Uranus in Northern Midspring: Persistent Atmospheric Temperatures and Circulations Inferred from Thermal Imaging

Images of Uranus were acquired in 2018 September and October using the mid-infrared VISIR instrument (Lagage et al. 2004; Kerber et al. 2014) at the European Southern Observatory's (ESO) Very Large Telescope (VLT, UT3). Two separate filters were used: Q2, with a central wavelength of 18.7 μm (534.2 cm⁻¹) and a full-width at half maximum of 0.88 μm; and NeII_2, centered at 13.0 μm (769.2 cm⁻¹) with a FWHM of 0.22 μm.

The 18.7 μm filter (Q2) is sensitive to the continuum emission from atmospheric hydrogen and thus serves as a probe of upper tropospheric temperatures with a peak contribution from 200 mbar at nadir geometry (contribution functions are provided in Appendix A, Figure 14). Images show enhanced emission at the equator and pole relative to midlatitudes, qualitatively consistent with inferences of midlatitudinal cooling in the upper troposphere dating back to the Voyager era (Flasar et al. 1987; Conrath et al. 1990; Orton et al. 2015).

In contrast, the 13.0 μm filter (NeII_2) is dominated by thermal emission associated with stratospheric acetylene with maximum contributions from pressures near 0.2 mbar. The 13.0 μm images show brighter emission at middle and high latitudes compared to the equator. Because C₂H₂ is a minor and likely variable species (Moses et al. 2018), variation in the 13.0 μm emission can be attributed to variation in either the C₂H₂ abundance, the stratospheric temperatures, or a combination of the two quantities.

Jointly, these images reveal tropospheric temperatures and stratospheric emission from Uranus' northern hemisphere in midspring (subsolar latitudes of 44°–45°), including unprecedented views of Uranus' north pole and the first-ever images of Uranus' rings in the mid-infrared. Analysis of the rings based on these images is presented in Molter et al. (2019). Data were collected in ∼45 minute blocks on six different nights (three nights for each filter) and reduced using ESO's pipeline of standard infrared chopping and nodding techniques. The resulting images (see Figure 1) were flux calibrated by comparison to observed standard stars. We assumed a systematic error of up to 30% in radiance due to fluctuations in sky brightness and up to 20% due to uncertainties in the stellar fluxes (Dobrzycka & Vanzi 2008). Random errors were estimated from the standard deviation of the background sky. To improve the signal-to-noise ratio (S/N) in each filter, the calibrated images from each night were weighted by the inverse of their squared random errors and combined to yield the final, absolutely calibrated mean image for each filter. A summary of our 2018 observations is provided in Table 1.

Zoom In Zoom Out Reset image size

Table 1.  Summary of 2018 Observations

Filter (Wavelength)	Date (UT)	Time (UTC)	Airmass	Calibration Star	Sub-observer (Solar) Latitude
Q2 [18.72 μm]	2018 Sep 4	8:14-9:03	1.249–1.316	HD 013596, HD 009692	453 (430)
	2018 Oct 3	5:15–5:54	1.236–1.255	HD 008498, HD 010380	444 (434)
	2018 Oct 13	5:32–6:04	1.240–1.272	HD 011353, HD 040808	440 (435)
NeII_2 [13.04 μm]	2018 Oct 09	5:13–5:55	1.238–1.245	HD 008498	442 (434)
	2018 Oct 14	4:51–5:39	1.234–1.237	HD 008498	440 (435)
	2018 Oct 20	4:49–5:36	1.233–1.271	HD 011353	438 (435)

Note. The VISIR instrument in 2018 had a plate scale of 0.0453 arcsec/pixel.

Download table as:  ASCIITypeset image

The random noise was significant in both images given the weak signal from Uranus' cold atmosphere. For individual images, we estimate an average S/N for individual pixels upon the disk to be as low as ∼1.2 at 18.7 μm and ∼0.9 at 13.0 μm. Combining the nights increased these S/Ns to ∼2.8 and ∼1.6, respectively, varying with signal across the disk. Zonal averaging and meridional binning were employed to significantly enhance the S/N in computed latitudinal profiles.

To evaluate temporal changes in the 18.7 μm images, we compared our new data to Voyager data (the oldest spatially resolved thermal data of Uranus available) following the techniques of Orton et al. (2015), as described in Section 3. In the case of the 13.0 μm stratospheric emission, the Voyager IRIS instrument was unfortunately not sensitive enough to provide measurements, and so changes in the NeII_2-band images were evaluated using VLT-VISIR images dating from near equinox (Orton et al. 2018; see Figure 2).

Zoom In Zoom Out Reset image size

The near-equinoctial data included two consecutive nights of 13 μm imaging from 2009, but no contemporaneous 18.7 μm images were available. We instead substituted a similarly sensitive 19.5 μm image (Q3 filter) to define the tropospheric emission when performing the stratospheric retrievals discussed in Section 3. Due to a previously greater plate scale on VISIR (0.075 versus 0.0453 arcsec/pixel), 2009 images were acquired at a coarser spatial resolution than the 2018 images, but they otherwise present equivalent measurements at a different time and viewing geometry. The latitudinal coverage and relative plate scale for both the 2009 and 2018 observations can be seen in Figure 3.

