The Optical Aurorae of Europa, Ganymede, and Callisto

All observations were made with the target satellites in Jupiter eclipse, which removes any ambiguity about the relative roles of photoexcitation versus electron excitation. Our model assumes that all auroral emissions are produced by electron-impact direct excitation or dissociative excitation of oxygen-bearing species. The model includes the parent species O, O₂, H₂O, and CO₂. The strength of the observed emission lines is a function of the satellite's atmospheric composition and density, and the density and energy distribution of the incident electrons. Our model is adapted from that described in de Kleer & Brown (2018) and uses experimentally derived emission cross sections. The electron energies and temperatures for Europa's orbital location are taken from Bagenal et al. (2015) and are derived from in situ measurements; we use an electron density of 160 cm⁻³ with Maxwellian-distributed energies, with 95% of the electron population centered at 20 eV and 5% centered at 250 eV. Note that previous observational studies and modeling papers that interpreted these observations used a four times lower electron density (e.g., Hall et al. 1995, 1998; Roth et al. 2016; Vorburger & Wurz 2021), and column densities from these works need to be scaled down by a factor of 4 for comparison with our results and with other more recent work (de Kleer & Brown 2018, 2019; Roth 2021). The model of Oza et al. (2019) uses an electron density of 70 cm⁻³, which is intermediate between the two values used in other work.

The electrons at Ganymede's orbit were similarly observed by Voyager to be composed of two populations, with a cold population at 30–60 eV and a hot population at 200 eV, constituting 10% of the total number density of 3–10 cm⁻³ (Sittler & Strobel 1987). However, Eviatar et al. (2001) demonstrated that the electron populations at Ganymede's orbit are insufficient to explain the brightness of the UV aurora and postulated that the aurora is excited by electrons that are locally accelerated to 100 eV. We follow these and other previous authors in adopting a Maxwellian distribution of electron energies centered at 100 eV, with an electron density of 20 cm⁻³ (but we note that some modeling work has used a higher density of 70 cm⁻³; Leblanc et al. 2017). An electron density of 20 cm⁻³ is consistent with recent in situ measurements from the Ganymede Juno flyby, which took place one day prior to our 2021 June 8 Ganymede eclipse observation (Kurth et al. 2022). Although the electron properties are uncertain and therefore the derived column densities should be viewed with caution, the ratios of the emission lines do not change substantially with the adopted electron properties, and the relative abundances of different species is therefore robust to this uncertainty.

For Callisto, information on the electron energies and densities is even more limited. Voyager measured an electron density of 0.1–1.0 cm⁻³ at Callisto's orbit (Belcher 1983), while Galileo found 0.01–1.0 cm⁻³ depending on the plasma sheet distance (Kivelson et al. 2004). The electron temperature estimates from these works range from 35 to 200 eV, with a suprathermal population. We adopt a density of 0.15 cm⁻³ at 35 eV following Belcher (1983). This is the same electron density as was adopted by Cunningham et al. (2015), but they included a hot population at 1.5 keV, constituting 27% of their total population. The cross-section measurements do not extend to this energy, but we tested adding a hot population at the same ratio at either 250 eV and 1.5 keV (extrapolating the cross sections beyond the measurement energies), and found that the derived column densities change by up to ±30% depending on the electron energies. This is well within the uncertainty in the overall electron densities, therefore we neglect the hot electron population in our modeling.

In addition to extending the model to Callisto, we update the model of de Kleer & Brown (2018) by adding CO₂ as a parent molecule. Emission cross sections for electron-impact dissociative excitation of CO₂ are not available for several of the UV/optical O i emission lines, and for CO₂, our model therefore includes only emission from the O(¹S), O(³S), and O(⁵S) states at 5577, 1304, and 1356 Å, respectively. We adopt the emission cross sections for O(¹S) recommended by Itikawa (2002), which are based on the measurements of LeClair & McConkey (1994). We adopt the cross sections for O(³S) from Ajello (1971), but note that their measurements disagree with those of Mumma et al. (1972) even after revised normalization of the latter (Itikawa 2002), and the cross sections are consequently uncertain at the ∼10% level over most electron energies and even higher at the low end of the electron energy distribution. Ajello (1971) also measured the emission from the O(⁵S) state at 1356 Å. Their emission cross sections are presented relative to those of 1304 Å and are therefore limited by the propagated uncertainties from the O(³S) measurements. Additionally, the authors note that the 1356 Å emission is a lower limit since the O(⁵S) atom may acquire excess kinetic energy during dissociation beyond the thermal velocity. The limitations on the above measurements and the lack of measurements for other emission lines highlights the need for future laboratory measurements of these cross sections for an interpretation of aurorae at CO₂-containing atmospheres throughout the solar system.

