PROBING FOR EVIDENCE OF PLUMES ON EUROPA WITH HST/STIS

Each observation resulted in a single file of photon events, which we expanded to include additional information such as new coordinate systems, tracking information, sky levels, the detector pixel flat-field value, and the location of Europa as a function of time. We refer to these expanded event files as "basetag" files. A transit image of Europa is derived by transforming the (x, y, t) values for each photon event into a frame in which Europa is at rest. We chose two primary coordinate systems: one fixed in detector coordinates, with the x, y values corrected for geometric distortion of the STIS FUV MAMA, and the Level 3 moving target drift removed, and the other, a frame in which Europa is oriented with the Europa North pole up and the spatial scale fixed at 35 km per pixel. These "35 km" images were the principal images with which we worked. In the 35 km pixel frame, the radius of Europa is 44.6 pixels. We emphasize that because the data are time-tag, no image resampling is involved in switching between coordinate systems: the different coordinates for each photon are stored as different columns in the basetag file (there is a small contribution to the noise from the digitization of the 12 mas pixel scale, which is negligible in our statistical analysis that uses binned data).

The task of accumulating images from the time-tag data involves first choosing a spatial coordinate system, based on a transformation of the original set of (x, y, t) events, and second, realizing that an image is a histogram of the index value corresponding to location (x, y) in the chosen coordinate frame. That is, if there are N events with unique coordinates (x, y), then the image intensity value is N at location (x, y). A subtlety arises when we additionally include the STIS flat-field response. Each photon event occurs at a known location on the detector and the detector pixel flat-field response ("p-flat"), or relative sensitivity, of that location is known. If the photon is detected in a region of low sensitivity, it needs to be given higher weight than a photon that arrives in a region of higher sensitivity. To do this, if the pixel p-flat value is f, then instead of counting "1" per event, we count "1/f" per event when accumulating the image from the time-tag file. This is equivalent to applying a standard flat field to the data in the case where there is no target motion or drift. To allow for various tracking solutions, we imbedded the photon events into images with pixel dimensions 4096 × 4096, which is double the size of the raw data coordinate system.

To create images in the reference frame of Europa, we compared the HST tracking solution available from the HST jitter files with the HORIZONS ephemeris. Using these two pieces of information, we predicted the (x, y) position of the Europa image in the geometrically corrected detector frame. In doing so, we noticed that a small error in the HST tracking was present: the track of Europa on the detector had a curvature resulting in deviations of order one pixel for some of the data, but less for the majority. Because we are working with time-tag data, this was completely removed by a coordinate transformation that followed the track of the Europa image accurately on the detector. This procedure provided a relative accuracy for the Europa location within a single observation of significantly better than a pixel. The position of the Europa image in geometrically corrected (x, y) coordinates was recorded as a polynomial in the basetag header.

The absolute position of Europa on the detector is subject to uncertainties in the guide star catalog, which can cause a shift of the image away from the nominal location by a few tenths of an arcsec. Hence, having obtained an image of Europa in corrected detector coordinates, nominally located at the image center, we measured its actual position in two ways. For the transit images the most effective procedure was to cross-correlate the images with a model of the image (Section 3.2), and then we adjusted the position slightly by eye if necessary so that any residuals in comparison to the model appeared symmetric. By comparing the cross-correlation results with estimates done by eye, we found that the image centering was good to approximately one pixel. Shifts of the model relative to the data by one pixel introduce noticeable asymmetries in their ratio. For out of transit images, we contoured the image with axis ratio and center as free parameters at a relatively low level compared with the peak count rate. This gives a highly precise measurement, but nevertheless could be slightly biased by an asymmetric light distribution. We found a consistent image elongation of ∼2%, which we attributed to the convolution of the asymmetric light distribution with the complex FUV PSF. We considered and rejected centroiding, as the trailing hemisphere is approximately a factor of two darker than the leading hemisphere; this biases the centroid toward the leading edge by several pixels.

