies K (1990) Ionospheric radio. Issue 31 of IEE electromagnetic waves Rawer K (2013) Wave propagation in the ionosphere, developments in electro- series. Peter Peregrinus Ltd., IEE, London magnetic theory and applications. Springer Science and Business Media, Ghezelbash M (2013) Occurrence and causes of F-region echoes for the Cana- New York dian PolarDARN/SuperDARN radars. M.Sc. thesis, University of Saskatch- Themens DR, Jayachandran PT, Nicolls MJ, MacDougall JW (2014) A top to ewan, Saskatoon, Canada. http://hdl.handle.net/10388/ETD-2013-03-949 bottom evaluation of IRI 2007 within the polar cap. J Geophys Res Space Gillies RG, Hussey GC, Sofko GJ, McWilliams KA, Fiori RAD, Ponomarenko P, Phys 119:6689–6703. https://doi.org/10.1002/2014JA020052 St-Maurice J-P (2009) Improvement of SuperDARN velocity measure- Themens DR, Jayachandran PT, Galkin I, Hall C (2017) The empirical canadian ments by estimating the index of refraction in the scattering region high arctic ionospheric model (E-CHAIM): NmF2 and hmF2. J Geophys Res using interferometry. J Geophys Res Space Phys 114:A07305. https://doi. Space Phys 122:9015–9031. https://doi.org/10.1002/2017JA024398 org/10.1029/2008JA013967 Themens DR, Jayachandran PT, McCafrey AM, Reid B, Varney RH (2019) A Gillies RG, van Eyken A, Spanswick E, Nicolls MJ, Kelly J, Grefen M, Knudsen D, bottomside parameterization for the Empirical canadian high arc- Connors M, Schutser M, Valentic T, Malone M, Buonocore J, St-Maurice J-P, tic ionospheric model (E-CHAIM). Radio Sci 54:397–414. https://doi. Donovan E (2016) First observations from the RISR-C incoherent scatter org/10.1029/2018RS006748 radar. Radio Sci 51:1645–1659. https://doi.org/10.1002/2016RS006062 Tindale E, Chapman SC (2017) Solar wind plasma parameter variability across Greenwald RA, Frissell N, de Larquier S (2017) The importance of elevation solar cycles 23 and 24: from turbulence to extremes. J Geophys Res Space angle measurements in HF radar investigations of the ionosphere. Radio Phys 122:9824–9840. https://doi.org/10.1002/2017JA024412 Sci 52:305–320. https://doi.org/10.1002/2016RS006186 Yeoman TK, Chisham G, Baddeley LJ, Dhillon RS, Karhunen TJ, Robinson TR, Hunsucker RD (1991) Radio techniques for probing the terrestrial ionosphere. Senior A, Wright DM (2008) Mapping ionospheric backscatter measured Springer, Berlin by the SuperDARN HF radars—Part 2: assessing SuperDARN virtual Jayachandran PT, Langley RB, MacDougall JW, Mushini SC, Pokhotelov D, height models. Ann Geophys 26:843–852. https://doi.org/10.5194/angeo Hamza A, Mann IR, Milling DK, Kale ZC, Chadwick R, Kelly T, Danskin DW, -26-843-2008 Carrano CS (2009) Canadian high arctic ionospheric network (CHAIN). Radio Sci 44:RS0A03. https://doi.org/10.1029/2008rs004046 Koustov AV, Ullrich S, Ponomarenko PV, Nishitani N, Marcucci MF, Bristow Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in pub- WA (2019) Occurrence of F region echoes for the polar cap SuperD- lished maps and institutional afliations. ARN radars. Earth Planets Space 71:112. https://doi.org/10.1186/s4062 3-019-1092-9 Kutiev I, Tsagouri I, Perrone L, Pancheva D, Mukhtarov P, Mikhailov A, Lastovicka J, Jakowski J, Buresova D, Blanch E, Andonov B, Altadill D, Magdaleno