Notice: Wiley Online Library will be unavailable on Saturday 30th July 2016 from 08:00-11:00 BST / 03:00-06:00 EST / 15:00-18:00 SGT for essential maintenance. Apologies for the inconvenience.
 We look at patches observed by digital ionosondes in the northern polar cap. These observations show that patches have the following properties. (1) Patches are relatively large, typically ∼500 km in the sunward-antisunward direction and ∼1000 km in the dawn-dusk direction. (2) When the IMF Bz has abnormally large positive-negative swings the patches show correlation with the IMF Bz swings. (3) For typical days the patches show no average correlation with IMF By and only weak correlation with negative IMF Bz. (4) Patches are slightly weaker in summer. The typical patch max-to-min Ne variation in winter is ∼2 × 1011 el/m3 and in summer it is ∼1.5 × 1011 el/m3. (5) There is restructuring of patches, particularly the small-scale structures, as the patches convect across the polar cap. (6) Patches are formed prior to entering the polar cap and their properties stay about the same thereafter. We explain the patch generation and most of these properties by low-energy precipitation into the returning flow from the nightside to the dayside around the dawn convection cell.
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 first detailed studies of polar patches were by Weber et al.  and Buchau et al. . Some degree of patchiness seems to be characteristic of the polar cap ionosphere. Patches have been observed from the ground using radio and optical methods and by satellites. Many of the possible patch generation mechanisms produce the patches before they enter into the polar cap. After entry these patches are simply conveyed across the polar cap by convection. In their passage across the polar cap it is generally assumed that the patches will retain their basic properties, although there may be some slight redistribution of ionization by, for instance, gradient-drift instability mechanisms operating on the patch to produce kilometer scale irregularities [see Gondarenko et al., 2003, and references therein]. The constancy of the large-scale patch characteristics during passage across the polar cap is because convection motion can only cause very slight changes in patch density. This is readily shown by putting some typical values into (expanded) dNe/dt = Ne(div(E × B)/B2). One gets a fractional change of the electron density, Ne, of the order of 4% during convective travel from cusp to central polar cap, much too small to be an important effect for patches.
 Most theories of patch formation assume that patches might be formed from the ionization entering the polar cap near noon by one of the following processes:
 1. Sudden or intermittent expansion of the polar cap/auroral region to lower latitudes changes the density of the stream of plasma entering the polar cap [Anderson et al., 1988; Carlson et al., 2004]. The change in polar cap size is in response to changes in the Interplanetary Magnetic Field (IMF) Bz component.
 2. IMF changes, particularly By changes, vary the polar cap convection pattern so that the incoming plasma can vary in density if it comes from locations that are, or are not, sunlit. [Cowley et al., 1991].
 3. IMF By changes could fracture a polar cap “tongue-of-ionization,” thereby producing patches [Sojka et al., 1993].
 7. Processes at lower latitudes could modulate the density of plasma moving toward the polar cap [Foster, 1989].
 Some of these processes that involve the IMF (processes 1, 2, and 7) would produce patches that enter the polar cap with a delay with respect to the causative mechanism, whereas other processes (processes 3, 4, 5, and 6) either take place within the polar cap or the patches would enter the polar cap immediately they are formed. There are only a few published studies that confirm that any one of these mechanisms was the causative mechanism for formation of observed patches [Valladares et al., 1999; Walker et al., 1999; Carlson et al., 2004; Lockwood et al., 2005; Oksavik et al., 2006]. These are all case studies looking at particular events, so it is not yet obvious what mechanisms are operating on typical patches. Of course there is no reason that there should be only a single mechanism that produces polar patches, and several of the proposed mechanisms would produce additive effects on patches that might have originally formed by a different mechanism.
 In this study we will examine patches on a number of days and attempt to show whether they could have been produced by any of the proposed mechanisms. We will also suggest an alternative mechanism that fits better with the observed patch properties.
 Ionospheric measurements in this paper use observations by the Canadian Advanced Digital Ionosondes (CADI) in the northern polar region. The location of the stations whose data will be presented in this paper is shown in Figure 1. Details of the CADI measurements can be found in the work of MacDougall and Jayachandran . The CADI measures both electron density via ionograms and convective motion via “fixed frequency mode,” and both types of measurement will be used in this study.