After accumulating a Europa transit image, relatively low-level band structures in the background image of Jupiter remain. To estimate the optical depth of any off-limb features of Europa, we divided the transit image by an empirically constructed model of the background Jovian light. In the FUV, the face of Jupiter in the equatorial regions is relatively smooth to begin with. Perpendicular to the Jovian cloud belts, the brightness variation had an rms brightness variation of ≈5%, while along the belts, the dispersion was consistent with Poisson noise. In the frame in which Europa is at rest, additional smoothing took place as Jupiter rotated behind Europa, smearing the Jupiter image relative to the Europa one. To model the appearance of Jupiter for the transit observation, we identified a point on the Jovian surface roughly centered on Europa during the transit and then constructed an image using coordinate transformations in which that point on the Jovian surface is at rest. In this frame, Europa drifted across the field of view during the observation. We then built a series of images in the Jovian frame each of duration 1 s and masked out the data within a circular region with diameter 1.2× the diameter of Europa. We also created an exposure time image, allowing for time intervals where data acquisition was paused due to buffer dumps, and omitted the path of Europa across Jupiter. Dividing the summed Jupiter images by the exposure time image, we obtained an image of Jupiter free of the presence of Europa, which we used as the basis of an empirical model image of Jupiter.

To determine the appearance of Jupiter in each Europa-frame transit image, we created an image stack with 1 s slices in the Europa rest frame using the Jupiter count rate image shifted according to the relative motion of Europa and Jupiter. We set buffer dump periods to zero, but otherwise used the entire Jupiter image. The sum of the image slices, divided by the equivalent exposure time, yielded a model of the smeared Jovian background for comparison to the transit observation. After division by this empirical model, the belt structure was not present at a level above the Poisson noise. Because these Jovian models and the actual Jovian images were smeared in the Europa frame due to the relative motions of Europa and Jupiter, and because they were intrinsically smooth relative to the noise level of the data, we concluded that they were not introducing fine structure around the Europa limb.

This process worked well for the majority of transit observations, though there are some caveats for those observations near the Jovian limb. The relative motion of Europa and the background Jovian cloud tops is dominated by Jupiter's rotation. Near the limb, however, the apparent movement due to Jupiter's rotation decreases, and additionally the surface brightness of a location on Jupiter changes as it moves either away from or toward the limb. The quality of the resulting models was somewhat inferior in these cases, although we were able to make the necessary models for all observations.

For all images, we measured the sky background in regions away from both Jupiter and Europa. The sky brightness was measured at 200 s intervals through the orbit and a total background sky level was subtracted from the images prior to processing (though included appropriately in the noise models). There are strong variations in the sky background as HST approaches the Earth limb.

To assess the likelihood of image artifacts arising due to the combination of albedo fine structure on the surface of Europa, and the optics of HST, we generated a series of models of Europa and convolved them with representations of the HST PSF. The phase angle of Europa varied from ≈3° to ≈11° which had a significant influence on the illumination pattern, which we also modeled.

Qualitatively, the Europa disk appearance in the FUV images is similar to the optical, with the same large scale overall albedo variations: a dark trailing hemisphere and a bright leading one. We did not see the albedo inversion shown in Figure 4 of McGrath et al. (2009, pp. 485–505)  and is described in more detail by Roth et al. (2014a), which is not unexpected since that appears at shorter wavelength.

The Europa model was derived from the 500 m resolution Galileo/Voyager (hereafter just "Galileo image") mosaic available from the US Geological Survey⁴ , which we mapped onto a sphere matched to the observing configuration for each visit. We neglected rotation of Europa during an observation. To allow for differences between the FUV and visible morphology, we adjusted the "contrast" of the Galileo image by a factor C such that $G^{\prime} =C(G-\langle G\rangle )+\langle G\rangle$, where G is the original image and $\langle G\rangle$, is its mean. These contrast adjusted Galileo images were then multiplied by an illumination function derived from the known geometry of the observation relative to the Sun, and the generalized Lambertian bidirectional reflectance distribution function described in Oren & Nayar (1994). This reflectance model is physically motivated, based on a macroscopic distribution of Lambertian facets (length scale ≫ wavelength), similar to the approach of Torrance & Sparrow (1967) for specular facets. The model is parameterized by its "roughness," which is the standard deviation of the angle distribution of facets. Rougher models have a flatter surface brightness profile with sharper edges.