Zoom In Zoom Out Reset image size

Data from each night were collected in blocks, reduced using the ESO pipeline, and combined to yield nightly averages. We initially attempted to calibrate these images using standard calibration stars, but we found resulting radiances to be 10%–40% less than the disk-integrated values from the more reliable Spitzer data acquired less than two years earlier (Orton et al. 2014a). Ultimately, given the uncertainty in the ground-based calibrations and the relative contemporaneity of the space-based Spitzer observations, we chose to scale the 2009 images such that their disk-integrated radiances equaled the Spitzer values at the equivalent central wavelengths. Details on these observations are found in Table 2.

The 2009 images reveal tropospheric temperatures and stratospheric emission near equinox (L_s ∼ 7°), with subsolar latitudes of ∼7°. S/Ns were ∼6–7 for the 19.5 μm and ∼3–4 for the 13.0 μm images. Though the systematic uncertainties for Spitzer are less than 10%, we conservatively estimate systematic errors as high as 20% to account for changes in viewing geometry and radiance in the intervening 1.5 yr.

To evaluate temporal changes in the tropospheric emission, we followed the method of Orton et al. (2015), which compared synthetic images, derived from Voyager/IRIS spectra, to a collection of ground-based imaging data. In that work, Q-band images (including 18.7 μm and 19.5 μm VISIR data from 2006 and 2009, respectively) were analyzed to assess changes in the meridional distribution of upper tropospheric temperatures between the 1986 Voyager encounter (roughly coinciding with Uranus' southern summer solstice) and the following equinox (southern autumn) in 2007. No significant changes were detected (Orton et al. 2015). Using a similar approach, we compared our 2018 images to the 1986 Voyager data to evaluate potential changes over a longer interval.

First, we forward-modeled emission from the temperatures inferred from the Voyager measurements to produce a synthetic image of Voyager-era emission with 2018 geometry—effectively what we would have observed at 18.7 μm in 2018 if the upper tropospheric temperatures near 200 mbar remained unchanged since 1986. These temperatures were taken directly from Orton et al. (2015), which reanalyzed Voyager 2/IRIS spectra acquired over the four days surrounding the closest approach to Uranus on 1986 January 24. These data covered both the southern and northern hemispheres as the spacecraft passed Uranus at distances ranging from 4.2 to 112 Uranus radii. Spectra were inverted to yield temperatures as a function of latitude and pressure from 70 to 400 mbar and then extended to lower pressures based on Spitzer observations (Orton et al. 2014a). We refer the reader to Orton et al. (2015) for further details on how the Voyager temperatures were retrieved.

We then convolved the forward-modeled images of radiance with point spread functions (PSFs) determined from the corresponding stellar images to model the effects of atmospheric seeing and instrumental diffraction. To account for any errors associated with imperfect navigation and stacking of our three images, we created a synthetic image for each of the three individual observations and combined them in precisely the same way as we had done for the data. Finally, zonal averages (binned 10° in latitude to improve the S/N of the data) were extracted from both the synthetic and real images, converted from units of radiance into brightness temperatures, and compared as a function of latitude (e.g., see Figures 6 and 7).

Because Orton et al. (2015) found no significant changes in the upper tropospheric temperatures between the 1986 solstice and the 2007 equinox (using a wider range of VISIR data and the same radiative-transfer code applied in the present study), we chose to only assess changes between the 1986 Voyager data and our 2018 18.7 μm data.

Evaluating temporal changes at the roughly 0.2-mbar heights sensed by the 13.0 μm (NeII_2) filter (see contribution functions in Appendix A, Figure 14) required a more complicated approach combining radiative-transfer forward modeling and retrievals. Because the Voyager/IRIS measurements were insensitive to the weak emission at 13 μm, we drew upon archival 13 μm images from 2009 for comparison (see Section 2.2). However, given the nine year difference between the 2018 and 2009 images, a direct, quantitative comparison required accounting for differences in the observing geometry that can affect the observed radiance as a function of emission angle. To account for this, we took the following approach: first, starting with a prior 1D model of Orton et al. (2014a), we retrieved zonally averaged temperature and acetylene profiles from the observations, accounting for the dependence on emission angle, and used these retrieved values to create a model of the atmosphere (i.e., the temperature and acetylene as a function of pressure and latitude). Then, as we had done with the Voyager temperatures, we computed the radiances emitted from this atmospheric model at different desired emission angles, resulting in a synthetic image (i.e., a forward model) of the planetary emission at a different viewing geometry.

