Simultaneous HF measurements of E- and F-region Doppler velocities at large flow angles

Data collected by the CUTLASS Finland HF radar are used to illustrate the significant difference between the cosine component of the plasma convection in the F-region and the Doppler velocity of the E-region coherent echoes observed at large flow angles. We show that the E-region velocity is 5 times smaller in magnitude and rotated by 30 clockwise with respect to convection in the F-region. Also, measurements at flow angles larger than 90 exhibit a completely new feature: Doppler velocity increase with the expected aspect angle and spatial anticorrelation with the backscatter power. By considering DMSP drift-meter mea- surements we argue that the difference between F- and E- region velocities cannot be interpreted in terms of the con- vection change with latitude. The observed features in the ve- locity of the E-region echoes can be explained by taking into account the ion drift contribution to the irregularity phase ve- locity as predicted by the linear fluid theory.


Introduction
The auroral ionosphere is filled with the magnetic-fieldaligned irregularities that can be detected by coherent radars (Fejer and Kelley, 1980;Haldoupis, 1989;Sahr and Fejer, 1996;Schlegel, 1996). At the F-region heights (above 130 km) these irregularities are believed to move with the velocity of plasma convection, V 0 =E×B/B 2 . Thus, the velocity of irregularity motion in the F-region is a measure of the electric field applied to the ionosphere. At the E-region heights of 100-120 km the irregularity velocity depends on several parameters; among the most important ones are the electric field magnitude E and the angles that the irregularity propagation vector k makes with the electron background drift V e0 (flow angle θ ) and magnetic field B (the complementary off-perpendicular angle is called the aspect angle α).
Correspondence to: R. A. Makarevitch (r.makarevitch@lancaster.ac.uk) In the E-region, the electron background drift V e0 to a good approximation equals the convection velocity in the F-region, V e0 ∼ =V 0 , since the electric field does not change significantly along the highly conductive magnetic field lines.
A number of recent studies successfully exploited the described above electrical link between the F and E-regions, to study the velocity of the E-region irregularities as a function of the ionospheric electric field. In the UHF band, Foster and Erickson (2000), in a unique subauroral experiment, studied the E-region Doppler velocity detected with the side lobe of the incoherent radar at Millstone Hill (440 MHz) in conjunction with the F-region electric field data from the radar's main lobe on the same magnetic L shell. These authors discovered an almost perfect linear relationship between the Eregion Doppler velocity and the electric field magnitude. In the VHF band, a comparison between F-region drifts observed by the European Incoherent Scatter (EISCAT) radar and the Doppler velocities measured by the Scandinavian Twin Auroral Radar Experiment (STARE) VHF radars (140 MHz) showed that the irregularity phase velocity for directions close to V e0 (θ = 0 • −60 • ) is significantly smaller than the projection of V e0 onto the line-of-sight (l-o-s), V los =V e0 cos θ , while for larger flow angles θ = 60 • -90 • , the Doppler velocity generally agrees with the electron drift velocity projection, implying the cosine rule for the Doppler velocity (Nielsen and Schlegel, 1985;Nielsen et al., 2002). However, Koustov et al. (2002) demonstrated that the cosine rule can be violated for the evening sector observations of the STARE Finland radar. Later, Uspensky et al. (2003) made a more general conclusion that for large drifts the cosine law can be used only as a first approximation even at large flow angles, because the ion motions contribute substantially to the irregularity phase velocity. In the HF band, Davies et al. (1999) reported smaller velocities of E-region decameter irregularities observed by the Co-ordinated UK Twin Located Auroral Sounding System (CUTLASS) Finland radar as compared to the F-region plasma drifts measured by the EISCAT radar.  Simultaneous observations of E-region velocities at HF and F-region plasma drifts are difficult to perform since currently available HF radars are positioned too far equatorward from the incoherent radar locations. In this respect, an interesting approach was undertaken by Milan and Lester (1998), who compared the CUTLASS Iceland E-region Doppler velocities with the F-region drifts measured by the same radar 100-200 km poleward and eastward of the E-region scatter area. The authors found that the average E-region velocity, at small flow angles was well below the l-o-s component of the F-region velocity though still correlated with it.