We tested a variety of PSFs: empirical and theoretically modeled using the TinyTim software package (Krist et al. 2011). The core of the PSF retains the diffraction limited width, but as aberrations increase relative to the wavelength, energy is redistributed from the core into Airy rings at greater radii. Hence, images retain the fine detail of the diffraction limit, but overlaid on that is a blurring from the broad wings of the PSF. We combined a PSF generated for the default focus position and STIS F25SRF2 filter parameters covering 15 wavelengths in the range 125–185 nm, which does not include a red leak element, with a 300 nm PSF to mimic the red leak, in the proportion 66%–34%, which we estimated this to be the proportion of counts from the Jovian reflected light spectrum blueward and redward of 200 nm, respectively, to approximately model the red leak. The actual value of the red leak is very uncertain. Empirically, the smooth appearance of Jupiter and the fact that the ice of Europa appeared darker than the clouds of Jupiter, argues that we were at least dominated by FUV photons in the images, consistent with this red leak estimate. At optical wavelengths, the clouds of Jupiter exhibit a much higher contrast and the albedo of water ice (on Europa) is extremely high.

Encircled energy data for the STIS F25SRF2 PSF are available in the STIS Instrument Handbook (Biretta et al. 2016; see imaging reference material). The residual HST rms wavefront error is in the range 25–70 nm, depending on field location, wavelength, focus, individual instrument optics, etc. (Krist & Burrows 1995; G.F. Hartig 2016, private communication), which for a low 25 nm error implies a Strehl ratio (the ratio of PSF peak to the theoretical diffraction limit) ≈33% at 150 nm or an encircled energy 28% of the total within the first Airy ring, radius 16 mas. The aberrations used by TinyTim yield a lower Strehl: for our composite PSF the encircled energy within a fiducial radius of 35 mas (∼5 pixel diameter) is ≈30%. The encircled energy within this radius for the red leak element is 39%, since the amplitude of the aberrations relative to the wavelength of observation declines to the red. We compared radial profiles of models of two of the out of transit Europa images convolved with the composite PSF and the individual red and blue components, and found that the data more closely match to the convolution with the red (sharper) PSF and that the composite PSF is slightly too pessimistic. This could be due to either a small mismatch between the assumed aberration model of TinyTim and the actual HST optics or to a higher fraction of red leak for the icy surface of Europa than we assumed. Hence, there is significant energy available at resolutions at or close to the telescope diffraction limit. Quantitatively, the amplitude of fine structure on these scales is diminished by approximately the fraction of energy in the PSF core ≈0.3–0.4, which we correct for in the Section 4 Results.

Empirical PSFs were also obtained from STIS F25SRF2 observations of NGC 6681, which contain numerous individual stellar images. Thirty-one separate observations used exposure times ranging from 110 to 2600 s. Stars were selected to be in the area of the detector where the Europa data were obtained and were ranked according to the HST focus model available at http://www.stsci.edu/hst/observatory/focus/FocusModel. We selected 10 relatively high signal-to-noise ratio (S/N) stellar images across a wide range of focus for exploration of the data, using as empirical PSFs either the one nearest in focus to the Europa observation, or the average of the PSFs within the focus range of the Europa observation. In the end, the empirical PSF proved too diffuse in comparison to the actual data, due to resampling necessary in their preparation, coupled to inadequate S/N. Hence, although it is an idealization with inevitable small mismatches to the real PSF, we used the composite TinyTim PSF described above as our input for the data modeling and assumed that it was unchanged within a visit and from visit to visit in the detector coordinate frame.