In practice, gradients in the observed emission in our data can be equally explained by gradients in the stratospheric temperatures or the acetylene mixing ratios. We performed retrievals in which one of the two parameters (either temperature or acetylene) was held fixed above the 1 mbar height (assuming values of Orton et al. 2015 and Moses et al. 2018) while the other parameter was free to vary. We found 1 mbar to be a reasonable pressure boundary for separating the 13 μm filter contribution from the deeper 18.7 μm contribution. Although the 13 μm contribution function showed some sensitivity to pressures greater than 1-mbar, enhancement in parameters at these deeper levels was found to produce excessive limb darkening, strongly inconsistent with the center-to-limb variation in the observations. Regardless, retrieved models of either stratospheric temperatures or acetylene worked equally well for forward modeling the emission.

All retrievals were performed on the contemporaneous image pairs using an optimal estimation retrieval algorithm, NEMESIS (Irwin et al. 2008). Calculations used collision-induced opacity based on Fletcher et al. (2018), C₂H₂ and other hydrocarbon line data from the GEISA-2003 compilation (Jacquinet-Husson et al. 2011), and precomputed k-distributions based on the VISIR filter properties. Additional details of the retrieval and forward modeling process are included in Appendix B.

With the retrieved parameters, we updated the prior 1D model (as a function of pressure) and created a 2D model (as a function of pressure and latitude). For latitudes beyond the observed domain, values were simply extended from the edge of the domain. The retrieved atmospheric models were forward modeled at different viewing geometries using NEMESIS and the same spectroscopic data as used for the retrievals.

To validate our modeling approach, synthetic images using unchanged viewing geometries were also created; these forward-modeled emission were convolved with appropriate PSFs, degraded with synthetic noise, and compared to the data. The added synthetic noise was modeled as an array of pseudorandom numbers from a normal (Gaussian) distribution with a standard deviation equal to the standard deviation of the sky in the images. As Figures 4 and 5 show, the synthetic images created from the forward models simulate the data very accurately, providing confidence in our characterization of the radiance, PSF, and random noise. The only significant deviation is near the limb in modeled 13.0 μm images, which appears only marginally brighter than the data. This likely suggests that the upper stratospheric temperatures are increasing too rapidly with increasing height in the adopted temperature model (Orton et al. 2014a) at altitudes above 0.1 mbar. Zonal averages derived from the data and models were also compared to demonstrate the validity of the modeled emission and the effects atmospheric seeing has on the inferred brightness temperatures (see Figure 6).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

In summary, models of atmospheric temperatures and acetylene were constructed from retrievals that accounted for the specific viewing geometry. These models were used to produce synthetic images that simulated the original data at alternative viewing geometries, enabling us to compare thermal emission from the different epochs.

We compared observations and zonally averaged 18.7 μm brightness temperatures from 2018 to 1986, as shown in Figure 7. The observed 18.7 μm image and the synthetic image derived from the Voyager temperatures appear remarkably similar after accounting for random errors and the effects of diffraction and atmospheric seeing. The 2018 data appear only very slightly dimmer than the Voyager data everywhere except along the north edge of the planet. This can also be seen in the meridional plots of brightness temperatures derived from the images (Figure 7).

Zoom In Zoom Out Reset image size

For the meridional plots, first the 2018 brightness temperatures were simply extracted directly from the observed 2018 image and the synthetic image that had been convolved with a PSF to mimic the effects of diffraction and atmospheric seeing suffered by our ground-based observations. By this comparison, the brightness temperatures in 2018 and 1986 are equivalent to within 0.2 K at most latitudes—roughly equal to the level of statistical uncertainty (3σ, estimated from the pixel-to-pixel standard deviation of the background sky divided by the square root of the number of pixels in our zonally averaged bins). Only along the southern flank of the equatorial maximum do differences slightly exceed these estimated uncertainties, but this sharp gradient is particularly sensitive to the image resolution and the modeled PSF.

To reduce potential differences due to imperfectly modeling the blurring, we additionally compared the 1986 brightness temperatures to equivalently blurred models of the 2018 data. These brightness temperatures were extracted from the blurred, modeled image derived from the 2018 retrieval (panel (d) in Figure 4 and the corresponding thick, solid curve in the upper-right panel of Figure 6). In this case, the 1986 and 2018 brightness temperatures are remarkably similar, with a difference of less than 0.1 K at most latitudes. Subtle differences remain near the southern edge of the equatorial maximum and the north pole. In either case, the north pole is just over 0.2 K warmer in the 2018 data compared to the 1986 data. Some of this polar brightening may be attributed to a contribution from the rings, which appear near the pole but are not simulated in our data; however, from modeling, we estimate that this contribution would account for less than 10% of the observed difference.

