Voyager 2 at Neptune: Imaging Science Results

Voyager 2 images of Neptune reveal a windy planet characterized by bright clouds of methane ice suspended in an exceptionally clear atmosphere above a lower deck of hydrogen sulfide or ammonia ices. Neptune's atmosphere is dominated by a large anticyclonic storm system that has been named the Great Dark Spot (GDS). About the same size as Earth in extent, the GDS bears both many similarities and some differences to the Great Red Spot of Jupiter. Neptune's zonal wind profile is remarkably similar to that of Uranus. Neptune has three major rings at radii of 42,000, 53,000, and 63,000 kilometers. The outer ring contains three higher density arc-like segments that were apparently responsible for most of the ground-based occultation events observed during the current decade. Like the rings of Uranus, the Neptune rings are composed of very dark material; unlike that of Uranus, the Neptune system is very dusty. Six new regular satellites were found, with dark surfaces and radii ranging from 200 to 25 kilometers. All lie inside the orbit of Triton and the inner four are located within the ring system. Triton is seen to be a differentiated body, with a radius of 1350 kilometers and a density of 2.1 grams per cubic centimeter; it exhibits clear evidence of early episodes of surface melting. A now rigid crust of what is probably water ice is overlain with a brilliant coating of nitrogen frost, slightly darkened and reddened with organic polymer material. Streaks of organic polymer suggest seasonal winds strong enough to move particles of micrometer size or larger, once they become airborne. At least two active plumes were seen, carrying dark material 8 kilometers above the surface before being transported downstream by high level winds. The plumes may be driven by solar heating and the subsequent violent vaporization of subsurface nitrogen.

fortably long exposure times for many ofthe sequences. This, in turn, required the use of a special sequence design to compensate for image smear caused by spacecraft motion. In addition, spacecraft engineering teams modified the attitude control software to provide further reduction of random spacecraft motion over that achieved earlier at the Uranus encounter (1). The success ofthese techniques is aptly demonstrated in the striking photographs that Voyager 2 has sent back to Earth from the very edge of our planetary system.

The Atmosphere of Neptune
Neptune's atmosphere provided many surprises during the Voyager encounter. The high wind speeds, the persistence of large oval storm systems, and the hour-tohour variability of small-scale features were unexpected in an atmosphere that receives 1/20 as much energy (power per unit area) from internal heat and absorbed sunlight as Jupiter and only 1/350 as much energy as Earth. Large features near the equator move westward relative to the interior at speeds up to 325 m s-l, making Neptune one of the windiest planets (with Saturn) in the solar system. Small-scale features appear to move at twice this speed. The Great Dark Spot (GDS), a weather system comparable to Earth in size, stretches and contracts as it rolls in a counterclockwise direction with a 16-day period. Bright clouds cast shadows on the main cloud deck 50 to 100 km below.
At some latitudes, the bright clouds resemble mountain lee waves, where largescale patterns remain fixed while small-scale elements move through them. Such patterns were not seen on the other gas giant planets. Although the atmospheres of the giant planets do exhibit some common phenomenacolored clouds and hazes, latitudinal banding, strong winds, for example-there remain many differences and they do not follow obvious rules.
Discrete cloud features and banded structure. The largest discrete feature in Neptune's atmosphere is the GDS, which resembles Jupiter's Great Red Spot (GRS) in several respects. Neptune's GDS has an average extent of 38°and 150 in longitude and latitude, respectively (Figs. LA, 2, and 3A), compared with 30°and 200 for the GRS (2).
The GDS is located at about the same latitude as the GRS (20°S) and seems to have a similar anticyclonic circulation sense (counterclockwise in the southern hemisphere). Evidence for the anticyclonic rotation of the GDS is based primarily on visual impressions gained from a time-lapse sequence, rather than on cirect measurement SCIENCE, VOL. 246 of the displacement of small features. However, there are significant differences between the GDS and the GRS. The circulation of the GRS modifies its surrounding environment, forming a turbulent-wake region to the west ofthe feature (2). Although the GDS on Neptune also appears to influence its immediate environment, the regions surrounding it appear morphologically more uniform compared to those on Jupiter. Furthermore, the GDS drifts rapidly westward at speeds over 300 m s-I relative to the planetary radio rate (3,4), while the GRS drifts westward at an average rate of only 3 ms-' (5).
The bright companion along the southern edge of the GDS (Figs. 1 and 3A) was detected in January 1989; it was the first discrete feature observed on Neptune by Voyager 2, and was also seen in groundbased images (4,6). This feature persistently appeared on the southern edge of the GDS, although its brightness and precise shape varied with time. In Fig. 3A, the bright companion is resolved into east-west linear structures.
Time-lapse images of the GDS and its vicinity (Fig. 4) suggest that the bright companion may be similar to orographic clouds observed on Earth, that is, clouds created by air being forced upward by the presence of a mountain. Specifically, the smallest features appeared to move relative to the structure as a whole. The role of topography, necessary for the formation of terrestrial orographic clouds, may be played by temperature and pressure anomalies associated with the GDS.
A bright feature near the southern pole (71°S) was first seen in April 1989 (Fig.  1B). This south polar feature (SPF) appears to be not one discrete feature but rather an active arc extending over some 900 of longitude and constrained to a latitudinal band spanning less than 5°. The intensity and distribution of brightness along this arc was seen to vary significantly during a single rotation of the planet.
The third bright feature discovered by Voyager (after the bright companion to the GDS and the SPF) was the "Scooter," a bright compact feature centered near 42°S (Fig. 1A); despite its name, however, it is not the fastest moving feature (4). The "Scooter" is composed ofmany small streaks stacked in latitude (Fig. 3C) rather than being a single round or oval storm system. The number and length of these streaks varies (causing the feature to change shape from round to square to triangular), but the composite structure identified as the "Scooter" persisted as a unit throughout the 80day encounter period.
Soon after the discovery of the "Scooter," a second dark feature (D2) was identified in the southern hemisphere at a latitude of 550S (Figs. 1, A and B). Several weeks after its discovery, a bright core developed at the center of the dark feature. The core remained visible for the remainder of the encounter, although its brightness varied. Small-scale cloud features were clearly visible within the bright core (Fig. 3B). Figure  5 shows three high-resolution images of the cloud structure in the core of D2. The size and shape of the local details varied on time scales of hours. On Jupiter, features similar to D2 rotate anticyclonically, but the sense of circulation has not yet been detected in D2.
Voyager images of Neptune's atmosphere (Fig. 1, A and B) revealed a banded zonal structure with the brightest region located near 20°S. Bands of lower reflectivity in the northern and southern hemispheres were located at 6°N to 25°N and 45°S to 70°S, that is, the brightness variation was not symmetrical about the equator. This asymmetry is also seen in Fig. 2, where the effect of viewing geometry was removed to show the brightness at normal incidence (7). Bright ephemeral streamers were seen both at the latitude of the GDS (Fig. 6) and at about 27N. These streamers were highly variable in both time and brightness. Methane-band images of the southern hemi- sphere revealed a bright band within 150 of the south pole. To within an uncertainty of 10, a small feature lies at the south rotational pole (Figs. 7 and 8). This structure suggests a well-organized polar circulation that has no analogue in the polar regions ofthe other giant planets.
Vertical structure fromfeature contrast. Analyses of ground-based observations of Neptune (8)(9)(10)(11) and analogous studies of Uranus (11,12) indicated the presence of three major particulate layers in the observable atmosphere of Neptune (Fig. 9). A photochemical smog layer was predicted at pressures starting at about 5 mbar in the lower stratosphere and extending to lower altitudes in the stratosphere and upper troposphere (8,10,13). This layer, composed of lower order hydrocarbons such as ethane, acetylene, and diacetylene, results from photochemistry of methane driven by solar ultraviolet radiation within the lower and upper stratosphere.
Two condensation cloud layers are thought to occur in the upper troposphere. Methane should begin condensing at about the 1.5-bar level (8)(9)(10)14) while a more optically thick cloud appears to exist near the 3-bar level (8)(9)(10). This deeper cloud may be made of hydrogen sulfide ice partides (8,10), but ammonia also is present (14).
The Voyager images provide both vertical and horizontal structural information (Fig.  1, A and B). We used several complementary approaches for estimating the absolute and relative heights of these features. In the first approach, we studied the wavelength dependence of the contrast of features. As shown in Figs. 8 and 10, the contrast of bright and dark features varies markedly with wavelength. For example, the "Scooter" displays the greatest contrast in the orange-filtered image and much lower contrast in both the UV and longer wavelength methane filter (designated MeJ, with center wavelength 619 nm). On the other hand, most other bright features, such as the bright companion to the Great Dark Spot, show an enhanced contrast in the MeJ band relative to orange. The Great Dark Spot has its maximum negative contrast in the blue filter.
We interpret the contrast variations in the following way: molecular Rayleigh scattering in the upper layers tends to mask deeplying features at the shortest wavelengths (the Rayleigh scattering optical depth is about 5 above the 3-bar cloud in the UV filter), while absorption by gaseous methane tends to mask deep features in the methane filters (especially the MeJ filter). Thus, the doud tops of the major bright features are not all located at the same altitude. In particular, the "Scooter" is situated at a much lower altitude than the bright companion of the GDS, as evidenced by the "Scooters" decreasing contrast at methaneband wavelengths relative to blue. Latitude is planetocentric and longitude is based on a rotation period of 17 hours 52 min, the "predict" period used to plan Voyager observation sequences. This is longer than the 16 I   I   I   I   I  I  I I more, the "Scooter" must be located below the base of the methane cloud, because it is masked in the MeJ filter (assuming there is little methane above the base ofthe methane cloud). Preliminary radiative transfer calculations that approximately reproduce measured contrasts substantiate these conclusions (8,15). The generally negative contrast of the GDS and the fact that its blue and MeJ contrasts are comparable might indicate that the atmosphere is relatively clear above the GDS; that is, the amount of scattering above the top of the 3-bar cloud is low (corresponding to a lower optical thickness of the methane cloud). If so, there is a gradual change from the outer to inner regions of the GDS, with the clearest atmosphere occurring toward its center. Altematively, the negative contrast might indicate that the "3-bar" cloud is actually at a deeper level in the center of the GDS. With the exception of the "Scooter," the discrete bright features are probably methane condensation clouds. These features are prominent, localized, and dynamically active. In contrast, a photochemically produced haze is likely to be spatially uniform and inactive. Such a haze layer is visible at the limb in the images taken through the MeJ and MeU filters (the latter with center wavelength 541 nm) (Fig. 6) and was inferred from ground-based observations of center-to-limb brightness profiles (8). In the troposphere, methane gas is probably abundant enough to produce a prominent localized cloud. The deep (labeled H2S?) condensation cloud (Fig. 9) is not a candidate for features that are bright in the MeJ filter, because that cloud has too much absorbing methane gas above it to appear bright at this wavelength. Figure 7 shows D2 and the SPF visible on Neptune's crescent in all filters from UV to orange. Such visibility of features on the crescents was not apparent at Jupiter, Satumr, or Uranus, and this is consistent with the idea that the bright white features on Neptune extend to considerable heights above most of the haze and atmospheric gases.