Our updated auroral model now also includes the O i emissions at 7774 and 8446 Å from the parent molecules O, O₂, and H₂O. Both emission features are triplets that are produced by the atomic oxygen transitions (3p ⁵P)→(3s ⁵S°) and (3p ³P)→(3s ³S°), respectively. Their radiative cascades to ground produce the well-known 1356 Å and 1304 Å UV lines, and so the cascade contributions into these UV lines is simply equal to the 7774 and 8446 Å brightnesses in Rayleigh units. The cross sections for both emissions following electron impact on O₂ are taken from Schulman et al. (1985), on H₂O from Beenakker et al. (1974) following the recommendation of Itikawa & Mason (2005), and on O from the excitation cross sections recommended by Laher & Gilmore (1990), which are based on the measurements of Gulcicek et al. (1988) and Gulcicek & Doering (1987).

The method of this emission model relies on some implicit assumptions that warrant a brief discussion. First, the approach assumes that all emissions are optically thin. Lines with the strongest Einstein A coefficients saturate first in their curve of growth, and 1304 Å has the strongest transition probability herein. The brightness of the HST/COS resolved 1304 Å triplet at Ganymede matches the optically thin ratio of 5:3:1 (Roth et al. 2021). Locally, Ganymede's emissions are bright relative to the other satellites, which ensures that opacity effects are negligible overall. A second assumption is that electrons remain warm enough to excite all lines at all altitudes; neutral collisions could cool magnetospheric electrons below the various thresholds to excite the different emissions. Cooling rates in Earth's thermosphere are dominated by vibrational excitation of O₂ near 1 × 10⁸ cm⁻³ (Pavlov & Berrington 1999). This density is expected only very near the surface. A third assumption is negligible collisional quenching of long-lived forbidden transitions. With a lifetime of 134 s (Wiese et al. 1996), O(¹D) can be collisionally depopulated before radiating. Thermal O₂ gas quenches O(¹D) at a rate of 5 × 10⁻¹¹ cm³ s⁻¹ molecule⁻¹ (Streit et al. 1976), and so the critical density where O₂ quenches atoms before radiative decay is 1.5 × 10⁸ cm³–again, a negligible effect over the atmospheric columns.

Table 4 gives the modeled emission rate coefficients for the UV and optical emission lines given the relevant electron energies for Europa, Ganymede, and Callisto. Table 5 gives the corresponding emission line ratios for comparison with observations.

We use the emission model described in Section 3.1 to find the best-fit atmospheric column densities for each satellite and date of observation. The model atmosphere is composed of O, O₂, and H₂O, and the free parameters in the fits are the disk-integrated column densities of each of these constituents. CO₂ is considered, but is ultimately not included in the fits because the cross sections are unavailable at all but one of the optical lines. The modeled auroral emissions are fit to the measurements and upper limits of the four oxygen transitions (at 5577, 6300/6364, 7774, and 8446 Å) and Hα (6563 Å). An MCMC algorithm is employed to obtain the joint probability distribution of the three atmospheric constituents and hence obtain the most accurate uncertainties on the free parameters. We use the emcee Python implementation (Foreman-Mackey et al. 2013) of the affine-invariant ensemble sampler for MCMC described by Goodman & Weare (2010). An example output of the MCMC algorithm is shown in Appendix A. In our results, we present our final model with uncertainties; upper limits are presented when a species is found to be present at the <2σ level. In the appendix we also report the single maximum likelihood model for each observation.

The strongest constraints on atmospheric composition come from simultaneous measurements that cover the largest range of emission lines. The atmospheric retrievals on data from a single observation have the advantage that all measurements are simultaneous. However, numerous measurements have also been made of the UV auroral lines (e.g., McGrath et al. 2013; Roth et al. 2016). We therefore also retrieve the atmospheric composition using the average measurements of auroral emissions at each transition, including previous UV measurements.