Retrievals from the data suggest these small differences in brightness temperatures could be produced by atmospheric temperature near 100 mbar changing by 0.3 ± 0.1 K. However, given the limited information on the vertical profile from a single image, slight changes in the retrieved profile at poorly constrained heights can partly offset values at the peak of the contribution function, leading to retrieval uncertainties that are comparable to or larger than these changes. Even for the directly observed radiance, it is also important to note that we are comparing ground-based imaging to PSF-convolved forward models based on retrievals from spacecraft data, and so small systematic errors may remain despite our attempts to account for the observational differences. Considering the uncertainties, these results are consistent with changes of no more than 0.3 K in the brightness temperatures between 1986 and 2018, similar to what Orton et al. (2015) determined evaluating images dating from near equinox (ca. 2007), but with possible warming at the north pole at the limit of our uncertainty.

A comparison of zonally averaged brightness temperatures at 13 μm from 2018 to 2009 images revealed a persistent peak in emission at northern midlatitudes with a minimum near the equator (Figures 6 and 8). The average 2009 distribution appears to be asymmetric with a minimum just south of the equator and a northern midlatitude peak in brightness temperature that is roughly 1.0 ± 0.2 K warmer than the corresponding peak in the south (see Figure 8). We note that the 2009 data showed zonal-mean variations in brightness temperatures of less than 0.5 K between the two consecutive nights, except at the equator where differences approach 1 K (Figure 8, left panel). These internightly differences are mostly consistent with the level of random noise in the images, but they may hint at coherent longitudinal variation (see Appendix C for more discussion on zonal variability).

Zoom In Zoom Out Reset image size

A comparison between the 2009 and 2018 brightness temperatures extracted directly from the data at overlapping latitudes (up to emission angles of 78.5°) show agreement near the center of the disk, but display discrepancies at higher latitudes. Our modeling shows that these discrepancies can largely be explained by differences in the viewing geometries (see Figure 8, middle versus right panel). We showed this by essentially remapping our 2009 temperature model onto the 2018 viewing geometry (such that matching latitudes have identical emission angles) and forward modeling the expected emission (see Figure 9). Zonally averaged brightness temperatures drawn from this forward model show no significant change in brightness temperatures between 20° and approximately 74° north latitude. North of these latitudes, the temperature and acetylene were not actually retrieved in 2009 because of close proximity to the limb, and therefore our model of the 2009 atmosphere simply assumes an extension of the retrieved temperatures and acetylene at ∼74°. With this assumed atmospheric structure north of ∼74°, the forward-modeled brightness temperatures are colder but still roughly consistent with the 2018 observations. However, south of 20° N, we find that our modeled brightness temperatures are significantly greater than the observations. Due to the degeneracy in our retrievals, we find the same result regardless of we vary temperatures or acetylene in our 2009 retrievals.

Zoom In Zoom Out Reset image size

If the southern hemisphere brightness temperatures from 2009 had persisted, our modeling indicates that we should expect to see a brighter southern limb in 2018, with blurring contributing to a brighter equator. As Figure 8 shows, this southern emission appears absent in the data. However, as Figure 9 shows, the difference between data and model demonstrates that this discrepancy exists along the entire limb and likely indicates a failure of the forward modeling of the limb rather than a physical change limited to the southern hemisphere. The modeled brightness along the limb is particularly sensitive to the assumed PSF (which convolves the disk and sky) and the vertical gradient in temperatures and acetylene, and so errors in either can produce modeling discrepancies. As we noted earlier, this can indicate that the upper stratospheric temperatures are increasing too rapidly in the adopted temperature model (Orton et al. 2014a) at pressures less than above 0.1 mbar. Temperatures (and acetylene) at these low pressures are not retrieved in our model because the contribution functions peak deeper in the atmosphere where the acetylene abundance is expected to be greater. Lower pressures are only sensed at the very edge of the disk, where even small amounts of uncorrected blurring between the planet and background sky can suppress observed radiances and lead to erroneously lower retrieved temperatures; hence, these locations (corresponding to emission angles beyond ∼72.5°, or μ < 0.3) were not included in our retrievals. Further investigations of the thermal center-to-limb behaviors and temperature lapse rates are beyond the scope of this paper but should be a goal of future work.

To investigate sources of the stratospheric emission, we performed retrievals of temperatures and acetylene from the data. If stratospheric temperatures were allowed to vary (while acetylene was held fixed), we found the observed meridional pattern can be reproduced by a temperature gradient of roughly 13 K (13.4 ± 2.8 K in 2018 and 12.4 ± 2.8 K in 2009), measured at 0.25 mbar—roughly the pressure at which the 13 μm contribution function peaks—from the near-equatorial minimum to the northern midlatitude peak at ∼40° (see Figure 10). In the southern hemisphere, the peak temperature was roughly 4 K less. The variation in radiances was equally reproduced by a 500%–600% increase in the acetylene mixing ratio at northern midlatitudes ((2.7 ± 0.7) × 10⁻⁶ volume mixing ratio (VMR) increase in 2018 and (2.0 ± 0.6) × 10⁻⁶ increase in 2009) relative to the expected values of Moses et al. (2018) and those retrieved at the equator. There was roughly 20% less at southern midlatitudes compared to the north, but these southern latitudes are only sampled in 2009 nearer the edge of the disk (see Figure 11).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