 There is always a certain amount of “patchiness” in the polar cap ionosphere. To be an “official” patch, the following are the requirements specified by Crowley : maximum patch electron density >2X background density and patch size >100 km. In terms of foF2 the requirement for an official patch is a 41% increase of foF2, which is equivalent to a doubling of the electron density. As will be seen, polar cap structures that meet the official patch requirements are relatively common. In fact a large number of polar cap structures easily meet the size criterion. At a convection speed of 400 m/s any structure that is seen passing over the station for more than 4 min has a size >100 km. As shown by Dandekar and Bullett , the frequency of occurrence of patches is statistically dependent on what one specifies as the minimum electron density change to call the fluctuation a patch. If the density change require for a patch is small, then patches are present 100% of the time. The requirements that are specified for a variation in electron density to be called a patch effect some of the statistics of patches. For instance, in summer the background density is higher, so a given-sized fluctuation in electron density would be a smaller percentage change and therefore might not be considered to be a patch. One needs to be aware of the patch definition that is being used when looking at, for instance, seasonal variations of polar patches.
 In this section we will first show examples of patches. The first example is a rather interesting but atypical day. The second example is more typical days. Following these examples, we show some of the patch properties that we deduce from our CADI observations.
2.1. Patch Examples: IMF Effects
 Our first example of polar patches, Figure 2, shows a whole day when the patches behaved in the expected way in that they essentially retained their basic properties as they convect across the polar cap. In general, patches are just fluctuations in electron density and are not isolated objects with any identifiable form. The fluctuations show statistiacally a linearly decreasing power as a function of frequency [MacDougall, 2006] with no significant power increase at discrete frequencies, so patches statistically are just random fluctuations. Figures 2a, 2b, and 2c show the foF2 observed at three polar cap stations which are approximately collinear: Alert (487 km north of Eureka), Eureka, and Resolute Bay (622 km south of Eureka) (see Figure 1). The sampling rate was different at the three stations, and this causes a difference in the appearance of the foF2 variations seen in Figure 2. Time delays as the foF2 variations convect along the line of stations can be seen in Figure 2 (see arrows): Around 0700 UT the antisunward convective motion is Alert⇒Eureka⇒ Resolute, and around 1900 UT it is the opposite direction. Figure 2d shows the Bz component of the IMF observed by the WIND satellite. The timescale of the IMF has not been adjusted to allow for transit time from the satellite (∼1 hour) to the magnetosphere. It can be seen that the patches show visual correlation with the IMF Bz. (The IMF By component on this day had mostly short-period fluctuations and was only very weakly correlated with the patches.) The IMF Bz variations on this day are unusually strong and well defined. Most days do not show such large clear IMF Bz variations. Figure 2e shows the convection velocity that the CADI measured at Resolute Bay. Comparing Figures 2d and 2e, one can note that, allowing for a time delay for the IMF, the velocities are larger when the IMF Bz goes more negative.
 To further show the relationship between IMF Bz and the patches for the day shown in Figure 2, we show, in Figure 3, the correlation of foF2 with IMF Bz and with polar cap convection speed for the whole day. As expected the correlation with the IMF has a peak (negative) at about −100 min (marked with arrow) which is close to the time delay from satellite to magnetosphere, plus the convection time from cusp to Eureka. The peak is negative because it is the southward (negative IMF Bz) that is associated with the maximum foF2 in the patches. For the correlation with the convection speed the peak is at ∼40 min (marked with arrow) which is the approximate convection travel time from cusp to Eureka. Since previous studies [MacDougall and Jayachandran, 2001, and references therein] have shown that the convection speed is well correlated with negative IMF Bz, and the relationship can be seen by comparing Figures 2d and 2e, it is not clear from these correlations whether the important parameter for patch generation is the IMF Bz or the convection speed.
 For the day shown in Figure 2 the patches showed very good interstation correlation as they traveled across the polar cap. The interstation correlations between Eureka and Resolute for this day are shown in Figure 4. The interstation correlations in 4-hour time intervals are shown, and the time of the middle of the 4-hour intervals is shown above each of the panels. The arrows mark where the expected correlation peaks should occur based on the average convection speed in each of the 4-hour intervals. It can be seen that the times of the correlation maxima agree well with the expected times for the maxima. Therefore on this day the patches are essentially just convecting across the polar cap.