For comparison to the data, for each visit we developed a complete model of the scene allowing for the (slightly) different view of Europa, the differing illumination pattern due to the Europa phase angle, the Jovian background and the convolving PSF. For all individual visits, we multiplied a contrast-enhanced Galileo image projected onto a sphere by the illumination pattern of known geometry and chosen roughness. The transit models included an empirically determined image of Jupiter smeared by the relative motions of Jupiter and Europa. The central surface brightness of the Europa model (R < 0.5R_E) was matched to the data self-consistently by convolving the model images of Europa and Jupiter (with a gap for Europa) separately, to account for the effect of the PSF spilling light from Jupiter across the Europa image. For the out of transit images, we assumed that the background sky was constant and tweaked the original measurement according to the residual sky measured in the range 1.5–3 R_E Figure 3 illustrates the construction of the models.

Zoom In Zoom Out Reset image size

For each observation, we generated a grid of models with a range of roughness in the range 0–1, a range of contrast enhancement from −1 to +3, and three PSFs: TinyTim, the empirical nearest neighbor, and the empirical range. We measured the reduced χ² for "data minus model" to derive a fit for roughness and contrast, and spatially cross-correlated the data and model using the best-fit parameters to adjust the position of the model relative to the data. Following this iterative step, we re-derived the reduced χ² as a function of roughness, σ, and contrast, C, and found best-fit values of σ = 0.57 ± 0.02 and contrast C = 1.61 ± 0.06 using the TinyTim PSF. The average reduced χ² ≈ 1.12 with a range 0.98–1.32, omitting the leading and trailing hemisphere images that have strong "edge effects" around the Europa limb. The departures from unity were likely dominated by mismatches between the assumption that the visual albedo is directly proportional to the FUV and that the PSF was well-represented by the TinyTim PSF. Nevertheless, the final models seemed acceptable, Figure 4.

Zoom In Zoom Out Reset image size

A value of σ = 0.57, corresponding to a standard deviation of slopes of 32°, would be considered very rough. It is not our intent to determine the surface characteristics of Europa from this analysis, as alternative scattering models may apply (Goguen et al. 2010). They find a comparable root mean square slope distribution for the lunar highlands of ≈35° derived from optical observations, but show that an equally good fit to the photometry can be obtained with flat regolith and multiple scattering between regolith particles on the microscopic scale. Here, we were less concerned with the physical interpretation of the model and were primarily concerned that the model empirically described the appearance of Europa to an acceptable degree of accuracy.

For the statistical analysis to examine evidence for plumes, we worked with the transit observations. We analyzed the data in a variety of different ways. For the transit observations, the goal was to determine a fractional absorption against the smooth background light of Jupiter. Hence, we worked with the data image divided by the convolved model, and for additional insights, we also inspected the data divided by a circularly symmetric azimuthally averaged version of the data itself (i.e., free of model assumptions). That is, we used ratio images that provide a measure of, or limit to, the optical depth of the Europa exosphere as Europa transits Jupiter, together with the original images that gave observed counts for deriving Poisson statistics.

If the background light intensity (of light scattered by Jupiter) is I₀(i, j) at each pixel in the image in the vicinity of Europa, and if the observed surface brightness is modified by a plume with optical depth τ(i, j), then the observed brightness will be I_obs(i, j) = I₀(i, j)e^−τ(i,j). For small optical depth τ, and omitting the (i, j) subscript for clarity, I_obs ≈ I₀(1–τ). If we bin the data using a box of side n pixels, i.e., a total of N = n² pixels, then to obtain the average optical depth $\langle \tau \rangle$ we take one minus the average of the ratio of observed flux to background flux $\langle \tau \rangle =1-\langle {I}_{\mathrm{obs}}/{I}_{0}\rangle$ (as opposed to $1-\langle {I}_{\mathrm{obs}}\rangle /\langle {I}_{0}\rangle$, which would give a luminosity weighted optical depth estimate, biasing the measurement to low values of τ). An exact estimate of the optical depth is $\tau =-\mathrm{ln}({I}_{\mathrm{obs}}/{I}_{0})$; however, the use of this quantity in the statistical comparisons would unnecessarily introduce a complex error distribution: the errors on I_obs are straightforwardly and correctly estimated from Poisson statistics. A first-order correction τ' to the optical depth may be obtained if $\langle \tau \rangle =1-\langle {I}_{\mathrm{obs}}/{I}_{0}\rangle$ as $\tau ^{\prime} =\langle \tau \rangle (1+\langle \tau \rangle /2)$.

