In this paper we concentrate on the plasma drift measurement performed by Digisonde Portable Sounder DPS 4 developed by the UMLCAR, Lowell. Special attention is paid to the quality control of data acquired prior to drift velocity calculations. Risks of carefree use of raw Doppler spectra for the plasma motion analysis have been recognized in the Digisonde community. In this paper we propose improvements of Digisonde drift data processing for velocity evaluation and provide a detailed step-by step description: (1) robust height range selection, (2) setting limits on the Doppler frequency shift, and (3) setting limits on the echo arrival angle. Raw data that passed our quality control were then used to derive three velocity components of the bulk plasma drift in the ionosphere. In the paper we discuss, in detail, application of the method on the DPS measurements. However, this technique can be efficiently applied on any other digital ionosondes drift data processing.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 The interest in the ionospheric variability has increased recently, partly due to the role that the ionosphere plays in the Earth's environment via coupling processes from above and below and partly due to the increasing use of telecommunications and satellite navigation systems. The need for information about the ionospheric state requires monitoring of the ionospheric plasma behaviour. Special campaigns have been organized in order to obtain information about plasma motion in the ionosphere [Sedgemore et al., 1998; Ujmaia et al., 1999] (among others), however regular Doppler frequency shift measurements of the plasma motion is necessary. Regular monitoring represents an important part of the ionospheric database.
 The Earth's ionosphere is a complex natural plasma system of extreme sensitivity to a variety of phenomena in the Sun-Earth environment. Changes in the ionospheric conditions are intimately linked with the fluctuations in the electromagnetic and corpuscular radiation from the Sun as well as with the state of Earth's neutral atmosphere. Since the inaugural measurements at Slough, England, 75 years ago, the ionosphere has been subject of an intensive research by means of remote high-frequency (HF) radio sounding. Recently introduced, advanced HF signal analysis allows monitoring of the ionospheric plasma drifts, which gives us new information about the processes that control the dynamic behaviour of the ionosphere. As another wave of scientific interest in the ionospheric processes rises due to increasing need of telecommunication and satellite navigation systems, knowledge of variations in the ionospheric plasma drifts as functions of height [Parkinson et al., 1997], latitude, geomagnetic activity [Scali and Reinisch, 1997], solar cycle, season and local time becomes a critical issue and enables development of efficient ionospheric forecast models [Mikhailov, 2005]. Dominant phenomena forcing plasma to move in the ionosphere are: diurnal variation in the production and loss of ionization; dayside thermal thermosphere expansion; transport effects due to electric fields, winds and waves in the neutral atmosphere.
2. Drift Measurement Using Digisonde
 Ionosonde measurements are based on the total reflection of HF waves at locations in the ionosphere where, for the ordinary wave component, the frequency of the transmitted wave is equal to the plasma frequency [Davies, 1990; Kohl et al., 1996]:
where N is the electron density, e is the electron charge, me is the electron mass and ɛ0 is the free space permittivity.
 When the transmitted signal illuminates a large area of the ionosphere, the receiving antenna field observes multiple vertical and oblique echoes arriving from the locations with matching electron density N(1) and local isodensity contour perpendicular to the wave propagation vector. For a perfectly smooth, horizontally stratified ionosphere only one such location exists, responsible for a single vertical echo. The off-vertical echoes occur due to ionospheric irregularities disturbing the density contours. Multiple echoes overlay each other as they arrive at the receiver location and remain inseparable for analysis by conventional ionosondes. The Digisonde uses spectral analysis to distinguish individual echoes with different Doppler frequency shifts. In the typical scenario of a large or medium scale travelling ionospheric disturbance (TID) passing the station with high horizontal velocity of hundreds m/s [Hocke and Schlegel, 1996], oblique signals from the opposite sides along the line of travel of the TID will have opposite Doppler shifts. Once signals in each antenna are separated by their Doppler frequency using the Discrete Fourier Transform, the multi-element antenna interferometry can determine the source location (zenith and azimuth of arrival signal) for identified echoes [Reinisch et al., 1998].