Vertical structure from limb scans and cloud shadows. Images obtained at high phase angles provide evidence for the existence of a stratospheric haze and also place constraints on its properties. Such images are particularly effective for detecting the presence of optically thin stratospheric hazes, since the submicronmeter-sized haze particles preferentially scatter sunlight at small scattering angles (large phase angles). Particle visibility is also enhanced by the long slant paths through the atmosphere that occur under these viewing conditions. Figure 1 1A shows several radial line scans across the planet's limb. The scans reach a nearly constant brightness at lower altitudes where the slant-path optical depth exceeds unity (or the vertical optical depth exceeds several hundredths).
On the basis of experience with similar images obtained at Uranus (11), we suspect that the transition to constant brightness (400 km relative alitude in Fig. 1LA) occurs within the bottom scale height of the stratosphere (80 to 100 km altitude in Fig. 9). Light scattered by a combination ofaerosols and gas molecules is detectable to altitudes above 550 km in Fig. 11A at least 150 km above the constant brightness portion. The sharp increase in brightness with decreasing altitude just above the constant brightness region indicates the existence of particles in this region of the stratosphere, since the brightness increases more rapidly than would be expected from molecular Rayleigh scattering alone. This point is further demonstrated in Fig. 1 1B, which shows an extinction profile derived by inverting one of the limb scans of Fig. 11A (11,13,16). The region with a rapid increase in the extinction coefficient may mark the altitude where temperatures become cold enough for gaseous ethane to condense into ice particles (13).
Cloud shadows provided additional data about vertical structure. As shown in Fig.  12, some discrete clouds that were observed 'v t~~~2 sis were consistent for observations made at both high and low phase angles (about 140 and 20 degrees, respectively), where the viewing geometry is quite diffcerent.
Large-scale variability offeatures. All of the larger-scale features tracked during observatory phase exhibited latitudinal drifting associated with changes ofwind speed (4). D2 exhibited the largest change of wind speed. When first measured, it was located at 550S and had a rotation period of 16.0 hours (Figs. 1 and 5). Then D2 drifted northward to 51°S, and its period increased to 16.3 hours. When the feature returned to 550S, the rotation period decreased to 15.8 hours.
Twenty-five days after the start of the cycle, the feature again drifted north with an associated increase of rotation period (4). The range of rotation periods spanned by D2 indudes the 16.05 hour period of planetary radio emissions (3).
The latitudinal drift of the "Scooter" was small in amplitude (less than 2 degrees) and was neither monotonic nor periodic. The feature was initially detected at 420S, where its rotation period was 16.74 hours; it remained at this latitude for more than three weeks. Then over the course of a week, the "Scooter" drifted north to 40°S, with a new rotation period of 16.76 hours. The "Scooter" then remained at this latitude for the remainder of the observations (Fig. 1A).
The Great Dark Spot drifted steadily northward at a rate of about 0.110 per day during most of the encounter period. During this time, its rotation period increased monotonically from 18.28 to 18.38 hours. For comparison, Jupiter's Great Red Spot displayed sinusoidal motion in longitude with a 90-day period and peak speeds between -1 and 5 m s-1 (17).
The Great Dark Spot exhibited a "rolling" motion around its circumference; the motion is anticyclonic and is best seen in timelapse sequences (Fig. 4A). The boundary between the darker blue of the oval and the lighter blue outside the oval (Fig. 1A) changed shape as if a two-lobed structure were rotating inside it. Every 10 or 11 rotations of Neptune the long axis of the structure pointed east-west, and the GDS reached its maximum longitudinal extent. This configuration appeared twice during each rotation of the structure, whose fundamental period is therefore about 21 Neptune rotations (21 x 18.3 hours at the latitude ofthe GDS, or approximately 16 Earth  days).
A rough estimate of vorticity within the GDS is 4'r divided by the fundamental period, or 0.9 x 10-5 s-1. In contrast, the vorticity of the ambient shear flow at the latitude of the GDS (assuniing a difference in velocity of 100 m s-over 10* oflatitude) is 2.3 x 10-5 s-1. The vorticity ofthe GDS appears to be less than that of the ambient shear flow-the opposite situation from that of the Great Red Spot of Jupiter.
The interpretation of the rolling motion as wind (that is, fluid motion) is ambiguous. For example, it may represent the propagation of a wave around the edge of the Great Dark Spot. Associated with the rolling motion, the Great Dark Spot exhibited other morphological changes. For example, dur-Flg. 4. Time-lapse sequences ofthe Great Dark Spot (GDS). (A) The "roling motion ofthe GDS can be seen in this 32-rotation sequence ofclear-filter images (top to bottom, left o right). The longitudinal extent ofthe GDS varies from 20 to 400 during a roll, while the latitudinal extent.can vary by 5°. The last three images are the first three images ofFig. 4B. Black regions on the right and in the upper righthand corners ofsome images are regions where no data were available. (B) This seveni-rotation sequence ofviolet images shows the dissipation ofa large western extension into a "string ofbeads" which moves westward relative to the GDS. The first image was taken on 13.3 August 1989, and the rotation period ofthe GDS is about 18.3 hours. Each imagc is centered near 22°S and extends ± 15°north and south of this latitude. Each frame includes 60°of longitude. (C) These six images represent our best highresolution tine sequence ofthe GDS (top to bottom, left to right). The first image was acquired at 20.3 August 1989. The time interval between images is one rotation. In the last two frames, a subde dark band appears extending horizontally across the center ofthe GDS. Each image extends in latitude from 7°S to 370S and covers 600 of longitude. Neptune. This color-enhanced image was created with images taken on 22.7 August 1989 through the orange, MeU, and MeJ filters. The images were assigned to the blue, green, and red channels of the false-color image, respectively. High clouds appear red, while low clouds appear blue. The companion to the GDS is bright in all filters and is a high cloud, although its appearance here is distorted by saturation during image processing (as is the appearance of the northern cloud band near the western limb). The high-altitude haze layer shows up as a red region on the limb; the haze is transparent when viewed at normal incidence. Subtle shades ofblue and green may represent differences in altitude, optical thickness, or composition of the cloud particles. [Image processing by L. K. Wynn] ing one particular rotation, a series of dark cloud features (a "string of beads") developed after the maximum western extension of the Great Dark Spot occurred (Fig. 4, B and C).
Ground-based images have shown bright cloud features on Neptune for more than a decade (18). By comparing spacecraft images with near-simultaneous ground-based images, we identified the discrete bright feature seen this year in 890-nm groundbased images as the bright companion at 330S associated with the southern edge of the Great Dark Spot (Fig. 8B). The GDS itself is not visible in the ground-based images. The bright features seen in previous years in ground-based images were located at 30S in 1988 and 38°S in 1986 and 1987 (18,19). Since the Great Dark Spot appears to drift in latitude by about 0.10 per day (4), 1428 it may have moved significantly over the last 3 years. If this is indeed the case, observers may have been detecting the Great Dark Spot companion as it drifted north and south, giving the Great Dark Spot a lifetime of more than 5 years. All ground-based images obtained prior to 1985 showed multiple bright features in both the northern and southem hemispheres (18).
The Great Dark Spot was not detected in ground-based images obtained at 550 nm, but a future search at 400 to 500 nm may have more success (see Fig. 10A). The Hubble Space Telescope should have sufficient resolution and wavelength coverage to detect the Great Dark Spot or any similar features. Other cloud systems that have been correlated with features in current groundbased images (6) are the south polar feature, the core of D2, some ephemeral bright streamers at the latitude of the Great Dark Spot and in the northern hemisphere, and the bright band surrounding the south pole (seen in MeJ images).
Zonal velocity profile. Hammel et al. (4) discussed the rotation period and zonal velocities of the four largest features visible in the Voyager images during the 80 days prior to encounter. Here we discuss the motions of smaller features observed during the last few days before closest approach. If the small-scale measurements and large-scale measurements were to disagree, we would suspect either that different altitudes were involved or that the patterns move relative to the wind, as with a propagating wave. At Jupiter and Saturn, the winds derived from tracking small features over short time intervals generally agree with those derived from the larger features when allowance is made for the tendency of larger features to move with the average flow in a latitude band. The measurement ofvelocity seemed to converge with smaller spatial and time scales. This The images were shuttered in temporal order: violet, blue, UV, clear, green, orange. These images show the bright core of D2, the south polar feature, and the symmetric structure immediately surrounding the south pole. The relatively high contrast of the features in these images indicates that they extend above most of the scattering haze and absorbing methane gas in Neptune's atmosphere. [Image processing by D.
A. Alexander] agreement suggests that cloud displacements are indicators of the wind at roughly the same altitude for both of these planets.
On Neptune, the small-scale features evolve rapidly and disappear quickly, so the interpretation of cloud displacements is less straightforward (Fig. 4C). Because of the rapid evolution of small-scale features, the best measurement strategy must be a compromise. On the one hand, reducing the time interval over which displacements are measured reduces the effects of evolution of the features, and this ensures that they are recognized on successive time steps. On the other hand, the error in velocity goes up as the time interval is reduced. This error is proportional to the resolution in kilometers per line pair divided by the time interval (in other words, the speed of a feature that moves 2 pixels in one time step). By using sequences of three or more images, we can shift the focus from one rapidly evolving feature to another while broadening the time base. Although the error is reduced by this technique, there is still no guarantee that one is measuring actual fluid motion.
Not all regions on Neptune have smallscale features suitable for tracking. At scales of50 to 200 km per line pair, there are eight general ypes ofdisernible cloud structures; (i) structure in the polar cloud feature at-71°S, (ii) variable brightening within the central region of D2 near 55°S, (iii) stnations in the "Scooter" at 420S, (iv) smal bright features similar to the "Scooter" that appear in the latitudinal range from 400 to 500S, (v) individual bright structures around the perimeter ofthe GDS, (vi) structure within the large-scale banding to the east ofthe GDS, (vii) details within the hazy bright equatorial patch located north of the GDS, and (viii) structure within the cloud bands at about 27N. Figure 13, A and B, gives the rotation periods and zonal velocities derived from both small-scale and large-scale features. Figure 13 reveals that measurements over different time intervals do not agree. The disagreement is particularly evident near 200S and 700S. The +'s and x's, for which the resolution per time step is less than 50 km per line pair, show less dispersion among themselves and better agreement with the solid curve. The diamonds, which use three or more images in a sequence, generally show less dispersion among themselves than the squares, which use image pairs only.