While an increase in stratospheric acetylene had no effect on the inferred underlying upper tropospheric temperatures, we found that stratospheric warming would also be detectable at 18.7 μm because of the broad contribution function of the Q2 filter. Therefore, we find that warming in the stratosphere must be compensated by cooling the troposphere to remain consistent with the observations. If the 13.0 μm emission is due to increased stratospheric temperatures, then the inferred tropospheric temperatures may be up to 1 K colder at the midlatitudes relative to those inferred from the Voyager/IRIS data (assuming the longer wavelength Voyager measurements were not equally influenced by the contribution of warmer stratospheric temperatures). This difference is nearly comparable to the potential uncertainty in retrieved temperature profiles, but it would be consistent with cooling of the northern (autumnal) hemisphere following equinox.

The 1986 Voyager data reveal a meridional cross section of Uranus' temperatures near the tropopause shortly after the southern summer solstice. If we assume the atmospheric temperatures respond to solar heating and the radiative relaxation times approached a full season, the 1986 Voyager data would have displayed the atmosphere's thermal response to solar forcing at the time of the preceding equinox in 1966 (Conrath et al. 1990). It then follows that the equinoctial data analyzed by Orton et al. (2015) would have provided a view of both hemispheres responding to asymmetric solstitial (southern summer, northern winter), and our 2018 observations would show the northern hemisphere's transition from winter to spring. The northern midlatitudes remained in darkness for nearly a decade past solstice in 1986, so it is plausible that cooling may have progressed beyond the 2007 equinox at these latitudes if the seasonal response lagged a full season behind the solar forcing.

Though uncertainties are significant, our data show that changes near the tropopause are consistent with atmospheric temperature changes no greater than ∼0.3 K since solstice, though perhaps slightly larger if potentially elevated stratospheric temperatures contributed to our retrieved temperatures (see Section 4.3). These values are consistent with the conclusion of Orton et al. (2015), who reported brightness temperature changes of less than 0.4 K between equinox and solstice.

The seasonal amplitude expected from radiative forcing will in general be reduced by a factor of $1/\sqrt{1+{r}^{2}}$, where r = 2πτ_R/τ_O and τ_R/τ_O is the ratio of the radiative time constant to the orbital period (Conrath et al. 1998). Conrath et al. (1990) showed that a seasonal forcing of 10 K and a radiative time constant of 130 yr should result in changes of 1 K or less. Using the updated, shorter time constants of Li et al. (2018), this value would be as large as 2–7 K with subseasonal lags (although this assumes the same 10 K forcing, which may not be consistent with the radiative heating applied by the authors). The amplitude and phasing derived from our data favor timescales of over 300 yr, which, considering the seasonal scales, amounts to essentially no seasonal variation due to solar forcing at midlatitudes. We emphasize that this is again based on seasonal radiative equilibrium temperatures from Conrath et al. (1990). Shorter radiative timescales would suffice if different heating rates and dynamical redistribution are assumed. Bezard & Gautier (1986) computed seasonal contrasts of 4 K or less in the radiative equilibrium temperatures (depending on the latitude dependence of the internal heat flux), which would be consistent with radiative cooling times on order of a century. Shorter timescales computed by Li et al. (2018) would require equilibrium temperature swings of under 1.5 K or simply a greater amount of meridional mixing to overwhelm the radiative response (Friedson & Ingersoll 1987). An updated radiative-dynamical model using the latest absorption coefficients and chemical abundances is needed to better evaluate the consistency of these radiative time constants with the observations.

Higher in the stratosphere, our comparison is limited to a span of nine years. This span is short compared to estimates of long radiative and dynamical timescales on Uranus (Conrath et al. 1990; Moses et al. 2018), and so it is unsurprising to find little change in the northern hemispheric emission at 13 μm. By the same token, it would be surprising if any physical process could produce the apparent discrepancies between the 2018 data and 2009 model, further suggesting that this is likely attributable to an error in the assumed profiles of temperatures or acetylene abundances at higher altitudes that go into producing the forward model.

The cause of the 13 μm emission distribution is unknown, but as discussed in Section 4.3, it could result from a regionally elevated temperature, acetylene, or a combination of the two. Though not obvious from each pair of images alone, maps of the data show a remarkable anticorrelation between the 13 and 18.7 μm radiances (see Figures 12(a), (b), and 13). Furthermore, while the midlatitudes are negatively correlated, both images show brightening at the poles. Together, these correlations suggest that a shared mechanism or mechanisms may be responsible for the emission in both filters, forming a dynamical link between stratosphere and upper troposphere.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