 The actual processing in the Digisonde depends on the mode of observation. In the ionogram mode, only the strongest O- and X-mode echoes in each range bin are recorded together with a coarse Doppler frequency and angle-of-arrival information. In the drift mode, source locations are calculated for all detected echoes, usually for a selectable subset of heights around a strongest echo. The optimal height selection is discussed in further detail below.
 In the Digisonde drift mode, the graphical presentation of the source (reflection) points is commonly called a skymap. Figure 1 shows a skymap with the source points plotted on a horizontal plane so that the vertical echoes are at the skymap centre. The corresponding values of the Doppler frequency shifts are distinguished by different colours of the plotting symbols. Each source point is defined by a specific source location, Doppler frequency shift, sounding frequency, the wave mode (O or X) and the echo amplitude. When the ionosphere is perfectly smooth and horizontally stratified, all echoes arrive vertically, which makes determination of the horizontal drift velocity from the line-of-sight Doppler shift impossible. An important characteristic of the skymaps is therefore “source coverage” that qualitatively describes distribution of sources over the zenith angle of arrival. Figure 1 shows a skymap with a good source coverage, well suited for an accurate calculation of the drift velocity.
 Identified source locations can also be used to calculate the bulk velocity of the ionospheric structure over the sounder location. Assuming the phase velocity of the signal to be the speed of light c, Doppler frequency shift for the i-th source point is given by:
where ki is a unit vector in the i-th source direction, f0 is the frequency of the sounding signal and vi represents the velocity vector in the i-th source location. In an ideal situation, when plasma moves with a single bulk velocity, for any choice of at least three linearly independent source points we obtain the same velocity that describes plasma motion. In reality, data are accompanied by noise and plasma motion (velocity vector v) is not exactly uniform for all source points. Hence, we can determine the velocity vector by a least squares fit of the source points:
where VN, VE, Vz are the velocity vector components (North, East, and vertical), and Θi and Φi are the zenith and azimuth angles for signals arriving from the i-th source.
 It is suitable to describe the ionospheric plasma motion using a single velocity vector v even if the real situation slightly departs from the condition of uniform plasma motion used in (3). Value of ɛ2 in equation 3 is influenced by uncertainty of the measurement. It also reflects how the real situation deviates from theoretical assumptions of uniform motion and indicates the quality of the fit for a particular data set [Potter et al., 2005].
 This technique is commonly referred to as the ‘Digisonde Drift Analysis’ (DDA method) and the resulting velocity is called the ‘drift velocity’ (see Reinisch et al.  for more details).
3. Skymap Processing
 The DDA method is implemented in the interactive analysis software tool “Drift Explorer” [Reinisch et al., 2005] commonly used by scientists. The automatic data processing software is distributed with the University of Lowell Digisondes where it serves for plotting real-time skymaps and automatical drift velocity computing. Automatically scaled data from a network of Digisondes are available through the webpage of the Lowell data centre (http://ulcar.uml.edu/stationmap.html).
 The initial drift observations at Pruhonice revealed problems in the automatic measurements, which were caused by the inclusion of multiple Es and 2-hop F echoes, and other erroneous signals with large Doppler frequency shifts. One may argue that such errors could have been eliminated by a more appropriate configuration of the Digisonde measurements. Indeed, correctness of the drift velocity calculated from raw data strongly depends on settings in the drift control program. In most cases it is possible to find the optimal settings for any given individual measurement, depending on current ionospheric conditions. However, those settings will not be optimal for some other measurements, and periodic manual interventions in the observational schedules of the sounder are hardly feasible. Simple manual measures (e.g., limiting of the height search to a tighter interval) may lead to a loss of relevant data. Further enhancements to the intelligent automated measurement configuration algorithm will have to rely on prediction of the optimal settings, which increases the risk of completely missing the signal during the most interesting, unpredictable periods of ionospheric activity. Instead, we propose to use the nominal settings for all measurements and process the data with additional steps described below.