The measurements of the south polar feature (70°S in Fig. 13) provide an example of how the dispersion arises. The measurements fall into three groups. Those with rotational periods near 16.0 hours are derived from data with 49-hour time separa-tion. Those with periods near 17.5 hours have 16-hour separation, and those with periods from 12 to 14 hours are from images separated by less than 2 hours. This grouping suggests that the observations in the first set track the large-scale structure measured during observatory phase (4), while the observations in the short-interval group track small-scale features that move through the larger structures.
Simple measurement error attributable to limited resolution is not the main source for the dispersion, particularly when three or more images are used in sequence. The dispersion is of order 300 m s-1, whereas the error is only 50 m s-at 70°S (Fig.  13B). Wind shear with respect to altitude is one possible source. However, the infrared interferometer spectrometer (IRIS) observations of temperature as a function of latitude (20) suggest that the vertical wind shear is small, on the order of 30 m s per scale height. For wind shear to be important, the structures must be distributed over a range of altitudes extending ten scale heights. The other possible source for the dispersion is propagation, either of the large-scale features or the small-scale features, relative to the fluid.
One interpretation is that the diamonds in . Contrast of the Voyager images was increased by a constant amount (a factor of -2.5). The ground-based image was stretched with a nonlinear function to enhance the doud feature relative to the disk. Quantitative measurements ofthe relative brightnesses ofspecific regions are shown in Fig. 1OA. Fig. 13 represent the true motion of the fluid, making Neptune the windiest planet in the solar system. Wind speeds would then be close to the speed of sound, which is about 560 m s-' at T-60K (Fig. 9). The solid curve might represent the slower motions at deeper levels or perhaps a large-scale wave. The other interpretation is that the diamonds and squares represent short-lived features that propagate relative to the flow. The solid curve, perhaps including the points at 5°N with periods near 19.5 hours and those at 27°N with periods near 17.5 hours, would then be our best estimate of the zonal wind. This is the most conservative interpretation, since the velocities of the large-scale features are the most certain.
Regardless of the interpretation of the points in Fig. 13, the periods of rotation at the equator are longer than both the radio period (3) and the periods at high latitudes. This fact puts Neptune in a class, with Uranus and Earth, ofequatorial subrotators. Venus, Jupiter, Saturn, and the sun are all equatorial superrotators, since the periods of the equatorial atmospheres are shorter than between the angular velocity of the interior and that of the equatorial atmosphere. In general, subrotation at the equator is easier to understand than superrotation. Contrast is defined as (IJlIb -1), where If and Ib are the brightnesses of a feature and its surrounding background, respectively. For the Great Dark Spot, two regions were measured, the central region and the outer edges. The contrast shown here for the bright companion was measured in low-resolution images. In higher resolution images, the feature was resolved into thin streaks (Fig. 3A), each ofwhich has even higher contrast than that measured for the overall feature. The Voyager filter passbands are indicated along the bottom axis. (B) Reflectivity as a function ofwavelength for the main cloud deck at different latitudes. The three curves show the wavelength-dependent absolute reflectivity (I/F) at latitudes 22°S, 33°S, and 42°S. The regions are at the same latitudes but different longitudes as the GDS, its bright companion, and the "Scooter," respectively. The feature contrasts shown in Fig. lOA were measured relative to these values. There are no significant differences between the measured reflectivities at these latitudes. Small differences in the reflectivity ofother latitude regions give Neptune a banded appearance. The Voyager filter passbands are indicated along the bottom axis.  Fig. 9. Vertical aerosol structure of Neptune's atmosphere. The zero of altitude corresponds to the 4-bar pressure level. The existence of the main cloud deck near the 3-bar level was first inferred from ground-based spectroscopic observations of hydrogen quadrupole lines (9). The large tropospheric concentrations of methane gas derived from ground-based visible and infrared observations suggest the presence ofa methane condensation cloud near the 1.2-bar level, but there is little direct observational evidence of a global methane cloud at this level. The bright cloud features seen in Voyager images may provide evidence for smaller-scale methane condensation clouds there.
The existence of the high-altitude hydrocarbon aerosol layers was first predicted by theoretical models of Neptune's atmospheric chemistry (13). Brightness variations seen in limb scans of highphase-angle Voyager images also suggest the presence of these aerosols. Without dissipation, rings of fluid circling the planet at a given latitude tend to conserve their angular momentum as they move from one latitude to another. Maintaining equatorial subrotation merely requires taking rings of fluid from higher latitudes and moving them to the equator. Maintaining equatorial superrotation requires some pumping of angular momentum into the equator by organized waves or eddies. The stability of a zonal flow depends not only on the velocity profile, but also on the density structure of the atmosphere. At the moment, we do not know the density field of Neptune's atmosphere over a sufficient range of depths and latitudes to do a comprehensive stability analysis. One necessary condition for stability, however, is that the angular momentum per unit mass, Qr2 cos2p, should decrease monotonically from equator to pole. Here Ql is the atmospheric angular velocity, r is the planetary radius, and up is the latitude. On Neptune, as on Earth and Uranus, the decrease of cos2wp outweighs the increase of Ql with latitude, and the condition is satisfied. This conclusion holds for all interpretations of Fig. 13 for which the equatorial period is less than 20 hours (21).
One of the most remarkable aspects of Neptune's zonal wind profile is its similarity to that found on Uranus (1). That these two planets, which have such different internal energy sources and such different obliquities, should have the same pattern of zonal winds requires an explanation that will certainly add to our understanding of atmospheric dynamics.
Fig. 13. Rotation periods (A) and zonal velocities (B) as a function of latitude. Velocity is measured with respect to the 16. 1-hour period derived from the planetary radio emissions (3). The solid line represents the motion of the largest features over the longest time intervals (4). It is composed of nine individual measurements, two from groundbased observations and seven from the four largest features seen in the early Voyager images (3 of these were measured at two different latitudes). The symbols are measurements of individual small-scale features, often at time steps less than 2 hours and resolutions better than 100 km per line pair. The diamonds and crosses are values calculated with software developed at the University of Wisconsin in which displacements are measured in a sequence of three or more images. The squares and +'s use software developed at JPL in which displacements are measured in pairs of images only. The resolution (in km per line pair) divided by the time step (in hours) is less than 50 for the crosses and +'s and is greater than 50 for the diamonds and squares. The error estimates are in the same units as the figures and are computed from the statistics of the observations.

The Neptune Ring System
Prior to the Voyager reconnaissance of Neptune, very little was known of its ring system. By August 1989, about 50 stellar occultations had been observed from ground-based observatories, representing 100 separate scans through the system. Although more than 90% of these observations yielded no detection, at least five occultations observed between 1981 and 1985 demonstrated with high confidence the presence of material in orbit around Neptune (22).
One of these observations was a detec-  Zonal wind (m/s) Error tion, observed by more than one telescope, of either a small satellite or an optically thick, azimuthally incomplete ring of about 80 km radial width. Several other detections, two ofwhich were confirmed, indicated narrower (approximately 15 to 25 km) but also discontinuous rings. Minimum inferred lengths were 100 km. Thus, the existence of relatively narrow "ring arcs" orbiting between 41,000 and 67,000 km from Neptune was commonly accepted. On the basis ofthese ground-based observations alone, however, it was impossible to distinguish between a family of permanent or transient arcs around Neptune or continuous rings of highly variable optical depth. Neptune's ring features became members of a sparsely populated class of ring structures including the narrow, azimuthally incomplete rings in Saturn's ring system [in the Encke gap, the Cassini Division, and around the F ring; see, for example, (23)] and possibly in the Uranian ring system as well (24).
The presence of short evolutionary timescales in ring systems is a well known and asyet unsolved puzzle (25). The time required for longitudinally localized material 20 km in radial width to spread 3600 as a result of differential rotation is only about 5 years. Stable, non-transient ring arcs would obviously require a longitudinal confinement mechanism (26,27); transient arcs require the continual creation, dispersal, and replenishment of local concentrations of ring material. We present here the initial results on the nature and dynamics of Neptune's rings from analyses of Voyager imaging observations, discuss the various theories suggested for the existence of arcs in light of our findings, and compare Neptune's system REPORTS 1431 with the other ring systems we see in the outer solar system. Radial distribution ofring material. The Neptune ring system, as seen in Voyager images, contains two narrow rings, 1989N1R and 1989N2R, at radial distances of 62,900 km and 53,200 km, respectively; a broad ring, 1989N3R, at a radial distance of 41,900 kin; a second broad ring, 1989N4R, extending outwards from 1989N2R to a distance of nearly 59,000 km; and an extended sheet of material that may fill the inner Neptunian system. (We will alternatively refer to 1989N1R, 1989N2R, and 1989N3R as the N63, N53, and N42 rings.) The N63 ring is outermost and includes three arcs of substantially greater optical depth than the ring average. The three arcs are clustered together within a total range of 33 degrees in longitude. 1989N1R and 1989N2R lie about 1000 km outside the newly discovered satellites 1989N3 and 1989N4, respectively.
Neptune's rings were most easily visible at the moderately high phase angles obtained after closest approach to the planet, and three images at these phase angles best characterize the overall distribution of material. Figure 14, a 111-second exposure, and Fig.  15, a composite of two 591-second exposures taken 1.5 hours apart, were all imaged through the clear filter of the wide angle camera and show the ring system at forward scattering phase angles of about 1350. Figure 14 clearly shows the arcs which are 12 data number units (DN, out of a total range of 255 DN) above background. In comparison, Fig. 16 is also a 111-second exposure taken through the clear filter of the Voyager wide angle camera, but at a phase angle of only 15.50. It shows the same three arcs within the optically thin N63 ring; the second, even fainter N53 ring is also visible. Though Fig. 16 is one of the best Voyager images taken of Neptune's ring system at low phase, the average brightness in the arcs measured only about 2.5 DN above background. The arcs were not captured in Fig. 15 because of the 50 degrees of orbital motion between the two frames. However, one can easily see the N63 and N53 rings as well as Clearly seen are the three ring arcs and the N53 and N63 rings. The direction of motion is clockwise; the longest arc is trailing. The resolution in this image is about 160 km per line pair; the trailing arc was imaged at this same time at a higher resolution of about 20 km per line pair (Fig. 18). 1989N2 can be seen in the upper right corner, streaked by its orbital motion; the other bright object is a star. The arcs are unresolved in this image (resolution of 37 km per line pair); their apparent width is due to image smear. N42 (1989N3R). This latter ring, which is seen at low phase angles only with great difficulty, is clearly resolved and has a full width at half maximum of about 1700 km. Also faintly visible in Fig. 15 is a sheet of material beginning midway between the two outer rings at approximately 59,000 km and possibly extending down to the planet. (We discuss this further below.) There are identifiable features within this sheet. The most prominent among them is the plateau, 1989N4R. A distinct feature or ring at 57,500 km, 1989N5R, on the outer edge of the plateau, can be seen above and below the ansa in Fig. 15. In addition, there are hints of other radial structure in the plateau.