For the optical transitions, the values in the average fits are averaged over the two dates for each satellite that had good observing conditions and low scattered light (i.e., 1998 November 15 and 2021 June 8 for Ganymede, and 2021 May 20 and 2021 June 21 for Europa; see Table 1). For the UV transitions at Europa, we use values of (40 ± 4) and (80 ± 8) R for the 1304 and 1356 Å emissions, respectively, based on 19 observations summarized in Roth et al. (2016). Roth (2021) also present measurements of the UV emissions from the subjovian hemisphere in and out of eclipse that are about half the average value cited above. While these data are a closer match to the viewing geometry of our observations, Roth (2021) show that the aurora are very similar in and out of eclipse, and we adopt the average values rather than the subjovian values because in the case of Europa, the time variability seems to be dominated by the plasma sheet distance and stochastic variability rather than hemispheric trends. Using the UV eclipse measurements instead of the average UV measurements in our modeling provides a much worse fit.

The situation is different for Ganymede, where the interactions between Ganymede's magnetic field and Jupiter's magnetosphere result in distinct auroral differences between the hemispheres (McGrath et al. 2013). In this case, we use the average of the two available subjovian UV measurements, giving values of (15 ± 2) and (36 ± 2) R for the 1304 and 1356 Å emissions, respectively (Roth et al. 2021). While one of these measurements was made in eclipse and one in sunlight, the values are almost identical. Averaging the leading, trailing, and subjovian hemisphere measurements from that work gives higher values of (23 ± 1) and (46 ± 1) R, which provide a worse fit when the retrievals are run jointly with the optical data.

Ly α emission has also been detected on both satellites; on Europa, a value of 70 ± 26 R was found based on two off-limb averages (Roth et al. 2014a). However, the authors note that these emissions may originate from an extended H cloud around Europa rather than directly from H₂O dissociation. At Ganymede, the H₂O-sourced Ly α emission level predicted by Roth et al. (2021) of 200 R would be inseparable from spatial variations in the surface reflections (Alday et al. 2017; Roth et al. 2021). We therefore do not include Ly α values in our average fits; including upper limits would not affect the fits because our Hα upper limits provide a much more stringent constraint on H₂O as a parent molecule because of the smaller uncertainties due to observing in eclipse.

Recently, Roth (2021) and Roth et al. (2021) postulated H₂O-dominated atmospheres on the trailing hemispheres on Europa and Ganymede. This claim is based in part on an observed 1356/1304 Å ratio that is lower for both satellites on the trailing hemisphere than on the leading hemisphere. While this difference had previously been attributed to a greater O abundance on the trailing hemisphere (Roth et al. 2016; Molyneux et al. 2018), the new observations included eclipse measurements of the subjovian hemisphere, permitting a stronger upper limit on the column density of O by placing a direct constraint on the contribution to 1304 Å from resonant scattering by O.

Our observations are all made in eclipse and therefore image the subjovian hemisphere. In the case of Europa, our upper limit for O of ∼1 × 10¹³ cm⁻² (see Table 2) is consistent with the upper limit of 6 × 10¹² cm⁻² derived by Roth (2021). However, our best-fit value for H₂O, 1.2 ± 0.5 × 10¹⁴ cm⁻², is a factor of 10 lower than the disk-averaged H₂O abundance of ∼1 × 10¹⁵ cm⁻² found by Roth (2021) for the sunlit trailing hemisphere. Note that while the column densities presented here depend on the assumed electron properties, we have adopted the same electron properties as used by Roth (2021) and Roth et al. (2021), so the factors by which the derived column densities differ between studies are not sensitive to uncertainty in electron density and are only weakly sensitive to uncertainty in electron temperature. Similarly, while the absolute abundances of species derived from our modeling depends on electron density, the relative abundances do not and are therefore not subject to uncertainty on this parameter.