 The present paper introduces the solutions for overcoming the difficulties by selecting appropriate source points from skymaps in order to obtain correct drift velocities using the DDA method. Our newly improved selection method consists of three steps: (1) robust selection of the height range, (2) setting limits on the Doppler frequency shift, and (3) setting limits on the zenith angle. The DDA algorithm is also part of the off-line interactive analysis software tool “Drift Explorer” [Reinisch et al., 2005], which we have used to optimise the drift analysis of the Pruhonice data.
3.1. Height Range Selection
 In the drift mode, the Digisonde stores complete Doppler spectra for each antenna, resulting in roughly two orders more data compared to the ionogram mode. In order to limit the data volume, the DPS control software automatically selects only useful range intervals for a small subset of operating frequencies. A search algorithm finds the strongest amplitude within the selected range window and assumes it to be the main echo (for either E or F region). The skymaps contain source points with a wide range of virtual heights, and some of these “apparent sources” may belong to different propagation modes, i.e., multiple hop propagation. In order to avoid problem, we first select only those source points that correspond to reflections taking place in a particular region of the ionosphere.
 Reflection points in the ionospheric F-region occur above 170 km and below 1000 km. F-layer parameters are subject to strong fluctuations according to geomagnetic situation, solar activity and state of the neutral atmosphere [Prölss, 2004]. Typical values of the F-layer peak density height (∼200 km–400 km in midlatitudes) may change dramatically during geomagnetic storms, passage of travelling ionospheric disturbances etc. Hence, severe restriction of the height range in the sounding setting may lead to the loss of reflections while still insufficient to remove source points belonging to different propagation modes.
 Nevertheless, in practice we carry on drift measurement for a narrower height interval according to the latitude and geomagnetic situation. For our initial F region studies we had selected the range interval between 150 and 540 km. However, very often this height interval covers not only direct F region reflections but also multiple-hop reflections from Es and F layers that travelled the path between the Earth and ionosphere more than once. To eliminate these wrong skymap source points we plotted the virtual heights of the skymap points as a function of the sounding frequency for comparison with the preceding ionogram data obtained immediately prior to the skymap measurement. Then we selected only those skymap points whose heights correspond to the height of F-trace on the ionogram.
 The reflection height of F region multiple-hop echoes appears to be above the F region peak. In this case, it would be possible to automatically select skymap points corresponding to F region reflections, it means the lowest group of registered source points. Well defined sporadic E layers (Es) that often occur around 100 km altitude can blanket the F layer, and usually produce multiple Es echoes with virtual ranges that fall in the height regime of the F layer. The autodrift mode then misidentifies these echoes as F reflections that are falsely included in the DDA analysis. Sporadic E layer is predominantly a summer phenomenon at middle latitudes, however Es also occurs during other seasons. Multiple Es reflections are often recorded at virtual heights corresponding to F reflection height interval, which is the major complication for a simple automatic selection strategy.
 Application of the height selection for F-region data is demonstrated using an example shown in the Figure 2. Figure 2a shows the virtual height of the source points as a function of the sounding frequency. From this plot and from the immediately preceding ionogram (Figure 2b) it is evident that raw skymap data (Figure 2c) contain multiple-hop reflection points that must be removed. Once they are removed, the corrected skymap (Figure 2d) shows different source points distributions. The velocity components derived from the raw and corrected skymaps differ significantly as shown in the Table 1. For the E-region drift measurements, the height interval from 90 to 150 km was specified; no multiple echoes fall into this interval and no correction is required.
Table 1. Velocity Components Computed for the Raw and Corrected Skymaps Presented on Figure 2a
Number of Points
More than two thirds of the points on the raw skymap correspond to second reflection in this case. Velocity estimation using least squares fitting on the corrected skymaps produces distinctly different results for the East component here.