 Using the time lags shown in Figure 3, we can estimate where the patches might have formed. The delay between IMF change observed at the WIND satellite and Eureka was ∼100 min. For the position of WIND on this day, and the solar wind speed, the delay from WIND to the magnetopause would be ∼54 min using the formulas in the work of Lockwood et al. . At 400 m/s (typical convection speed) the travel time from cusp (∼78°) to Eureka would be ∼51 min. The sum of these times is compatible with the 100 min delay from the IMF Bz change at WIND and to when the patch was seen over Eureka if the patch formed promptly at the cusp when the IMF Bz change arrived there. Mechanisms that would take more time, such as processes 1 and 7 in the Introduction, are incompatible with this timing. These timings should probably be somewhat adjusted because the patches are associated with intervals of higher convection speed (see Figure 3b). A higher speed would give additional time for the formation mechanism to operate and still stay within the ∼100 min overall time. It appears that there would then be an interval of the order of 10 min after the IMF changes arrive at the magnetosphere within which the patch is generated.
 The day shown in Figure 2 was quite unusual because of the very large swings in IMF Bz. In fact, one can visualize the patch mechanism on this day as being a chopping of the convection entering the polar cap. Mechanisms that could be operating under these unusual conditions to produce patches have been described by Valladares et al.  and Idenden . In fact, the Figure 2 day was the only day in many years of observations that showed such clear behavior.
 Our second example is for days with smaller IMF Bz swings (the usual case) which show much less clear patch behavior. Figure 5 shows observations for 11 days in January 1995. As can be seen, there are constant fluctuations of foF2, many of which would meet the 41% variation of foF2 requirement to be an official patch. The two stations show a daily variation of foF2 with minimum 5–10 hours and maximum 17–23 hours. This is mostly the effect of the average density of incoming plasma being highest in the North America sector where sunlight reaches higher magnetic latitudes during winter due to the offset of the geographic and geomagnetic poles. It is also possible to detect a minor reduction in mean foF2 values (shown by the heavy horizontal lines in Figure 5) of Eureka with respect to Resolute Bay at noon when convection is from Resolute to Eureka, and at midnight there is a slight reduction of mean foF2 at Resolute Bay relative to Eureka.
 The correlations for this foF2 data with IMF are shown in Figure 6 for Eureka. For these correlations the data was low pass filtered to remove periods shorter than 15 min, since short period structures are small scale (see below), and the foF2 were also corrected for the average diurnal variation of foF2 before doing the correlations. The solar wind data has been delayed to allow for the approximate travel time to the magnetosphere. The correlation of patch foF2 with IMF Bz is seen to be relatively weak, although there is a low correlation at delay −50 min which is typical convection time delay from cusp to Eureka. The correlation of foF2 with IMF By shows no significant average correlation. These low correlations seem to indicate that the IMF plays only a minor role in patch formation on most days. This implies that processes that depend on the IMF magnetic field (processes 1, 2, and 6 in Introduction) are not usually important in forming patches.
Figure 7 shows the cross correlation between the foF2 at Resolute and Eureka in 4 hour time segments for 16 January 1995, one of the days plotted in Figure 5. It can be seen that there is moderate correlation, and the correlation maxima occur roughly at the expected time shifts due to antisunward convection marked by the arrows. However, in some 4-hour segments there is an obvious time discrepancy between the time of correlation maxima and the time marked by the arrows. The low correlation could be due to temporal changes in the convection direction so that patches are not carried over adjacent stations. For this day the convection direction was very steadily antisunward with only minor deviations in direction so the patches should be steadily convected over the station pairs. However, the interstation correlation would be reduced if the convection speed varies during the 4-hour sample interval. For the day used for this sample the convection speed variations were less than the variations for the day used for Figure 4 which showed good interstation correlation. Therefore the fact that peak interstation correlations are only moderate, even for times when one expects that the convection will carry the same patch directly over the two stations, seems to imply that the patches must be undergoing a certain amount of restructuring during their travel between Resolute and Eureka. Since some restructuring processes such as the gradient drift instability might have more effect at shorter scales (still many kilometers), we tried filtering the 4-hour intervals of data before doing the correlation. This filtering seemed to confirm that the shorter period structures in the patches were less well correlated than the longer periods. However, the 622 km spacing between stations might itself cause poor correlation for the short period structures which have smaller spatial scales (see next section), so these filtered correlations seem inconclusive and are not shown here.
2.2. Patch Sizes and Shapes
 Using data from the station pair, Resolute Bay and Eureka, it is possible to get some information about the sizes and shapes of polar patches. Of course some information about the size can be had from just one station observations using the autocorrelation along with the measured convection speed. For instance on 11 February 1995, the day shown in Figure 2, the average autocorrelation function for both stations and for the whole day was essentially exponential in shape and decreased to correlation = 0.5 for a lag of 27 min. The average convection speed on this day was 306 m/s so the average patch width, to correlation of 0.5 was 2 × 27 min × 306 m/s = 991 km. Average widths on other days varied somewhat but were generally of comparable size as will be shown below. Other studies of patch sizes using a similar method [Dandekar and Bullett, 1999] also found patch sizes of the order of 1000 km.
