CADI ionosonde estimates of the F region plasma convection over Resolute Bay (NWT, Canada, MLAT = 83.5°) are compared with nearly simultaneous ion drift measurements on board DMSP satellites passing in the vicinity of the station. First, CADI vectors are projected onto the cross-track directions of DMSPs for 141 events and compared with ion drifts measured by DMSPs in these directions. The best linear fit line is found to have a slope of 0.75 and a shift of −13 m/s indicating good correspondence between the instruments with CADI showing minor underestimation. Then, a fully two-dimensional comparison is performed by considering (1) magnitudes and azimuths and (2) eastward and northward components of the CADI and DMSP vectors. Reasonable agreement is shown as well. For both comparisons, the data spread is small; the correlation coefficients are of the order of 0.7–0.8. The strongest differences between the instruments were found to occur for the periods with spotty enhancement of the plasma flow detected by one or both instruments.
 Monitoring of plasma convection in the ionosphere is of fundamental importance for understanding plasma circulation in the entire magnetosphere-ionosphere system. Over the years, data on convection have been provided by a number of instruments; one of these has been digital ionosondes with Doppler capabilities, such as CADIs [e.g., MacDougall and Jayachandran, 2001, 2006]. A CADI ionosonde produces one convection vector assigned to the zenith of the station every 30 s. For a convection vector estimate, CADI obtains optimal average velocity value for multiple echoes received from the entire sky around the zenith of the instrument for the period of ∼1.5 s [e.g., Grant et al., 1995; Scali et al., 1995].
 Despite the suggested reliability of CADIs as a convection monitoring instrument, not much work has been done on assessing their performance and making intercomparisons with other instruments measuring convection. Several results published in the past suggest basis for concern. Digital ionosonde at Tromso (whose method of measurements is similar to CADI's method) showed occasional, significant inconsistencies with E × B plasma drift measured concurrently by the EISCAT incoherent scatter radar [Sedgemore et al., 1996, 1998]. CADI convection and HF SuperDARN radar velocity vectors do not agree perfectly [Grant et al., 1995] and, moreover, cases were presented for which systematic and consistent differences between these two instruments were observed [Xu, 2003]. Recently, Morris et al.  reported persistent differences in the velocity measurements between a CADI ionosonde and a portable digital ionosonde during their concurrent operation at the Cassey station in Antarctica. The differences were up to ∼200 m/s for typical ionosonde drifts of 500 m/s, with the CADI velocity magnitude being consistently smaller.
 This study compares CADI-inferred plasma convection vectors with data from an independent instrument that has not been considered so far for the CADI assessment work, the ion drift meter on board DMSP satellites passing the vicinity of the CADI ionosonde.
2. Data Selection and Analysis
 In this study we consider only one CADI instrument, the one located at Resolute Bay (RB), NWT, Canada (74.7N, 265.0E, MLAT = 83.5°). There are several reasons for this choice. Firstly, this ionosonde is conveniently located very deep inside the polar cap where convection is much more uniform than in the auroral zone so that the convection estimates are expected to be more reliable. Secondly, this ionosonde is considered to be an important complementary instrument for work with the recently installed SuperDARN HF radar at Rankin Inlet. Thirdly, data from the RB ionosonde have been extensively used in the past for various studies and reliability of the published results need to be established in view of potential uncertainties with the quality of CADI data.
 The DMSP series satellites are on circular polar orbits locked to certain MLT sectors. Over one day, the RB ionosonde rotates with the Earth and is eventually located close to the footprint of a DMSP trajectory at least once for each satellite. Kihn et al.  presented an excellent diagram illustrating typical trajectories for F12, 13 and F15 DMSP satellites over an extended periods (their Figure 2); this diagram shows that the satellites can be in the vicinity of Resolute Bay (MLAT = 83.5°) mostly on the dayside. We note that F12 and F15 footprints are closer to the magnetic meridians than F13. DMSP data for this study were obtained from the University of Texas at Dallas Website. We first consider the cross-track ion drift measurements as these are more frequently of satisfactory quality [Drayton et al., 2005] and compare them with CADI observations. We then make a fully two-dimensional (2-D) comparison by involving the DMSP along-the-track ion drifts. Only those DMSP points were considered that had the quality flag 1 [Kihn et al., 2006; Drayton et al., 2005], indicating the most reliable points.
