Images of Saturn's aurora at the limb have been collected with the Advanced Camera for Surveys on board the Hubble Space Telescope. They show that the peak of Saturn's nightside emission is generally located 900–1300 km above the 1-bar level. On the other hand, methane and H2 columns overlying the aurora have been determined from the analysis of FUV and EUV spectra, respectively. Using a low-latitude model, these columns place the emission layer at or above 610 km. One possibility to solve this apparent discrepancy between imaging and spectral observations is to assume that the thermospheric temperature in the auroral region sharply increases at a higher pressure level than in the low-latitude regions. Using an electron transport code, we estimate the characteristic energy of the precipitated electrons derived from these observations to be in the range 1–5 keV using a low latitude model and 5–30 keV in case of the modified model.
 Images collected with the Hubble Space Telescope (HST) have shown that Saturn's auroral morphology is characterized by a dynamic aurora ring located between 70° and 80° which responds to the solar wind dynamic pressure [Gérard et al., 2004; Clarke et al., 2005; Grodent et al., 2005]. The energy of the primary auroral electrons exciting the H2 emissions in Saturn's ultraviolet aurora has so far only been determined indirectly using two distinct spectral methods. The first approach makes use of the FUV color ratio, taking advantage of the strong wavelength dependence of the methane absorption cross section between 130 and 145 nm. When the emission layer is located below the homopause, the altitude where the molecular and eddy diffusion coefficients are equal, wavelengths <145 nm are partly absorbed by the overlying hydrocarbon column. Ultraviolet spectra of Saturn's aurora at ∼3 nm resolution were first obtained with the UVS instrument during the Voyager 1 and 2 flybys of the planet [Broadfoot et al., 1981; Sandel et al., 1982]. Broadfoot et al.  noted that limb scans with the UVS slit locate the auroral emission approximately 800 km above the limb. Sandel et al.  found that, unlike the Jovian aurora, most spectra of Saturn's aurora did not indicate the presence of absorption by hydrocarbons. A particularly bright spectrum was best fitted with a CH4 column of 8 × 1015 cm−2, suggesting an electron energy of ∼10 keV. Six auroral spectra obtained with the Space Telescope Imaging Spectrograph (STIS) were analyzed by Gérard et al.  who found indications of a weak absorption by methane. Using Moses et al.'s  low-latitude atmosphere model and an electron energy – H2 column relationship, they derived a primary electron energy of 12 ± 3 keV. Recently, Gustin et al.  analyzed a Cassini-UVIS FUV spectrum moderately absorbed by a vertical CH4 column of 1.2 × 1016 cm−2, corresponding to electron energies near ∼10 keV. Other UVIS spectra were found to be unabsorbed by methane (J. Gustin, private communication, 2008), suggesting that the maximum electron energy is ∼15 keV, which corresponds to an altitude of ∼620 km in the low-latitude atmospheric model by Moses et al. . A second approach is based on the presence of self-absorption of H2 lines below 120 nm. Transitions connecting to the ground state v″ = 0 and 1 vibrational level may be self-absorbed by the overlying H2 gas, leading to a weakening of EUV specific lines and an increase of intensities at longer wavelengths [Gustin et al., 2004]. Sandel et al.  found that one UVS spectrum was best fitted using a spectral model with an overlying H2 column of 1 × 1020 cm−2. Gustin et al.  used a detailed spectral model to make a thorough analysis of spectra at ∼0.2 Å resolution collected with the Far Ultraviolet Spectroscopic Explorer (FUSE) satellite. An excellent fit to the line intensity distribution found a rotational temperature of ∼400 K, in agreement with the 420 ± 50 K obtained by Melin et al.  from ground-based H3+ IR spectra. A foreground vertical H2 abundance of 3 × 1019 cm−2 derived from the FUSE spectra corresponds to a pressure level of ∼0.1 μbar, independently of any atmospheric model, and to an altitude of ∼660 km in the Moses et al.  model.
2. Observations and Data Analysis
 The HST FUV images used for this study were taken during the HST-Cassini campaign (GO 10862 program) between January 13, 2007 and February 16, 2008 when the subsolar latitude ranged from 14.2° S to 8.3° S. They were obtained with the photon-counting Multi-Anode Micro-channel Array (MAMA) of the High Resolution Camera Solar Blind Channel (SBC) on the Advanced Camera for Surveys (ACS). The point spread function is <2 pixels FWHM. Images were taken both with the F125LP filter (sensitive to the H2 Lyman and Werner bands but excluding the H Lyman–α line) and with the F115LP filter (which includes Lyman–α). In this program, a typical orbit sequence consists of five 100-s exposures taken with the F125LP filter (Ly-α rejected), followed by nine 100-s exposures with the F115LP filter (Ly-α included) and finally another five 100-s exposures with the F125LP filter. The field of view of 35 × 31 arcsec2 is wider than Saturn's apparent equatorial diameter of 10 arcsec, and thus also includes part of the ring system of the planet.
