We investigate the fine-scale structure of the subauroral electric field in the vicinity of the polarization jet and subauroral ion drift (SAID) using coherent E region backscatter observed with the Millstone Hill 440 MHz UHF radar during a moderately disturbed period (Kp = 5) on 13 November 1998. We use a combination of data obtained at high time and spatial resolution and radar response modeling, incorporating magnetic aspect angle sensitivity and radar antenna beam shape. The modeled radar response function depends not only on the magnetic and radar pointing geometry but also on the altitude and extent of the irregularity layer. In addition to modeling of the radar system response, an optimal, regularized deconvolution technique is used to resolve fine-scale spatial structure from the relatively long pulse length used. At the 440 MHz Millstone radar frequency, coherent backscattered power is a linear function of electric field amplitude, and we are able to relate changes in the range variation of deconvolved power to spatial/temporal structure in the electric field. We observe the polarization jet to have superimposed multiple instances of narrow (0.1°) intense SAID configuration electric fields having lifetimes as short as 1.5 min and gradients on their equatorward edge as large as 4 mV/m per kilometer. Simultaneous DMSP satellite observations of westward ion drift across our experimental field of view confirms our interpretation of the radar data in terms of a subauroral polarization jet structure spanning over 3° latitude moving poleward at 250–400 m/s velocity (cross-L-shell) past the radar beam with superimposed intense SAID-like electric fields.
 A region of poleward directed electric field forms at subauroral latitudes in disturbed geomagnetic conditions. Strong sunward (westward) ionospheric convection accompanies this polarization jet [Galperin et al., 1974; Yeh et al., 1991], which is most pronounced in the premidnight sector. The polarization jet is roughly L-shell-aligned, 3°–5° in width, and is seen as a separate region of sunward plasma convection offset from the two-cell auroral convection pattern by 2°–10° latitude [Yeh et al., 1991]. The polarization jet electric field is of magnetospheric origin and is generated as freshly injected ring current ions penetrate more deeply into the inner magnetosphere than do the electrons. Narrow regions (<1°) of intense electric field (>50 mV/m) and westward ion drift (>1 km/s) are seen in this region and have been termed subauroral ion drifts (SAID) [Smiddy et al., 1977; Anderson et al., 1993]. Because of its proximity to Millstone Hill the polarization jet has been observed with the Millstone Hill incoherent scatter radar for many years, and a number of studies of its characteristics and space weather effects are underway. This paper introduces coherent backscatter analysis as an ionospheric diagnostic technique of considerable use for high-temporal/spatial-resolution polarization jet studies.
 The Millstone Hill 440 MHz UHF radar system, when directed toward the north, is sensitive not only to weak incoherent scatter returns from the ambient ionosphere but also to intense echoes from 34 cm irregularities at aspect angles near perpendicular to the geomagnetic field [St.-Maurice et al., 1989; del Pozo et al., 1993]. The experimental geometry most favorable to reception of these echoes occurs when low elevation angles (4°–10°) are used, creating near-zero magnetic aspect angles approximately 5°–8° poleward of Millstone's subauroral location (288.5°E geodetic longitude, 42.6°N geodetic latitude, 55Λ). During geomagnetically disturbed periods, ambient E region electric fields can exceed 15–20 mV/m, and subsequent electron drift speeds can become greater than the local ion acoustic speed, leading to the creation of two-stream Farley-Buneman instabilities [Farley, 1963; Buneman, 1963], which produce the irregularities responsible for auroral radar backscatter. A comprehensive review of the current understanding of E region irregularity and radar backscatter production was recently presented by Sahr and Fejer . Millstone's high power (2–5 MW peak), narrow beam (1.2° full width at half maximum), and receiver sensitivity down to incoherent scatter levels (−170 dBW) give it a unique capability for wide dynamic range, high-resolution E region backscatter measurements. Several previous studies using Millstone data have probed the relationship between E region backscatter cross section, phase velocity, and, more recently, electric field strength [St.-Maurice et al., 1989; Foster and Tetenbaum, 1991, 1992; Foster and Erickson, 2000].