−3.1 ± 15
61.0 ± 45
−55 ± 130
20.3 ± 9
69.0 ± 25
−220 ± 35
3.2. Restricting Doppler Frequency Shift
 Skymaps are frequently comprised of two populations of sources. The first population consists of a cluster located in the central part of the skymap with small zenith angles. Representative Doppler frequency shifts of such points typically fall within a relatively narrow interval and form a peak around zero Doppler frequency shift on the histogram. Furthermore, the angular distribution of frequency shifts on the skymap shows a distinct bipolar pattern that corresponds to the bulk horizontal motion of the ionospheric plasma structure over the sounder location. The characteristic range of the Doppler shift values and the bipolar pattern of their distribution for the first population of skymap sources remain fairly consistent over consecutive skymaps.
 The second class of sources is characterised by a higher Doppler frequency shift. This population usually has no clear pattern of the Doppler frequency distribution and rarely forms a cluster in the central part of the skymap, covering the entire sounding range at various zenith angles. When viewed in a time sequence, this group shows highly variable, inconsistent behaviour (see Figure 3).
 We believe that the first population of radio sources is of the primary importance to the determination of the bulk ionospheric plasma motion. The source points of this population form the cluster with significantly bipolar pattern of the Doppler frequency shifts that corresponds to the bulk plasma drifts with predominantly horizontal component.
 We think that the second group of radio sources is related to random external interference present in the receiver bandwidth that can be powerful enough to survive amplitude thresholding used for the source detection in the Doppler spectra. Commonly such interfering radio emissions can be identified by their consistent presence in a particular Doppler line for all ranges, because they are not associated with the Digisonde pulse.
 Inclusion of the radio sources from interferers distorts calculations of the bulk movement velocity. Figure 3a shows a sequence of three raw F-region skymaps at 15 minute cadence (top) and corresponding Doppler frequency shift histograms (bottom). Table 2 lists drift velocities for all three times calculated using the least squares fitting. Results show large differences in both amplitude and direction of the drift velocity.
Table 2. Velocity Components Computed for the Raw and Corrected Skymaps Presented on Figure 3a
Number of Points
In the contrast to velocities calculated from the raw data, results for corrected skymaps are similar. Velocity estimation using least squares fitting to raw data produce incorrect results mainly in cases where is number of corrected points distinctive less then number of points on raw skymap (here skymaps for times 02:05 and 02:20).
Jan 16, 01:50
2 ± 20
−70 ± 130
−120 ± 100
4 ± 5
−20 ± 30
−115 ± 25
Jan 16, 02:05
27 ± 20
100 ± 140
−360 ± 200
2 ± 6
−17 ± 30
−100 ± 40
Jan 16, 02:20
−70 ± 30
−320 ± 70
90 ± 170
15 ± 4
−23 ± 10
−115 ± 20
Figure 3b shows selected sources corresponding to first population for all the three skymaps. It is evident that the skymap pattern is similar for all three skymaps. Note that the color scale is different from Figure 3a for a better readability. Velocity components in Table 2 show results derived by the least squared fitting for the corrected skymaps. Quality of the fit is much better for the corrected data than for raw skymaps, and calculated velocity components do not vary much with time, as it should be for a quiet ionosphere (Kp during studied period is 1-).
 In general, it is not difficult to separate these two groups of points on the skymaps for most cases. We select only those skymap points forming a pronounced separated peak around the zero frequency shift on the histogram (see Figure 3). This selection step is important for both E and F region drift data.
 Configuring drift measurements for a narrower interval of the Doppler shifts (i.e., larger integration times) does not eliminate the interferers but rather lets them fold into the Doppler spectrum at smaller frequency shift values, thus reducing the errors that they introduce to the plasma velocity calculations. However, making the Doppler range too small will leave out possibility of registering truly high velocities in the ionosphere. Narrowing down the Doppler shift interval also results in generally longer measurement times and therefore reduced time resolution. Instead, we suggest keeping the range Doppler shift measurements high enough to accommodate dramatic events during the storm times and select subsets of observed frequency shifts in the post-analysis by singling out the main peak in the frequency histogram. In this case the interfering sources appearing at higher Doppler shifts are effectively eliminated in the post-processing. Further discussion of the optimal balance for selecting the Doppler measurement settings can be found below in the section “F region drift measurements at Pruhonice observatory”.