 Using data from the pair of stations, it is also possible to get information about the average patch shape. The simplest procedure is to use just the cross correlations between the two stations for 0 time lag. For instance if the 0 lag correlation between Eureka and Resolute Bay was 0.7 when the relative locations of the two stations was in the antisunward direction, then the distance along this direction for a correlation of 0.5 would be (using an exponential for the correlation shape) = 622 km × ln(0.5)/ln(0.7) = 1209 km. Doing this calculation for various times throughout the day, one can plot the average patch shape and size. Figure 8 shows the average patch shape calculated by this procedure for 11 February 1995. During the day the patches varied greatly (see Figure 2), so treating the whole day as having similar statistical properties to produce Figure 8 obviously leads to a patch shape that is not the neat “elliptical” shape expected for statistical stationary irregularities. However, the Figure 8 pattern does have a width in the sunward-antisunward direction that is comparable with the 991 km width calculated above. Also the Figure 8 patch appears to be elongated in the dawn-dusk direction. In fact for this day the dawn-dusk size is comparable with the polar cap width, so these patches appear to stretch completely across the polar cap.
 Another way of getting information about the patch shape is to use the autocorrelation and convection speed, as above, to get the sunward-antisunward size and then use the cross correlation from the station pair when the two stations are aligned in the dawn-dusk direction to get the dawn-dusk size. We assume that these measurements of the sunward-antisunward size and dawn-dusk size are the two axes of an elliptical shaped patch. On Figure 8 the ellipses calculated in this manner for dawn and dusk are shown. It can be seen that they are roughly compatible with the patch shape drawn using the 0 lag cross correlation method for the whole day. In particular the patch ellipses also show a similar dawn-dusk elongation.
 This “ellipse” method is easier to apply since there is often good data for a 4-hour dawn and/or dusk interval, but there might not be good data for the entire day which is required for the 0 lag method. From results using the “ellipse” method a histogram plot of the lengths of the sunward-antisunward axis and dawn-dusk axis lengths is shown in Figure 9 for a small sample of days. The histograms show that the typical sunward-antisunward size of patches is of the order of 500 km, and the typical dawn-dusk size is about twice as large. The patches are therefore relatively extended in the dawn-dusk direction, the typical dawn-dusk length being comparable with the width of the cusp. Patch formation therefore must take place over a meridionally extended region. Some caution should be used regarding the actual lengths that are shown in Figure 9. To get the sunward-antisunward size, an average convection speed for the 4-hour interval was used and often the convection had large variations in this time interval. Also on some days the ionograms which were scaled for foF2 values showed traces which appeared to come from adjacent patches and when this happened there was some subjectivity in selecting the appropriate trace which could lead to foF2 differences at the two stations which would reduce the correlation and thereby give a shorter dawn-dusk size.
 It is of interest to look at the patch sizes/shapes for different periods. If one filters the data and selects the longer periods (>15 min), then the size/shape is essentially the same as for the unfiltered data. This is because the longer periods tend to dominate the power spectrum of the patches. If the filtering selects short periods (<15 min), then the patches are relatively small and quasi-circular. Obviously, when the patch is much smaller than distance between Eureka and Resolute (622 km) then the correlation does not provide a useful way of determining size.
 A more satisfactory way of getting shapes of patches is to use all-sky imaging at 630 nm. However, there has been little published using this method. Studies by Fukui et al.  and Steele and Cogger  showed results that are comparable with those found in the present study. In particular they found that smaller patched had a dawn-dusk elongation.
2.3. Multistation Observations of the foF2 Patch Variations
 With the array of CADI stations it is possible to look at the foF2 variations at several sites to see whether the average properties of the patches change as they convect across the polar cap. In Figure 10, data for three stations are presented for the days 5–12 December 1995. Figure 10a is for Rabbit Lake which is at 68° magnetic latitude and is thus well outside the polar cap, particularly at noon. Rabbit Lake shows a relatively smooth daily variation peaking at ∼6.5 MHz at noon. The Rabbit Lake foF2 gives a value for the maximum densities of the patches if the patch formation mechanism is the first method mentioned in the Introduction. Some of the other patch methods would produce enhancement or depletions of this foF2. Since the interval of the foF2 maximum shown for Rabbit Lake is only ∼5 hours wide any influx at a time very different from local noon could bring in much lower density plasma. Since one might expect that large values of + or − IMF By might move the influx time to a time significantly away from local noon, we looked at the correlation of the patch foF2 with the absolute value of the IMF By component. The correlation did not show any clear evidence to support the idea that By produces patches by significantly moving the local time of influx. This moving of the influx time to generate patches is essentially method 2 mentioned in the Introduction.
