A middle ultraviolet imager (235–263 nm) on the Midcourse Space Experiment (MSX) has obtained the first large-scale, two-dimensional maps of polar mesospheric clouds (PMCs). The lower parts of individual images, which lie below the ozone horizon, are mapped onto an ellipsoidal shell at 83 km altitude and then combined to establish the transpolar PMC field in two spatial dimensions across the entire polar region. At all latitudes where they appear, the PMCs clearly evidence a “patchy” structure as opposed to a uniform layer. Among the interesting features of these cloud patches are zonal alignments, arcs, and repetitive structures characteristic of waves. Whether random or repetitive, the cloud structures exhibit scales ranging from several hundred kilometers down to tens of kilometers.
 Polar mesospheric clouds (PMC) were discovered by satellite [Donahue et al., 1972] and are now considered to be the poleward extension of noctilucent clouds (NLC) observed from the ground for decades. Ground observations at low polar latitudes employ traditional optical methods including cameras [Fritts et al., 1993; Warren et al., 1997], spectrometers and spectrographs [Fogle and Rees, 1972; Hecht et al., 1997] and polarizers [Witt, 1960]. These traditional observations have established the whispy nature of the clouds and their small-scale, local structure [Fritts et al., 1993]. The height of these clouds and their scattering properties require, for optical observations from the ground, lighting and contrast conditions only available at twilight during the polar summer and at latitudes below 70° [e.g., Gadsden, 1982]. In recent years, lidars have accurately measured the altitudes, thicknesses and local time variations of the clouds under non-twilight conditions, and radars have measured polar mesospheric summer echoes apparently associated with noctilucent clouds [Thayer et al., 1995; Von Zahn et al., 1998; Cho and Rottger, 1997]. Instruments on rockets have attempted with some success to directly sample the cloud particles by flying through the clouds [Goldberg et al., 1991, 2001].
 While important, all these observations are limited to local samples of polar mesospheric clouds. Satellite instruments can readily sample on a much larger scale that can potentially extend across an entire polar region. Photometers on the OGO6 satellite first observed polar mesospheric clouds from space in 1969 and immediately showed that the clouds extended entirely across the polar region above 80° latitude [Donahue et al., 1972]. Spectrometers on the SME satellite established the statistical climatology of the clouds: they appear above 55° latitude at altitudes within 80–85 km [Thomas and Olivero, 1986], last from ∼20 days before summer solstice to ∼60 days after [Thomas and Olivero, 1989], and result from particles with effective radii of ∼70 nm or less [Rusch et al., 1991]. Currently, the Student Nitric Oxide Explorer (SNOE) satellite offers the potential to continue the long-term observations of PMC climatology using techniques similar to those of OGO6 and SME [Bailey et al., 2001].
 Satellite imagers can also effectively map the instantaneous transpolar configuration of polar mesospheric clouds, thereby establishing their global morphology (as opposed to climatology) in much the same way as similar imagers have mapped the instantaneous aurora [e.g., Holzworth and Meng., 1975; Frank and Craven, 1988; Liou et al., 1997]. (Here, “instantaneous” means on timescales much less than those expected for the subject phenomena to change on a global scale.) This investigation reports observations of PMC morphology made by a middle-ultraviolet imager (235–263 nm) on the Midcourse Space Experiment, an experimental defense satellite [Mill et al., 1994]. The orbit of the satellite carried the imager over the polar caps and allowed instantaneous, large-scale PMC observations during the Antarctic summer of 1997–1998 and the Arctic summer of 1999. Because the lower portions of the UV images lie below the UV horizon, pixels can be unambiguously mapped onto an ellipsoidal shell at the 83 km altitude of the clouds. The combination of many mapped images reveals the first complete transpolar pictures of the PMC field.