3.3. Choice of the Maximum Zenith Angle
 Horizontal distance between individual source points shown on a skymap can reach values of 200 km in E-region and several hundreds kilometers in the F-region. The DDA method assumes a uniform velocity within the entire area illuminated by the transmitted signal. In general, plasma velocity can vary in the ionosphere (primarily during disturbed geomagnetic conditions). In that case the pattern of clustered points on a skymap is not strictly bipolar. We argue that the most reliable evaluation of the drift velocity above the Digisonde location is achieved by using only those skymap sources whose zenith angle is relatively close to vertical. We do realize that selecting sources with smaller zenith angles reduces the accuracy of horizontal velocity computation that rapidly decreases when zenith angle goes to zero. When the skymap pattern is not quite bipolar, it is necessary to think about the trade-off between imposing a limit on the zenith angle and sustaining the accuracy (sensitivity) of velocity computation.
 Hence, in order to analyze the skymap Doppler patterns that are far from being bipolar, we select source points only in a reduced range of zenith angles. In our analysis we typically use source points with the zenith angles below 20° for F-region measurements and 30° for E-region measurements.
Figure 4 illustrates the choice of the maximum zenith angle for E-region data. Left panel corresponds to the raw skymap that shows a changing pattern in its Northern sector. That Northern pattern is the reason why we select only sources from the central part of skymap, where the pattern remains bipolar. The right panel of Figure 4 shows a corrected skymap for which we set maximal zenith angle to 25°. Comparison of velocities calculated for the raw and corrected data is shown in Table 3.
Table 3. Velocity Components Computed From the Raw and Corrected Skymaps Presented on Figure 4a
Number of Points
Points with small Doppler frequency shift in north part on the raw skymap decrease amplitude of estimated north velocity component.
−34 ± 4
−44 ± 15
−40 ± 30
−34 ± 4
−74 ± 10
−41 ± 25
 For the most cases, limiting the zenith angle does not substantially influence results of the F region drift measurements. Standard setting to 20 or 30 degrees guarantees good quality of results.
 For the E drift measurements we often (over 30% of cases) register good quality of skymaps with a consistent pattern across all zenith angles. For the best quality of calculated drift velocity it is better to use all points if possible. However, relatively often (20–30%) we register skymaps with a good coverage and large number of points, but with a changing spatial pattern. This situation corresponds to a lateral variation of drift velocity over the observatory location. In such cases, optimizing the zenith angle limit becomes important.
3.4. F Region Drift Measurements at Pruhonice Observatory
 In January 2004, a Digisonde DPS-4 was installed in Pruhonice Observatory (geographic coordinates 50.0N, 14.6E). The Pruhonice Digisonde routinely records ionograms every 15 minutes. The ionogram autoscaling process “ARTIST” automatically finds the F region critical frequency foF2. The F-region drift sounding follows 5 minutes after the ionogram sounding. We have programmed the DPS to automatically select four frequencies fi spaced by 200 kHz, with fi < foF2.
 The F region drift measurements in Pruhonice are made with the search interval for potential echoes limited to 150 to 540 km of the virtual height. Inspection of ionograms recorded at the Pruhonice Observatory since 1990 reveals that the maximum virtual heights of signal reflection in the F region exceed 540 km very rarely. Unfortunately, this selected height interval is wide enough to frequently accommodate second hop F and Es layer reflections. Even though it is the strongest echo that is automatically located in the given search interval, it may occasionally correspond to the multiple hop mode of propagation. Decreasing the upper height limit (e.g. to 400 km) might lead to a considerable loss of valid measurements when the F layer is high. Valid skymap heights have to be selected manually by referring to the immediately preceding ionogram to determine the actual virtual height of the F layer, h'F. While Digisonde control software automatically selects optimal operating frequencies for the drift measurements using the autoscaled foF2 value (autodrift mode), the height search interval remains constant for all measurements. A fully automatic real-time selection algorithm could be devised that would determine the type of the reflection (ordinary or multiple) from the autoscaled ionogram and then avoid ranges that correspond to the multiple reflections.