Figure 1 gives an example of DMSP F13 cross-track ion drift measurements for one pass over the northern high latitudes. The dots are the footprints of the satellite trajectory in the ionosphere. In terms of time, the separation between the measurements at neighboring locations (dots) is 4 s. This implies that the original data with 6 measurements per second were averaged over 4-s intervals. In Figure 1, thin vectors represent the magnitude of the E × B ion drift in the direction perpendicular to the satellite motion. One can see relatively smooth variation of the cross-track ion drift; this was certainly not the case for all events.
 For initial results that are shown here it was assumed that the E × B drift does not change with height. Estimates by Sofko and Walker  show that the convection magnitude reduces by ∼11% at 300 km (CADI measurements) as compared to the magnitude at 840 km (DMSP measurements). In obtaining this estimate, the effects of geomagnetic field change with height (the field magnitude and divergence of the field lines) have been considered. We later use this 11% correction to comment on our statistical results.
 One can notice that when the satellite passed over RB it did not pass exactly through its zenith, which is a very typical situation. In this study, we selected a criterion of suitable satellite proximity to RB as being not farther than 200 km from the zenith of the station. This number is not drastically different from the size of the effective area of data collection for CADI. Finally, a thick vector in Figure 1 is the plasma convection vector inferred from CADI measurements. One can notice reasonable agreement between the magnitude of the CADI vector and typical DMSP vector; the CADI vector itself is slightly larger than the DMSP vectors, but its projection onto the DMSP direction of measurements is very comparable to the DMSP vectors.
 To make a quantitative comparison, a search algorithm was developed that allowed one to select those DMSP passes for which the satellite footprint point closest to RB was separated from the RB zenith by less than 200 km. CADI measurement performed at the time corresponding to the satellite's closest proximity to RB was selected as CADI-measured convection. Typically, there were several DMSP measurements of good quality at other locations along the satellite trajectory, still in close vicinity of RB. To take into account these data, the DMSP measurements were averaged over 30-s intervals centered around the times of the selected CADI measurement. At this stage, commonly 5 DMSP points were available. To have a representative statistics, all possible CADI and DMSP measurements in 2001 were considered; 153 events were identified out of observations in January–March and December. We should note that additional reason for limited statistics was that, for some periods, CADI was detecting echoes from both E and F regions and because quality of convection estimates for such periods is questionable, they were excluded from the consideration. We also noticed that DMSP measurements for some passes were very erratic, perhaps indicating high variability of the convection, and we tried to avoid such events because one would not expect compatibility of the measurements under these conditions.
 We first consider the cross-track ion (plasma) drift measured by DMSP and the projection of the full CADI convection vector onto the corresponding directions of the DMSP ion drift measurements. We will call this the projection comparison.
Figure 2 shows results of the projection comparison in a form of scatterplots; Figures 2a, 2b, and 2c are for F12, F13, and F15 satellites, respectively, while Figure 2d combines data for all satellites. Total number of events is 141. Twelve events were excluded from the comparison because DMSP measurements had strong variability. The standard deviation over ∼5 points for these 12 events was more than 200 m/s and we judged this as being too uncertain to compare with CADI data. Similar criterion was adopted for SuperDARN validation work [Drayton et al., 2005]. Vertical and horizontal bars in Figures 2a–2c indicate standard deviation for each measurement (error in measurements). For DMSP, these were either standard deviation for ∼5 points used in averaging or simply the error given in the standard DMSP data file.
 If the agreement between the instruments were perfect, all the points would be located on the bisectors shown by dashed lines. For all three data sets, corresponding to different satellites, the points are scattered around the bisector but not far from it. The correlation coefficients are in the range of 0.7–0.9 indicating good agreement between the instruments. We also present the linear least squares fit line in the form VCADI = aVDMSP + b for the entire data set, Figure 2d. Errors in measurements were considered in calculating the line. Parameters a and b were found to be 0.75 and −13 m/s, respectively. We note that similar values of a and b were obtained for individual satellite data sets. One can conclude that the agreement between the DMSP and CADI data is reasonable. The data for F12 (F15) are clustered somewhat better (worse) than for other satellites. We noticed that F15 satellite made measurements at a substantial angle with respect to the L-shell direction more often than F12 and F13. Since the usual CADI measured flow was along the L shells, then small errors in the CADI measurements projected to the F15 directions gave larger uncertainties for the comparison. This effect might be one of the reasons for somewhat poorer agreement between the F15 and CADI measurements. We also noticed that for quite a few F15 cases, there were only 1–2 reliable points, and these might not well represent the average flow in the area of measurements.