 In this paper, we use Millstone observations during a 13 November 1998 high-resolution experiment in a new way which exploits the use of E region coherent backscatter measurements as an ionospheric diagnostic. Specifically, we will show that radar backscatter cross-section values, in combination with modeling of aspect angle sensitivity and other experimental parameters, can yield detailed information on the temporal and spatial variations of subauroral disturbance electric fields associated with the polarization jet. We find that the equatorward edge of the polarization jet electric field becomes highly structured on ∼10 km spatial scales, leading to the formation of intense, discrete regions of Farley-Buneman irregularities and patches of auroral radar backscatter. Figure 3 of Foster and Erickson  shows that both log power and coherent phase velocity are directly proportional to electric field strength at 440 MHz when the driving electric field exceeds 15–20 mV/m, and we use this relationship to interpret the spatial/temporal variations of the coherent scatter parameters in terms of the variability of the electric field in the region of the polarization jet. The Millstone Hill radar is a particularly sensitive and effective system for conducting these types of studies.
2. Experimental Technique
 Real-time monitoring of interplanetary conditions which drive magnetospheric activity enables us to efficiently use the Millstone Hill system for studies of disturbance-related phenomena. In particular, the Millstone radar was activated from 1800 UT on 13 November 1998 to just before 0000 UT (1300–1900 magnetic local time (MLT)) on 14 November, in response to a strong postnoon convection bay. Interplanetary magnetic field conditions as observed by the Advanced Composition Explorer (ACE) satellite prompted the commencement of operations, with Bz steadily negative (−15 to −20 nT) from 1400 to 0000 UT on that day.
 Millstone is most sensitive to E region echoes at ranges where the radar beam is approximately perpendicular to the local magnetic field (i.e., zero aspect angle) at 110 km altitude [Foster and Tetenbaum, 1991]. Using an elevation angle of 6° to observe conditions poleward of the radar, this perpendicularity criterion is met at two azimuths to the northwest and northeast. Magnetic aspect angle increases rapidly with azimuth away from these points, and radar sensitivity is sharply reduced. For the experiment presented here, the 46 m steerable antenna was directed to the northwest at an azimuth of 300°, equatorward of the region of maximum sensitivity. Geomagnetic disturbance was moderate, with a daily planetary A index of 60, a planetary K index of 5 at 1800–2100 UT and the midlatitude Fredericksburg K index at 4. The Polar satellite visible and infrared imager showed evidence of enhanced substorm activity from 2000 MLT to past 0400 MLT during 1800–2100 UT.
 The transmitted pulse used was a 2000 μs long single pulse, selected to provide adequate SNR for F region incoherent scatter (to determine ambient electric fields) simultaneously with strong E region coherent echoes, in a manner similar to the 27 August 1998 experiment described by Foster and Erickson . However, the 300° azimuth used in the November 1998 experiments described here resulted in a different radar sensitivity pattern (see section 3). Furthermore, the 300 km spatial extent of the transmitted pulse caused considerable smearing in range of the E region echoes, making a careful deconvolved analysis of backscattered power profiles necessary to determine fine-scale coherent scatter features (see section 4).
 A radar data acquisition system optimized for raw data taking (VAMPIRE) recorded complex in-phase and quadrature receiver voltages with a 20 μs sampling period (3 km range spacing), allowing determination of backscatter power at any desired time integration down to individual pulses, selectable after the experiment. A total of 10 Gb of raw receiver voltage data was collected using the VAMPIRE system. The high-resolution data were analyzed by computing power profiles for each interpulse period (IPP) recorded and postintegrating for 1 s. These values were then deconvolved to produce final high-resolution (3 km) power profiles (see section 4).