 Since 2004 we have tested several measurement settings with different Doppler frequency shift ranges and resolutions:
 1. Originally, the pulse repetition rate was set to 200 pulses per second, corresponding to the coherent integration time of 5.12 s, Doppler frequency resolution of 0.195 Hz and Doppler range from −12.5 to +12.5 Hz. Short integration time has a practical advantage for campaigns requiring high cadence of measurements (e.g., under 1 minute). However, raw skymaps acquired at such wide Doppler range were practically always contaminated by incorrect sources with higher Doppler shifts that dramatically affect calculations of the drift velocity. Applying step 2 to raw skymaps is critically important in this case.
 2. By setting the pulse repetition rate to 50 Hz, we increased the integration time to 20.48 s, with corresponding Doppler frequency resolution of 0.049 Hz and Doppler range +−3.1 Hz. As we were expecting a decreased number of sources in this configuration, we chose to set the height step to 2.5 km instead of original 5 km for a better height resolution of the reported data. In some cases we obtained skymaps with a good coverage of correct sources, which guaranteed consistently correct velocity interpretation. However, such ideal situation did not occur too often. Besides, measurements at this setting took longer time to complete and had to be made with smaller number of repetitions, which consequently meant worse representation of sources in the skymap.
 3. In our opinion, the optimal setting for the F-region drift measurement in middle latitudes is: height resolution −5km, number of heights – 128, rate −100 pulse/s, number of pulses per frequency – 128, four fixed frequencies. Corresponding Doppler frequency shift spectral resolution is 0.098 Hz and Doppler frequency shift range is +−6.3 Hz. This setting yields skymaps with good skymap-points coverage and sufficient number of correct sources (in the order of hundreds). In this case we also need to apply step 2 for all measurements. Similar settings for the F-region drift measurements were recommended and used in other midlatitude Digisonde locations, Athens and Ebro observatories. More information about Digisonde settings and other technical details can be found on the Digisonde web page (http://ulcar.uml.edu/digisonde_dps.html).
3.5. E Region Drift Measurements
 Since May 2005, the Pruhonice Digisonde also measures E region drifts every 15 minutes, using four fixed frequencies between 2.0, 2.2, 2.4, and 2.6 MHz. Our experience shows that the optimal settings are: height resolution −2.5km, number of heights −256 and pulse repetition rate −50 pulses/s. Resulting Doppler frequency shift spectral resolution is 0.049 Hz and Doppler frequency shift range +−3.1 Hz. These settings usually provide us with skymaps with good skymap-points coverage and a sufficient number of correct sources (in the order of several thousands). In about half of all measurements we observed the number of incorrect source points to be sufficiently low, with a negligible influence on the resulting drift velocity.
 Proper selection of the radio sources on Digisonde skymaps is critically important for correct interpretation of drift measurements and for the accurate velocity estimation. We described a selection method that has been used to reprocess drift data measured at Pruhonice Observatory during 2006 (so far we have completed recalculation of 21,712 F-drift and 13,812 E-drift measurements). Figure 5 shows an example one-day plot of plasma velocity for February 3, 2006, derived directly from the raw data with no correction applied. The upper and lower panels of Figure 5 show F-region drift (auto mode) and E-region drift velocities, respectively. No smoothing filters were applied to the data. It is evident that estimated F-region velocities have markedly larger fit errors indicating presense of noise, and velocity values rapidly fluctuate with time.
Figure 6 represents the same data set for February 3, 2006, but with the skymaps reprocessed using the new selection method. As in the previous Figure 5, the upper panel corresponds to the F-region drift, while the lower panel shows the E-region drift. Clearly, the standard deviations represented by the error bars have significantly decreased. Here we observe a continuously changing velocity with only few exceptions. Occasional outliers are caused by particular conditions (low skymap coverage, fewer reflection sources).