Our initial goal for each image was to test whether the data are statistically consistent with the observation model described above, given the Poisson noise statistics appropriate to the image. The sky-subtracted data images in 35 km coordinate space were therefore divided by the model of the observation. The model included a reconstruction of the appearance of Jupiter, which flattened the off-limb zone, and yielded the optical depth estimate. Because we were testing for departures of the data I_obs from the model image I₀, we used the noiseless model image, with the addition of the measured sky pedestal, as the basis of estimate for the Poisson count uncertainty. We tested the significance of the departure of the ratio data-to-model by calculating an estimate of the uncertainty σ on the average $\langle {I}_{\mathrm{obs}}/{I}_{0}\rangle$. The uncertainty on $\langle {I}_{\mathrm{obs}}/{I}_{0}\rangle$ is given by the assumption that the uncertainty on an individual pixel is σ²(I_obs) = (I₀ + S), where S is the sky level per pixel. Then, it follows that within a box of side n pixels, the uncertainty on the mean within the box $\sigma (\langle {I}_{\mathrm{obs}}/{I}_{0}\rangle )=\tfrac{1}{n}\sqrt{\langle 1/{I}_{0}\rangle +S.\langle 1/{I}_{0}^{2}\rangle }$. Hence, we derived a z statistic $z=(\langle {I}_{\mathrm{obs}}/{I}_{0}\rangle -1)/\sigma$, which we assumed was normally distributed to test for departures from a ratio value of one, i.e., optical depth zero. For the transit observations, I₀ was typically 15–20 counts per pixel, so the Poisson distribution was adequately approximated by the normal distribution, even for no binning. It was very well approximated by the normal distribution for modest amounts of binning (counts in the hundreds) as shown in Table 2. We excluded pixels interior to the limb of Europa in the binning of points close to the limb.

To test the assumption of Poisson (normal) statistics, we established a sparse grid with sampling points spaced by the binning box size plus one, in the region of the image between 2.0 and 4.0 R_E. The means and standard deviations of the z statistics for these grids are reported in Table 2.

If the statistically significant regions are due to plumes, then we can quantify their properties with the caveat that these numbers are to be treated with caution, and apply only if the features are actually plumes and not artifacts. Table 2 gives the mean dispersion in the bright region surrounding Europa (i.e., the transit image divided by the model of Jupiter). This may be interpreted as the "1σ" optical depth limit, for which the average value is 0.056 and 0.04, respectively, for 5 × 5 and 7 × 7 binning. The individual best limit (1-σ) is 0.033 for 7 × 7 binning in oc7u02g2q, which also has the second best spatial resolution and is one of the images which appears to show statistically significant off-limb features. The actual candidate features were seen at ≈4.5σ, corresponding to a lower detection limit of τ ≈ 0.15–0.25. There is an additional correction factor due to the broad wings of the PSF, as discussed in Section 3.2. This correction factor is size dependent, larger features requiring smaller correction. By simulating a grid of Gaussian patches of absorption with a range of optical depth and length scale, we found that for the length scales of interest, a factor of ≈2.3–3.3 to scale the as-seen integrated extinction over the patch, to the underlying total, was appropriate, corresponding to a PSF with ≈37% of its energy in the compact core (see Section 3.2). Hence, taking the mean factor 2.8× and correcting the optical depth, we derived a lowest detectable limit of intrinsic optical depth of τ ≈ 0.4. We illustrate one of these models below, applied to the data of 2014 March 17.

To quantify the amount of water required to produce a patch of absorption, we looked at both molecular and ice particle cross-sections. The STIS FUV MAMA detector has a significant red leak. Using the component level software package pysynphot, with a Solar spectrum, we calculated that the fraction of detected photons with wavelength longer than 200 nm is approximately 34%. This acts to dilute the effective cross-section of absorbing molecules, typically highest in the FUV region, which represents ≈66% of detected photons. The actual value of the red leak is very uncertain; this value was obtained using pre-launch component throughput curves and it has not been calibrated on orbit or as a function of time. Qualitatively, however, the smooth appearance of Jupiter, and dark appearance of Europa indicate that the image is dominated by the FUV spectrum and not the red leak.