Figure 10b is for Cambridge Bay which is at 77° magnetic latitude and thus at local noon on a typical day is showing the patches when they are at the cusp. Cambridge Bay is located most of the time on the border of the polar cap and only clearly become a polar cap station near local midnight. Looking at the noon interval for Cambridge (around the time marked by the horizontal 2-hour bar at 1900 UT), the plasma has already become structured into patches with maximum densities slightly higher than at Rabbit Lake and lowest densities significantly lower.
Figure 10c shows Resolute Bay foF2 for the same days. Resolute Bay is 720 km poleward of Cambridge Bay. In the noon time interval one sees patch fluctuation that are of similar magnitude to Cambridge Bay, although the densities have decreased somewhat during their travel time, presumably by recombination.
 Unfortunately, the data for Eureka, 622 km further poleward from Resolute Bay, was not good for the days used for Figure 10. Plots of foF2, similar to Figure 10, for Resolute Bay and Eureka (see Figure 5) show little obvious change in the patches properties but do show further small decrease of densities due to the additional travel time from Resolute Bay to Eureka. At midnight, (∼0700 UT) the direction of patch travel is opposite: Cambridge Bay → Resolute Bay → Eureka and for midnight the patches look similar at the three stations but again with small decreases of density with travel time. Remember that at midnight (∼0700 UT) in winter the densities are lower than at noon (∼1900 UT), not due to travel time but because the incoming plasma is from a lower-density region. This is clearly seen in Figure 5 for Eureka which, being approximately in the center of the polar cap has approximately equal travel time from the cusp at all times. For Eureka at 0700 UT the 2-hour average foF2 is 3.5 MHz and at 1900 UT it was significantly higher at 4.5 MHz. Thus incoming plasma in the American sector has higher average density than in the European sector.
 The progressive decay of density with travel of the patches means that one cannot easily form high-density patches with closed convection circulation unless some way of increasing the densities is operating. The plasma passing through Cambridge Bay would be significantly lower in density by the time it travels across the polar cap and then back to Cambridge Bay without any enhancement. Thus to have higher-density patches, one either needs to bring in “fresh” higher-density plasma from regions equatorward of the polar cap or enhance the plasma densities in the return flow region by exposure to sunlight or precipitation. We discuss this further in a later section of this paper.
 One final question here is whether the magnitude of the patch fluctuations of foF2 stays the same all the way across the polar cap. Of course one needs to allow for decrease of density due to recombination or other processes. For the data shown in Figure 10, and the 2-hour midday interval between Cambridge Bay and Resolute Bay the average foF2 decreased from 4.9 MHz to 4.3 MHz. The standard deviation of foF2 due to the patches was 0.91 at Cambridge and 0.88 at Resolute, a slight decrease as would be expected from recombination effects. Thus to a first approximation the patch fluctuations stay about the same after they leave the cusp region.
2.4. Seasonal Variation
 Some processes for generating patches will be affected by seasonal variations. In Figure 11, multiday foF2 values are shown for 12 days of June at Cambridge Bay and Eureka. At Eureka the patches form a band with maximum foF2 ∼5.3 MHz and minimum foF2 ∼3.8 MHz. This range of foF2 is comparable with the patches seen at Cambridge Bay in Figure 11a during hours around midday (1900 UT) and hours around midnight (0700 UT). Comparing Figure 11 with Figures 5 and 10 for winter, one sees that summer patch activity as a percentage fluctuation of average foF2 is somewhat smaller than winter patch activity. However, if one calculates the fluctuation in electron density (∇Ne) in the patches, the summer electron density variations are not much smaller than in winter. It is the high average electron densities in summer that cause the patches to be a smaller percentage foF2 fluctuations. Note that because of these high summer average foF2 values, much of the structuring shown does not have sufficiently large percentage foF2 fluctuations to meet the 41% increase of foF2 requirement to be an “official” patch.
 For later analysis it is of interest to estimate the typical maximum-to-minimum difference of foF2 electron density (∇Ne) in the patches. For winter (Figures 5 and 10) ∇Ne ≈ 2 × 1011 el/m3 and for summer (Figure 11) ∇Ne ≈ 1.5 × 1011 el/m3.