2. Instrument and Geometry
 The UVISI instrument suite [Carbary et al., 1994] includes one narrow-field imager that responds to middle ultraviolet radiation from 235 nm to 263 nm. This imager uses an intensified charge-coupled device that can accommodate the strong intensities of the terrestrial dayglow and light scattered from PMCs. The imager's electro-optics reduces light from longer wavelengths to a factor of ∼10−7, which effectively eliminates visible-light contamination (“red-leak”) and reduces off-axis and Rayleigh scattering. An imager frame rate of 2 Hz diminishes smearing caused by spacecraft motion. The imager pixels have an effective size of ∼85 × 110 μrad within a field of view (FOV) of 1.6° × 1.2°.
Figure 1 illustrates the geometry of PMC observations. From a circular orbit at 900 km altitude, the imager viewed the terrestrial horizon from a slant range of ∼2800 km. Approximately half of the field of view encompassed below-the-horizon polar mesospheric clouds at an altitude of h ≈ 83 km. The line-of-sight of a single pixel typically intersected the PMCs at the near-field and the far-field. Confusion between the near-field and far-field was eliminated by choosing only pixels below the mid-UV or “ozone” horizon. A full UVISI image consists of 256 × 244 pixels, and the extraction of a sub-image of 101 × 94 pixels, which lie well below this horizon, ensured the selection of near-field radiance. The pixel limit defined by this horizon was determined by inspecting PMC “strip” profiles [Carbary et al., 1999].
 A typical sub-image pixel measured ∼3.0 km along-track (including the effects of smearing caused by satellite motion) and ∼2–3 km across track. For the center pixel in a sub-image, the observer's line of sight made an angle of about 84° to the zenith of the PMC. The solar zenith angle was essentially constant: 67°–71° in the south and 69°–75° in the north. The scattering angle varied slightly across a pass from 67° to 72° for southern passes and from 129° to 144° during northern passes. The MSX satellite provided an ultra-stable platform with absolute inertial pointing accuracy of ∼50 μrad (about half a pixel) and a precision (jitter) of ∼10 μrad, both of which were less than the pixel size of ∼100 μrad.
3. Data Set
 UVISI observations took place in discrete sessions called data collection events (DCEs). Thirteen PMC-dedicated data collection events took place during the Antarctic summer of 1997–1998, and ten took place during the Arctic summer 1999. A complete DCE typically covered a transpolar distance of ∼7000 km and took about 28 minutes min, although PMCs were not necessarily observed throughout this entire interval. Scene background observations and instrument calibrations occupied a portion of this interval, and data gaps at times interrupted measurements. Nevertheless, the 23 passes recorded a total of 25,431 images suitable for PMC mapping. Because of the imager sampling schedules, about two thirds of this total number come from the northern polar region and one third comes from the southern. To reduce the processing burdens, this investigation surveyed 1121 southern images and 1225 northern images distributed evenly over each of the 23 passes.
 The transpolar mapping of the two-dimensional PMC field had required new techniques for background subtraction and pixel mapping. These new techniques are described in some detail before presenting the actual cloud maps.
4. Background Subtraction
 Counts from the selected foreground and background sub-images are converted on a pixel-by-pixel basis to radiance (photons/cm2secsr) using standard UVISI processing that includes photocathode response, filter throughput, field of view, uniformity, dark level removal, image intensifier gain, and accumulation time. The pixels of the sub-image are then mapped onto the surface of an ellipsoid having an altitude of 83 km above the mean Earth surface [e.g., Carbary et al., 2000]. This mapping assumes the clouds have a common altitude and establishes the longitudes and latitudes of all pixels in the sub-image. This step in the processing requires considerable computational time, and the mapped sub-images are stored in binary files for later processing.