To obtain an effective cross-section for H₂O, in light of these considerations, we derived a count rate-weighted mean cross-section using the pysynphot spectrum together with the cross-section as a function of wavelength. The resulting wavelength averaged weighted cross-section was σ(H₂O) ≈ 1.8 × 10⁻²² m². For a patch of area A₃₅ pixels of 35 km size, and optical depth τ_v the corresponding implied mass in water vapor due to molecular absorption is M_v = 0.2 × 10⁶τ_v A₃₅ kg.

The plumes of Enceladus comprise a mixture of water vapor and ∼micron sized ice grains (Dong et al. 2015). The ratio by mass of grains to vapor was variable, but in the range 15%–25% typically. The ultraviolet scattering cross-sections for solid column (and other) ice particle habits are given by Key et al. (2002). Converting their parameters into mass and optical depth, we derive a mass in grains of Mg(H₂O) = 0.64 × 10⁶ τ_g A₃₅ a_μm kg, where τ_g is the optical depth due to grains, A₃₅ is the area of the patch in 35 km pixels, and a_μm is the grain size in microns. The total optical depth is the optical depth due to molecular absorption (vapor) and grains. If the mass ratio of grains to vapor is α ≈ 0.2, the total inferred mass in water from a patch of intrinsic optical depth τ and area A₃₅ pixels is M_tot(H₂O) = 0.2 × 10⁶τ A₃₅ (1 + α)/(1 + α/3.2 a_μm) kg, from which it is evident that the presence of grains in any scattering ice plumes makes a difference only of order 10% (12.9% specifically for α = 0.2). More subtly, however, is that grains scatter across the entire spectrum, including the redder regions for which the PSF is sharper, raising the possibility that the correction factor due to PSF blurring may be (slightly) reduced. In the discussion that follows, given these offsetting small corrections, we set α = 0.

At face value, inspecting the 2014 January 26 image, there are three patches of absorption in the latitude zone where Roth et al. (2014a) found evidence of an emission line plume, though with sublongitude ≈181° rather than ≈93°, Figure 21. We defined the potential plume region as those points within the Roth latitude range (polar angle 220°–260°), and probability of chance occurrence p < 0.01 based on the probability image for 5 × 5 binning, for the quantitative analysis that follows.

Zoom In Zoom Out Reset image size

Converting the statistical data to physical quantities, the average optical depth, defined as I = I₀e^−τ, in the plume region as described in the previous paragraph, is $\langle \tau \rangle \approx 0.15$ (15% absorption) averaged over 45 pixels (55,125 km²) with a peak τ_max ≈ 0.25. For an average measured optical depth $\langle \tau \rangle \approx 0.15$ the implied intrinsic optical depth is $\langle \tau \rangle \approx 0.42$ for a blurring correction factor of 2.8, and the average column density is $\langle N\rangle \approx 2.3\times {10}^{21}\,{{\rm{m}}}^{-2}$ which is higher than the H₂O column density inferred by Roth et al. (2014a) of ≈1.5 × 10²⁰ m⁻². Nevertheless, the total number of H₂O molecules in the two-plume model of Roth et al. (2014a) is 1.3 × 10³², while the implied total number of H₂O molecules (column density times projected putative plume area) of our new observation is also 1.3 × 10³², or ≈3.9 × 10⁶ kg, in good agreement, though the uncertainty is significant.

Hence, at face value, the absorbing features we see imply a similar order of magnitude of material to the emission features of Roth et al. (2014a) and arise in the same latitudes, but with physical characteristics derived in a completely independent fashion using a very different observing strategy, albeit with the same detector.