3. Modeling of Coherent Echo Sensitivity
 Since the backscatter radar cross section for E region irregularities is large, the Millstone antenna receives echoes not only from irregularities in the main beam but also from off-beam (sidelobe) returns, which appear at the same range. The 46 m diameter antenna has 42 dB of gain with a 1.2° full width at half maximum beam width, and the amplitude of the sidelobe returns is reduced by the two-way off-beam antenna factor. However, in certain circumstances, sidelobe returns can dominate main beam incoherent scatter. Figure 1 presents a schematic view of the geometry involved in a coherent/incoherent scatter experiment, including the effects of sidelobe coherent contamination. Coherent two-stream irregularities are confined to a limited altitude extent near 110 km and only occur in regions where the electric field strength in the E region exceeds the Farley-Buneman irregularity threshold. For a uniform layer of E region irregularities confined to a narrow altitude, intense scatter occurs at those ranges where the main beam penetrates the E region layer, while strong sidelobe echoes occur at further ranges where the unobstructed radar line of sight intersects the layer at a favorable magnetic aspect angle. The sidelobe echoes at further ranges are reduced in magnitude by the two-way off-beam antenna attenuation factor and the magnetic aspect angle sensitivity. Sidelobe coherent echo contamination is confined to ranges less than 1300 km for the Millstone system, as beyond this point the limb of the Earth shields the antenna from direct observation of the irregularity layer.
 The absolute backscatter cross section for 440 MHz coherent returns can be up to 90 dB greater than that for incoherent scatter [Foster and Tetenbaum, 1992]. Because of this, we cannot assume that coherent returns received at a given range originate solely within the main beam of the antenna. In order to interpret the observed range returns correctly, we use a model which simulates the radar range-intensity profile that would be observed from an assumed irregularity latitude and altitude layer structure, on the basis of radar pointing azimuth and elevation, the 46 m antenna beam pattern, and magnetic aspect angle sensitivity. Such a model has been described by Foster and Tetenbaum  and has been used for several studies of coherent echoes [Foster et al., 1992; Foster and Erickson, 2000]. In the current version of the model we integrate over the antenna beam pattern out to 10° from the main beam pointing direction and determine the contribution to the total power at a given range from every point at which the constant-range surface intersects the irregularity layer. A 440 MHz magnetic aspect sensitivity of −15 dB/deg is used for aspect angles in the range 0°–3°, and −10 dB/deg is used for aspect angles >3° [Foster et al., 1992]. The center altitude, altitude distribution, and latitude extent of the assumed irregularity layer can be varied, providing an estimate of the radar range response for specific geophysical conditions.
 We assume that the irregularity layers are longitudinally aligned. Furthermore, for the azimuths relevant to this study, the radar response is assumed to be invariant to changes in flow angle (i.e., the angle between the radar line of sight and the E × B direction). Assuming that flow occurs along L shells, flow angle for the E region experiment geometry under study varies from 25° to 45°, and prior studies have suggested that coherent backscatter cross section is nearly invariant (to within 2–3 dB) over this range of angles at UHF frequencies [Haldoupis et al., 1990].
 Relevant to this study is the difference in response as the altitude and extent of the E region layer is varied. The left plot of Figure 2 shows the modeled Millstone response as a function of range over 30 dB of power variation for a latitudinally uniform layer (3 km Gaussian in altitude) at 110 and 116 km, respectively. At the pointing position of 300° azimuth and 6° elevation used in this study, the received response is sharply peaked at between 670 and 720 km, depending on layer height, and is down by over 30 dB by 800 km range. This corresponds to a mapped spatial sensitivity confined to a longitudinally aligned spot centered at (45.4°N, 281°E) geodetic and approximately 1° wide in latitude and 2° wide in longitude. By contrast, the right-hand plot of Figure 2 shows the modeled response for latitudinally thin scattering regions (0.15° latitude, 2 km Gaussian altitude) at 110 km altitude located at 45.2° and 45.5°N latitude, respectively. The response curve is much sharper than for the latitudinally uniform layers, and the relative peak amplitudes follow an envelope dictated by the latitudinally uniform layer response curve. These results show that coherent backscatter layers, within the narrow spatial sensitivity dictated by the antenna pointing position chosen, can be differentiated in type between latitudinally thin and uniform scattering regions. However, accurate spatial localization of the irregularities requires an a priori assumption about mean layer altitude, since the overall uniform layer response is sensitive to that parameter.