 These results do not seem to agree with some of the other studies of seasonal patch behavior. Dandekar and Bullett  and Dandekar  looked at the statistics of foF2 fluctuations in various months and found a significant decrease of large foF2 fluctuations in nonwinter months. Since the large foF2 fluctuation (see these same references) are significantly less frequent than smaller foF2 variations, if one computes the average deviation of foF2 in the patches, as we have done, the small seasonal variation that we found is not incompatible with these other seasonal variation studies.
 Some of the patch properties that a successful theory must explain are as follows:
 1. Patches are relatively large, typically ∼500 km in the sunward-antisunward direction and ∼1000 km in the dawn-dusk direction.
 2. When there are very large swings of IMF Bz (atypical days), the patches show correlation with the IMF Bz swings.
 3. For typical days the patches show no average correlation with IMF By and only weak correlation with negative IMF Bz. Some portion of patch generation appears to be affected by IMF Bz. From the time of the maximum correlation, the effects of IMF Bz on patches generation appears to take place within a short time (<=10 min) after IMF changes reach the Earth.
 4. There is seasonal variation; patches are slightly weaker in summer. The patch foF2 max-to-min electron density change in winter is ∼2 × 1011 el/m3 and in summer it is ∼1.5 × 1011 el/m3. Other studies [e.g., Dandekar, 2002] show that the larger foF2 fluctuations are significantly less common in nonwinter months.
 5. There is probably some restructuring of patches, particularly the small-scale structures, as the patches convect across the polar cap.
 6. Most patches are formed prior to entering the polar cap and their properties stay about the same thereafter.
 Patches could be relatively complex phenomena so that any given theory of formation would only explain some of the patches. It can be seen by comparison of the patch properties listed above, and the previously suggested theories listed in the Introduction, that many of the patch properties only imperfectly fit with the proposed theories. In particular most patches do not show good evidence for generation processes driven by IMF Bz or By.
 In order to understand the generation of polar patches, it is necessary to examine the time history of polar cap plasma. The IMF controls the convective circulation into the polar cap. The standard convection patterns show the high-latitude convection as closed circulation loops, although the requirements for mathematical closure in some of the studies has possibly eliminated convective input from lower latitudes. Figure 12 shows, at the top, electric potential plots from Weimer . These are for northern hemisphere winter and when local noon is located over North America. Potential contours from other studies show similar features. Comparing the two convection cells, the dusk cell (on the left) is seen to be larger and stronger than the (right-side) dawn cell. As the IMF By changes, the relative strength of the two cells varies, with the dawn cell becoming slightly stronger for more negative IMF By (top left). It would appear that most of the convective polar cap flow would be around the dusk convection cell (the dotted circle show the size of a 13° wide circular polar cap). This viewpoint, however, becomes considerably altered when corotation effects are included.
 In order to see which convection cell is moving plasma into the midday cusp, it is necessary to also include corotation. The corotation acts against the convective flow around the dusk cell and can actually stagnate the flow around this cell. However, the return flow around the dawn cell is enhanced by corotation. The bottom shows the effects of corotation for the same conditions as at the top. It can be seen that almost all the convective flow is actually around the dawn cell. Because of the offset of the geographic and geomagnetic poles, one can only correctly show the corotation effect for any single instant of UT. In these plots the offset of the geomagnetic pole has been ignored, so these plots show an “average” corrotation effect.
 The plasma motion, relative to the local time frame of reference, will follow these potential lines and the plasma speed is inversely proportional to the spacing between the potential lines. We will be considering two processes that increase the density of the plasma during its return from nightside to dayside entry into the polar cap. The first process is exposure to sunlight during the passage of the plasma through the sunlit region near the cusp. The second process is energetic particle precipitation during the return convection. The overall enhancement of plasma is the sum of these two processes. We first consider the plasma enhancement due to sunlight exposure. Sunlight exposure will not obviously produce patches but will enhance the average plasma density.
 To have higher average electron densities in the central polar cap, more extended exposure to sunlight is required. Comparing the plots for By < 0 and By > 0, one sees that for By > 0 some of the plasma has spent slightly more time in the sunlit region (unshaded) prior to entering the polar cap (note the larger convection loop in the sunlit region for IMF By > 0 compared with By < 0).
