Observations of Loki Patera by Juno during Close Flybys

A. Conrad et al. (2015) showed that detailed views of Loki Patera, captured through ground-based direct imaging, reveal emissions from two distinct areas within the patera’s floor. According to their findings, the feature in the southwest corner likely represents a persistent high-temperature source, while the emission in the southeast is probably due to radiation from the nearby cooling crust. However, the observations at much higher spatial resolution from JIRAM reveal a more complex morphology. The maps from 2023 May (panels (A) and (F) of Figure 1) show a maximum radiance in the northern part of the lake, which partially saturates the M-band channel. We interpret this as relatively young crust floating above underlying magma and cooling. In the maps from 2023 December and 2024 February, which have very good spatial resolutions (orbits 57 and 58; panels (C), (D), (I) of Figure 1), the crust appears to have cooled, but there is a maximum radiant emission near the patera’s wall to the south, like what was observed in NIMS data (R. M. C. Lopes et al. 2004). In the same maps, some regions of intensification of the brightness are visible close to the promontories that are at ∼15°N and ∼16°N to the east of the lake; these are more evident in Figure 3. Additionally, two clear signatures of newly exposed lava were observed in 2024 April (panels (E) and (L) of Figure 1): one to the west near the patera’s wall and one to the east close to the island’s shore. In the earlier observation from 2004 February, a minor lava feature was visible to the west as well, but not exactly at the same location (panels (D) and (I) of Figure 1).

The presence of intense radiance near the patera’s wall has been found to be common in many lava lakes (A. Mura et al. 2024b) and is suggested to indicate exposed lava due to the breakup of the crust near the patera’s wall. A possible reason proposed for this phenomenon is a piston-like movement of the entire lake’s surface, without lava spilling out of the patera’s rim. This idea is supported by the observations made by the Voyager 1, Galileo, and Cassini spacecraft, that the shape of the low-albedo region was unchanged between 1979 and 1997 (e.g., A. S. McEwen et al. 1998; D. R. Allen et al. 2013). However, given Loki Patera’s size—it is more like a sea than a lake—it is also reasonable to consider an alternative hypothesis where the surface does not rise and fall uniformly. Instead, the phase of this movement could vary across the lake. Indeed, in orbit 57, there is an intensification of the radiance close to the patera’s wall on the south side, while in orbit 58, there is an intensification close to the patera’s wall on the north side, near the promontories. This variation may be partially due to the observation geometry, as the patera’s walls can be very steep and high, potentially blocking part of the radiance from the lake sections closest to the walls (A. Mura et al. 2024b). It is also notable that the eastern edge, which should be visible without obstruction in orbit 58, does not show any significant excess of radiance, as if there is no exposed lava close to the walls in that part of the lake. One possibility is that, instead of the piston movement, a sloshing-type movement (characterized by alternating upward and downward displacements on opposite sides of the surface) occurs in Loki Patera. In this case, at a given time, some part of the crust might move vertically while other sections remain relatively stable, thus not producing crust disruption and lava exposure. This model is not yet tested but may be viable, given that Io’s intense tidal forces applied to such a large lake could potentially drive such a phenomenon (tides at Loki Patera have been discussed by D. L. Matson et al. 2006; see their Figure 14). Another possibility is that the temperature difference is due to the nonuniformity in the lava’s volatile content and/or the crust’s bulk density across the patera, as proposed by K. de Kleer et al. (2017). In a final scenario, this signature could reveal lava erupting from source regions in either the southwest or northwest, then extruding out and sweeping across the floor. Thermally, all these scenarios may appear similar, so distinguishing between them is challenging.

We start the analysis by noting that the three zones introduced in Section 2 (Figure 1) have temperatures that are systematically different: zone 1 is always colder, and zone 2 is always hotter. L-band and M-band derived temperatures are in very good agreement: the correlation coefficient is greater than 0.99, there is no systematic bias between the two (they have the same mean value). We conclude that the possible presence of small, cold islands in the lava lake (which is indeed true, as discussed later), or of hot cracks, or, finally, of some calibration residual of the instrument, is not affecting the estimation of the temperature in a significant way.