 For the fixed antenna position used during this experiment, a map, spanning 55 dB dynamic range, of the sensitivity of the radar to backscatter from coherent irregularities is presented in Figure 3. Lines of both constant range (solid lines) and invariant latitude (dashed lines) are shown, an altitude of 112 km is assumed for the center of the irregularity layer, and the fine structure of a short transmitted pulse is used. At ranges between 600 km and 800 km the dominant sensitivity is controlled by the antenna function, and scattering sensitivity is concentrated in the main radar beam direction. Beyond 1000 km range, however, the strongest signals originate from antenna sidelobes at the latitude where the radar line of sight is approximately perpendicular to the magnetic field at E region heights. We will concentrate in this study on the radar response in the central sensitivity region within the 600–800 km range.
4. Backscattered Power Deconvolution Technique
 To make use of the fine resolution backscattered power determined by the VAMPIRE system, the narrow range sensitivity results of section 3 combined with the large transmitted pulse size make deconvolution essential for determining size and distribution of the E region scattering layers. We have sufficient information to accomplish this since we know the range extent of the transmitted pulse (i.e., the convolving function shape) and the VAMPIRE received power values oversample the pulse length by a factor of 100. However, deconvolution is by nature quite sensitive to measurement noise [Press et al., 1992], and therefore it is important to use a technique which can incorporate as many a priori constraints on the results as possible.
 The deconvolution problem we wish to solve can be written as
where s(r,t) is the desired deconvolved power value at range r and time t, P(r) represents the range-smearing effects of the transmitted pulse and receiver response, n(r, t) is added measurement noise power, and m(r, t) is the resulting measured power. Fourier-transform-based techniques for optimally solving this equation to produce s(r, t) have been described in the literature [Nussbaumer, 1982; Tolmieri et al., 1997]. However, use of linear inversion techniques [e.g., Menke, 1989] provides an easy way to perform the deconvolution while incorporating any desired constraints or a priori knowledge. For the deconvolutions performed in this paper, we use the GULIPS general purpose linear inversion matrix package [Lehtinen et al., 1998] within the MATLAB analysis environment. To further limit noise arising from numerical instabilities inherent in deconvolution, we use regularization techniques to gain stability. In particular, we can enhance the stability of the result in two ways.
 First, we can apply a constraint equation to limit the point-to-point variation of the final signal; this takes the form
where si ≡ s(ri, t) and i indexes over the number of range samples available. In practice, the equation is combined with the main deconvolution equation (1) using a regularization variance which can be adjusted to control the allowed size of peaks in the final solution. For the data presented in this paper, we selected a variance value of 1 as producing reasonable results, with values in the range 0.5–2 producing nearly identical solutions.
 Second, since the final power values si must be nonnegative, we incorporate this a priori knowledge through a second constraint equation
In practice, this procedure is done iteratively. At each iteration step, negative power values are constrained to be zero using the above equation, and the linear inversion process is repeated. This continues until 12 iterations have been performed or until the current solution has not significantly changed in a least squares sense.
Figure 4 presents an example of the deconvolution algorithm results for a 1 s average power profile from 1947:51 UT during the 13 November experiment. The left-hand profile shows the measured power profile with 3 km range spacing, while the right-hand profile plots the corresponding optimally deconvolved power profile. The deconvolution results reveal that the large, nearly flat, distributed range return from 730 km to beyond 1000 km is created primarily by a compact layer of coherent backscatter at 730 km convolved with the 300 km long pulse used. (The very narrow, weak “layer” appearing beyond 800 km is a residual deconvolution artifact rather than an actual scattering region, since the model of Figure 2 for the experiment pointing geometry predicts very weak sensitivity to coherent backscatter at that range.)
5.1. Backscattered Power
 The E region irregularities observed during the 13 November 1998 late afternoon geomagnetically disturbed period presented here were highly variable in both space and time. Figure 5 presents a range-time-intensity (RTI) image of coherent E region backscatter power over 50 dB dynamic range during the period from 1920 to 1955 UT, obtained from 1 s resolution VAMPIRE power profiles using the deconvolution techniques of section 4. The range span of 550–850 km plotted is centered around the region of maximum coherent echo sensitivity at the fixed azimuth of 300° (compare Figure 2). The relative echo strength varies widely and is strong enough from 1930 to 1938 UT that it suppresses the normal background noise seen above and below the central range region due to receiver dynamic range limitations.