We turn now to the compact nub of 2014 March 17, on the trailing limb. Quantitatively, the average optical depth over the central 22 pixels is $\langle \tau \rangle \approx 0.21$, reaching a peak τ ≈ 0.34 at a significance level of 4.4σ, with formal probability of being a random event ≈4.8 × 10⁻⁶. The corresponding column density for a pure water event would be 3.3 × 10²¹ m⁻² corresponding to a total of 1.8 × 10³² H₂O molecules, or ≈5.4 × 10⁶ kg. The two polar region patches of absorption in this image have an implied corresponding column density 2.7 × 10²¹ m⁻² corresponding to a total of 2.1 × 10³² H₂O molecules, or ≈6.6 × 10⁶ kg.

Figure 22 shows an example of an idealized model, with a Gaussian σ = 2.5 pixels (≈90 km) absorption patch, peak optical depth 0.75, with Poisson noise added, not intended to be a rigorous fit to the data, but in quite good qualitative and quantitative agreement with the data, also shown. In this example, the peak implied intrinsic optical depth is ≈2× the peak observed optical depth and the total intrinsic absorption of the model is 2.3× the face value extinction in the data.

Zoom In Zoom Out Reset image size

Examining the image of 2014 April 4, the average optical depth is 0.04 over 42 pixels, reaching a peak τ ≈ 0.3 at a significance level of 4.5σ, with formal probability of being a random event ≈2.9 × 10⁻⁶. The corresponding average column density for a pure water event would be 0.7 × 10²¹ m⁻² corresponding to a total of 0.5 × 10³² H₂O molecules, or ≈1.4 × 10⁶ kg.

Given large correction factors due to pixel blurring, red leak and the balance of ice particle to molecular scattering, the numbers above are intended to provide only an approximate order of magnitude to assess plausibility. Since the total amounts of implied material are of the same order as those of Roth et al. (2014a), we conclude that it is possible water plumes could be producing the apparent absorption features.

The original hypothesis being tested by Roth et al. (2014a) was that, following the example of Enceladus, the plumes were modulated on the orbital period of Europa (3.55 days) and that maximum activity would occur at apoapsis. The true anomaly values for the three transits potentially indicative of activity, are intermediate between apoapsis and periapsis (see Table 1 and Figure 1). These three candidate images are grouped together covering a time period of ∼2.5 months, though with a gap on 2014 March 24 (Figure 16), hinting at a protracted period of activity. If this were to represent a timescale of activity, the two month period is ∼19 orbits of Europa about Jupiter, many times the Europa orbital period (3.55 days); hence the initial expectation of an activity cycle that correlates with Europa's location in its orbit, similar to Enceladus and as proposed by Roth et al. (2014a), is not valid.

Also, despite almost contemporaneous observations during early 2014, Roth et al. (2014b) failed to detect any plume activity. There are possible physical explanations: the plumes may be genuinely transient if their activity is governed by tidal stresses as Europa orbits Jupiter. The observing programs of Roth et al. (2014a, 2014b) were designed to test this hypothesis, and they concluded that other factors must be involved. Variations in the plasma environment cause large variations in auroral activity, and even if plume activity is fairly constant, or slowly varying, rapid fluctuations in the ambient plasma can result in substantially different detectability of the plume emission sought by Roth et al. (2014a, 2014b). The upper limits of Roth et al. (2014b) are typically a factor of a few less than their detection, which in principle could be due to decreased excitation rather than diminished plume activity. Our technique is insensitive to the plasma environment.

The feature of 2014 March 17 is close to the crater Pwyll, but is significantly displaced to the north, by approximately 65 of latitude, or 177 km. Figure 23 plots an ellipse of dimensions 83 × 34° centered at longitude 27568, latitude 164S, corresponding to the approximate uncertainty in the location of the feature from the HST image. The apparent extent of the feature along the limb is ≈6.6 pixels and we assume an uncertainty of 2 pixels orthogonal to the limb. Fagents (2003) summarizes a wide variety of models that could explain cryogenic volcanism on the surface of Europa, though acknowledging that despite the geological youth of the surface, there may be other explanations for the features considered. There is a great deal of complex terrain within the ellipse shown in Figure 23, and several features reminiscent of those discussed in Fagents (2003) in the context of cryovolcanism. However, we consider discussion of detailed physical processes premature until the plume activity is confirmed (or shown not to exist).

Zoom In Zoom Out Reset image size