Although one gets the impression from Fig. 15 that material extends continuously inward from 59,000 km, this is difficult to confirm because of the uncertainty in the distribution of scattered light. Figure 17 is a radial profile of the intensity seen in the left hand frame of Fig. 15. In this scan, a smooth function has been subtracted to remove the scattered light from the planet; this has the effect of tapering off the brightness distribution to zero at small radii. Though this removal of scattered light is uncertain, the N42 ring does appear to be embedded in material which extends out to the N53 ring, with brightness only slightly less than that of the plateau region but having a local minimum at 52,000 km, approximately the orbit of 1989N3. A similar configuration was observed in the Uranian system between Cordelia and the A ring (1986U1R) (1).
In other high-phase frames, a narrow, clumpy ring (as yet unnamed) is also visible just interior to 1989N1R. This feature is not seen in any low phase angle images, and appears to lie at about the same radius as satellite 1989N4. Imaging observations of the arcs and the satellites within the ring region indicate beyond doubt that the direction of orbital motion is prograde. Because of remaining uncertainties in the Laplacian plane pole orientation, we cannot at the present time completely rule out very small ring eccentricities and inclinations relative to this plane, which presumably is identical to that of the inner, regular satellites.
Voyager observations have put strong constraints on the existence of a potential polar ring system (28); several high and low phase angle images of Neptune's north and south polar regions, covering a radial range out to several hundred thousand kilometers from the planet, reveal nothing. Although the upper limiting optical depth for broad sheets of material is roughly 10-5, it is still impossible to rule out the existence of narrow clumps in polar orbits with somewhat higher optical depths. 15 Fig. 17. This radial profile ofthe brightness ofthe rings as seen in Fig. 15, in units of DN (data number) above background as a fimction oforbital radius, was obtained by averaging all points lying within narrow radial bins and excluding the stars. A smooth background has been subtracted out to account for glare from the planet; since the true background is uncertain, it is not possible to say whether the rings extend all the way in to the planet.
High-resolution imaging coverage of the rings, excluding the arc retargeting that is discussed below, was made mostly at sky plane resolutions of 15 and 40 km per line pair. Due to the smear resulting from the long exposures, we have been unable to resolve in the radial direction either the N53 ring or the non-arc part of the N63 ring. Our highest resolution images, obtained within 13 hours ofclosest approach, were of the arcs and were retargeted approximately 4 days before closest approach, based on earlier imaging observations of arc locations and orbital motion.
The single highest resolution frame, FDS 11386.17, has a sky plane resolution ofonly 3.0 km per line pair. At the geometry of this observation, radial foreshortening results in a ring-plane resolution of 15 km per line pair. The trailing arc has a full radial width at half maximum that is close to the resolution of the camera. However, in our narrow angle outbound retargetable image (Fig.  18), measurements of variations in the ring width indicate that the ring may be just barely resolved, implying a radial width of about 15 km. This value is in good agreement with the groundbased and Voyager photopolarimetry stellar occultation measurements (22,29).
Longitudinal distribution ofmaterial. On longitudinal scales of a radian, all three rings and the plateau region are continuous around the planet. This is very clearly seen in forward-scattering geometry (Fig. 15), but it is also evident, though with greater difficulty, in backscattered light for the N53 and N63 rings. On intermediate scales, the only prominent structures are the three bright arcs seen in the N63 ring (Figs. 14 and 16). The azimuthal lengths of these arcs, measured to be the distance between the half- intensity points in azimuthal brightness scans in Fig. 14, are approximately 40, 40, and 100 for the leading, middle, and trailing arcs, respectively. The distances between the midpoints of these features are about 140 (leading-middle) and 120 (middle-trailing). To within the measurement uncertainties, the same values are obtained from low phase images. Given the signal-to-noise ratio of the data, we find no convincing evidence at this time for longitudinal structure on scales greater than about 50 in either of the two remaining rings or in the remainder of the outer ring in either forward or backscattering geometry.
Small-scale azimuthal structure is most easily seen in the retargeted image of the trailing arc taken at high phase (Fig. 18). Several long linear features are apparent, which we believe are formed by discrete clumps in the ring, trailed out by a combination oforbital motion along the ring and the motion of the spacecraft across the ring.
These features are unresolved and appear to be associated with microscopic particles because of their enhanced brightness in forward scattered light (phase angle of 135 degrees). One of the retargeted images taken of the trailing arc at low phase shows structure similar to that seen in Fig. 18 when the image contrast is strongly enhanced. Though it is not yet certain if these are the same clumps seen in Fig. 18, it is noteworthy that the typical separation between these clumps, approximately 0.10 to 0.20, is the same in both frames. A similar object within the leading arc is seen in another low phase image not reproduced here. These features may be large embedded ring particles or associated clumps of debris similar in morphologv to the discrete features found in the F-Ring or Encke gap ringlet of Saturn (23).
Moon/let search. A search for additional small satellites orbiting near the rings of Neptune is still being conducted (30), but the lack of confirmed sightings implies that few or no additional satellites larger than 12 km in diameter with assumed geometric albedo of 0.05 are orbiting in the ring region. To discriminate actual sightings from noise, we required each candidate object to be in a circular, equatorial orbit. Because of this requirement, our limiting radii are roughly twice as large for satellites with orbits inclined by more than 100 or eccentric by more than 0.1.
Ritng photometry atnd particle properties. The ability of spacecraft to observe a system over a range of viewing angles is of great importance to our understanding of planetary ring systems, since the scattering behavior of particles of microscopic and macroscopic sizes differs dramatically with phase angle. Observations at high phase angles (>1500) have been important in establishing that the microscopic "dust" particle fractional area in the main rings of Saturn and Uranus is quite small-between 0.01 and 0.001 by area (31). In other rings, the dust fraction can be considerably larger; for example, Saturn's F ring, a narrow, clumpy ring, and Saturn's E ring, a broad, diffuse ring, are both visible primarily because of microscopic particlesdust fractions greater than 80% (32). In the case of Neptune, the spacecraft trajectory did not allow any extremely high phaseangle observations; however, the rings themselves compensated for this lack of observational sensitivity by being extremely "dusty." Our preliminary photometric analysis relies on only twvo phase angles (approximately 140 and 1350) for the following regions: the middle arc of 1989N1R, wide azimuthal averages in the N42, N53, and N63 rings (excluding the arc material) and the constant brightness region ranging between 54,500 '434 and 57,500 km from Neptune. In order to suppress the effects of possibly variable smear in the long exposures that were used, the radial integral of the brightness profile, or "equivalent width," was employed (33). Figure 19 shows the observations for the different regions, converted to the quantity (,PfTdr, as functions of phase angle (33). The fact that all regions are more reflective at high phase angles is direct evidence for a substantial dust population, since the brightness of macroscopic objects depends primarily on the fraction of the visible illuminated area and decreases by an order of magnitude as phase angle increases over this range. We have modeled the particle properties in these regions to obtain the optical depth of macroscopic and microscopic material, using assumptions as to the individual particle properties based on prior experience. We have assumed that the large macroscopic particles in the Neptune ring system have the Uranus ring particle phase function and a Bond albedo between 0.01 and 0.02, which brackets the Uranus ring particles, Phobos, Diemos, and Amalthea. This com-  19. In this figure, observations of the radially integrated brightness of different regions of the Neptunian rings are shown as functions of phase angle. The brightness has been converted to the "equivalent width" product .a0PfTdr, in units of kilometers, where fTdr is the "equivalent depth" and T is the normal optical depth. The regions are the N42, N53, and N63 rings (not including the ring arcs), the plateau region averaged between 55,000 and 58,000 km, and the middle ring arc (plotted at one-third of its actual values). bination results in a geometric albedo of about p = 0.05, similar to that seen for the newly discovered Neptune satellites.
To bound the properties of the microscopic dust component, we chose coal dust (which is used to model the Uranus rings) and a typical silicate (which is used to model the Jupiter ring). Mie scattering, as modified by a simple irregular particle algorithm, was used to obtain the range of dust particle albedo and phase functions for material with this range of composition. We assumed a power law size distribution with index of 2.5, such as has been observed to characterize the dust in the Uranus and Jupiter rings; microscopic dust in ring systems tends to have a somewhat flatter size distribution than typical comminution products due to the size dependence of removal processes. The results of this preliminary photometric modeling are shown in Fig. 20.
The N42 ring and the plateau are clearly low optical depth structures, although two orders of magnitude more substantial than the Jupiter ring or the E and G rings of Saturn and about one order of magnitude more substantial than the Uranus dust bands. The optical depth of the middle arc, about 0.04 to 0.09, is in excellent agreement with ground-based values when appropriate diffraction corrections are made. Within the uncertainties, the N42, N53, and arc regions have the same dust fraction (0.5 to 0.7), which is about twice as large as the fraction found in the N63 ring and the plateau regions, and significantly larger than found in the main rings of Saturn or Uranus (10-3 to 10-2).