 For all but the brightest clouds, the background signal dominates the PMC signal in the sub-images by a factor of ∼10, so careful background subtraction must be performed. The background in the sub-images consists of Rayleigh scattering from the atmosphere and out-of-field radiance from the bright Earth. Models have proven inadequate in calculating both the Rayleigh and out-of-field backgrounds when folded into the geometrical response of the imager, so the background must be determined from actual observations. Thus, the background B(t) at any time t during a pass is calculated using a three-point LaGrange interpolation:
where B0, B1, and B2 are the observed background images at times t0, t1, and t2 for which no PMCs are detected in the sub-image. Non-PMC images were determined by inspecting transpolar intensity profiles as was done, for example, by Carbary et al. . Care is taken that these background times are appropriately spaced across a single polar pass.
 The corrected image Ic(t) at time t is the difference between the uncorrected image Iu(t) and the background at t:
This correction is performed on a pixel-by-pixel basis and does not involve any models of the atmospheric background. However, the correction does assume that the background varies no more quickly than second order across the polar region. This assumption appears reasonable because the observing geometry as well as the atmospheric structure remain essentially constant over the relevant time and distance scales. Imager observations made during the PMC “off-seasons” clearly support the assumption about the spatial variability of the background across the polar regions.
Figure 2 illustrates background removal process for one sub-image. The left frame displays the clouds plus the background, while the middle frame shows the interpolated background. The corrected sub-image on the right reveals numerous horizontal “streaks,” which are the clouds observed from a high observer zenith angle. The graininess of the corrected image illustrates the relatively low contribution of PMC signal to the total signal. This graininess can be overcome by co-adding overlapping sub-images.
 After background subtraction, the pixel radiances are corrected for solar zenith angle and observer zenith angle using the cosine approximation. This is a source correction that effectively adjusts for the “slant path” of oblique viewing.
5. Transformation to Polar Rectangular Coordinates
 After background correction of all pixels in the selected image, the east longitude φ and the latitude λ are mapped into a “polar-rectangular” coordinate system according to:
where Rp is the polar radius of the Earth, and the sign [s] is negative in the southern hemisphere and positive in the northern hemisphere. In these coordinates, the −y axis (downward) marks the prime meridian (0° longitude). In the northern hemisphere, the east longitude φ is measured counterclockwise when looking down on the north pole. In contrast, the east longitude is measured clockwise when looking down on the south pole. This mapping preserves the usual appearance of polar maps without distorting the spatial aspect of the clouds. Furthermore, the mapping allows the immediate and undistorted realization of linear distances on a polar scale. Pixel mapping errors in x or y of up to ∼30 km will result if actual PMC altitudes are 3 km higher or lower than the assumed altitude of 83 km. Unless this extreme variation occurs within one cloud group, which is unlikely, the qualitative conclusions of this paper will be unaffected by such errors.
 The individual pixels of many overlapping sub-images are accumulated and mapped into polar-rectangular coordinates. The pixel-by-pixel display of multiple images gives a grainy but suggestive view of the PMC field, as illustrated in the left panel of Figure 3. A much clearer picture of the clouds is obtained by averaging the pixel intensities into 5 km × 5 km bins, as shown in the right panel of the figure. The bin map reveals features having scales from ∼100 km down to the bin limit of ∼5 km. The following sampling of maps, all based on the 5 km × 5 km binning, illustrates some of the more interesting PMC macro-structures observed from a satellite platform.
6. Zonal Features
 One of the most interesting characteristics seen in both the northern and southern PMC fields is a tendency of the clouds to be zonally aligned, or aligned along parallels of equal latitude. Figure 4 exhibits two instances of such alignment in the northern polar region, while Figure 5 displays an alignment in the southern polar region. The left panel summarizes an entire pass, while detailed bin-maps at higher resolution appear in the right panels. The intensities have been logarithmically scaled to enhance less intense features.
 In the upper panel of the northern pass, bright clouds are aligned along the 74N parallel of latitude. The features are brightest toward the east (counter-clockwise in north) and less intense toward the west. In the lower panel, the southern edge of a bright cloud is aligned along the 85N parallel. The cloud features in both panels apparently extend beyond the field of regard of the imager. In the southern pass of Figure 5, the bright feature is aligned with the 87S meridian, and the intensity decreases toward the west (counter-clockwise in south). This feature is embedded in a larger cloud field that again extends outside the field of regard of the imager. Note that two of these alignments appear within ∼500 km of the pole at very high latitude.