 Comparison of E-region measurements calls for a more detailed explanation. We record the E-region drifts at fixed frequencies (2–2.6 MHz) during the entire day regardless of the possibility for presence of the E or Es layer. As seen in the lower panel of Figure 5, incorrect velocities do occur when the E or Es layer is not present or its critical frequency falls below 2 MHz, the first measurement frequency. When all of the sounding frequencies are above the E-layer critical frequency (foE), some sources can still be selected for the skymap, but derived velocity would have no physical meaning. Skymaps for all such cases contain only noise and should be removed in selection process (step 2). Correspondingly, only those measurement periods in the E region are significant when the sounding frequency can be reflected by ionospheric layer. Comparison of the lower panels in Figures 5 and 6 shows that standard deviations of velocities estimated from raw data are slightly larger then those estimated from the corrected data. For all valid measurements (when the sounding frequencies were below foE), the vertical velocity components are not affected much by the reprocessing, whereas horizontal components display significant differences between 7 and 11 UT for this particular day. In general, reprocessing of the E region measurements have smaller effect compared to the F-region measurements.
 Our method represents a suite of techniques that improve quality of data used for analysis of the ionospheric plasma drifts based on Digisonde measurements. Step by step, we choose only relevant reflection sources detected by the Doppler skymap technique in order to describe the representative bulk plasma motion across the sounder location. Our selection method is based on robust height range determination, setting limits on the Doppler frequency shift, and setting limits on the echo arrival angle.
 The method in its present state requires manual exclusion of the irrelevant reflection sources that are entering the standard “DDA” analysis. Therefore, it cannot be applied routinely for unattended processing of the Digisonde measurements. However, it is very useful when accurate and robust results are required. We have demonstrated the importance of the careful source selection. We have also shown that the resulting velocity can be strongly affected by using an incorrect set of reflection sources for the DDA analysis.
 The proposed method consists of three steps. Their application on the raw data may significantly affect resulting estimated velocity since it eliminates incorrect reflection points. Step 1, selection of valid reflection heights, is absolutely necessary during the summer when probability of the Es layer occurrence is high. Importance of removal of multiple F-reflections increases mainly during the night when focusing conditions can favor the second hop propagation in the F layer.
 Steps 2 and 3, limiting Doppler frequency shifts and zenith angles, can be important depending on the Digisonde measurement settings. Our experience shows that our standard setting, which is also frequently used by other mid-latitude stations, benefits from Steps 2 and 3 filtering. In most cases 2nd step of the selection procedure is especially important for obtaining correct results in the F-region. As for the E-region measurements, an optimal choice of the automated selection parameters can be achieved during particular periods of daytime that results in a low number of incorrect source points (about 20–30% of measurements). During such periods of time, velocities estimated from raw data and reprocessed filtered data are certainly similar, as in the Figure 5 and 6.
 Ionospheric drift measurements have been performed using different equipments (Dynasonde, Ionosonde, EISCAT etc.). Comparison of the plasma motion estimation measured by Dynasonde and EISCAT shows good agreement [Sedgemore et al., 1998] in polar region. Also a fair agreement between F-region drift measurements using ionosonde and incoherent scatter radar at magnetic equator [Woodman et al., 2006] has been reported. Observations of F-region drift measured using DPS and CADI ionosondes however shows substantial differences between the vertical and horizontal drift velocities [Morris et al., 2004]. Our paper shows the possibility of the improvement of the data reliability. We hope our results will rejuvenate ionospheric plasma motion research using the HF skymap/drift technique.
 The authors would like to thank Bodo Reinisch of UMLCAR for helpful comments and suggestions. This work was supported by the Grant Agency of the Czech Republic (projects 205/06/1619 and 205/06/1267), the Grant Agency of the Academy of Sciences of the Czech Republic (project IAA300420504) and European Union project COST 296 (MIERS).