In this study we expand the approach of Milan and Lester (1998) with the goal to explore the relationship between the E-region HF Doppler velocity and the plasma drift in the Fregion at large flow angles. Earlier studies by Makarevitch et al. (2002a, b) and Milan et al. (2003) have indicated that the cosine law might not be appropriate at HF even at large flow angles, but the results were not well substantiated in the sense that no reliable convection measurements were available. In this study we use F-region convection measurements by the CUTLASS Finland HF radar supported by the Defence Meteorological Satellite Program (DMSP) data on the ion drift velocities in the upper F-region and compare them with the E-region Doppler velocity measurements.

Observations
The CUTLASS Finland HF radar (62.3 • N, 26.6 • E), together with the Iceland HF radar, forms the easternmost part of the Super Dual Auroral Radar Network (SuperDARN) system of paired coherent HF radars and is designed to monitor the large-scale convection patterns at F-region heights in the high-latitude ionosphere (Greenwald et al., 1995). The radar is agile in frequency (8-20 MHz) and measures the Doppler velocity, backscatter power and width of ionospheric echoes in 45-km steps (180-3500 km) for each of the 16 radar beam positions separated by 3.24 • in azimuth. Beam 0 (15) of the radar corresponds to the westernmost (easternmost) direction.
We concentrate in this study on a four-hour interval between 13:30 and 17:30 UT on 31 March 2000. During this event, from 13:30 UT to 16:30 UT, the radar observed a stable band of E-region echoes at slant ranges r=350-750 km. After 16:30 UT the echoes became weaker, the band narrowed (r=400-600 km) and eventually disappeared. During almost entire period (13:40-17:10 UT), the radar observed simultaneously F-region echoes at farther ranges r>750 km. Co-existence of the E-and F-region echoes, even in spatially separated areas, provided an opportunity for velocity comparisons in the E-and F-regions, because DMSP measurements during the period under study did not indicate significant latitudinal variation of the convection intensity, as shown below.
Throughout the period the IMF was very stable and southward, B z =−3 nT; B y =−5 nT; the K p index was also stable and around 3. The level of ionospheric absorption and magnetic perturbations as measured by the IRIS imaging riometer at Kilpisjarvi (69.1 • N, 20.8 • E) and magnetometers of the IMAGE and SAMNET networks of high-resolution fluxgate magnetometers within the radar near FoV, was low, 0.1-0.2 dB and 100-250 nT, respectively, and exhibited little variation with time, indicating that the ionosphere was in a quite stable state. Under the above conditions one would expect that the convection pattern consists of 2 global cells with the stable L-shell-aligned flow in the afternoon sector of the magnetosphere (MLT ∼ =UT+2 for the radar FoV). Indeed, the SuperDARN global convection maps (not presented here) based on the Ruohoniemi and Baker (1998) technique demonstrate the high degree of L-shell alignment of the flow for the interval under study and latitudes of interest. Figure 1 shows the location of the Finland radar and the radar FoV at near slant ranges r<1215 km. The PACE magnetic parallels or L shells (Baker and Wing, 1989) are indicated in Fig. 1 by black curves. In addition, Fig. 1 presents ion drift data for two DMSP passes over the radar near FoV during the interval under study. The vector length at each point of the path corresponds to the measured ion velocity perpendicular to the path. The scale for the ion velocity is indicated in the top right corner of the diagram. Data from an DMSP altitude of 810 km were projected to the E-region height of 110 km along the magnetic field line. The Doppler velocity averaged over a 10-min interval, 14:40-14:50 UT (corresponding to the first pass of DMSP), is indicated by the color according to the scheme given in the left bottom part of the diagram as a color bar. The velocity scale for ranges 180-765 km (765-1215 km) is indicated by the digits to the left (right) of the color bar. Cells filled with horizontal lines correspond to the positive velocities. The slant range marks of 270,360,495,630,765,900,1035, and 1215 km are indicated by the dashed circular lines. We also show, by yellow curve 0, the line of perfect aspect angle at 110-km altitude (α=0), assuming that the radar rays undergo refraction based on typical ionospheric profiles, as described by the model (Bilitza, 2001) and using a simple geometric optics approach (Uspensky et al., 1994). Curves 1 and 2 represent the zero aspect angle for the electron density reduced by 25% and 50%, respectively. Finally, the thin ragged solid black line close to the yellow curve 1 shows the location of the power maxima along each radar beam.