One possible explanation for the observed emission may be elevated temperatures resulting from the adiabatic compression of regionally subsiding gas. Just as the meridional temperature structure in the upper troposphere has been interpreted as a consequence of large-scale vertical motions (Flasar et al. 1987), similar but separate circulation cells could possibly be at work higher in the stratosphere, rotating in an opposite sense to produce downwelling at midlatitudes and upwelling at the equator. The meridional position of this stacked circulation would not be merely coincidental, as it could be linked to the underlying meridional temperature gradient. While the driving force behind the upper tropospheric circulation is unknown (Flasar et al. 1987), the meridional temperature structure it produces has a maximum temperature gradient between 15° and 20° latitude in both hemispheres (with dT/dy < 0). It is conceivable that this upper tropospheric temperature gradient could be geostrophically balanced by an unobserved stratospheric tropical jet aloft via the thermal wind relationship. Potential dissipative processes weakening the thermal wind with height could then lead to a mass-balancing meridional circulation with descending, warming air on the poleward side of the jet (Conrath & Pirraglia 1983). Following Flasar et al. (1987; their Equation (3)), we can relate the desired steady-state temperature differences to a requisite differential vertical velocity by the expression

$$\begin{eqnarray}&&{\rm{\Delta }}w={\rm{\Delta }}{TR}/{\tau }_{R}{{HN}}^{2},\end{eqnarray} \tag{ 1 }$$

where ΔT is the meridional temperature difference, R is the gas constant (∼3614.9 J kg⁻¹ K⁻¹), H is the pressure scale height (∼47 km), N is the buoyancy frequency (0.0048 s⁻¹), τ_R is the radiative time constant, and (Δw) is the vertical velocity difference. Assuming τ_R ∼ 49 yr at 0.2 mbar (Li et al. 2018) and ΔT ∼ 13 K from our retrievals, we calculate Δw ∼ 2.8 × 10⁻⁵ m s⁻¹—five times greater in magnitude than what Flasar et al. (1987) computed for the tropospheric vertical velocity differential.

While adiabatic warming associated with thermal winds could potentially explain the enhanced emission at midlatitudes, a different mechanism would be needed to explain the pattern in emission at higher latitudes. The apparent second peak in emission at the pole would be inconsistent with a thermal wind-driven circulation as the tropospheric temperature gradient is in the opposing direction at high latitudes (i.e., dt/dy > 0.) A separate mechanism such as wave heating or a larger, independent circulation would be needed to heat the pole. Ultimately, assessing the dynamical feasibility of this scheme requires evaluating the thermal wind equation and the corresponding mass conservation for zonal, meridional, and vertical components of the winds, which is beyond the scope of the present data and this paper.

Alternatively, a regional enhancement in acetylene could satisfy the 13 μm emission with a much simpler and coherent dynamical circulation. If the putative midlatitude upper tropospheric upwelling simply extended farther into the stratosphere (∼6 scale heights above the tropopause), tropospheric methane could potentially be mixed to lower pressures where it would photolyze to produce acetylene. As long as chemical conversion timescales were short compared to timescales for meridional transport, the concentration of acetylene would peak in the region of upwelling and diminish quickly to the north and south. Moses et al. (2018) suggest acetylene loss timescales are roughly 2 yr, with net lifetimes of 40 yr at 0.2 mbar. Timescales of meridional transport are uncertain but potentially much larger, with Conrath et al. (1998) suggesting a dynamical timescale of 700 yr in the stratosphere.

The resulting enhancement would be consistent with inferences of latitudinal variation in stratospheric hydrocarbons from the Voyager UVS data (Yelle et al. 1989; McMillan & Strobel 1992). Yelle et al. (1989) reported a factor of 2–3 reduction in the observed reflectance in the 1338–1583 Å spectral band at midlatitudes relative to the pole and interpreted this signal as the relative depletion of hydrocarbons at the poles and enrichment due to extended upwelling at lower latitudes. For consistency, it then follows that the brightness detected at the poles at 13 and 18.7 μm could both be potentially attributed to adiabatic warming in the sinking branch of this extended cell.

Regional enhancement of acetylene would also imply that methane would be similarly enhanced at midlatitudes. Though not detected in near-IR data (Karkoschka & Tomasko 2009), regional enhancements in methane are indirectly consistent with inferences of supersaturated and spatially variable methane from Herschel PACS spectra (Lellouch et al. 2015), which found supersaturated methane mole fractions of ∼9.2 × 10⁻⁵ at the tropopause—roughly six times larger than the value inferred from Spitzer (Orton et al. 2014b). The authors noted that the measurements could be reconciled if the methane profile decreased by a factor of 5 from 100 to 2 mbar, but such a profile would be difficult to explain. They argued that the discrepant supersaturated mole fraction may have instead indicated spatial heterogeneities in stratospheric methane abundance or temperatures, noting that the Herschel data probed more global conditions while the Spitzer data were biased toward warmer temperature regions. Interestingly, our data indicate a factor of 5–6 enrichment in acetylene in precisely the coldest regions (i.e., midlatitudes), while mole fractions at warm equatorial regions remain in strong agreement with values derived from Spitzer (Orton et al. 2014b; Moses et al. 2018), consistent with the proposed explanation for the observed discrepancy.