3. Two-Dimensional Comparison
 As a second step, we perform a full two-dimensional vector comparison by adding DMSP ion drift data along the satellite track. The same (originally selected) 153 events are considered. It turned out that even though the same amount of events was available (as for the projection comparison), the variability of ion drift data along the DMSP track was large (the standard deviation was more than 200 m/s for quite a few of them) and such events have been excluded from the statistics to match similar criterion adopted for the projection comparison. As a result, data for only 89 passes were available. This still gives reasonable statistics.
 Two-dimensional comparison of DMSP and CADI data was performed by considering magnitudes and azimuths (clockwise from geographic North) of the convection vectors. Results are presented in Figure 3 as scatterplots of CADI quantities versus DMSP quantities. Bars around individual points reflect the uncertainty of measurements as inferred from original errors provided by the instruments. Median value of errors for the velocity magnitude and azimuth are ∼80 m/s and ∼10° for CADI and ∼70 m/s and ∼15° for DMSP.
 One can see in Figure 3 that the agreement is reasonable for both plots; the correlation coefficients are 0.69 and 0.88 for the magnitude and azimuth, respectively. There are measurements with significant differences, both for the magnitude and for the azimuth, but the majority of points are located near the bisectors of expected agreement between the instruments.
 We also compared CADI and DMSP vectors by projecting them onto two directions, geographic north (N) and geographic east (E). We found that the agreement is reasonable for both components with the correlation coefficients of 0.65 and 0.85 for the N and E directions, respectively. We note that, for most of the events considered, the dominating component of the flow was eastward so that the better agreement was for the predominant direction of the flow.
4. Discussion and Summary
 The present study was motivated by recent observations indicating uncertainty in the quality of F region plasma convection estimates with a CADI ionosonde in Antarctica [Morris et al., 2004]. According to Morris et al. , the CADI instrument was persistently (over 2 year period) showing smaller convection magnitudes as compared to a co-located digital portable sounder which used basically the same procedure as CADI to estimate convection vectors. Also, earlier comparisons of Tromso digital ionosonde data and EISCAT convection measurements revealed differences, at least for small integration time comparisons [Sedgemore et al., 1996, 1998]. Differences in the velocity magnitude for some individual points were as large as 200 m/s in all the above papers and points with different flow polarities can be identified in the data published.
 In a case of another HF system is used to establish the quality of CADI measurements, such as in observations by Morris et al. , the question as to how similar are the collecting areas of the systems (for signal detection) is the one that requires serious scrutiny, and the velocity differences can be solely attributed to the difference in the location of the collecting areas under the condition of strongly nonuniform plasma flows. This is one of the conclusions by Grant et al. , who compared CADI and SuperDARN HF data. Involvement of incoherent scatter radar [Scali et al., 1995; Sedgemore et al., 1996, 1998] seems to be a better choice because propagation effects are eliminated at least for one instrument, but still the sizes of the signal collecting areas of the instruments are significantly different since ISRs (HF systems) have narrow (broad) antenna beams.
 In a case of fast temporal variations of ionospheric flows, difference in integration time of systems used might also complicate comparisons. This is important for CADI/ISRs work since ISRs have integration time of one or two minutes in a best case while digital ionosondes average data over several seconds.
 In this study we involved a new kind of instrument into CADI assessment work, the ion drift meter on board DMSP satellites. We attempted to make integration time comparable by averaging DMSP data over 30-s chunks centered around the times of successive CADI measurements. Such a procedure does not mean that the signal collecting areas were overlapping, and in fact, for some passes, differences of up to 200 km were allowed. To diminish the effects of nonuniformity of ionospheric flows, we selected CADI observations within the central polar cap where the flows are expected to be generally more uniform than in the auroral zone.
 Presented data show remarkably good agreement of convection velocities obtained by CADI and ion drifts measured by DMSPs. The data clustering for the present comparison is better than for the SuperDARN radar line-of-sight (HF system) – DMSP comparison [Drayton et al., 2005]. Comparison of DMSP ion drifts and SuperDARN convection vectors (2-D comparison) shows even stronger scatter [Xu et al., 2007].