An important feature of the F-region echoes detected at farther ranges (r=765−1215 km) is their negative velocities up to −750 m/s in the western part of the FoV and positive velocities up to +500 m/s in the eastern part of the FoV. The change in the Doppler velocity sign occurs at beams 9-10, corresponding to the direction nearly perpendicular to magnetic L shells, which is indicative of the L-shellaligned nature of the flow at far ranges. At closer ranges (r=180−765 km), where E-region echoes were detected, no reversal of the velocity sign is observed. The E-region velocity magnitude is maximized for beam 0 (red area with V =200 m/s). It gradually decreases to 50 m/s (blue area) with the beam number increase being still negative for beam 15. The range location of the power maxima (the black ragged line in Fig. 1) follows closely the aspect angle curve 1. Since it is widely accepted that the echo power should be at maximum for the range of perfect aspect angle (Fejer and Kelley, 1980;Haldoupis, 1989;Sahr and Fejer, 1996), from hereafter we take the curve 1 from Fig. 1 as a model for the aspect angle within the radar FoV.
Curiously enough, in Fig. 1 at beams 13-15 the velocity magnitude is smaller at r=360−495 km, close to the range of the perfect aspect angle, as opposed to the velocity magnitudes at beams 0-5, where it is clearly at maximum at r ∼ = 495 km. To make this point clearer, we present the data of Fig. 1 using a different format. The solid thick (thin) curve in Fig. 2 shows the averaged (for the 14:40-14:50 UT interval) Doppler velocity (backscatter power) slant range profiles for beams 0, 9, and 15. For slant ranges 270-765 km the averaged L-shell angle φ, defined as the angle between the radar look direction and the L-shell direction in the western sec- tor (assumed direction of the plasma flow), is given in the right top corner of each panel. The scale for the backscatter power (Doppler velocity) is shown on the left (right) vertical axis. Also shown by a dotted line is the model aspect angle α as a function of slant range for each beam. The scale for aspect angle in degrees is given by small ticks ±10, ±5, 0 on the right axis. One can notice that for each radar lookdirection the average echo power has a clear maximum at a range r 0 corresponding closely to the range with perfect aspect angle, α(r 0 )=0. The velocity magnitude profile for beam 0 shows quite a similar slant range variation with maximum at r 0 . However, in beam 15 there is no maximum of velocity at r 0 , but distinct minimum. At intermediate beam 9 the velocity change with slant range is not strong, with weakly-pronounced local minimum and maximum at 540 and 630 km, respectively (both not at r 0 ).
To explore this effect in more detail we present the same data using a different approach developed by Makarevitch et al. (2002b). Figure 3 shows the Doppler velocity versus L-shell angle for all 16 radar beams and all 23 slant range bins for each measurement within the 10-min period under consideration. We coded the aspect angle for each point of E-region measurements by color and marked all F-region measurements by the black dots. The color scheme for the aspect angle is shown at the bottom right part of the diagram. One can notice right away that F-and E-region echoes exhibit strikingly different velocity variations with the L-shell angle. The F-region echoes are steadily increasing in velocity from negative to positive values with L-shell angle; the reversal of the velocity sign occurs at φ ∼ =90 • . The E-region echoes are also progressively less negative with φ increase, but no reversal in velocity sign occurs.
One of the most remarkable features of the E-region points is the presence of V-like structures. For example, the Doppler velocity first increases in magnitude for φ=51 • −53 • and then decreases for φ=53 • −56 • with the smaller aspect angle points at the bottom of V-structure. We interpret this feature as associated with the aspect angle attenuation of the phase velocity (Makarevitch et al., 2002b). Each V-structure is in fact the data from one of the radar beams. As the range and L-shell angle changes along the beam, so does the aspect angle, reaching at some range minimum where the phase velocity is expected to have maximum. The V-structures in Fig. 3 are thus just another form of the Fig. 2a presentation. For high-number beams, for example, beam 15, the inverse-Vor -structures are seen instead, with the good aspect angle points at the top of the -structures, which is rather unexpected but in agreement with the presentation of fitted two cosine law functions of the form V 0 cos(φ + φ 0 ) to the E-region data (considering only red crosses) and Fregion data (all black dots), shown in Fig. 3 by the thin and thick line, respectively. The fitted parameters, velocity V 0 and L-shell angle φ 0 , signify the flow velocity and deviation of the flow from the L-shell direction, respectively. One should note here that in the above cosine model the fitted velocity V 0 is assumed to be negative, simply to reflect predominantly negative Doppler velocities in our observations, so that a negative flow velocity V 0 at φ=0 implies westward (away from the radar) convecting plasma. One can notice that in the F-region the flow is indeed highly L-shell-aligned, φ 0 =−2 • . It is not true, however, for the E-region; the L-shell angle of the velocity reversal is shifted about 31 • to the east from the perpendicular direction φ=90 • , so that the azimuthal difference between directions of the flow in the F and E-regions is quite large, −2 • −(−31 • )=29 • , and so is the ratio between fitted velocities 1141/203 ∼ =5.5.