Comparison ofground-based and Voyager ob- A precise determination of the arcs' mean motions must necessarily await careful arc position measurements in images spanning as large a temporal range as possible, and the fitting of such measurements with a general model describing an eccentric and inclined orbit, as well as an evaluation of the pole of Neptune's Laplacian plane, which is independent of the assumed ring arc model. A preliminary value was obtained, however, by assuming the arcs' orbits to be identical, circular, and equatorial [Neptune rotational pole orientation, (a, 8)1950 = (298.904, +42.841)], and by measuring the beginning and ending longitudes of each arc in smoothed, radially averaged, azimuthal scans taken from at most five images spanning a total of about 6.5 days. For each of the three arcs, a weighted least-squares fit to the measured locations versus time was used to determine the mean motions; the final value and its uncertainty, 820.12 ± 0.06, are the mean and the standard error of the mean, respectively, of these three numbers. Using this value, we projected forward from the longitudes of arc detection in the three reliable ground-based stellar occultations in the N63 region mentioned above (35) Fig. 14. (Although the ground-based longitudes are measured in the Neptune-Triton invariable plane, and the Voyager longitudes in the Neptune equator plane, the difference amounts to at most a few degrees.) We consider these results convincing evidence that the Voyager arcs themselves were the occulting material for all ground-based occultations at this radius and, therefore, that the arcs are stable over intervals of at least 5 years. Future work in this area, taking into account the geometrical considerations mentioned above, may in fact allow us to determine with high probability which of the three arcs was observed in each of these three ground-based occultations and to predict the locations of these features at the times of upcoming Neptune stellar occultations observable from the ground or from the Hubble Space Telescope. It is not surprising that ground-based observations, in general, do not detect material in the non-iS DECEMBER i989 Fig. 20. In this figure, we have estimated the total optical depth and fraction in microscopic dust particles from the data of Fig. 19, making certain assumptions as to the phase function and albedo of the ring particles. The total normal optical depth, plotted vertically, has been obtained from the equivalent depth by dividing by a physical width of 15 km for N53, N63 and the arc, by the 1700-km full-width at half maximum for N42, and by the 3000-km width of the plateau. The fraction of this total represented by "dust" is plotted horizontally. In all cases, the patches are bounded on the lower right and upper left by large particle single scattering albedos of 0.02 and 0.01 respectively, and on the upper right and lower left by microscopic material composed of coal and of silicates, respectively. The large particle phase function and the dust particle size distribution are those ofthe Uranus ring particles. Although these specific values are clearly modeldependent, the relative differences between the regions (that is, corresponding corners of the patches) are real. arc regions of the N53 and N63 rings-the estimated optical depth of 0.01 to 0.02 in these regions is below the ground-based threshold. However, it is worth noting that (unconfirmed) ground-based stellar occultation observations, revised by using Voyager improvements for the Neptune pole orientation (35), indicate events at around 42,000 kilometers and 55,000 kilometers-opening the possibility that clumpy material may now, or did then, exist in regions where we now observe broad, diffuse belts of material (see Fig. 21).
One other important result concerns the identity of the object that was responsible for the first occultation observation of material orbiting Neptune (22). Using the mean motions for the small satellites determined from imaging observations, we obtained their positions at the time ofthe 1981 event. Since this event did not occult the planet, some astrometric uncertainty remains in addition to the uncertainty in the mean motion of the candidate satellites. However, we find that 1989N2 falls within the total uncertainty (about 8°of orbital longitude or less than 1 arc second) of the location of the event, and that the only other possible candidate (1989N4) is on the other side of the planet. This excellent positional agreement, combined with the fact that the 1981N1 occultation was completely opaque and 180 km across and 1989N2 has a diameter of about 200 km, makes us confident that 1981N1 and 1989N2 are one and the same object (Fig. 21).
Discussion of ring observations. Despite the singular nature of Neptune's system of relatively large satellites (in particular, retrograde Triton   rings, like the Uranus X ring (1986U1R) and the Saturn F ring; diffuse dusty rings, perhaps similar to the Jupiter ring and Saturn G ring; azimuthally confined arcs embedded with a ring, reminiscent of Saturn's F and Encke rings; and possibly a broad sheet of dust like that which is seen around Uranus at high phase angles. An examination of Neptune's system of satellites and its distribution with orbital radius (Table 1) supports the generality that, as the distance from a giant planet decreases, there is a gradual transition from large, isolated satellites to families of more numerous, smaller satellites and ring material. The presence of relatively massive objects (for instance, satellites at Neptune, rings at Saturn) well within the outer planets' Roche limits, where structural stability depends on internal strength but where accretion to radii of tens of kilometers is difficult or impossible, supports the idea that the parent objects of these outer planet ring-satellite systems migrated into their respective Roche zones from elsewhere long ago (37).
In comparing only Uranus with Neptune, we find that the reflectivities of the surfaces of their inner satellites are similar and very low. Also, the agreement between our derived Neptune ring arc optical depths and those obtained from ground-based stellar occultation measurements supports equivalently low albedos for the ring particles, comparable to those found for Uranus. Compositionally, therefore, the Neptune and Uranus ring-satellite systems appear to be quite similar, suggesting chemical origins and/or evolutionary histories that proceeded in tandem, histories that may well characterize the outer solar system beyond Saturn.
However, the dramatic differences also call for our attention. The amount ofmass in Neptune's rings is approximately 10,000 times less than that at Uranus and many orders ofmagnitude less than that at Saturn, yet the inner Neptune satellites are significantly larger, and presumably more massive, than the bodies in similar locations in the ring-satellite systems ofthe other giant planets. At Neptune, the five satellites 1989N2 through 1989N6, with diameters ranging from approximately 55 to 190 kin, all fall within the Roche "liquid" limit at roughly 77,000 km; at Uranus, the nine satellites falling within its Roche limit, 1986U1 through 1986U9, range from about 25 to 110 km in diameter (38). If we assume the overall satellite-size distributions to be similar among the outer planets, the relatively large number ofbig bodies close to Neptune might lead one to expect a proportionately large abundance of smaller moonlets in the ring region. Yet, our search for satellites to date does not support the existence of more than two objects (1989N5 and 1989N6) with diameters less than 100 km.
Evidently, the distribution ofmass among the inner satellites of each planet, and between each planet's satellites and rings, is very different. The absolute amount of mass within these zones is also notably different: All the mass in Saturn's rings, which fill Saturn's Roche zone, can be contained within an icy body approximately the size of Mimas, 195 km in radius; in Neptune's Roche zone, the rings' and satellites' masses can be contained within an icy body 130 km in radius; and for Uranus, within an icy body 75 km in radius. It is interesting to compare these differences with the variation in present-day cratering rates on the inner satellites of these three planets: The ratio for Saturn/Neptune/Uranus is roughly 3/50/100 (39). It would appear that where bombardment is greatest, there is less overall mass. However, the present-day differences in the distribution of mass around the giant planets may reflect, in part, the varying degrees to which bombardment and collisional processes have combined to shape their ringsatellite systems.
The presence and distribution of dust in these systems may provide a direct indication of the relative importance of these processes today. The relatively large number of microscopic particles spread throughout the Neptune rings (Fig. 20) is not unique; the Jupiter ring and the Saturn E ring contain a fractional optical depth of 50% to 80% in dust. However, the absolute abundance of dust in the Neptune system, which is about two orders of magnitude larger than the Jupiter and Saturn counterparts, presents a serious problem. Because microscopic particles are very short-lived (40), they must be continually replenished. When material is in a state of dynamic balance, equilibrium abundances are maintained by equal rates of creation and destruction. Different removal processes dominate in different environments: In the Uranus rings, microscopic material is removed primarily by gas drag (41) and, in the Jupiter rings, by plasma drag (40). In the Neptune rings, in which dust has a relatively large optical depth (approximately 10-4) and which are relatively free of plasma or neutral gas, simple sweep-up on the surfaces of macroscopic particles dominates dust removal.
If the source of the dust is meteoroid bombardment, as believed for the Jupiter ring, the creation and removal processes are both proportional to parent body optical depth. Consequently, the dust optical depth resulting from meteoroid bombardment is independent of the large particle optical depth and depends instead on the meteoroid flux and impact yield parameters (40). For heliocentric distances less than 15 to 20 AU observed by the Pioneer 10 and 11 dust detectors, a value of interplanetary meteoroid flux of about 10-16 g cm-2 s-' is generally accepted. Given this value and ejecta yields of about 104 [see, for example (42)], "Jupiter ring" dust optical depths on the order of 10-6 are easily obtained. However, dust optical depths in N42 and the plateau (and the Uranus dust bands) are around 10-4 (Fig. 20), requiring a bombarding flux roughly two orders of magnitude larger than found at Jupiter and Saturn. This larger dust abundance is qualitatively consistent with the previously mentioned larger estimated projectile population at Uranus and Neptune. Although the estimate falls short by a factor of 3 to 10, this may be within the uncertainty in the Neptune dust optical depths and in the estimated projectile populations.
However, the dust within the N53 and N63 rings is orders ofmagnitude larger than can be explained by the meteoroid bombardment mechanism. Therefore, we suspect that it is most likely generated locally through vigorous collisions between larger, unseen particles. In this mode, the equilibrium dust optical depth depends in a more complicated and model-dependent way on the optical depth of the parent bodies that create it. The dust optical depth does tend to increase with that of the colliding parents as long as the latter is much less than unity and the mass injected per collision is a constant. This creation of dust through interparticle collisions, assuming relative velocities con-sistent with radial excursions as large as the observed ring widths, may explain the generally large dust abundance in the Neptune rings, given reasonable assumptions about the yield per impact (43). The brightening at 57,500 km (1989N5R) near the outer edge of the plateau is one Neptune ring feature which, by analogy with Uranus, may be the manifestation of a moonlet belt (1). The lack ofother noticeable fine structure, in contrast to the nearly 100 belts revealed in the Uranus rings or the internal structure of the Saturn D ring, may be due to long exposure times of the images and the corresponding smear rather than to actual absence. Of course, greatly increased optical depth will diminish the dust population because the attendant large collision rate tends to damp the relative velocities. For instance, in the large optical depth rings of both Saturn and Uranus, the dust fraction is extremely small. For this reason, it is not clear whether the amount of dust in the Neptune arcs is consistent with a relatively simple moonletbelt model or whether additional stirring of the ring arc material by unseen perturbing bodies is implied. Nonetheless, it appears overall that the moonlet-belt hypothesis that has been proposed for the Uranus rings (1) and recently modeled in more detail (44) may go a long way toward explaining some of the global characteristics seen in the Neptune ring system.
While it seems clear that the 1989N1R arcs are stable over an interval of at least 5 years, a search for dynamical relationships between the Neptune rings, ring arcs, and the new satellites found within the ring region has demonstrated that none of the current hypotheses based on the combined action of satellite corotation and Lindblad resonances can explain the persistence of the arcs. Confinement of the ring arcs through a combination of corotational and Lindblad resonances with a single satellite (27)  ring, which might account for the locations ofthe inner edges of these rings in much the same way as Cordelia shepherds the inner edge of the Uranus e and probably X (1986U1R) rings (45). Though both inertial and acoustic waves in Neptune were proposed as a possible mechanism for the azimuthal confinement of arc material, these mechanisms can now be discounted because the required planetary wave amplitudes would have to be impossibly large to produce arcs with the observed longitudinal scale of 120 (46). Moreover, acoustic mode corotation resonances at Neptune do not fall outside 28,000 km (47). At the present time, therefore, there are no theories explaining ring arcs that are verified in Voyager imaging data. The effort to date that has focused on the details of a longitudinal confinement mechanism for the arcs, while now clearly justified, does not address the larger question of the origin of rings and ring arcs themselves. This question will require studies of catastrophic disruptions and the subsequent dispersal and distribution of collisional fragments, as well as the study of the coupled behavior of ensembles of moonlets and ring material, both under the influence of a variety of ongoing processes, like tidal evolution and meteoroid bombardment, which continually sap orbital energy and angular momentum from the system. However, knowing as we do now the basic properties of the four ring-satellite systems of the outer solar system, we can begin to explore comprehensive models of their undoubtedly complex evolution.