A vertically extended circulation cell has already been suggested for Neptune to explain correlations between observations in the stratosphere and troposphere (de Pater et al. 2014), and was previously proposed for Uranus by Yelle et al. (1989) and McMillan & Strobel (1992). It is perhaps unsurprising that a similarly extended circulation may be found in the stratosphere on both planets, given their qualitatively similar, broad tropopauses (de Pater et al. 2014), despite the weaker vertical mixing and stronger seasonal forcing on Uranus. In either case, the upward flux of methane may have to exceed the presumed VMR limit imposed by the estimated equilibrium saturation vapor pressure at the tropopause (i.e., the cold trap). Evidence of potentially supersaturated mixing ratios of methane on Uranus and especially Neptune suggest that vertical mixing may indeed overcome the cold trap limitation (Lellouch et al. 2015), though the mechanisms of this process remain speculative (Lunine & Hunten 1989; de Pater et al. 2014). Inversely, the greater vertical extent of the circulation cell would mean that upwelling gas would experience a greater and more extended flux of high-energy photons along its extended path compared to a circulation cell that ceased nearer the tropopause; this would serve to further deplete the gas of chemicals destroyed by photolysis, meaning only the most stable species would be present in the downwelling branches.

A strong coupling of the troposphere and stratosphere would also help to physically link the apparent hemispheric asymmetry seen in the tropospheric temperatures and 13 μm emission. Both Voyager and equinoctial 13-μm imaging data show the equatorial maximum/minimum to be slightly offset to the southern hemisphere (Orton et al. 2015, 2018). Likewise, Voyager spectra and ground-based 18 μm imaging showed the midlatitude temperature anomaly in the northern hemisphere to be roughly 1 K colder (i.e., a ΔT of ∼4 versus 3 K ), while the 13 μm emission from the northern hemisphere in 2009 appears roughly 1 K brighter than the southern hemisphere (a retrieved acetylene VMR of 3.0 × 10⁻⁶ versus 2.3 × 10⁻⁶). Both of these represent a roughly 30% difference that could be explained by upwelling that is greater in the north than in the south. Considering this asymmetry and the potential dynamical link, the apparent absence of emission along the southern limb in 2018 data may be an indication of an asymmetric extent of the upwelling, such that the acetylene mixing ratio and the contribution function peak at greater pressures and hence become less visible in the extended optical paths along the limb. Or perhaps the strength or position of the upwelling has changed in time (e.g., a slight southward shift of the southern hemisphere upwelling) due to deeper dynamical processes. Unfortunately, the southern hemisphere will soon be unobservable from Earth for several decades, so testing theories and confirming changes will have to remain a goal of future work and potentially visiting spacecraft.

In contrast to the elevated temperature scheme due to downwelling, upwelling would actually serve to reduce the temperature at midlatitudes through adiabatic cooling. Therefore, the alternative explanations of stratospheric downwelling and hydrocarbon upwelling would serve to produce temperature anomalies in opposite directions, although the cooling would be relatively minor due to the much weaker updrafts expected at these heights (Flasar et al. 1987). Independent observational constraints on the meridional temperature structure in the stratosphere can come from sensing the H₂ S(1) quadrupole line and thus could help discriminate between these two theories. Trafton et al. (2012) note that 2007 Gemini TEXES observations show a bimodal but asymmetric pattern that is brightest toward the north, but further work is needed to determine the consistency with the above models.

One potential flaw with the extended upwelling mechanism is the apparent inconsistency at the pole and the equator. If downwelling produces adiabatic heating in the upper troposphere (i.e., at 18 μm) at the poles and equator, why does stratospheric downwelling appear to heat the stratospheric pole but not the equator? Even if the effect of the adiabatic heating was offset by the reduced acetylene, it is not obvious why the equator and pole should be different. Considering mass conservation, the considerably lesser surface area of the polar regions relative to the equatorial regions may naturally be a factor in producing greater adiabatic heating at the pole, but then we would also expect a similar effect in the deeper 18.7 μm emission unless the area of subsidence expands with increasing depth. One possible solution is that the pole is also heated by independent dynamical mechanisms such as breaking waves or greater concentrations of absorbing aerosols, but this remains purely speculative. More detailed dynamical modeling will be necessary to help evaluate the circulation in the stratosphere, its link to the troposphere, and its overall changes in time.