 Does the polar cap plasma density that is observed agree with what might be expected from these patterns of plasma motion? First, we will look at the plasma in the centre of the polar cap as viewed over the station Eureka (89° magnetic latitude). Figure 13a shows the foF2 for 11 days in January 1995. Solid curves are for daily averaged IMF By > 0, and dotted are for daily averaged IMF By < 0. It is apparent that foF2 is generally lower for By < 0 conditions, and in particular there are many short time very low foF2 values when By < 0. The large foF2 fluctuations will be discussed later; here we are considering the average foF2 values.
 In Figure 13b, scatterplots of daily averaged foF2 versus daily averaged IMF By and Bz are shown. As expected from Figure 13a, the daily average foF2 shows an increase for more positive daily average IMF By values. There is a correlation of 0.56 between the average foF2 and By.
Figure 13c shows that there is also a relationship between daily average foF2 and daily average IMF Bz, with higher foF2 for more negative Bz. The correlation between foF2 and Bz is −0.51.
 The correlation of central polar cap foF2 with IMF By is expected based on our above discussion about sunlight exposure. How does one explain the correlation with negative IMF Bz? For this there are at least two effects that will increase the polar cap electron density. The polar cap size is known to be related to IMF Bz (see the plots in the work of Wiemer ) and a larger polar cap means plasma trajectories go into lower latitudes and hence more sunlight. A second effect is that the plasma motion is faster so that there would be less recombination before the plasma reaches the central polar cap.
 To further study the polar cap plasma, we show in Figure 14 the foF2 at Rankin Inlet (72.5° magnetic latitude) and Cambridge Bay (77° magnetic latitude) for a number of days. Unfortunately, the intervals of good data were not the same for these two stations. Again the foF2 for daily average IMF By > 0 are shown solid and for By < 0 are dotted. The means are shown by the heavy dashed curves. For Cambridge Bay the foF2 went below the lowest frequency for good ionograms (∼2 MHz) during hours around 0800 UT, so the data, and the averages, at this time are only showing days when the FoF2 is higher than usual. The daily average IMF By was mostly negative for these foF2. It is not immediately obvious that there is any significant difference between the By > 0 and the By < 0.
 In Figure 15 we show the Weiner convection with average corotation for By = −2, Bz = −2 which is typical for the average data shown in Figure 14. Also shown are the MLT positions for Rankin (outer circle) and Cambridge (inner circle) with the circular arc thicknesses showing the average foF2 values from Figure 14 (see figure caption). It can be seen that in general the arcs showing the highest foF2 are where one would expect influx of plasma which has had long sunlight exposure. As the plasma travels across the polar cap, its density decreases (compare these densities with those for Eureka shown in Figure 13) due to recombination, possibly augmented by polar outflow effects [Cully et al., 2003]. We can estimate the recombination coefficient from the data in Figures 13 and 14. The travel time across the 12 degree polar cap width for plasma entering at Cambridge Bay and passing over Eureka would be typically ∼12 × 116 km/400 ms−1 ≈ 1 hour. At time 1900 UT for Eureka average foF2 = 4.5 MHz = 2.5 × 1011 el m−3; at Cambridge Bay average foF2 = 6.3 MHz = 4.9 × 1011 el m−3. The recombination coefficient is therefore −ln(2.5 × 1011/4.5 × 1011)/3600 s = 1.6 × 10−4 s−1. Typical values of recombination coefficient from Rishbeth and Garriott  are comparable with this value. Therefore the main reduction of polar cap ionospheric densities appears to be recombination rather than polar outflow.
 Although the main features of the foF2 shown in Figure 15 agree with the plasma flow pattern, there are nevertheless some obvious disagreements. In particular during morning hours from about 0200 MLT to 1100 MLT the observations from both stations are of the ionosphere in the dawn convection cell. Since this cell is closed entirely in darkness, one expects very low densities in this cell. In fact this would be a likely location for a “polar hole.” However, the densities very obviously increased here relative to what they were in the region where the plasma exits the polar cap around midnight. This increase is particularly obvious for the Cambridge Bay average foF2 in Figure 14. The only explanation for this increase that we can suggest is that there is sufficient low-energy particle precipitation in this region to increase the ionospheric foF2 in spite of the lack of sunlight.