The Satellites of Neptune
Prior to the Voyager 2 encounter, Neptune's known satellite system consisted of one large retrograde satellite, Triton, a smaller satellite, Nereid, in a direct but highly eccentric orbit, and the tentatively identified satellite (22) in the vicinity of the ring arcs. Triton was discovered in 1846 by Lassel, immediately following the discovery  (48). Of the six newly discovered satellites, two (1989N1 and 1989N2) were imaged with sufficient resolution to study their surfaces.
Its proximity to Neptune and low relative brightness make ground-based observations of Triton difficult; many of its fundamental properties (diameter, albedo, mass, atmospheric pressure) remained obscure prior to Voyager's encounter (49). Spectroscopic studies established the presence of methane (CH4) ice or gas (or both) but showed no water-ice absorption features which are so prominent in the spectra of most outerplanet satellites. These studies also tentatively identified molecular nitrogen (N2) in the gaseous and condensed states on the basis of a weak absorption band at 2.15 ,m (50).
Triton's physical properties. Triton's radius (1350 + 5 km), determined from limb measurements on the Voyager images, and mass, obtained from analysis of radio tracking data (14), yield a density of about 2.075 ± 0.019. With the exceptions of the rocky satellites Io and Europa, Triton's density is the highest observed for an outerplanet satellite and is very similar to that of the Pluto/Charon system; see Fig. 22 (51,52).
If these bodies are assumed to be composed chiefly of silicates and water ice, the rock/ice mass fractions can be derived from the bulk density by using interior structure models (53). Two extreme cases are examined here: differentiated (silicate core and ice mantle) and homogeneous (uniform mixture ofsilicate and ice or clathrate). A silicate density of 3.361 g cm-3 was used (54); a "chondritic" value of 3.6, used in some other studies, would result in slightly lower silicate mass fractions than those quoted here.
In the case that Triton was melted and  Table 2, column 1). As the interior cooled, a layer of Ice-II may have developed due to the very low surface temperature. The existence and thickness of such a layer strongly depend on the details of the Ice-I/Ice-II phase boundary, poorly known for these low temperatures. Ice-II also will not be present if the ammonia hydrate/water eutectic is reached and some of the interior is partially molten. Column 2 of Table 2 gives the characteristics of models for a mixture of Ice-I and Ice-II by means of the phase relation given by Lupo (54). A homogeneous undifferentiated model ( Table 2, column 3) would result in a layered structure of silicate mixed with Ice-I, Ice-II, and Ice-VI in the deep interior. The differentiated models are clearly the most plausible. Triton is large enough that radiogenic and accretional heating alone likely resulted in differentiation. In addition, if Triton is a captured satellite, tidal evolution of its orbit also would have produced significant heating and melting (55,56).
Triton's extensive resurfacing is also inconsistent with an inactive, homogeneous interior.
Triton's estimated silicate fraction is higher than those of the large icy satellites of Jupiter, Saturn, and Uranus, but similar to that of the Pluto/Charon system; see Fig. 23 (53). This is consistent with the hypothesis that Triton was formed in the solar nebula and subsequently captured by Neptune (55). These high densities have important implications for the carbon chemistry ofthe outer solar nebula. If the nebular carbon were mostly in CO gas, the H20 abundance would be depressed relative to that of the silicates (53). Modifications to the simple concept of CO-rich versus CH4-rich nebulae may be required, however, in response to recently suggested revisions of the cosmic carbon abundance (57) and to the lack of detected CO on Triton or Pluto.
Another important aspect of these models is the relatively high interior temperatures expected, even at the current epoch.  (53,54). Two types of satellite interior structure are given, a fully differentiated body with a silicate core and an ice mantle, and a homogeneous undifferentiated mix of water ice and silicates. Radiogenic heating is taken into account in the current thermal structure of the interiors. Also shown are approximate values for silicate mass fraction for bodies formed in CO-rich and CH4-rich nebular conditions and the system averages from several satellites as compiled in (51 would be reached at a depth of only 200 to 300 km; the melting point of water ice would be near the core-mantle boundary. Convective heat transport in sub-solidus convection would cool the interior more rapidly than conduction alone, and more complicated models need to be considered. Local topographic relief on cliffs, ridges, knobs, pits, and craters commonly exceeds -1 km over most of Triton's surface. This implies that a rigid material, which would not flow at the 40 to 50 K near-surface temperatures over billions of years, is required to support them. The rheologies of solid N2 and CH4 are not well known at these temperatures but it seems unlikely that a 1-km-high cliff could be supported over geologic time in a material held at half to two-thirds of its melting temperature. Such relief could easily be supported in water ice or water-dominated ammonia-water ice, however. We suspect that water ice is in fact the primary component of the near-surface crustal materials overlain by thin veneers of nitrogen and methane ices and their derivatives. Triton spectrophotometry. Clear-filter, narrow-angle images acquired during the last 2 weeks of approach were analyzed to determine Triton's light curve (Fig. 24). Integraldisk brightness measurements were scaled to constant distance from Triton and corrected for the disk-averaged phase curve described below. Confirmed by preliminary geodetic control measurements, these data indicate synchronous rotation. A Cassini state 2 rotation state (rotation axis not perpendicular to the orbital plane) was suggested as a theoretical possibility based on dynamical arguments (58) but is ruled out by the data. The light curve agrees well in magnitude and sense (leading side brighter-90°longitude, Figs. 24 and 25) with broadband V filter measurements, but telescopic measurements at 890 nm show no light curve larger than about 2% (59). This suggests that the surface contrast at 890 nm is less than at visible wavelength and implies that the chromophores (possibly silicates or hydrocarbons or both) responsible for albedo and color variations on Triton are more absorbing at shorter wavelengths.
Triton was imaged at phase angles between 12 degrees and 156 degrees through the violet, green, and clear filters. These data are combined with ground-based V bandpass data (similar to the green filter in effective wavelength) in Fig. 26. Also shown are fits to the green and violet data using Hapke's photometric model; see (61) and Table 3. The phase curves display remarkable wavelength-dependent differences. At small phase angles Triton is brighter in the green filter than in the violet; at large phase  Table 3. Triton photometric parameters. The V band value is from a ground-based measurement (60). The data were derived from clear-filter narrowangle camera. The period of revolution was assumed to be synchronous.
angles the reverse is true. Clear-filter data are intermediate.
At large phase angles (>150°) the violetand green-filter data both deviate from the fitted Hapke functions, with the deviation being larger in the violet filter. This may result from the effects of atmospheric scattering at large phase angles, which are not accounted for in our models. Atmospheric scattering is expected to be greater at large phase angles and at shorter wavelengths. The more positive value of g (indicative of forward scattering particles) for the violet filter Hapke fit is consistent with the presence of atmospheric scattering.
Triton's global-averaged, single-scattering albedo is among the highest of the outerplanet satellites studied to date. Only values for Enceladus (w 0.998) and Europa (w -0.97) are comparable (62). The Hapke parameters can be used to estimate violetand green-filter geometric albedos (p), phase integrals (q), and spherical albedos (A = pq). These values are given, along with the ground-based V filter value ofp in Table 3. The greater-than-unity phase integrals are unusual for icy, airless satellites, but such values are expected for an object covered by transparent grains offrost or terrestrial snow (63).
The green-filter Hapke parameters were used to derive crude normal albedos for various regions imaged at high resolution through the green filter (Table 4). Greenfilter data were chosen as they are less 15 DECEMBER I989 influenced by atmospheric scattering than are the violet data and because both groundbased and Voyager-imaging observations can be used to constrain global photometric parameters. Figure 27A compares Triton color data derived from Voyager imaging observations with contemporaneous ground-based observations (64) and with ground-based observations acquired about a decade earlier (65,66). The Voyager data were reduced to normal albedos by using the global-average phase function discussed above. Both the Voyager imaging and contemporaneous ground-based measurements of Tholen (64) show Triton to be substantially less red than indicated by the earlier ground-based observations. Because both the 1979 and the 1989 measurements have each been confirmed by two sets of independent observations, it is clear that Triton's global color changed in the intervening decade. This is not surprising. The subsolar latitude ranges between about 55°N and 55°S on time scales of several 100 years (67). Groundbased measurements indicate solid nitrogen and methane on Triton's surface and Voyager UVS observations show these compounds are dominant in its atmosphere (50,68). These ices are quite mobile, even at the 37 to 39 K surface temperature of Triton. It is quite plausible that volatile transport, combined with the 100 change in the subsolar latitude between 1979 and 1989, could account for the large change in the diskaveraged color.
Comparison of the colors of several large albedo units and streaks (Fig. 27, B and C) reveals that the central polar unit is slightly reddish, similar in color to the dark streaks. The bright equatorial collar along the outer fringe of the cap is neutral and very bright, suggestive of freshly condensed frost. The redder areas may be contaminated with darker, redder material. The reddest areas measured are in the plains north of the bright collar that extend all the way to the terminator in the north. Even Triton's "darkest" regions are in actuality quite bright in comparison to most outer-planet satellites. Even the "dark" streaks show normal reflectances in the range of 0.40 to 0.75. Only a few small spots in the northern plains have normal albedos with values as small as -0.20.
One explanation for Triton's reddish color is that the surface is dominated by a physical mixture or solid solution of methane in nitrogen and that organic polymers produced by photolysis and charged-particle bombardment of methane are responsible for the red coloration (69). Although alternative explanations do exist (for instance, that the coloring agent is derived from primordial organic material or from material that continues to accrete onto Triton's surface), it is certain (i) that methane is present on the surface and in the atmosphere and (ii) that methane will be polymerized by interaction with cosmic rays, UV photons, or charged particles, or all of these.
Triton's geologic processes and evolution. Voyager 2's highest resolution view of Triton was of the hemisphere that faces Neptune in synchronous rotation. Figure 28, the first  The solid line represents the best-fit I rameters for the violet filter; the dashed the green filter (see Table 3). Both cun Bo = 0 (no opposition surge).
image to clearly show surface feat Fig. 29, a mosaic of higher resolut ping frames, both show this hei Less than 40% ofTriton was imagc resolution; it should be kept in r our understanding of Triton's geol cesses and history is limited to thi By comparison to most planetar) Triton's appears geologically youn and Europa, heavily cratered ter absent. All other outer-planet sate play regions of heavily cratered ter evidently date back to the early p tional bombardment. Triton's surfa sistent with an object that was ge active and resurfaced well after the heavy bombardment.