With this premise, we can say that the cooling of the crust, observed for 330 days, aligns with the temporal model proposed by K. de Kleer & I. de Pater (2017), which we generalize and discuss in Appendix A.2. The model is represented by the three black dotted curves in Figure 2. The t₀ date of the model (i.e., the beginning of lava cooling) is not known, and it is found by best-fitting the model against the data. The three onset dates with respect to the last observation on 2024 April 9 (t₀) are: t₀₁ = t₀ − 481 days for zone 1, t₀₂ = t₀ − 373 days for zone 2, and t₀₃ = t₀ − 418 days for zone 3. In other words, at the time of the last observation, the crust was 481, 373, and 418 days old in the three zones, respectively. These results indicate that the temporal sequence of cooling initiation in the three zones is: zone 1 (south), then zone 3 (east), and finally zone 2 (north). This could be the effect of the propagation front of a resurfacing wave that moves in counterclockwise direction, the same as observed in the Galileo PhotoPolarimeter/Radiometer (PPR) data (J. A. Rathbun et al. 2002). The phenomenon is further discussed in Section 3.4, where we use the terms “resurfacing wave” and “phase velocity” to indicate the movement of the ideal instantaneous location where the surface of the lake starts to cool down, without implying any specific physical model. Alternative models are introduced in the discussion (Sections 4 and 5).

In Appendix A.2 we describe two possible scenarios for the cooling of the crust. Scenario A assumes the same initial conditions as K. de Kleer & I. de Pater (2017) (i.e., fully molten magma); scenario B refers to a case in which 5 m of magma are cooling on top of some preexisting crust. While scenario A provides an adequate fit to the temperature data, scenario B does not. As summarized in Appendix Figure A3, the cooling in scenario B occurs too rapidly to match the observed temperatures. We therefore conclude that hypothesis B (lava cooling on top of preexisting crust) is not compatible with the observations unless the lava is thick; at least 5 m of crust would have been produced for each cycle in that scenario.

The cyclic brightening of Loki Patera is thought to result from large-scale resurfacing processes within the lava lake; it is estimated that the entire cycle lasts approximately 540 days, aligning with observed brightening events before 2001 (J. A. Rathbun et al. 2002; D. L. Matson et al. 2006). I. de Pater et al. (2017) and K. de Kleer & J. A. Rathbun (2023) show that recent activity exhibits a slightly shorter periodicity (490 days) compared to what J. A. Rathbun et al. (2002) observed. Our observations do not cover a full cycle, but they are also consistent with this shorter period. In fact, the bright area is observed in the southwest as in previous observations; the hotspot observed in the western edge of Loki Patera in 2024 April is very close to the location where PPR observed the beginning of an eruption in late 1999. Furthermore, the observation of 2024 April was obtained approximately 490 days after our calculated previous eruption in 2022 December, suggesting that this spot is the beginning of a new eruption event.

There are different alternative explanations for the cyclic activity of Loki Patera, other than that it is driven by episodic lava flows (D. L. Matson et al. 2006). T. K. P. Gregg & R. M. Lopes (2008) proposed that a brief eruption would deposit minimal fresh lava on Loki Patera’s floor; most new lava would form a shallow intrusion via channels and tubes, with a significant portion possibly draining back through the main vent, which corresponds to the thermally bright zone observed in the NIMS data. This model helps explain why lava has not overflowed the patera rim despite multiple brightening events, interpreted as eruption episodes. J. A. Rathbun et al. (2002) proposed that the brightening cycles are due to periodic crust overturns rather than lava flows resulting from cyclic eruptions. They also suggested that such periodic overturns can happen if specific conditions are met. The periodicity occurs because, initially, the solidified crust on the lake must be stable, but as it thickens and cools, it may become unstable. Sinking slabs of crust may thus initiate convection within a magma “sea” (D. L. Matson et al. 2006). This shift in stability might occur due to thermal contraction alone. In the case of typical terrestrial basalt, the solidified crust has a density significantly higher than that of the liquid beneath it, but the crust’s porosity might contribute to its initial stability, preventing immediate overturn. K. de Kleer et al. (2017) modeled their observations according to this scenario and pointed out that the flooding one is not very plausible, because that would require two distinct pulses of magma reaching different parts of the patera at separate times to explain the ground-based observations by Keck. A foundering crust model makes it challenging to retain numerous islands that have remained unchanged for decades, unless the lava is excessively fluid (R. R. Howell et al. 2014), which is possible with ultramafic lavas, which have been proposed but not confirmed for Io (L. Keszthelyi 1999). We note that recent observations of Loki Patera with the JunoCam visible camera reveal a widespread bright, specular reflection across the patera floor at low phase angle, consistent with a fresh and regionally uniform and flat lava surface, typical of lava lakes but perhaps possible with extensive flows (M. Ravine et al. 2024).