 Initially, the irregularity power increases slowly until 1926 UT, when a very strong diffuse scatter develops, filling the entire main sensitivity region between 1930 and 1935 UT. This rapidly evolves in under 1 min to narrow (10–30 km range extent) discrete coherent backscatter regions. The discrete regions appear at close range and move away from the radar along the sensitivity curves in Figure 3 until they disappear at approximately 780 km range as the scatterer locations pass out of the main radar sensitivity region. At times, two to three discrete regions remaining visible from 1 to 4 min can be identified simultaneously (e.g., 1943 UT), with up to 30 dB of dynamic power variation from appearance to fade-out. (Because of the focused main portion of the radar sensitivity pattern the observed maximum lifetime of 4 min must be considered a lower bound.)
 Whether diffuse or localized, the backscatter signatures are confined to follow the radar sensitivity envelope as they appear and move poleward, away from the radar. Figure 6 plots the log of backscattered power versus range over nearly 20 dB of dynamic range for the irregularities seen from 1939 to 1941:30 UT, when narrow features were observed. We also plot as a dashed line the modeled sensitivity curve for a uniform E region layer at 116 km altitude and a solid line representing the modeled response to a latitudinally thin 116 km layer 0.15° wide centered at 42.5°N latitude. Both the modeled layer altitude and the latitudinal extent of the thin scattering region were selected a priori as best representing the data. As expected, the irregularity power always remains within the modeled sensitivity curve, and the range extent of backscatter is clearly narrower than the pattern which would be produced by a uniform irregularity layer (dashed line).
 Similar results were obtained for the rest of the poleward moving discrete scattering regions observed in Figure 5 between 1937 and 1950 UT, and Table 1 summarizes their characteristics, sorted by total amount of discrete region motion. Each discrete region is corrected using the modeled response curve of section 3 (which assumes longitudinally extended features) with identical input parameters, with the exception of mean irregularity layer altitude. For each discrete feature observed, we list the maximum corrected log power, center invariant latitude, and invariant latitude motion seen from appearance to fade-out in both degrees covered and meters per second (assuming the motion is entirely across L shells). As before (e.g., Figure 2), the irregularity layer shape is modeled as a Gaussian 2 km thick in altitude. The 2 km altitude thickness was selected a priori as best matching the observational data, since thicker modeled layers gave too broad a response for more diffuse scattering events (e.g., 1922–1927 UT in Figure 5).
Table 1. Discrete Coherent Backscatter Region Statisticsa
116 km Layer: Max log Power / Center InvLat, deg
InvLat Span, deg
Poleward Motion, m/s
InvLat, invariant latitude.
13.19 / 56.3
11.61 / 56.6
10.82 / 56.8
11.60 / 56.5
 A mean layer altitude of 116 km (compare Figure 6) is somewhat higher than that used in typical models of E region irregularity backscatter [Foster and Tetenbaum, 1991]. However, the actual E region irregularity mean altitude is an uncertain parameter, as it is an a priori input for the modeled response from section 3. In particular, at the 6° elevation angle used in this experiment, the Millstone antenna beam may be subject to variable amounts of tropospheric refraction. This can be responsible for up to 0.5° of downward deflection, resulting in an effective increase of 5 km or more in the apparent irregularity layer altitude [Foster and Tetenbaum, 1991]. A change in apparent altitude can also occur if the discrete scattering region with the largest contribution to total received power, taking into consideration magnetic aspect angle, lies poleward of the main beam [Foster et al., 1992].
 Latitude motion of the discrete scattering regions ranges from very localized (0.20°; region disappears temporally) to an upper bound of 0.66°, which covers the entire main sensitivity curve of the Millstone antenna. Finally, within the radar sensitivity region, the poleward motion of the four individual scattering regions (assuming exclusively cross-L-shell velocity) is quite similar, ranging from 250 to 430 m/s.
5.2. Electric Field Gradients
Foster and Erickson  have shown that both the log of backscattered power and the coherent phase velocity vph are directly proportional to ∣E∣ at 440 MHz. Although not shown here, a determination of simultaneous F region incoherent scatter measurements of the E × B drift velocity (and hence ∣E∣) at ∼60Λ and E region coherent backscattered power observed in the zero-aspect angle sidelobe at 1150 km range plotted in Figure 2 has the same relative slope value as that presented by Foster and Erickson . We conclude therefore that use of the slope values presented in that work is fully valid in interpreting the November 1998 observations presented here.