Albedo patterns. Triton's color at patterns can be broadly divided int of the brighter polar cap that occu the entire southern hemisphere somewhat darker and redder pl extend roughly from the equator n to the terminator (Fig. 25). Data c show that all ofthese surfaces have albedos (0.6 to 0.9); all are likely c substantially of volatiles.
In most cases the color/albedo not correlate with geologic terrain graphic features, suggesting they thin veneers draped over the ten (Fig. 30). For example, clusters albedo patterns with irregular, dark centers surrounded by brigh' occur near the eastern limb (Fig.  right). A few impact craters can be and on them but they themselves c obvious topography. The interic polar cap displays discrete bound irregular, patchy areas that appc "windows" in the bright reddish ic (Fig. 30, lower left). There are no ing cases of relief along these discri which is again suggestive of thin below the limits of detection. liolet fit The slightly darker streaks scattered over Jreen fit the interior of the ice cap resemble the ubiquitous martian "wind streaks" attributed to eolian erosion and deposition. The Triton streaks range in length from a few IQo -10's up to about 100 km. Often small, o darker, irregular patches a few kilometers \ across occur at the heads of the streaks; less commonly a bright region also appears near 150 the head. Although between roughly 100 and 30°S, the streaks are preferentially orireen-, and ented toward the northeast; farther to the ignitude is south the directions are highly irregular icqalbedo with streaks crossing one another. In a few Hapke pa-cases streaks of opposite direction originate l line is for from the same point. ves assume Whether the streaks are modern or ancient features is difficult to say from their surface patterns alone. It seems unlikely that ures, and they are older than very many Triton years tion map-(that is, a few thousand years), because the misphere. dark particles of which they are probably ed at high formed would migrate deep into the ice nind that deposits after multiple cycles of pole-to-pole logic pro-migration of the volatiles. Lssample.
Transport of material by the winds in y surfaces, Triton's tenuous atmosphere seems required ig; like Io as part of the explanation for the streaks. As rrains are mentioned earlier, some methane will inev--llites disitably be converted to particles of dark com--rains that plex hydrocarbons that will be found in ost-accre-Triton's surficial deposits. Those particles ace is con-that are fine enough (a few micrometers or ologically less) can be suspended in the atmosphere period of and carried substantial distances downwind. For Triton dust settling occurs in the Epnd albedo stein or kinetic regime as opposed to the o (i) units Stokes or viscous regime as the mean free ipy nearly path in the tenuous nitrogen atmosphere is and (ii) large compared to the particle size. From lains that simple momentum transfer arguments, we iorthward estimate, for instance, that a 1-,um particle craters. Our descriptions of Triton's surface morphology are confined to the terrains to the north (Fig. 31). An extensive unit termed the "cantaloupe" terrain (ct) dominates the western part of the equatorial region. It consists of a dense concentration of pits or dimples that are crisscrossed by ridges ofviscous material  (64,65). Plotted with the ground-based data are the Triton disk-averaged colors as obtained from Voyager narrow-angle camera images. Note the substantial color change from 1979 to 1989 shown by these data. (B) Normal reflectance of selected areas on the encounter hemisphere of Triton are shown (see text). (C) Normal reflectances of selected dark areas on Triton are shown. The brighter background spectrum is an average of bright areas adjacent to the dark streaks whose normal albedos are also plotted. The darkest spot is located north of the equator in the darker, redder plains. erupted into grabens. The grabens are global in scale and organization, continuing into other terrains to the east and south. Most of the dimples fall into two roughly uniform size classes of about 5-km and 25-km diameter; the smaller ones are found dominantly in the western part of the terrain. Both sizes are often organized into linear, equally spaced sets. It is possible that none of these features is of impact origin. In fact, recognition of any impact craters in the cantaloupe terrain is extremely uncertain. Seen at the highest resolution (Fig. 30, upper left) the cantaloupe terrain displays an extremely rugged and mottled texture. Some combination of viscous flow and collapse and deterioration of the landforms by extensive sublimation of surface materials may have been responsible for the complex landscape observed here. A morphological sequence in the degree of eruption into the grabens can be discerned. The sequence begins with the Ushaped floor at the Y-shaped branch in the top of Fig. 30, lower left. Farther northwest along the valley floor, a narrow ridge is visible that erupted along a section of the valley floor and along an intersecting fault, crossing out into the plains to the south. Even farther north in the cantaloupe terrain and to the east limb along the equator, viscous material welled up in the valley floors forming ridges that stand above the adjacent plains (Fig. 29).
In contrast to the cantaloupe terrain, the eastern plains are dominated by a series of much smoother units. The first of these is the floor material (sv) in two "lake-like" features a few hundred kilometers across located near the terminator (Figs. 29 and Fig. 30, upper right). The floors are extremely flat compared to other plains units, embaying the scarp-rimmed margins and surrounding numerous hills and knobs that protrude through the deposit. The floors are terraced, occurring at several levels separated by scarps a few hundred meters high, suggesting multiple episodes of emplacement. A cluster of small irregular pits surrounding a large central pit, likely associated with an eruptive vent, is found in each of the floors. A single impact crater about 15 km in diameter shows that material that is rigid on geologic time scales makes up the bulk of the floor material.
Another smooth plains unit dominates the region just south of the two flat-floored lake-like depressions. It occurs as highstanding smooth plains (sh) that appear to have erupted from large quasi-circular depressions; one can see that strings of irregular rimless pits occur commonly in this unit (Fig. 31). Like the flat-floored deposits, the high-standing plains are quite smooth and IS DECEMBER I989 have a very low density of superimposed impact craters. In contrast, however, these deposits stand as thick masses extruded onto and standing above preexisting terrains. The deposits terminate with rounded flow margins partially burying subjacent landforms. In places these deposits appear to be a few kilometers thick. Eruptions of comparable style, recognized on Ariel, are thought to have been formed by highly viscous material, such as partially crystalline ammoniawater mixtures (70).
The third plains unit in the eastern region is a hummocky plain (th). It appears to have been formed by extensive eruption of material along sections ofthe grabens that flowed out onto adjacent plains forming hummocky, rolling deposits and obliterating sections of the graben entirely. Figure 29 shows this unit to have the greatest abundance of impact craters.
In general, Triton's global fault system evidences a tensional regime. Several features in the transition zone between the lake-like features and cantaloupe terrain, however, suggest lateral displacement along strike-slip faults. Some show an offset of a few kilometers, others up to 30 km. The best example is an irregular depression 100 to 120 km (Fig. 32, left). It resembles many other depressions so common in this region; by contrast its margins appear to have been offset along ridge-and-groove lineaments. We speculate that the feature was deformed by two subsequent strike-slip motions, each with an amplitude of about 30 km. When the image is sheared along the hypothesized fault lines, the outline of the feature is restored to one typical of other cantaloupe Fig. 28. Color image ofTriton acquired at a range of 530,000 km with a resolution of 10 km per line pair. Narrow angle images obtained with green, violet, and ultraviolet filters were depressions (Fig. 32, right). Other truncated features also appear to line up after restoration.
Impact crater abundances and relative ages. At the resolution of the global mapping images of Fig. 29 (1.5 to 3 km per line pair) impact craters are generally rare. Craters with diameters from the limit of resolution up to about 12 km display sharp rims and bowlshaped interiors. Larger craters, ranging up to the largest (27-km diameter) are complex, having flat floors and central peaks. The transition diameter from simple to complex and depth/diameter ratios for Triton's craters are similar to those observed on other satellites whose crustal materials are thought to be dominantly water ice. Ejecta blankets are not visible, probably due to inadequate resolution. The craters also do not have rays, probably because veneers of mobile surface volatiles mask them.
Impact crater statistics were collected for four areas (Fig. 33, top). Area 1, the most heavily cratered region, occurs in the leading hemisphere ofTriton's rotation-locked, synchronous orbit. Area 2, the least cratered, coincides with units sv and sh. Area 3 is in the cantaloupe terrain. Area 4 is a part of the ice cap interior. Owing to difficulty of recognizing impact craters with any confidence, crater statistics for the cantaloupe terrain have been excluded. Figure 33, bottom, compares the cratersize frequency distribution for areas 1, 2, and 4 with those of the lunar highlands, typical lunar maria, and the fresh crater population on Miranda's rolling cratered plains (1,71). Triton's most heavily cratered terrain displays a population similar to that used as the red, green, and blue components. High frequency information from a clear image, containing the greatest detail, was merged with the color data. fact seen in the Triton images. These include clouds above the limb and extending into the terminator and an extensive optically thin haze that appears to be uniformly distributed around the disk. The haze is difficult to detect in limb images due to scattered light from Triton's bright surface, but it is easily seen in crescent images where it shows an extension of the cusp beyond the terminator. It apparently extends to an altitude of about 30 km. more than two scale heights, and has an optical depth of about 2 x 10-4.
We interpret it to be composed of photochemically generated smog-like particles described above.
Several clouds are seen in backscatter terrain-ct, smooth floor material-sv, high-standing smooth matenials-sh, hum rolling plains-th, linear ridge materials-ri, bright spotted polar unit-bs, bright streaked polar unit-bst, bright rugged polar unit-br. I,o I 0 above both the east and west limbs (Fig.  34). In places these clouds are clearly detached; in others they appear to extend to the surface. The limb clouds appear above diffuse bright regions seen on the limb, suggesting that they are composed of bright particles that can be seen both in backscatter as well as in projection against the disk. All ofthe limb clouds so far detected are located over the subliming south polar ice cap.
Brightness scans perpendicular to the limb for which data were collected indude: area 1, the most heavily cratered; area 2, lightly cratered; area 3, cantaloupe terrain, and area 4, a plains region just outside the inner cap zone. The dash-endosed areas (6, 7, and 8) were analyzed to detect a possible gradient in the crater flux (see text).
(Bottom) Triton crater size/frequency distribution for area 1 (Triton HC), area 2 (Triton LC), and area 4 (Triton SH) compared with the fresh crater population on Miranda, the lunar highlands, and the lunar post-mare. The curves were normalized to a standard -2 cumulative distribution power law as in (1). on one of these images (Fig. 35) show a cloud that extends from close to the surface to an altitude of about 4 kmi, above which its brightness sharply decreases. We interpret this feature to be a nitrogen condensation cloud. The ubiquitous diffuse haze can barely be discerned in the scans extending from the surface to an altitude of almost 30 km. Because they are thin, the optical thickness of the limb clouds can be estimated from their observed brightness. Assuming isotropic scattering (both of the direct and surface-reflected sunlight), we estimate optical depths of about 10-3. If more realistic particle phase functions are used, the estimated optical depths are several times larger (73). Also, in some cases the brightness of the discrete clouds is as much as three times larger than shown in Fig. 35. Together these factors imply optical depths ranging up to 10-2.