 Zonal features such as these appear in approximately 20% of the areas covered by the 23 passes made by the UVISI imager. Unfortunately, the narrow width of the imager track prevents a more accurate assessment of the length of these features, and the limited sampling precludes a better estimate of their occurrence frequency. The UVISI data do suggest, however, that zonal alignment of polar mesospheric clouds is common. The alignment of cloud features could represent smearing caused by zonal winds known to exist in the mesosphere [Nastrom et al., 1982; Berger and Von Zahn, 2002], although no specific evidence exists that these features are actually caused by wind smearing.
7. Isolated Features
 Some clouds exist in apparent isolation from larger masses of clouds. Figure 6 exhibits an unusual “curved” cloud isolated in the northern polar region. The arc described by this cloud extends about 160 km in length and about 40 km in width. Figure 7 presents two isolated features near 80N. The clouds have amorphous shapes measuring ∼100 km in scale size. Like many PMCs, the cloud features in both figures display some internal structure. For example, a “channel” of low intensity separates the two clouds in Figure 7.
 Patches of clouds such as these were typically observed in every polar pass made by the satellite. They appear at all polar latitudes and in both hemispheres, in seemingly random places, with scale sizes similar to those of the previous figures.
8. Repetitive Features
 Periodic or quasiperiodic structure is commonly observed on a global scale in PMC fields [Carbary et al., 2000]. Unfortunately, the narrow swath of the imager's field of regard across the polar region compromises a wave analysis. The use of two-dimensional maps eases this problem somewhat, although the observation of repetitive cloud features is still serendipitous. The following two examples are advanced as features that are “possibly” repetitive and that suggest a large-scale organization of the clouds.
Figure 8 exhibits an example of repeating, if not periodic, cloud features. (In this case, the intensity scale is linear rather than logarithmic.) The left panel summarizes half of a pass between the south pole and 75S latitude, and the right panel exhibits four cloud features along a span of ∼500 km. A great deal of structure is evident here, but the features marked “A”, “B”, “C”, and “D” seem to repeat along the pass. These features do not have a common shape, but do display “traits” that repeat with a scale of ∼100 km along the track of the satellite. Such a scale length is common in a one-dimensional periodogram analysis of clouds [Carbary et al., 2000].
 A second example of repetitive structure appears in Figure 9, which details another southern polar pass. Three elongated features extend across the track of the satellite. These are labeled “A”, “B”, and “C” in the upper panel, The “B” feature is less intense than the other two, which brighten toward the east (counter-clockwise in south). The scale of this repetition is again ∼100 km. Another interesting characteristic of this pass is seen in the summary frame in the bottom panel. The long bluish swath (between data gaps) indicates that no clouds were detected for ∼450 km along the pass.
 The synthesis of hundreds of middle-ultraviolet images has generated the first global-scale pictures of polar mesospheric clouds. This new view has revealed a variety of interesting morphological features at high polar latitudes. The clouds do not exist as a uniform layer but rather occur in groups or as isolated clouds with scale sizes of ∼100 km. The clouds are often zonally aligned. Some cloud features also appear to repeat on scales of >100 km, and there exist smaller-scale features (at least to the bin resolution of 5 km) within the larger cloud structures. These morphological features are present in both the northern and southern polar regions.
 This survey represents a brief introduction to the global morphology of polar mesospheric clouds. The picture of the global distribution of PMCs is becoming clearer with the advent of satellite observations, and future investigations of the clouds will certainly profit from knowledge of the large-scale structure of the clouds.
 The authors gratefully acknowledge the support of the Ballistic Missile Defense Office and data processing personnel at the Applied Physics Laboratory. This research was carried out under National Aeronautics and Space Administration Office of Space Science Grants NAG5-8062 and NAG5-9240.