 One very obvious feature shown by all the foF2 plots in Figures 13 and 14 is the large foF2 variation. The stations in Figure 13 show, during hours around 1900 UT, the plasma that is just about to enter the polar cap, and one can see that it is already “patchy” without requiring any new patch generation processes(s). Part of the patchiness is obviously old patches that are still present but in a weakened form because of recombination. However, there appears to be vigorous generation of new patch ionization, particularly during the time that the plasma is passing around the dawn convection cell. This can be seen in Figure 14 for the hours 1200–1500 UT for both Cambridge and Rankin when very substantial plasma density fluctuations are observed. We therefore conclude that most plasma patch structuring actually takes place in the return flow region on the morningside. This explains why the correlation with IMF is generally low, since these processes take place over extended time intervals as the plasma returns from the midnight outflow region to the dayside inflow region. Also, it is not at all obvious in the size of the fluctuations shown in Figure 14 (compare solid and dotted curves) that there is a significant IMF By difference in the amount of patch generation during hours 1200–1500 UT.
 Is there sufficient energetic precipitation during the return of plasma from midnight to noon to give the observed patches? Looking at the maps of precipitation presented by Newell et al.  in the dawnside return region, where we speculate that the main patch generation occurs, there is obviously a great deal of low-energy precipitation which would produce F region density enhancements. One can do a very crude estimation of the amount of F region electron density enhancement that could be expected due to the precipitation. We use production curves from Niciejewski . These show that at 250 km height (typical F region peak height) for 0.5 keV electron energy the production is 1 × 109 ions m−3 s−1 for 3.13 × 1012 eV cm−2 s−1 energetic electron flux. Using just the electron precipitation shown by Newell et al.  (this precipitation has a suitably low energy for F region ion production comparable with 0.5 keV), the energy precipitation is ∼1 × 1011 eV cm−2 s−1. This flux would therefore produce 3.2 × 107 ions m−3 s−1. We assume that return travel time through this region is 2 hours, the same travel time as the typical travel time across the polar cap. The peak electron density production is 2 × 3600 × 3.2 × 107 = 2.3 × 1011 el/m3. This value is very compatible with the ∼2 × 1011 el/m3 that we found is typical for patches. The reason for the high densities that are produced by the precipitation is the long time exposure to the precipitation during the return plasma travel. Of course shorter intervals of strong precipitation in the cusp region can produce an enhancement [e.g., Walker et al., 1999], but unless there is unusually strong precipitation, or other unusual factors [Oksavik et al., 2006], the time exposure to precipitation as the plasma convects through the cusp is general too short to produce a strong patch [Weber et al., 1985].
 A final part of this patch explanation would be to show that the precipitating flux during the return travel time has a suitable time structuring to produce patches. Of course the Newell et al.  plots do not give information about whether the time structuring of the precipitation is suitable for producing patches since their plots are averages based on periodic satellite sampling. In further studies of the patches we plan to compare patches with ground optical measurements about the precipitation in the dawn convection cell region to see if the precipitation has a suitable time variation to produce patches.
 There can also be generation of additional patch plasma in the cusp region. The correlation (low) of patch foF2 with negative IMF Bz that we observe may be due to processes, such as cusp precipitation, taking place more vigorously when IMF Bz is more negative. Our observations indicate that these additive cusp processes are less significant for patch generation than are processes in the region of the dawn convection cell.
 Simultaneous digital ionosonde observations from a number of sites in the polar region have enabled us to more clearly define the properties of polar patches. These observed properties such as the weak dependence on IMF Bz component, no average dependence on IMF By, a patch electron density enhancement that remains relatively constant as the patches convect across the polar cap (allowing for recombination), and the large size of the patches determined using correlations show some points of disagreement with some of the published patch formation mechanisms. We propose a mechanism for patch generation in which the majority of patch electron density enhancement is due to relatively low-energy electron precipitation as the plasma returns from nightside to dayside around the dawn convection cell. The majority of returning plasma passes around the dawn convection cell due to the corotation effect. The enhancement is possible because of the relatively long exposure to precipitation during the return travel. This mechanism fits well with most of the observed properties of patches.
 The production of patches is obviously involves multiple processes. Sunlight exposure of the plasma is involved, mainly to increase the average plasma densities. Precipitation during the plasma return from nightside to dayside around the dawn convection cell produces patchy enhancement of the foF2. In the cusp/cleft region various processes such as flow channels and cusp precipitation will superimpose additional foF2 structuring. Finally, during convection across the polar cap plasma instability processes will cause more restructuring, mainly at smaller scales.
 Operation of the CADI digital ionosondes was supported by the Canadian Space Agency. The Rankin CADI was installed and operated by G. Hussey from the University of Saskatchewan. We acknowledge research support from the National Research Council of Canada. IMF data were obtained from CDAWeb. Code for computing the polar cap potentials was obtained from D. Weimer.
 Wolfgang Baumjohann thanks Lie Zhu and another reviewer for their assistance in evaluating this paper.