In Fig. 4 we further investigate the relationship between the E-region velocity and the F-region plasma drift by showing the fitted velocities V 0 (blue) and L-shell angles φ 0 (red) in the E (thin line) and F-regions (thick line) as a function of time for 10-min intervals during the entire period of interest, 13:40-17:10 UT. Also shown are the spans (minimum to maximum) of the DMSP ion drifts for the two passes over the radar FoV. Again, thick (thin) line represents measurements for the latitudes that correspond to the F-(E-) scatter region. One can estimate that the obtained ratio of ∼5 is very typical for the entire period under study.

Discussion
The fact that the E-region velocity is ∼5 times smaller than the F-region E×B drift velocity, together with a ∼30 • shift in the azimuth of velocity reversal in the E and F-regions, can be explained in two ways. One can think that since the F-and E-region scatter areas were separated spatially by 300-400 km in our observations (Fig. 1), it is possible that it is the electric field (that mainly determines the Doppler velocity) at latitudes of the E-region observations that was smaller in magnitude and rotated by some angle with respect to that at latitudes of the F-region observations. However, the DMSP measurements of the F-region ion drifts presented in Fig. 1 do not support this scenario, at least in terms of the convection magnitude. One can see that the DMSP ion velocity does not change much in a broad range of latitudes. There is perhaps a 20% reduction in the ion drift at latitudes of the E-region observations, but one cannot expect the decrease in the velocity magnitude of several times. It is impossible to say whether this reduction was due to slight convection rotation or a simple decrease in the magnitude. Because the DMSP drift (one component of the ion drift vector) does not change much from one point to another, we believe that no significant changes in the direction of the total ion drift vector were taking place. Thus, the convection magnitude at ranges of the E-region observations is comparable to the one measured at the ranges of F-region scatter, Figs. 1 and 4. In support of our conclusion, we would like to stress the fact that our convection estimates from the CUTLASS data are in good agreement with DMSP ion drifts at ranges of F-region observations, Fig. 4. The CUTLASS/DMSP comparisons for two passes give us confidence that the convection estimates for the ranges of E-region observations and other periods are also reliable.
The region of interest was also monitored by several stations of the IMAGE magnetometer network located within the CUTLASS near FoV. The IMAGE magnetometers measure the north (X), east (Y ), and vertical (Z) components of magnetic field perturbations with 10-s resolution. Figure 5 is the magnetogram of the IMAGE X-component ( X) for five IMAGE stations located along the CUTLASS beam 8. On the right axis we indicated the station three-letter abbreviations and geographic latitudes and longitudes. One can notice that the X variation with time is very similar for all five stations and that it is more or less stable, especially during the last two-thirds of the period under study. The X time variation also resembles the F-region flow intensity variation from Fig. 4, which was gradually decreasing towards the end of the period. The X magnitude somewhat increases with latitude; thus X∼150 nT (300 nT) at the most equatorward (poleward) station OUJ (SOR) during a 10-min interval, 14:40-14:50 UT, the data from which were featured in Figs. 1-3. This ∼2-fold increase with latitude in the level of magnetic perturbation could, in principle, imply the corresponding doubling of the electric field intensity at farther ranges of the CUTLASS FoV, which is somewhat a greater increase than that in the DMSP data, but still well below a factor of 5 for the typical ratio between F-and Eregion velocities observed. One should also keep in mind that the magnetic perturbations do not necessarily represent the "true" electric field variation, since the electrojet current intensity is also controlled by the plasma density distribution in the E-region.  One can conclude that significant reduction (perhaps up to 5 times) of the E-region velocity, as compared to the F-region plasma drift, is a real effect associated either with the plasma physics of decameter irregularity generation or with the specifics of the backscatter signal formation at HF. A similar conclusion has been made by Koustov et al. (2002), who found that E-region HF velocities can be comparable with the VHF STARE velocities which were observed at large aspect angles and thus, were strongly reduced as compared to the plasma convection. Milan and Lester (1998) and Foster and Erickson (2000) reported only a factor of 2 for the ratio between F-and E-region velocities, but these observations were performed along the flow. The velocity difference was interpreted in these two studies in terms of the ion acoustic saturation for the Doppler velocity of type 1 irregularities Schlegel, 1983, 1985;Robinson, 1986Robinson, , 1993Nielsen et al., 2002;St.-Maurice et al., 2003). In the present experiment, however, the observations were performed across the flow. For these directions, type 2 irregularities are expected to be seen and thus a new explanation is needed.