The presence of H₂O in our modeling is only indicated at the 2.4σ level and only after averaging data sets together with the UV; for individual nights, we find only upper limits on H₂O in the 0.8–1.1 × 10¹⁴ cm⁻² range. For all UV and optical oxygen lines except 5577 Å, the emission rates are at least ten times higher for O₂ than for H₂O, so that these lines are not strongly sensitive to H₂O (see Table 4). However, for 5577 Å, the emission rates are comparable between the species. This line, combined with Hα, therefore provide the strongest constraints on water abundance. In particular, the H₂O disk-integrated column density from Roth (2021) would produce 70 R of emission at Hα, compared to the 2σ upper limit of 12 R from our Europa observations, in addition to predicting O i line ratios inconsistent with the optical and UV lines.

To determine whether an atmosphere containing O, O₂, and H₂O is preferred over an O₂-only atmosphere, we fit the data using both the three-species model and the O₂-only model and find that the fit is only marginally better when all three species are included. Given the lack of strong preference for the H₂O-containing model and the fact that H₂O is only preferred by our average fits (and only at 2.4σ), and not on individual nights, we consider the detection of Europa's H₂O at all to be tentative in our data. Moreover, we strongly rule out a water abundance of 2.5 × 10¹⁴ cm⁻² or higher, or alternatively an H₂O/O₂ ratio above 0.5, on the eclipsed subjovian hemisphere.

For Ganymede, our observations on the two nights that had clear sky conditions as well as low Jupiter scattered light conditions (see Table 1) place an upper limits on H₂O of 3 × 10¹³ cm⁻², or a maximum H₂O/O₂ ratio of 0.06. In contrast, the modeled ratio for the center trailing hemisphere from the UV data was found to be 12–32, and for the leading hemisphere, it was found to be 2–5 (Roth et al. 2021), with the disk-averaged ratios a factor of a few lower. This H₂O abundance contributed 1 R to the disk-averaged 1304 Å emission, but should contribute 23 R and 46 R to the optical emission at 5577 Å and 6563 Å, respectively, both of which are strongly ruled out in our observations.

There are several differences between the optical and UV observations that may be relevant to these different results, in particular, the low H₂O/O₂ ratios (<1) derived from our optical eclipse data of both satellites, and the much higher ratios of 10–30 determined for the trailing hemispheres in the UV.

First, our observations were made with the targets in shadow, while the UV observations were made with the targets in sunlight. For an H₂O atmosphere sustained by sublimation, the column density should be lower in eclipse due to the lower surface temperature. Leblanc et al. (2017) modeled Ganymede's H₂O exosphere in eclipse and calculated a collapse by four orders of magnitude from a peak column density in the 10¹⁴ to 10¹⁶ cm⁻² range. However, in this case, the UV auroral brightnesses would also be lower in eclipse than in sunlight, whereas the observed auroral brightnesses of the two satellites are comparable in and out of eclipse (Roth 2021; Roth et al. 2021). A sublimation atmosphere that partially collapses in eclipse therefore cannot reconcile the UV and optical data sets. For Europa, even the O₂ atmosphere is modeled to drop by an order of magnitude in eclipse (Oza et al. 2019), which is not consistent with the UV observations either.

The observations also differ in that we target the subjovian hemisphere, while the UV-derived H₂O abundances were highest on the trailing hemispheres of both Europa and Ganymede. For Europa, the UV line ratio is also just as low, and hence as inconsistent with pure O₂, on the subjovian hemisphere as it is on the trailing hemisphere (Roth 2021). Previous UV data and our optical results together are therefore hard to reconcile with the presence of significant O or H₂O on Europa, which seems to require a new explanation for the low 1356/1304 Å ratio at least on the subjovian hemisphere.

For Ganymede, the UV line ratio is higher on the subjovian hemisphere than either leading or trailing (Roth et al. 2021) and is consistent with pure O₂. The difference between the UV and optical data for Ganymede can thus be attributed to an H₂O atmosphere that is only present on the leading/trailing hemispheres, and not on the subjovian, regardless of whether it is sunlit or eclipsed.

Given the clear differences between hemispheres on the two moons, the upper limit derived for O on the subjovian hemispheres might not be applicable to the trailing hemispheres, and hence, both O and H₂O remain candidates to explain the low UV ratio on the trailing hemispheres. If atomic O is primarily a product of electron-impact dissociation of O₂, it is similarly predicted to be most abundant at the center of the trailing hemisphere for Europa (Cassidy et al. 2013), so either species would produce a 1356/1304 Å ratio that increases from center to limb of the trailing hemisphere, as observed.