Table 1 presents the total power emitted by Loki Patera. The blackbody radiant emittance (in W m⁻²) is derived from the brightness temperatures using the Stefan–Boltzmann law and integrated over the entire lake. This calculation can be performed using either the M-band or L-band brightness temperatures (fifth and sixth columns). As shown, the two values are in close agreement (a nonuniform temperature across the crust, such as in the case of frequent cracks, would result in a substantial difference between the M- and L-band temperatures: see the calculation in Appendix A.4 about the central island). The single-band power output (in both the M and L bands, third and fourth columns) does not correlate well with the total power, as previously observed by A. Mura et al. (2024a).

Once the temperature temporal profile is modeled, and the phase of the onset across the lake is determined from data, calculating the total heat output from Loki Patera becomes straightforward, as demonstrated by D. L. Matson et al. (2006). We replicated this specific theoretical calculation based on our data, and the result does not differ substantially from that of D. L. Matson et al. (2006). Moreover, we note that the empirical values of total power output we observe (Table 1) are in very good agreement with the theoretical model. Additionally, it is possible to calculate the time-averaged total power, yielding a value of ∼10 TW, which aligns very well with previous literature (e.g., I. de Pater et al. 2017, Table 9).

K. de Kleer & I. de Pater (2017) modeled the overturn process proposed by J. A. Rathbun et al. (2002), finding a propagation speed of 1 km day⁻¹. Such speed is highly sensitive to the properties of the magma or the physical process by which the crust recycles, which suggests it could differ between brightening events (J. A. Rathbun & J. R. Spencer 2006). The varying resurfacing rates between the northern and southern regions suggest differences in magma properties, such as composition or gas content, rather than local topography, since the floor of the patera is relatively flat; the southern part of the patera has a resurfacing rate twice that observed in 2001, indicating more variable lava properties (R. R. Howell & R. M. C. Lopes 2007). In our study, we do not observe such a difference, instead the northern part has a faster resurfacing rate as discussed in Appendix A.3.

K. de Kleer et al. (2017) suggested two overturn waves. The first one was moving in a clockwise direction from the northwest at 1 km day⁻¹, while the second one was moving in a counterclockwise direction from the west at 2 km s⁻¹ and started 70 days later. The two fronts should meet approximately in the east or southeast, where our zone 3 is. However, the times at which the resurfacing wave arrives at the three locations in this study indicate that only a single wave, moving counterclockwise from the eastern part of the lake, was present (this is also apparent by looking at radiance maps, where the north, and not the east, is the brightest region). The distance from zone 1 to zone 3 is approximately 100 km, and it is 120 km from zone 3 to zone 2. Hence, the timings are consistent with a phase velocity of ∼2 km day⁻¹ between zone 1 and 3 and ∼3 km day⁻¹ from zone 3 to zone 2 (there are, in fact, ∼65 days between t₁ and t₃ and ∼45 days between t₃ and t₂). This implies that either the clockwise wave was not present in the event we observed or it started much later than the counterclockwise one (and not earlier as reported by K. de Kleer et al. 2017). We conclude that each resurfacing event may have its own characteristic as proposed by J. A. Rathbun & J. R. Spencer (2006); this is also in agreement with I. de Pater et al. (2017), who reviewed 30 years of observations of Loki Patera and noted a reversal in the direction of propagation of resurfacing waves. Since 2009, the resurfacing wave has moved in a clockwise direction, beginning in the north or northeast corner and spreading toward the southwest. In contrast, during the Galileo era, the resurfacing wave propagated counterclockwise, starting in the southwest and moving toward the east. We are now once again observing a counterclockwise propagation.