The fact that in Fig. 4 the fitted E-region velocity shows good temporal correlation with that in the F-region leads us to a conclusion that the E-region Doppler velocity is proportional to the F-region convection velocity, but strongly  . 6. Phase velocity versus L-shell angle. By the blue (green) line we show the fitted curve for F (E) region from Fig. 3. The predictions of the linear fluid theory calculated by assuming the ion drift V i0 that is ten times smaller and rotated by 90 • with respect to the electron drift V e0 (which is estimated from the Fregion velocity measurements, blue curve) and for anisotropy factors =0.1, 0.5, 2.0, 5.0, and 10.0 are shown by the red curves. depressed below it. The reason for this effect is not entirely clear, and we consider here two possibilities. Makarevitch et al. (2002b) and Milan et al. (2003) proposed that low-velocity HF echoes are coming from the bottom of the unstable E-region, where the Doppler velocity is smaller because of increased collisions with neutrals. Let us comment on how this hypothesis helps in understanding the data presented in this study.
According to the linear fluid theory of electrojet irregularities (e.g. Fejer and Kelley, 1980), the phase velocity at a direction of wave propagation vectork ≡ k/k is given by where the anisotropy factor (Sahr and Fejer, 1996) is a function of aspect angle α, collision frequencies of ions and electrons with neutrals (ν i , ν e ) and ion and electron gyrofrequencies ( i , e ): (2) The expressions for both electron and ion background drift velocity (V α0 , α=e, i) can be readily obtained from the zeroth order momentum equations (e.g. Schlegel and St.-Maurice, 1981): In the E-region (say, below 120 km), ν e e , ν i i , and which means that the ion drift velocity is much smaller than that of electrons and rotated with respect to it by ∼90 • .
We note that Eq. (1) predicts proportionality between the irregularity velocity and the electron plasma drift, since the first term dominates in a broad range of flow angles with the exception of observations close to the perpendicularity to V e0 . An increase in collision frequencies ν e , ν i results in the corresponding increase in the anisotropy factor and decrease in the irregularity phase velocity, because of the 1+ factor in the denominator. The calculations based on the formulas for collision frequencies by Schunk and Walker (1973) and Schunk and Nagy (1978) give ν i ∼1.5·10 4 s −1 , ν e ∼10 5 s −1 and hence, an anisotropy factor of (α=0)=ν i ν e /( i i )∼1 at an altitude of 95 km. Thus, we can estimate that this purely collisional depression of the Doppler velocity in the lower E-region (below 95 km) can be as large as 2 times, which can partially explain our observations.
However, this is still less than the reported factor 5. We think that another effect might be involved, namely the effect of echo reception from a range of altitudes with quite different aspect angles, as described recently by Uspensky et al. (2003). According to the model of Uspensky et al. (1994Uspensky et al. ( , 2003, auroral backscatter is always nonorthogonal since the purely orthogonal component, coming from just one height, constitutes only a fraction of all echo power. This means that observations at any spot of the ionosphere can be characterized by some finite effective aspect angle. An increase in effective aspect angle α results in the corresponding increase in the anisotropy factor , Eq. (2), and decrease in the irregularity phase velocity. Estimates for the STARE radars showed that the effective aspect angle can be in the range of 0.8 • -1.0 • . If one adopts effective aspect angles comparable to the ones expected for STARE, one can explain additional velocity attenuation by a factor of 2-3.