A comparison of the mid-infrared images to visible and near-IR imaging, dominated by absorption from methane and the scattering from clouds and hazes at greater pressures, shows some possible signs of correlations between the stratosphere and troposphere (Figure 12). Although the 13 μm emission extends ∼15° farther to the south, the boundary of the brighter polar northern haze layer in a near-IR Hubble Space Telescope (HST) image is at roughly the same latitude where the 13 μm emission peaks (∼40°). Intriguingly, the bright discrete cloud features near 40° in the HST image also roughly correspond in latitude to the brightest discrete signals seen in the nightly averaged 13 μm images. The tenuous correlation might suggest that tropospheric vortices generate localized hydrocarbon enhancements through upwelling (de Pater et al. 2014) or localized heating via breaking waves, although a thorough analysis of the image noise suggests that these radiances would constitute noise at the ∼2.5σ level and hence be inconclusive (see Appendix C). Establishing a link between upper tropospheric vortices and the stratosphere will require greater signal-to-noise imaging in future work.

If we assumed hemispheric symmetry in the meridional structure of the 13 μm polar emission, we can compare our thermal model to images from Voyager. The polar aerosols seen from this perspective show an enhanced brightness very near to the pole in Voyager's orange (590–640 nm) and methane (614–624 nm) filters, with an extent that is similar to the pattern seen in our models derived from the stratospheric emission. The sharper increase at the pole is not evident in the Voyager/IRIS thermal spectra, which show a more gradual warming roughly coincident with the polar haze region. We note that we do see a similar sharper feature at the north pole in our 18.7 μm 2018 data (see Figure 4) even though it appears absent at the synthetic images generated from the Voyager spectra. If linked across the troposphere and stratosphere, these observations may indicate that stratospheric downwelling or photochemistry at the pole affects the albedo or abundance of the tropospheric haze layer ∼9 scale heights below. This may potentially be accomplished by methane-depleted air settling from above, combined with photochemical products produced in the higher ultraviolet flux found in the stratosphere. Assuming the parent chemicals were derived from upwelling gases, the aerosols or particulates would have to be stable enough to survive meridional transport from midlatitudes to the pole and down to the cloud level, potentially over centuries (Conrath et al. 1998).

Considering the recent detection of hydrogen sulfide above the cloud layer (Irwin et al. 2018), a possible candidate for stratospheric aerosols may be stable sulfur allotropes, investigated as a possible source of haze in exoplanetary atmospheres (Zahnle et al. 2016; Gao et al. 2017). Trace amounts of upwelling H₂S could photodissociate in the stratosphere and combine with free H atoms (produced from the photolysis of CH₄) to eventually form stable octasulfur allotropes (S8) capable of surviving further UV radiation and the range of temperatures found above the clouds. If carried with the circulation and concentrated at the pole before settling down upon the clouds, these yellow particulates (Meyer et al. 1972; Eckert & Steudel 2003) could potentially explain the scattering preferentially seen in the orange (614 nm) filter relative to the blue (477 nm) and violet (431 nm) filters (Smith et al. 1989). It is, however, worth noting that this is purely speculative, and several other species of hydrocarbon polymers absorb in the violet and may potentially explain the visible spectra (Baines & Bergstralh 1986; Pollack et al. 1986).

Polar aerosols could also potentially explain the slightly warmer summer temperatures at the poles in the lower stratosphere seen in the Voyager data (Orton et al. 2015). If this were due to solar heating, such changes would only be expected from the radiative-dynamical model of Conrath et al. (1990) if radiative time constants were significantly shorter than even those of Li et al. (2018) at stratospheric heights. This could potentially be caused by unaccounted presence of absorbing aerosols. Likewise, the broader polar brightness seen in the model 18.7 μm emission derived from the Voyager IRIS spectra may indicate broader subsidence or additional heating associated with overlying hazes or deeper tropospheric aerosols seen in the Voyager imaging. However, in terms of seasonal changes, the vertical velocities implied by the temperature variations (∼10⁻⁵ m s⁻¹) are still comparable to or smaller than the expected fall velocities of photochemical aerosols (Toledo et al. 2019). Therefore, if estimates of settling times and vertical velocities are correct, downwelling may not significantly aid the transport of material from the stratosphere to the cloud level on timescales short enough to explain seasonal changes in the appearance of polar regions. Therefore, any apparent correlations in structure may be signs of the mean annual signatures of transport imprinted on the seasonally varying albedo of the haze layer, itself controlled by mechanisms that have yet to be fully explained. If seasonal changes are related to temporal changes in the methane mixing ratio above the cloud layer, it is conceivable that this process may be aided or triggered by a seasonal increase in the rate of subsidence. If true, this process may be detectable by measuring changes in the polar thermal emission, although as our data show, it is challenging to measure the polar regions near equinox given the geometry seen from Earth.

If variations in the seasonal albedo are related to changes in temperatures—whether through changes in condensation, subsidence, or convective stability—these temperature changes are not apparent in our upper tropospheric or stratospheric data. Li et al. (2018) computed radiative time constants as short as a decade near the polar cloud tops (Sromovsky et al. 2019), but the variation of cloud layer temperatures have yet to be measured. We have shown that radiative time constants at higher altitudes are either longer than expected or mixing can effectively dampen seasonal temperature changes. If meridional mixing is present in the variable cloud layer as well, it apparently does not traverse the boundary at 40° latitude.