To further analyze the propagation of the resurfacing wave, we present a similar calculation in Appendix A.3, but repeated across 15 different regions. The results are evident: the progression of the resurfacing wave is consistently revealed by the data, indicating an average speed of 2–3 km day⁻¹, with a slightly higher velocity observed in the northern region. The crust thickness at the time of the last observation ranges from 5 to 6 m.

A persistent island in the middle of Loki Patera has been observed since the Voyager 1 flyby of 1979 March, remaining stable despite surrounding lava eruptions. This island is likely an anchored feature prominent above the floor and the lava surface, rather than a floating raft, and similar islands are seen in other paterae such as Tupan (R. M. C. Lopes et al. 2004). The JIRAM data reveal no visible channels or cracks on the central island, except for one located in the south, which appears in the M-band radiance maps from both orbits 57 and 58. This feature resembles a thin arc extending from the northwest to the southeast. Aside from this, the island appears largely featureless. In contrast, the image in the visible spectrum displays several channels where thermal emission was previously detected in the infrared (R. Lopes-Gautier et al. 2000) and could potentially be detectable by JIRAM, assuming the channels still exist. We propose that one possible explanation for JIRAM’s inability to detect these channels or cracks is their depth, combined with the misalignment of their axes relative to the line of sight during the observation. Estimating the depth of these features is challenging, as it requires knowledge of the channel width and the inclination of its walls—parameters that remain unknown. In Appendix A.4 we show that, in fact, there is some extra emission from some of these channels, although very faint.

The combined analysis of the L- and M-band radiances in a fashion similar to that performed in Figure 2 (that is, defining a zone inside the island and calculating the temperature as a function of time) is quite challenging because the radiance is very close to the noise level. However, we note that, on average, the L-band derived temperatures are about 20% higher than the M-band ones (255 K versus 210 K); a possible explanation is that there are many small areas (subpixel) that contribute to the average radiance without being visible in the image (as detailed in Appendix A.4). This hypothesis is supported by the fact that the central island is significantly warmer than the region that surrounds Loki Patera, as would be expected if the island were being heated by the active lava lake.

Numerous smaller colder features were observed by Voyager 1 in different locations across the lake, but it was uncertain whether they were actually anchored to the floor. The comparison with recent JIRAM data shows that at least 20 of them are most likely real anchored islands since they could not have remained in the same position if they were floating. Some minor evidence suggesting that these features are anchored to bedrock, rather than floating crust, is the fact that their temperatures match the coldness of the surrounding plains.

The short distance between islands (separation is ∼10 km), combined with their small size, points to a lake level that has not changed since 1979. This favors the hypothesis of the periodic overturns of part of the crust over that of the periodic flows, because the latter would constantly increase the level, except if there were a complete drainback. The supposed vertical up-and-down movement, proposed for other lava lakes on Io (A. Mura et al. 2024b) and necessary to produce the crust breakup close to the patera’s walls, may not be sufficient to submerge all the island completely. In fact, it seems that in the subsequent observation (orbit 58, panel (D) of Figure 1) the larger islands are still visible; the smaller ones may not be revealed due to the lower spatial resolution of this map.