 Considering data for individual satellites, Figure 2, one can notice much smaller data spread for F12 though this might be owing to rather limited number of events; there were almost 3 times more measurements for F13. We note that the data for F15 were obtained mostly in the prenoon sector while for F12 and F13 there were points for the afternoon sector as well. In terms of the correlation coefficients, there was not much difference between the data for individual satellites, Figure 2, once again pointing at robustness of the results. Obtained slope of the best linear fit line of 0.75 (Figure 2d) would improve to ∼0.83 (11% increase) if one considers the fact that DMSPs measures the ion drift at the height of ∼840 km while CADI measurements (echo heights) are referred to the height of ∼200–300 km.
 An important question is the reasons for data spread in Figures 2 and 3. One thing that immediately comes to mind is the fact that although the DMSP data were averaged over chunks of the trajectory near Resolute Bay, this is not equivalent to signal averaging with CADI which collects signals from various parts of the ionosphere with off-zenith angle as large as 20°. Since the flow is not ideally homogenous, small differences are inevitable.
 An attempt has been made to investigate the quality of the data for both instruments for the points with significant differences. Figure 2a shows one point that is far away from the bisector with VDMSP = 800 m/s and VCADI = 200 m/s. Examination of the DMSP records showed that, for this event, the DMSP value was based on only one measurement and all other measurements, which could have been considered in finding the average DMSP cross-track velocity, were classified as not reliable because of the problems with measurements of the along-the-track ion drift component. This point still had been accepted for a comparison because it satisfied all adopted criteria. The rejected DMSP measurements for this event (with the quality flag of 4) show the cross-track component of the order of 300 m/s which is close to the CADI value of 200 m/s. Figure 2c has another anomalous point with VDMSP = 800 m/s and VCADI = 150 m/s. For this event, CADI measurements prior to the event and after it showed much larger values, 300–500 m/s, which would make disagreement smaller if CADI data were averaged over several 30-s intervals. In addition, for all CADI observations in this event, the number of echo sources contributing to the convection determination was of the order of 20 which we consider as a rather small number since for most of other measurements this number was above 30. It has been found that, in general, larger number of echo sources gives more reliable measurements. We also investigated the data for some points with different CADI/DMSP polarity in Figure 2. It turned out that for most events either CADI or DMSP measurements or both were quite variable for “neighboring” periods of time implying that more smoothing would have eliminated the reported polarity differences.
 Good agreement of CADI and DMSP measurements shown in this study suggests that the reasons for persistent differences between CADI and digital ionosonde measurements in Antarctica [Morris et al., 2004] are probably related to local arrangements for the experiment. One issue is low sensitivity of the antennae used by CADI so that the convection-derivation algorithm did not work well in the presence of noise.
 Finally, results reported here are based on not very large statistics. We should mention that building up of a common DMSP/CADI database is not a simple task, as in the best case, one can get 1–2 passes during a day, and more importantly, CADI operation was often intermittent for various reasons. Recent installation of an HF radar of SuperDARN style at Rankin Inlet provides ample opportunities for investigation of velocities measured by both systems and this work is under way.
 Results obtained in this study can be summarized as follows:
 1. The magnitude and direction of the high-latitude F region plasma convection inferred from CADI observations correspond reasonably well to DMSP-measured ion drift. The comparison is made by integrating DMSP data over ∼30 s and considering one CADI measurement corresponding to the center of a ∼30-s period. The relationship between the CADI velocity vectors projected onto the direction of DMSP measurements in the cross-track direction can be written in a form VCADI = 0.75 · VDMSP − 13 (m/s) implying that CADI velocities are slightly smaller than the DMSP velocities. The correlation coefficient between the CADI ionosonde and DMSP data is of the order of 0.8. The DMSP convection velocity decrease as it is mapped down to CADI heights from the DMSP heights and accounting for this decrease makes the slope of the best fit line closer to one. Fully two-dimensional comparison between CADI and DMSP vectors shows reasonable agreement as well with correlation coefficients of ∼0.7 for the velocity magnitude comparison and ∼0.8 for the azimuth comparison. Data presented give additional evidence on the reliability of CADI convection estimates within the polar cap where the flow is often spatially homogeneous and changes slowly with time.
 2. The reported minor differences between CADI and DMSP velocity measurements might be related to several reasons out of which the most likely one is local nonuniformity of plasma flows. This nonuniformity is often identifiable by excessive variability (spottiness) of the convection pattern according to either one or both instruments.
 We thank the University of Texas at Dallas DMSP team for providing free access to the DMSP ion drift data through their website (M. Hairston). This work has been supported by NSERC (Canada) and CSA (Canada) grants to A.V.K. and J.W.M. Comments of both referees are appreciated.