A sputtered H₂O atmosphere has also been proposed for both satellites (Johnson et al. 1981). For Ganymede, recent modeled sputtered H₂O column densities are in the 10¹¹–10¹³ cm⁻² range, while modeled sublimated column densities are on the order of 10¹⁶ cm⁻² (Marconi 2007; Leblanc et al. 2017; Vorburger et al. 2022). Sublimated column densities on this scale are ruled out in our data, but the expected sputtered column densities are low enough to be below our detection limits. For Europa, sublimated H₂O should be roughly two orders of magnitude lower than for Ganymede based on their surface temperatures, while sputtered H₂O is limited to 10¹³ cm⁻² as for Ganymede (Shematovich et al. 2005; Smyth & Marconi 2006; Feistel & Wagner 2007; Plainaki et al. 2013). Both predictions are near our detection limits, so we cannot provide strong constraints on a potential sputtered H₂O atmosphere. The expected sputtering production is also longitudinally variable. For Ganymede, Leblanc et al. (2017) showed that the H₂O sputtering production should be highest on the leading hemisphere due to the magnetic field geometry, while for Europa, Cassidy et al. (2013) showed that the sputtering rate of all species is highest on the trailing hemisphere.

While Ganymede's aurora is produced by complex interactions between the incoming plasma and Ganymede's magnetic field, Europa's auroral brightness should be a more straightforward product of the local electron density and the column density of species in its atmosphere. In the UV, the aurora is found to vary by a factor of 5–10 across 71 total observations (Roth et al. 2016); there is a correlation with Europa's distance from the Jovian plasma sheet, although there is significant variability between individual exposures as well as between observations at similar plasma sheet locations. There is no indication that there is a difference in brightness between sunlight and eclipse.

Our individual integrations have a sufficiently high signal-to-noise ratio in the 6300/6364 Å emission for us to analyze the changes in the aurora through each eclipse, during which the plasma sheet distance is changing, but longer-timescale variations in either Europa's atmosphere or in the plasma conditions are not a factor. Figure 7 shows the brightness of the O(¹D) transition, the sum of the 6300 and 6364 Å lines, as a function of plasma sheet distance (as defined in Phipps & Bagenal 2021) during each observation. There is an apparent correlation with plasma sheet distance on 2021 July 16 and 2021 June 21, although we note that there is no apparent correlation on 2021 May 20, and even when present, the scatter is significant and the correlation reverses slightly right at the equator. We fit a model of the form $B(z)={B}_{0}{e}^{-{\left(z/H\right)}^{2}}$, where B₀ is the brightness as Europa is crossing the centrifugal equator of the plasma sheet (Phipps & Bagenal 2021), z is the absolute-value distance from plasma sheet in R_J , and H is the scale height (Hill et al. 1974; Roth et al. 2016). The free parameters are B₀ and H, and the best-fit model is shown in Figure 7. The best fit scale height of 1.67 R_J is in the vicinity of the scale height 1.6 R_J found for the UV aurora (Roth et al. 2016), as well as the plasma torus scale height at Europa's orbit 1.7 R_J from the model of Bagenal & Delamere (2011).

Zoom In Zoom Out Reset image size

The UV and optical observations thus collectively demonstrate that Europa's auroral brightness correlates with magnetic latitude, although there is large scatter about this correlation and other sources of variability may dominate on timescales of minutes to days. We note that Io's aurora also varies by a factor of ∼3 even at a given plasma sheet location (Oliversen et al. 2001; Roth et al. 2014b; Schmidt et al. 2023), and that this is indeed expected based on observed variations in the electron density (Bagenal et al. 2015).