 After dark current subtraction and flat-field and geometric corrections, the final image plate scale is 0.0301 arcsec/pixel in each direction. The first step in the analysis consists in the accurate determination of the planetary center. Since the HST pointing accuracy is limited by the onboard guide star catalogue that may include a 1-arcsec (∼30 pixels) uncertainty, the planetary central pixel must be determined using the image itself. Given the Earth-Saturn distance for the time of the observation and the plate scale, our automatic method fits elliptic ribbons to the Kronian A, B and C rings. The center of these ellipses provides an estimate of Saturn's center position in the field of view of ACS with an accuracy of ∼1 pixel. The plate scale is then directly converted into an altitude scale so that the location of the 1-bar level may be determined. To improve the signal to noise ratio, images have been co-added by sets of 5 consecutive images obtained with the same filter. Radial cuts of the planetary disk at different angles are then generated and are rebinned to increase the signal to noise ratio.
Figure 1 shows an example of an ACS exposure obtained on January 21, 2007 at 0322 UT with the F115LP filter clearly showing a brightness gap between the limb and the auroral emission. The altitude of the 1-bar level calculated from the camera plate scale is indicated by the white contour. It is seen that this contour agrees within one pixel with the position of the observed sunlit limb, corresponding to the 1-bar altitude level. Figure 1 also shows the observed light curve along a radial cut through the auroral region. The location of the calculated 1-bar level altitude is shown by the vertical dashed line. The brightness decreases rapidly above the limb but increases again beyond 650 km. The light curve shows a peak near 1150 km, corresponding to the auroral emission layer observed on the image of the planetary disk.
3. Altitude of the Auroral Emission Peak
 To analyze the dataset, an automatic algorithm performs radial scans from the planet's center with 0.1° steps to cover the entire southern auroral region. For each step, the algorithm determines the maximum in the light curve between 0 and 2000 km. A median filtering over 20 profiles (i.e. ∼2°) is then performed and the code identifies the longest sequence of points above 500 km. This sequence retains only the points where the light curve gap is observed under the auroral emission. We then compute the maximum peak altitude in this point sequence with the median filter. This method provides the actual altitude of the emission layer assuming that the aurora is located in the plane of the observed planetary limb and provides a lower limit otherwise. Our statistical study is based on 836 individual images leading to 176 light curves. We find that the average emission peak is 1111 ± 347 km for exposures taken with the F115LP filter, 1181 ± 251 km with the F125LP filter, leading to an average value of 1145 ± 305 km when all images are considered. The 70 km difference between the two filters is not significant, considering the value of the standard deviations.
 We now examine the implications of this altitude determination and how they fit with the other constraints based on analysis of FUV and EUV spectra. To summarize, the FUSE spectra indicate that the auroral emission originates from a level of ∼0.1 μbar in a region where the temperature is ∼400 K, a value incompatible with 135 K at 0.1 μbar provided by the low-latitude model of Moses et al. . Figure 2 compares observational results with the pressure-temperature relationship in the low-latitude model by Moses et al.  adapted to auroral latitudes by adjusting the value of the gravity. The 0.1-μbar pressure determined from the FUSE spectra is indicated by a horizontal line and the auroral temperature of 400 K derived from the H3+ and the H2 spectra is indicated by a closed circle. Another disagreement between auroral observations and the Moses et al. model is clearly apparent in Figure 3 showing the altitude-pressure relationship. The vertical lines at 0.1 μbar show the emission level based on the FUSE spectra and the horizontal bars define the range of variability of the auroral emission peak centered on 1150 km. The circle locates the central point meeting both constraints. It is situated at a pressure level approximately two orders of magnitude larger than the Moses et al.  model at 1150 km. Inversely, in the Moses et al. model, the 0.1 μbar pressure is at an altitude of ∼600 km, 550 km below our auroral limb observations. We thus conclude that the low-latitude model cannot match the observational evidence derived from a recent set of auroral spectral observations. In particular, Figure 2 shows that the strong evidence (from both IR and FUV spectra) of an auroral temperature of ∼400 K is met at pressures less than ∼2 × 10−2μbar. To reconcile both datasets, we speculate that the thermal structure of the auroral upper atmosphere of Saturn is modified by the presence of auroral particle and Joule heating whose effect is to deposit energy and heat up the region located below the 10−2μbar level. This heating does not appreciably increase the local exospheric temperature since meridional transport efficiently redistributes heat from the polar to the equatorial regions. The role of transport was demonstrated by Smith et al.  using a three-dimensional atmospheric circulation model. As an example, Figure 2 presents an alternative model meeting the observational constraints listed before. We arbitrarily modified the temperature between 103 and 10−4μbar in such a manner as to reach a value of ∼ 400 K at the 0.1 μbar level. The functional dependence used by Yelle et al.  for Jupiter's thermosphere is adopted with a mesospheric temperature T0 = 160 K, an α parameter = 0.03, and an exospheric temperature T∞ = 420 K reached at zm = 650 km. Between 1 and 10 mbars, the temperature profile is a linear combination between Moses et al.'s  values and the expression given by Yelle et al. . From this modified pressure-temperature relationship, the new pressure-altitude curve is derived assuming hydrostatic equilibrium and shown in Figure 3. Incidentally, we note that the faster increase of temperature above the 1-bar level probably implies a lowering of the altitude of the homopause compared to the low latitude value. In this case, the full circle corresponding to a pressure of 0.1 μbar and an altitude of 1150 km falls near the center of the rectangle defined by the observational constraints from the FUSE spectra and the HST limb images, while meeting the 400 K determination from the EUV and infrared auroral spectra.