Another consequence of the large anisotropy factor is an increase in the ion motion V i0 contribution to the phase velocity and rotation of the direction of maximum phase velocity away from the electron flow V e0 . The latter effect, reported by Uspensky et al. (2003) as the 10 • -20 • azimuthal difference between F-and E-region velocity vectors seems to find some confirmation in the data of the present study. Indeed, Fig. 4 shows that while the F-region flow was more or less L-shell aligned (|φ 0 |<8 • ), the Eregion flow was not, with the typical azimuthal difference between the flows larger than 15 • (except of the last 30 min). Figure 6 shows the result of phase velocity calculations based on Eqs. (1)-(3) for different anisotropy factors (red lines). For these calculations we assumed that the ion drift is ten times smaller and rotated by 90 • clockwise with respect to the electron drift, which is typical for the central part of the E-region (105-110 km), ν i ∼ =1800 s −1 and V i0 /V e0 = i /ν i ∼ =1/10. The electron drift was estimated from the F-region velocity measurements; we show the fitted curves for the F-(blue) and E-regions (green) from Fig. 3. One can see that all theoretical red curves intersect at one point and, remarkably, this is also the point where fitted experimental curves intersect. The theoretical line for =5 agrees well with the fitted E-region velocity curve, suggesting that the anisotropy factor was indeed quite large. If so, one might wonder why exactly the anisotropy factor was so large. As we argued, the enhanced collision frequencies at the bottom of the E-region coupled with the effectively nonorthogonal scatter can be a reason, but in this scenario the ion drift would be too small (of the order of 1/100 V e0 only) to cause any significant shift in the E-region velocity reversal direction from 90 • of the flow angle (e.g. similar to those in Fig. 6). This is why in the reasoning above we assumed that the E-region echoes originated mainly from the electrojet center.
An important new result of this study is that for large flow angle observations (φ>90 • ), the E-region velocities were at minimum at a range r 0 where the echo power was at maximum and where the model aspect angle was around zero (Fig. 2c and right part of Fig. 3, where instead of V-structures, -structures were observed). This velocity decrease near r 0 is very unlikely to be caused by the decrease in the electric field intensity at these ranges, since the DMSP convection component is quite stable over this area, Fig. 1. To some extent, this result reminds us of observations of Makarevitch et al. (2002a), who reported the absence of the aspect angle attenuation for the Doppler velocity at large L-shell angles φ ∼ =90 • . Makarevitch et al. (2002a) explained their result in terms of the ion drift contribution to the irregularity phase velocity at large flow angles. We believe that a similar explanation can be applied to the present observations, as outlined below.
Equation (1) can be rewritten in terms of the flow angle [θ ≡ cos −1 (k · V e0 /kV e0 ] as One should note here that Eq. (6) is more appropriate for discussion of the aspect angle effects, since only the first term is dependent upon and hence, upon the aspect angle α.
The simplest formulation of the linear fluid theory that takes into account only the electron motions predicts that the velocity should change with θ as V e0 cos θ/(1+ ) and fall off to zero with the aspect angle α increase. If the ion motions are included into calculations, the situation changes. In this case the phase velocity at large flow and aspect angles will be determined mostly by the ion drift l-o-s component or the second term in Eq. (6), since the first term is reduced drastically because of the enhanced anisotropy factor at large aspect angles.
We illustrated this point in Fig. 7 which shows the results of the phase velocity calculations based on Eqs. (1)-(3). The phase velocity is shown in Fig. 7 as a function of the aspect angle for various flow angles. As before, the typical (for the central part of the E-region) collision frequencies ν i =1800 s −1 , ν e =15 000 s −1 were assumed, as well as the exact L-shell aligned electron flow, V e0 =1100 m/s. For each panel of Fig. 7, the L-shell angle is constant (we indicated this angle at the top, for example, φ=40 • for the leftmost panel), while the aspect angle is changing from −5 • to +5 • . Thus, each panel models radar observations in one beam position. Let us first have a look at the left part of the diagram (φ=40 • −70 • ). The phase velocity magnitude maximizes at zero aspect angle and decreases with the aspect angle increase. The velocity at zero aspect angle is approximately equal to the l-o-s component of the electron drift velocity, and we connected points with perfect aspect angle (shown by grey circles) with the grey thick line, which is essentially a simple "cosine law" curve V e0 cos θ , similar to the fitted F-region cosine curves in Figs. 3 and 6.