A model of periodic lava flow is inconsistent with JIRAM’s observation of small islands in the middle of the lake, where the flow should have passed. The lava is estimated to grow at least 5 m per cycle (K. de Kleer et al. 2017). If multiple layers are flooded and if cycles have occurred throughout all the time from Voyager’s observation to JIRAM’s, the total thickness over 45 years would be at least 150 m (R. R. Howell et al. 2014), which is not a negligible figure compared to the small size of the islands. However, it is worth noting that our analysis cannot determine the height of the islands, and whether any islands have disappeared or how many might have, as it focuses solely on counting similar features. In addition to the current analysis, it is pertinent to consider the characteristics of very small, closely packed islands. Are they anchored to the floor of the lava lake? We suggest that this is the case but, if so, why have the subsurface parts of each island not been thermally eroded so that they have collapsed during the multiple resurfacing cyclesʔ Superficially, these islands exhibit similarities to those found in terrestrial lagoons, which are typically situated in shallow waters. Alternatively, they may be analogous to pillars (“inselbergs”) formed in terrestrial karst topography near the coast (M. Veress 2020). This observation raises a hypothesis regarding the temporal variability in the depth of Loki Patera’s lake, but this has yet to be quantified.

The imaging channel of JIRAM has two filters: the M-band filter (∼4.6 to ∼5.1 μm) and the L-band one (∼3.3–3.6 μm). The spectrometer mode is not used in this study. The imager has an angular resolution of 237 μrad (instantaneous field of view (iFOV)); JIRAM can take only one image and one spectrum for each rotation of Juno (∼30 s). The FOV of JIRAM is 266 by 432 pixels, or 3$\mathop{.}\limits^{\unicode{x000b0}}$5 by 5$\mathop{.}\limits^{\unicode{x000b0}}$9; the boresight is perpendicular to the spin axis of Juno. Hence, Io can be observed only when it crosses the spin plane of Juno (or when Juno is reoriented to make it so). The despinning mirror of JIRAM was not functional at the time of the observations. We used very short integration times (2 ms on average) to keep the smearing low (0.8 pix ms⁻¹); residual smearing is corrected via software by applying a Lucy–Richardson algorithm (L. B. Lucy 1974; implemented by MATLAB).

The NAIF-SPICE software (C. H. Acton 1996) is used to map the images onto the surface of Io. For each orbit, we select the images with best quality and spatial resolution. Then, to increase the accuracy of geometric reconstruction, the coalignment of these images is checked and, if necessary, corrected. Finally, to reduce the noise, coaligned images are stacked (producing a 3D cube) and the median value along the third dimension is taken, so to obtain one 2D superimage for each orbit and each band (as done by A. Mura et al. 2020, 2024a). After this procedure, the uncertainty about the geometric reconstruction of JIRAM images is estimated to be equal to, or less than, one pixel.

In Table A1 we report the date of observation, minimum distance (rounded), the pixel size at the surface (not considering possible smearing and the point-spread function of the instrument, so technically that is a spatial sampling), and the number of images used for each band. The pixel size is calculated as iFOV × distance, which is the best resolution virtually obtained from that distance, and hence it is an indicative value.

To analyze the data set provided by the temperature as a function of time, we introduce a model that simulates the cooling and crystallization of magma. As the magma cools from the surface, it begins to solidify, forming a crust, while the deeper layers remain molten. The 1D model employs the finite-element method to solve the heat diffusion equation in the crust, following the approaches detailed by R. W. Lewis (1996) and G. Strang & G. Fix (2008). The physical assumptions and constants in our model closely align with those described by K. de Kleer & I. de Pater (2017). In both models, the magma and crust density (ρ) are assumed constant at 2800 kg m⁻³, with a latent heat of crystallization (H_L) set at 4 × 10⁵ J kg⁻¹. The thermal conductivity (k) is 0.9 W m⁻¹ K⁻¹, and the specific heat capacity (c_p) is 1500 J kg⁻¹ K⁻¹, ensuring that the fundamental physics governing the cooling process in both models remains essentially identical. A minor difference is that, while our model assumes constant magma and crust densities with temperature, K. de Kleer & I. de Pater (2017) consider temperature-dependent density. However, they also note that density variations have a minimal effect on the cooling dynamics and overall system behavior. Therefore, this approximation is not expected to significantly impact the model’s accuracy or its predictions regarding cooling and crust formation.