 The power/electric field relationships enable us to examine the spatial/temporal variations seen in backscattered power in terms of the structure of the driving ionospheric electric field. Coherent backscatter parameters depend on the total electric field E* = E° − vN × B, and we have no measure of vN which could be significant in a region of fast ion motion and intense driving electric field [Codrescu et al., 1995]. However, relative variations in electric field (whether E° or E*) are accurately determined by our technique. In particular, the sharp onset of coherent backscatter as a function of range during the period of time associated with poleward moving discrete regions, along with the 3 km range spacing of the VAMPIRE observations, allows determination of the electric field gradient associated with the equatorward edge of these discrete-scattering regions. Figure 7 presents raw (i.e., not deconvolved) VAMPIRE power versus range at three times when discrete scattering regions were observed. We use the raw rather than deconvolved power since the raw power does not contain additional noise introduced by deconvolution and the discrete regions are compact enough in range that they do not significantly smear out the leading edge of the raw power. We observe gradients averaging 20 dB per 10 km range. Foster and Erickson  have found that a 10 mV/m change in ∣E∣ results in a 5 dB change in coherent power. Thus the observations of Figure 7 indicate very large spatial electric field gradients of up to 4 mV/m per kilometer and a 40 mV/m total electric field increase associated with the equatorward edges of fine-scale polarization jet structure.
5.3. Comparison With DMSP Satellite Observations
 Since the radar observes only in a single line of sight during this experiment, it is difficult to determine whether the structures seen in Figure 5 are indicative of temporal variations (i.e., scattering which appears and disappears on 1–4 min timescales), spatial variations, or a combination of the two. To gain further insight, we compare our radar observations with simultaneous data obtained by the ion drift meter (IDM) instrument on the Defense Meteorological Satellite Platform (DMSP) satellites F13 and F14. The IDM is sensitive to localized intensifications of the cross-track (westward) velocities, driven by poleward electric fields, at its 850 km orbital altitude. In particular, during this experiment, the DMSP F13 and F14 satellites were making an ascending pass through the Northern Hemisphere eastward of Millstone Hill at 1800 MLT (F13) and 2100 MLT (F14).
Figure 8 plots a comparison of DMSP F14 westward ion drifts from the IDM (lower plot) during a ascending pass eastward of Millstone Hill at 1946:05–1947:14 UT (2042 MLT; 3.5°–1.4° geodetic longitude) with the relative electric field (upper plot), derived from VAMPIRE measurements, during the 1920–1955 UT period imaged in Figure 5. We derive the upper plot by determining at each time the maximum backscattered power from 625–750 km range (which incorporates both diffuse and discrete scattering configurations) and then converting to an electric field strength relative to the UHF coherent irregularity threshold of 15–20 mV/m, again using the Foster and Erickson  relation of 10 mV/m change in ∣E∣ per 5 dB change in coherent backscatter power. By plotting this parameter reversed in time, we yield a curve in reasonable agreement with the simultaneous DMSP westward velocity (poleward electric field) measurement. This result strongly suggests that the structures seen moving poleward in the RTI plot in Figure 5 are, in fact, mostly spatial (latitudinal) variations in the electric field and that the entire structure (approximately 3° wide) is to first order moving poleward past the fixed radar beam in a “slit-camera” fashion.
 The Millstone Hill measurements, however, reveal much more detail about the process than the simple DMSP snapshot can provide. This is evidenced by the radar's capability to determine not only the electric field gradient on the equatorward edges of the more intense scattering regions but also the temporal variation of fine-scale structure as some features, e.g., 1937 UT on Figure 5, appear and fade within the radar field of view. We also are able to observe the entire meridional electric field structure move northward at approximately 250–400 m/s cross-L-shell, considering the range rates of the discrete scattering regions; compare to Table 1. At this rate the entire electric field structure moves poleward in the 20–25 min observed by the radar by approximately 3° invariant latitude, in excellent agreement with the DMSP observation in Figure 8.