 We now examine what may be inferred from the altitude of the emission layer about the characteristic energy of the auroral electrons. For this purpose, a series of simulations has been made with an electron transport code to calculate the vertical distribution of the H2 FUV volume emission rate. These calculations are based on a direct simulation Monte Carlo method solving the Boltzmann equation described by Shematovich et al.  for the Earth's thermosphere. It was subsequently applied by Bisikalo et al.  to an H2-dominated upper atmosphere by replacing cross sections by those appropriate to proton and electron collisions with H2 and H. Figure 4 illustrates the altitude distribution of the total H2 Lyman and Werner band emission calculated for a series of monoenergetic electron beams ranging from ∼750 eV to 30 keV with an isotropic pitch angle distribution. The dashed line curves have been obtained using the Moses et al.  model adapted to gravity in the polar regions. To generate an emission peak at 1100 km, simulations show that electron energies need to be about 2 keV. This value is close to the that obtained in recent models [Cowley et al., 2008, and references therein] as the minimum field-aligned voltage required to produce the current density compatible with Cassini observations.
 The simulations also show that electrons about 20 keV are required to generate an emission peak at 0.1 μbar, independently of the temperature profile. The consequence is that in order to produce auroral emission at 1100 km, the pressure levels must be located at a higher altitude than in the low-latitude model. For comparison, calculations performed with the modified model are illustrated in Figure 4 by the solid lines. As expected, all curves for a given energy are now located at a higher altitude. In particular, electrons with energies of about 20 keV produce an emission peak at 1100 km, the auroral altitude determined in this study.
 We also note that this result depends somewhat on the adopted pitch angle distribution. For example, if the precipitation is field-aligned, the pressure reached by 20 keV electrons would be less than doubled, changing from 0.06 to 0.1 μbar. In other words, the energy required for field aligned electrons to reach a given pressure level would be approximately divided by 2, corresponding to an energy of ∼10 keV.
 Analysis of over 800 images of Saturn's FUV aurora above the nightside limb indicate that the maximum H2 emission is observed at an altitude of about 1100 km. A set of independent spectral observations indicate that i) the temperature prevailing in the region of emission is close to 400 K, ii) the auroral electron energy is occasionally high enough to produce a weak absorption by methane, iii) the H2 column overlying the emission layer corresponds to a pressure level of about 0.1 μbar. Electrons of ∼15–20 keV deposit their energy at this pressure level, independently of the detailed thermal structure. We find that the set of different observations can only be reconciled if the thermal structure of the high-latitude thermosphere is different from the low-latitude reference model available so far. We estimate the characteristic energy of the precipitating electrons giving a luminosity peak near 1100 km to be in the range 1–5 keV using a low latitude model and 5–30 keV in case of the modified model.
 The occasional presence of hydrocarbon absorption in the FUV spectra may result from sporadic hardening of the energy spectrum of the precipitated electrons. The results of our numerical simulations in Figure 4 indicate that >10 keV electrons may produce emission near the homopause region located at ∼800 km in the low-latitude model. However, ∼100 keV electrons would be needed in our modified profile, in which the homopause altitude is close to 500 km, in order to generate a significant absorption in the FUV. Further study of the circumstances when hydrocarbon absorption is observed should help to clarify this issue.
 This work is based on observations with the NASA/ESA Hubble Space Telescope, obtained at the Space Telescope Science Institute (STScI), which is operated by the AURA, Inc. for NASA. JCG and DG acknowledge support from the Belgian Fund for Scientific Research (FNRS). Partial funding for this research was provided by the PRODEX program of the European Space Agency managed in collaboration with the Belgian Federal Science Policy Office and by FNRS grant 4.4508.06. This work was also supported in part by grant HST-GO-10862.01-A from the Space Telescope Science Institute to Boston University and V.I.S. and D.V.B. were partially supported by RFBR Grant 08-02-00263.