At the surface, the Neumann-type boundary condition is governed by the Stefan–Boltzmann law, which describes radiative heat loss as a function of temperature. The heat flux is

$$\begin{eqnarray*}&&{\phi }_{q}={\varepsilon }{\rm{\sigma }}{T}^{4},\end{eqnarray*}$$

where ${\varepsilon }$ is the emissivity of solid crust (0.9), σ is the Stefan–Boltzmann constant, and T is the temperature at the surface. The numerical solver uses a Galerkin method to iterate along the domain of time; the time step is chosen to maintain the required stability with the criterion that the dimensionless parameter β is less than 0.5:

$$\begin{eqnarray*}&&\beta =\frac{k}{\rho {c}_{p}}\frac{{dt}}{{{dx}}^{2}}\lt 0.5;\end{eqnarray*}$$

the model simulates a 10 m deep column, with a 10 cm node spatial step (dx).

In the first scenario, the entire column consists of hot magma cooling only from the surface; the domain is initially set at a uniform temperature of 1475°C. The release of latent heat during crystallization slows the cooling process as the top layers solidify. The simulation lasts for 1000 days, which is longer than the proposed period for Loki’s activity (∼500 days).

In summary, the first scenario follows K. de Kleer & I. de Pater (2017) and produces the same results; only the numerical solver is different. However, an alternative scenario is modeled with altered initial conditions, where the top 5 m consist of freshly deposited magma at 1475°C, while the bottom 5 m represent preexisting solid crust at 270 K (a conservative value: the lowest measured temperature is less than 270 K, and at the time of our last observation the crust was still cooling). This setup simulates the case of magma flooding over a cooler crust. As in the first scenario, the latent heat of crystallization is included, slowing the cooling rate during the solidification process.

The results of these two scenarios are presented in Appendix Figures A1 (temperature as a function of depth and time, and crust thickness) and A2 (surface temperature as a function of time). Scenario A remains virtually out of equilibrium even after 1000 days, with molten magma still present at the bottom of the simulation domain (this suggests that simulating depths greater than 10 m is unnecessary).

Zoom In Zoom Out Reset image size

For scenario B, we observe that after t₀ + 100 days, the top 5 m of magma have solidified and the latent heat has been fully released. From this point onward, cooling proceeds much more rapidly, as the absence of latent heat allows for faster decreases in temperature. Initially, the two temperature profiles remain similar, but they diverge significantly after 100 more days (t₀ + 200 days). At this point, the cooler crust below continues to act as a heat sink, accelerating the cooling of the system. After a further 200 days (t₀ + 400 days), the surface temperatures in the two cases differ by 200 K or more, allowing us to compare the model predictions with observational data to assess which scenario is in better agreement with the temperatures measured by JIRAM.

Zoom In Zoom Out Reset image size

The results of fitting both scenarios to the data are presented in Figure A3. The left panel (A) shows the same result as Figure 2 in the main body of the article. The only difference here is that, instead of displaying absolute time on the x-axis and showing the three models, we plotted relative time (showing only one model) and adjusted the data by subtracting the t₀ values derived from the best fit. The right panel (B) presents scenario B with adjusted t₀ values, though the results remain inconsistent with the data. The issue lies in the fact that the data exhibit a cooling rate of only 0.1 K day⁻¹ in the region below 300 K, while the model predicts approximately double that rate.

Zoom In Zoom Out Reset image size

Incidentally, we also tested a third case in which the starting condition is a hot column of solid crust. Regardless of the initial temperature, this simulation produces results that are entirely inconsistent with the data. This simulation does not have a clear physical interpretation, as it does not adhere to any particular model for Loki Patera; it was performed solely to demonstrate that latent heat is the only factor capable of explaining the temporal evolution of temperature.