Bright clouds extending beyond the terminator are visible in a long sequence of outbound images of the opposite hemisphere to that seen in the mapping coverage (Fig. 36). The terminator is at roughly 45°S; again the clouds are seen above the south polar cap. From the distance they extend beyond the terminator, their altitudes are estimated to be in excess of 13 kilometers. These clouds remained stationary relative to the surface for the 2-day period as they rotated through the terminator region. They appear to correlate with dark surface markings seen in the low resolution approach images 3 days prior to closest approach.
Several east-west, elongated clouds can also be seen over the illuminated part of the crescent in a color image set acquired soon after Triton closest approach at a phase angle of approximately 1400 (Fig. 37). Both clouds and surface detail can be discerned. The clouds apparently cast shadows, giving a very rough estimate of their altitudes of a few kilometers. Like the clouds associated with the geyser-like plumes described below, these elongated clouds are roughly 100 kmn long and 10 km wide. They may also be associated with erupting plumes, but no direct evidence exists for plumes connecting them to surface sources.
Triton's geyser-like plumes. The Voyager 2 images of Triton were acquired over a sufficiently wide range of viewing angle that many regions can be studied stereoscopically. The last global view of Triton, acquired just prior to closest approach to Neptune, was imaged from a sub-spacecraft latitude of about 15°S. Subsequent higher resolution frames of the mapping coverage were acquired from latitudes ranging from about 100 to 250N at about the same longitude.
Through stereoscopic examination and by projecting these images to a common perspective onto a sphere, it was possible to study topographic landforms and search for material aloft. In this way two active geyserlike plumes have been confidently identified; several others are suspected. All are located well inside the complex central zone of Triton's south polar cap.
The first of the two well-observed plumes (termed here the west plume) is situated at 3340E, 50°S (Fig. 38, top). West plume is visible in at least four views, which have emission angles of 370, 620, 670, and 750 at its location. The feature appears as an 8-km- tall narrow, dark stem rising vertically from a dark spot on the surface; the upper end terminates abruptly in a small dense dark cloud. A more diffuse cloud, appearing as a narrow dark band, can be seen extending westward for at least 150 km in the highest emission angle view, at which point it diffusely disappears. The cloud band maintains a very narrow projected width (approximately 5 km) along its length. In the later, higher resolution and higher emission angle images, the shadow of the band of cloud is visible; the sun was nearly directly above the west plume (incidence angle approximately 100). The east plume (Fig. 38, bottom) is located at 12°E, 57°S, and is also visible in at least four views with emission angles near the plume of 530, 720, 76°, and 77°. Like the west plume, it rises to an altitude of about 8 km where a small, dark, dense cloud is formed. In contrast to the west plume, however, its dark cloud ofmaterial is generally denser and diverges as it extends westward. The cloud is also quite dense in the vicinity of the source. Hence, although the rising column appears to be tilted slightly westward, its geometry is less obvious than in the case of the west plume.
In both cases the plume material rises roughly vertically to the 8-km altitude before being carried downwind. The geometry suggests stratification of the atmosphere may be controlling the form. Perhaps the 8km altitude represents an inversion at the tropopause; as the plumes rise under some combination of momentum and buoyant force, they cease to rise above this altitude. The cloud geometry also suggests vertical structure in the wind speeds, in which the winds aloft increase abruptly at this altitude, although this dearly depends on the timescales for rising, suspension, and downwind transport.
It seems likely that the active dark plumes are related to the ubiquitous dark streaks scattered over the south polar region described earlier. The observed variation in the two plumes in terms of cloud density and degree of divergence is consistent with the variance seen in the collection of dark streaks.
Most plausible mechanisms to drive the plumes involve the venting ofsome gas from the surface, entraining fine dark particles. The particles are carried by some combination of ballistic and buoyant forces to altitudes where they are left suspended in the thin atmosphere to be carried downstream by the complex, sublimation-driven winds. The most likely driving gas is nitrogen, although models involving concentrated methane gas rising buoyantly are conceivable. Secondly, different types of energy sources can be considered, including insolation and geothermal energy sources either localized, for instance by intrusion of cryogenic lavas into areas ofvolatile material, or more broadly distributed through Triton's overall radiogenic heat flow. The fact that the active plumes are near the current subsolar latitude argues for a solar-driving mechanism. We discuss here one model out of many conceivable, simply to demonstrate  38. Profile view ofTriton's active geyser-like plumes. These two high-emission-angle views are of the regions near the southern limb; south is up, west to the right. In both views the spacecraft was about 15°above the horizon as seen from the base of each plume near its source. The plumes are about 8 km tall; the east-west dimension ofthe two views is about 150 km. The west plume is shown in top image; the east plume in the bottom image. that simple, plausible explanations do exist. The model we describe is for solar-driven nitrogen gas geysers.
The insolation-driven mechanism involves a "greenhouse" effect. A way to construct an extremely efficient greenhouse is to cover a dark absorbing layer with a relatively transparent layer, which we propose to be a layer ofnitrogen ice, that is both volatile and has a low thermal conductivity. Groundbased spectroscopic observations of Triton show an absorption feature at 2.15 pm that was attributed to nitrogen, requiring a path length in the nitrogen equivalent to a meter or more ofsolid or liquid (50). In this model the radiation is absorbed in a dark substrate beneath the nitrogen layer. The temperature will rise until the thermal gradient reaches a point where the excess heat is conducted and reradiated back to the surface.
The vapor pressure of nitrogen ice increases rapidly with increasing temperature. A temperature rise of 10 degrees above Triton's 37 K surface temperature results in roughly a one hundred-fold increase in pressure. If the layer ofnitrogen is thick enough (>1 or 2 meters), sufficiently transparent, and locally seals off the subsurface, the subsurface vapor pressure will increase, filling permeable subsurface reservoirs. Ifthe seal is ruptured or if the pressurized gas migrates laterally to an open vent it will rapidly decompress, launching a plume of nitrogen gas and ice, entraining dark particles encountered in the exit nozzle, and carrying them to altitude.
Small satellites. Of the newly discovered satellites (Table 1), only 1989N1 and 1989N2 were well enough resolved to see surface features (Fig. 39). As resolution was sufficient to measure radii directly for 1989N1, 1989N2, and 1989N3, their albedos are also reasonably well determined. One color sequence acquired of 1989N1 shows it to have a flat spectrum; violet-, green-, and clear-filter albedos all fall within 10% of each other, similar to Voyager photopolarimetry observations for 1989N2 (29).
The low resolution view of Nereid provided only an approximate albedo; no surface or limb features were detectable (Fig.  41). Nereid's spectral reflectivity (geometric albedo in the violet, green, and clear filters) is flat within 10%. Clear-filter images used to derive Nereid's phase curve between 250 and 550 (Fig. 40) yield a linear phase coefficient of 0.021 magnitudes per degree. No rotational effects were detected in the Nereid images; the amplitude of its light curve is less than 10%. Inasmuch as the orientation of Nereid's pole is unknown, we cannot derive limitations of its shape from these data. The Voyager viewing direction differed only by approximately 250 from that of recent ground-based observations (74) that suggested a large amplitude light curve. We see no evidence for an amplitude of more than 10%. The rotation period remains unknown.
1989N1 and 1989N2 are irregularly shaped; 1989N1 displays craters, one near the terminator about 150 kilometers across.
1989N1 was resolved at several longitudes and is slightly elongated; topography of about 20 km can be seen on its limb. The closest view weakly shows linear features; no substantial albedo features are seen. The irregular shape of 1989N1, which had a diameter of 400 kilometers, might seem surprising. However, the internal strength required to support such topography falls within the expected range even for an icy object (75).
Collisional histories of satellites. Bombardment by comets over the last -3.5 billion years probably accounts for most of the craters observed on Triton and for the origin of the small satellites 1989N5 and 1989N6 by collisional fragmentation. The comet flux close to Neptune is dominated by comets captured into relatively short-period Neptune-crossing orbits by encounters with that planet. A rough estimate of the number of these Neptune family comets can be made by extrapolation from the observed Jupiter family of short-period comets (76,77). On the basis oftheoretical studies of the capture process (78), we infer that there is a fairly steady population of the order of a million Neptune family comet nuclei with absolute B-magnitude greater than 18 (this corresponds approximately to diameters greater than about 2.5 kilometers). This estimate is conservative; if most short-period comets have been derived from a region lying a moderate distance beyond Neptune or between Uranus and Neptune (79), the population of Neptune family comets might be 10 to 20 times higher. Long-period comets also strike the satellites ofNeptune, but their impact contributes, at most, only a few percent to the estimated production of craters on Triton and on the small regular satellites.  The present cratering rates on the satellites, estimated from methods described earlier (76) and from the conservative estimate ofthe Neptune family comet population, are given in Table 5. At Triton, the conservatively estimated present production of craters >10 km in diameter is about half the present rate on Earth (80) and similar to the average rate on the Moon over the last 3.3 billion years. At this rate, the craters observed on most of the sparsely cratered terrains (marked sv and sh in Fig. 31) of Triton could all have been generated in the last billion years or so.
Even at the conservatively estimated cratering rates, the small, innermost satellites 1989N5 and 1989N6 are not likely to have survived intact over the last 3.5 billion years. These satellites are probably fragments pro-duced by catastrophic disruption of a larger body in the last 2 or so billion years, possibly even in the last half-billion years. 1989N2, 1989N3, and 1989N4, on the other hand, might have escaped destruction by comet impact at the present rate, but would almost certainly have been destroyed by collisions during an early period of heavy bombardment. The rate of cratering on these satellites by comet impact is about three to five times higher than on 1989N1; this result is independent of the estimate of the comet population. The large craters observed on 1989N1 suggest a fluence of impacting bodies that would have destroyed 1989N2, 1989N3, and 1989N4. Most likely these satellites are the product of a disruption of a body comparable in size to 1989N1 near the end of the heavy bom- Table 5. Estimated present cratering rates and past production of large craters on the satellites of Neptune. The entries are as follows: P > 10 km is the estimated present rate of production of craters >10 km diameter in units of 10 to 14. Dmax is the diameter of the largest crater likely to have been formed in 3.5 billion years at the current rate ofcratering. Frsat is the frequency ofproduction ofcraters with diameters larger than the radius of the satellite over a time period of 3.5 billion years. The column headed D > Rs.t gives the number of craters with diameters greater than the radius of the satellite that would have been formed while the observed craters on Ni were formed.