 The overall polarization jet structure observed and reported here extends over more than 3° of invariant latitude with a maximum electric field amplitude ≥80 mV/m, similar to earlier ground-based studies [Yeh et al., 1991]. Both the DMSP F13 (at 1812 MLT; not shown) and F14 (compare to Figure 8) platforms measured westward drifts in excess of 2500 m/s at 53–55 Λ, equatorward of significant ion and electron precipitation and associated with a deep, narrow convection-induced ionospheric trough. The narrow (∼0.1°) latitude span of the SAID-like velocities which appear superimposed on the polarization jet, and which are confined to the equatorward side of the enhanced subauroral convection, have been seen as well in the DE 2 observations of Anderson et al. [1991, 1993] along with earlier satellite studies referenced therein. We also agree well with Karlsson et al.'s  statistical survey of subauroral electric fields from the FREJA satellite, which finds half widths of SAID velocity channels to range from 0.1° to 0.4° for driving electric fields <100 mV/m, corresponding to total potential drops of 1–10 kV. We note that the SAID drifts exceeding 3 km/s (>120 mV/m) in the DMSP data of Figure 8 at 53 and 53.7 Λ are not reproduced in the Millstone Hill observations, which level out at ∼80 mV/m. However, the relationship between coherent backscatter power and ambient electric field of Foster and Erickson  may become nonlinear for electric field strengths beyond 80 mV/m [also, cf. Foster and Tetenbaum, 1992]. Variations with either altitude or longitude of the intense electric fields driving SAID flow could also exist.
 Typical radar observations of F region ion velocities [e.g., Yeh et al., 1991] smear out important features of polarization jet electric field perturbations due to low temporal and spatial resolution, but our technique avoids these problems and allows the determination of several key parameters characterizing the overall polarization jet/SAID structure. One new result is our observation of the entire polarization jet moving poleward at a uniform speed of 250–400 m/s during the ∼30 min that it remains in the radar sensitivity region. This suggests that the observed motion could be an E region signature of a change in the magnetospheric electric field configuration driving the polarization jet. In addition to the overall polarization jet motion, we observe extremely large electric field spatial gradients of 4 mV/m per kilometer accompanied by temporal variation on 1–3 min timescales on two occasions (1937 and 1942 UT in Figure 5), as the features disappear after covering approximately half of the total radar sensitivity region. This is considerably less than recent models of SAID development driven by inward moving injected plasma fronts, which evolve structure over ∼10 min timescales as they develop sharp gradients [de Keyser, 1999].
 The Millstone Hill coherent backscatter observations of midlatitude irregularity cross section during a moderate postnoon geomagnetic disturbance on 13 November 1998, with their fine spatial and temporal resolution, give additional insights into fine-scale polarization jet and SAID event structure by providing a detailed look at the structure and variability of polarization jet electric fields. We accomplish this through detailed modeling of the radar response function to Farley-Buneman irregularities, combined with the linear relationship between UHF coherent backscattered power and electric field reported by Foster and Erickson .
 We observe the subauroral polarization jet, extending over more than 3° invariant latitude, to move poleward at cross-L-shell velocities of 250–400 m/s, with superimposed multiple instances of latitudinally narrow electric fields in a SAID configuration. The polarization jet reaches maximum electric field strengths of ≥80 mV/m, and the SAID intense electric field structures have extremely large electric field spatial gradients of up to 4 mV/m on their equatorward edges with temporal lifetimes as short as 1.5 min. Such observations and interpretation establish the Millstone Hill UHF radar, when operated in a high-resolution coherent scatter mode, as a sensitive diagnostic of the intense electric fields and dramatic variability which characterize the midlatitude polarization jet.
 Observations and analysis were supported by the National Science Foundation through a Cooperative Agreement with the Massachusetts Institute of Technology. We thank the members of the Atmospheric Sciences Group at the Millstone Hill Observatory for many useful discussions, F. J. Rich of Phillips Laboratory, Hanscom Air Force Base, for kindly providing the DMSP drift meter and particle data, and Invers Oy for helpful discussions and technical assistance concerning their GULIPS analysis package.