We extend the analysis presented in the main part of this paper and illustrated in Figure 2 by dividing the lake into 15 distinct regions instead of the original three. These regions are numbered sequentially in a counterclockwise direction, beginning from the southwest (right panel of Appendix Figure A4). Region 3 corresponds to the southern part of the lake, region 9 to the eastern part, and region 15 to the northern part. The regions are spaced with margins from the patera’s wall to exclude extra radiance originating from near the walls. For each of these regions, four or five mean temperatures in both bands were derived from radiance maps across four or five different epochs/orbits (during orbit 57 regions 4–7 were not observed in the M band, and the whole lake was not observed in the L band). For each epoch and region, the average between the L-band and M-band derived temperatures was calculated. Using the method detailed in the main text, we estimate the cooling onset time, or crust age, by fitting the temperature data to a time-dependent model. The estimated crust ages are shown in the left panel of Appendix Figure A4, and the right panel illustrates the locations of the 15 regions. The regions are of approximately equal in size (∼1000 km²) and spacing (∼20 km). The temporal progression in the left panel of Appendix Figure A4 suggests a faster cooling wave beyond region 9 (east). The total distance covered by the cooling wave between regions 1 and 15 is ∼300 km over ∼110 days, resulting in an average phase velocity of slightly less than 3 km day⁻¹.

Zoom In Zoom Out Reset image size

The uncertainty associated with the estimation of the crust ages is not easy to assess, because several factors may play a role. We assume that the uncertainty on the input data (i.e., the temperature) is small because, as mentioned, L-band and M-band derived temperatures align well. The uncertainty arising in the best-fitting method can be estimated by running a simple Markov Chain Monte Carlo procedure instead of performing a single best fit of the crust age. The standard deviation of the output is about 20 days (it changes slightly from zone to zone), and it is used as a reference value for the error bars in Appendix Figure A4.

Appendix Figure A5 presents the radiance map of the central island in Loki Patera, captured in the M band. This map represents the same data as in Figure 1 but with a different zoom and color scale. In visible light, the central island appears to be traversed by several cracks or channels. However, these features are barely discernible in the infrared, despite the resolution seemingly being sufficient to detect them. As illustrated in Figure 1, Juno’s viewpoint is considerably inclined. In orbit 57, the emission angle is 56°, with a view from the northwest of the lava lake. In orbit 58, the angle decreases to 32°, and Loki is observed from the west.

Zoom In Zoom Out Reset image size

In neither case are the channels strongly emitting in the infrared, except for an arc in the southwest, which appears to correspond to a visible feature but does not match exactly. This arc is clearly visible in orbit 57 and can also be detected, with slightly lower radiance, in orbit 58. Since the viewing angle in orbit 57 is more aligned with the direction of this supposed channel, we infer that it is deep. The more aligned the line of sight is with the channel, the less shadow the walls cast on the warmer, lower channel floor compared to the rest of the island.

A horizontal feature is observed in orbit 58, though it may be a JIRAM artifact, as it is aligned with one of the sensor’s axes. A barely noticeable diagonal feature, running from southwest to northeast and more visible in the raw data (Appendix Figure A6), aligns remarkably well with a feature observed 45 years ago in Voyager 1 images, which was thought to be one of the wider channels. In summary, there is sufficient indication that the bottom of these channels is emitting at higher temperature. We also conclude that the channels visible in the optical spectrum are likely deep, such that they can only be discerned in the infrared when viewed at a low emission angle, or with a line of sight that is parallel to the direction of the channel.

Zoom In Zoom Out Reset image size

The island appears hotter in orbit 57 than in orbit 58, suggesting the possibility of a temperature-versus-time analysis for this region similar to that in Figure 2. However, the low average radiance makes the data quite noisy, which could result in a misleading analysis. Also, at very low temperature the subtraction of reflected sunlight is more challenging and may leave some residual contribution.

Notably, temperatures derived from the L band are consistently higher than those from the M band by about 20%. A possible explanation for this discrepancy—aside from the obvious hypothesis of calibration residuals, which, however, have been ruled out by previous analyses in this study—could be a nonuniform temperature distribution across the island. This may result from the presence of subpixel hotspots. For instance, if most of the island is at 200 K, with just 0.01% consisting of hotter features at 600 K, the overall radiance would correspond to that of a blackbody at 215 K in the M band and 255 K in the L band, thus explaining the 20% difference. Although the actual scenario is likely more complex, with many possible physical configurations, this example illustrates the reasoning behind such a hypothesis.