Callisto's atmosphere is the most poorly understood of the Galilean satellites. The presence of an ionosphere at Callisto was indicated by Galileo from the plasma density measured during Callisto flyby (Gurnett et al. 2000) and from radio occultation observations (Kliore et al. 2002). Under the assumption of a predominantly O₂ atmosphere (which is supported by modeling work by Liang et al. 2005), the radio occultation measurements were used to infer an O₂ column density of 4 × 10¹⁶ cm⁻² (Kliore et al. 2002). However, the ionosphere was only detected on the sunlit trailing hemisphere, which is the side impacted by Jupiter's corotating plasma, and not on the sunlit leading hemisphere. It was also denser on the sunlit than on the night side of Callisto's terminator, suggesting that photons play a role in producing it. Kliore et al. (2002) hypothesized that the atmosphere is generated by sputtering, but it has been suggested that the presence of the ionosphere should divert plasma around Callisto (Strobel et al. 2002). If this is the case, it may be that sputtering on the night side generates the atmosphere, which becomes ionized by photoelectrons when it moves into sunlight.

The first detection of Callisto's aurora was made with HST/COS of the 1304 and 1356 Å emission lines in sunlight on the leading/Jupiter-facing hemisphere (Cunningham et al. 2015). The observed brightnesses of (3.3 ± 2.8) and (3.2 ± 1.6) R for the two emissions correspond to an O₂ column density of 4 × 10¹⁵ cm⁻² assuming excitation by photoelectrons. The authors argue that photo- rather than magnetospheric electrons are exciting the emissions, on the basis of the fact that photoelectron-excited emission is modeled to be ten times brighter, assuming the magnetospheric electrons penetrate the atmosphere at all instead of being diverted around Callisto (Cunningham et al. 2015). This atmospheric density is five times higher than the CO₂ column density of 8 × 10¹⁴ cm⁻² derived from a Galileo off-limb airglow measurement (Carlson 1999), but ten times lower than the trailing hemisphere O₂ density inferred from the ionospheric measurements.

We present the first detection of Callisto's optical aurora from 2021 July 4, specifically, the O i emissions at 6300 and 6364 Å with brightnesses of 6.1 ± 1.7 R and 3.3 ± 1.2 R, respectively. When a purely O₂ atmosphere is assumed, these emissions correspond to a column density of (4.0 ± 0.9) × 10¹⁵ cm⁻², which is identical to the column density inferred from previous UV observations, although our derived column density depends on the assumed electron energies and densities at Callisto, which are poorly constrained at the current time and are likely variable. The UV observations were made in sunlight, and the derived column density was based on the assumption that the aurora was primarily excited by photoelectrons rather than magnetospheric electrons. Our observations were made in eclipse, and the detected emissions must therefore be excited by magnetospheric electrons, removing this ambiguity in interpretation. The match between our column density and that inferred from the past UV data may support the interpretation of the excitation mechanism for the UV aurora. However, the large uncertainty on the electron density, and the unknown variability of Callisto's atmosphere and the electron density at Callisto's orbit, prevent firm conclusions. When we assume a purely O₂ atmosphere and excitation by magnetospheric electrons alone, our measured optical emissions would correspond to emission of 0.5 R at 1304 Å and 1.2 R at 1356 Å (see Appendix B). These values are lower than the UV aurora measurements of 3.3 ± 2.8 R and 3.2 ± 1.6 R, which suggests that photoelectrons are indeed needed to excite the observed UV emissions, although the measurements are still consistent within 1–2σ with magnetospheric electron excitation due to the large uncertainties. In addition, the column densities inferred from aurora for the leading/subjovian hemispheres are an order of magnitude lower than the estimate from radio occultation for the trailing hemisphere of 4 × 10¹⁶ cm⁻² (Kliore et al. 2002; Cunningham et al. 2015), which could be due to enhanced atmospheric generation via sputtering on the trailing hemisphere if confirmed by measurements of both hemispheres using the same technique.

The detection of combined 6300 and 6364 Å emission at 9.4 R along with the upper limit on 5577 Å of 2 R places a lower limit on the O(¹D)/O(¹S) ratio of 4.75. This rules out H₂O as the dominant parent molecule, as it would produce an emission ratio of 1.2 (see Table 5). Atomic O was previously ruled out as the dominant parent species due to an early nondetection of 1304 Å, which is produced in part by resonant scattering (Strobel et al. 2002). CO₂ is known to be present in Callisto's atmosphere, but the CO₂ column density of 1.1 × 10¹⁶ cm⁻² that would be required to produce our measured auroral emissions is two orders of magnitude higher than the detected CO₂ atmosphere (Carlson 1999). We conclude that molecular O₂ is the most plausible parent molecule for the aurora we observe at Callisto.