At large aspect angles, however, velocity is not zero but some finite value, which is determined by the second ion term in Eq. (6). For our observations, the latter can be easily estimated from Eq. (4); the electron drift is ∼1100 m/s and hence, the ion Pedersen drift is ∼110 m/s and directed northward (along the poleward electric field in the afternoon sector) that is away from the radar, providing a negative offset of ∼110 m/s in the E-region velocity, consistent with our observations. Importantly, the phase velocity calculations show that for certain flow angles (as for φ=90 • in Fig. 7), the phase velocity is less in magnitude at perfect aspect angle than at larger ones. This is simply because at these directions the ion motions become more significant at large aspect angles than the electron motions. This effect can explain the reported minimum in the high-number beam velocities at ranges near r 0 that correspond to the power maximum and to the minimum achievable aspect angles ( Fig. 2c and right part of Fig. 3).
One can also notice that although the velocity decrease with the aspect angle is seen in Fig. 7 for φ=90 • , the range of the L-shell angles with the -structures in Fig. 3 is much wider and extends from φ ∼ =85 • to 113 • . Another discrepancy between Figs. 3 and 7 is that the maximum Doppler velocities recorded were much smaller in magnitude than, for example, −500 m/s that should be observed near φ=60 • . Following Uspensky et al. (2003), we argued that this reduction can be related to the HF backscatter signal collection from the range of heights with the different finite aspect angles, which is equivalent to assigning some effective nonzero aspect angle to a specific range.
We can now estimate the effect of the aspect angle "finiteness" on the flow angle dependence using the presentation of Fig. 7. We indicated by vertical lines the aspect angle of −1 • , by black dots the theoretical velocities which correspond to this aspect angle at φ=60 • , 90 • , and 120 • , and connected these points by the thin black line. The phase velocity is greatly reduced, especially for directions away from φ=90 • . For example, for φ=60 • , it changes from ∼500 to 200 m/s. Another interesting feature is that the black curve intersects the zero velocity line to the right from the grey curve, meaning that the phase velocity should be reversed at flow angles of more than 90 • , consistent with our measurements. On the other hand, in our observations the E-region velocity reversal was not seen anywhere in the radar FoV, even at the easternmost beam 15 (φ∼110 • ), while according to Fig. 7 this reversal should occur somewhere between 90 • and 110 • . One should note here, however, that in the reasoning above we used a minimum aspect angle of −1 • , consistent with an estimate for the STARE radar (144 MHz) obtained by Uspensky et al. (2003), who adopted an aspect sensitivity of 10 dB per 1 • of aspect angle. Haldoupis (1989), by considering the publications on the aspect sensitivity at various radar frequencies, concluded that the 50-MHz E-region echoes are perhaps less aspect sensitive than those at 150 and 400 MHz, with the typical values in the range 1-3 dB/ • . Thus, if one assumes a slower rate of the power decrease with the aspect angle for observations at lower HF frequencies, say, 2 dB/ • reported recently by Makarevitch et al. (2002a), one can argue that the effective aspect angle at HF should be larger and hence, one can reach better agreement between observations and theory.
On a more critical note, from Eq. (5) the flow angle corresponding to a maximum of the phase velocity (or, in other words, the deviation of the flow from the E×B direction) is given by tan θ max = i /ν i and hence for fixed collision frequencies/altitude should not depend upon the flow intensity V e0 . Experimentally, Fig. 4 shows that this deviation tends to decrease in magnitude with a drift magnitude decrease, suggesting that perhaps other (than Pedersen motion of ions) factors can contribute to the total ion drift vector, for example, the neutral wind. The detailed discussion of this issue is, however, beyond the scope of the present paper. We would like only to point out that even a simple ion Pedersen drift interpretation explains reasonably well most of the Doppler velocity features observed in this study (small magnitudes, deviation from the E×B direction, and the spatial anticorrelation with the backscatter power).

Summary and conclusion
A comparison of the F-and E-region HF Doppler velocities observed by the CUTLASS Finland radar shows that the Eregion velocities are typically several times smaller and do not exhibit a change in their sign (within the radar FoV), as opposed to the F-region velocities. The E-region velocity variation with L-shell angle φ can be described by a shifted cosine function V 0 cos(φ+φ 0 ), with the typical values of V 0 = − 200 m/s, φ 0 =−30 • . For the high-number radar beams, the Doppler velocity is minimized at slant ranges that correspond to the power maxima and minimum aspect angle, as determined from the model. The observed difference between F-and E-region HF Doppler velocities, as well as the velocity magnitude increase with the aspect angle at φ>90 • , cannot be explained by the electric field magnitude and direction variation with latitude, but most likely is a result of the non-orthogonality of backscatter coupled with the ion motion contribution to the E-region irregularity phase velocity significant at large flow and aspect angles.