Increasing the Usability and Accessibility of Voyager 2 Images of Triton

Much of what we know about Neptune’s moon Triton was inferred from the analysis of images returned by the Voyager 2 mission, the only spacecraft to have visited that putative ocean world. Unfortunately, the highest-resolution images (scales < 2 km pixel−1) are difficult to use because they are only available in nonstandard formats, and the locations of the images on Triton’s surface are incorrect by up to 200 km. Although image mosaics of Triton are publicly available, these do not include the highest-resolution data. Here we describe our effort to improve the usability and accessibility of Voyager 2 images of Triton. We used the USGS’s ISIS software to process 41 Triton images, including geometric calibration, radiometric calibration, and reseau removal. We improved the image locations using a photogrammetric control network with 958 points and 3910 image measurements. Least-squares bundle adjustment of the network yielded rms uncertainty of 0.50, 0.52, and 0.51 pixels in latitude, longitude, and radius, respectively, and maximum residuals of −4.21 and +3.20 pixels, respectively. Image-to-image alignment is therefore vastly improved. We have released these processed images as cloud-optimized GeoTIFFs in orthographic projection at the original pixel scale of each image. Associated mosaics have also been created and released to provide geologic context for the individual images. These products provide the science community with analysis-ready data that enable new investigations of Triton, increase accessibility to this unique data set, and continue to enhance the scientific return from the Voyager 2 mission.


Why Triton?
The ice-covered moons of the outer solar system have surfaces that are morphologically distinct from those of the terrestrial planets.Although each has a distinct style of surface morphology, most involve cratered terrains that are overprinted by greater or lesser degrees of tectonism.An important exception is Jupiter's famous moon Europa, which has an extremely young surface (100 Myr; Bierhaus et al. 2009) with very few impact craters.The youthfulness of Europa's surface and its distinctive features are consistent with the presence of a liquid water ocean beneath its icy exterior (Pappalardo et al. 1999), a hypothesis strongly supported by measurements of Europa's induced magnetic field (Kivelson et al. 2000).Similarly, Neptune's large moon Triton is nearly free of craters, and its surface may be even younger (6-50 Myr) than Europa's (Schenk & Zahnle 2007).Instead of cratered terrains disrupted by tectonic features, Triton's surface includes cantaloupe terrains (Croft et al. 1995; Figure 1(a)), massive double ridges (Prockter et al. 2005; Figure 1(a)), large calderalike walled plains (Croft et al. 1995; Figure 1(c)), and frostcovered southern hemisphere terrains (Figure 1(b)).Triton is also known to be one of the few bodies in the solar system to be geologically active, with geyser-like plumes emanating material up to 8 km above the surface, which is then carried "downwind" over 100 km in the moon's tenuous (1.4 Pa) nitrogen atmosphere (Smith et al. 1989;Hansen et al. 1990; Figure 1(b)).Whether Triton's plumes are linked to a subsurface volatile reservoir like at Enceladus remains unclear (see summary in Kirk et al. 1995;Hofgartner et al. 2022).
The apparent youthfulness of Triton's surface, unusual terrain morphologies, and geologic activity indicate that Triton is a dynamic world, leading to the suggestion that, like Europa, it may contain a subsurface ocean (e.g., Hansen et al. 2021).Its inclined, retrograde orbit and the lack of other large moons in the system suggest that it is a captured Kuiper Belt object (KBO;e.g., McCord 1966;McKinnon 1984;Agnor & Hamilton 2006).Yet its surface is unlike that of the similarly sized KBO Pluto (Triton's radius of 1352.6 km is comparable to Pluto's of 1188.3 km; Archinal et al. 2018).Capture into the Neptune system would have generated internal heating sufficient to drive differentiation early in solar system history (Ross & Schubert 1990), and under some scenarios that ocean could persist to the present day (Gaeman et al. 2012).However, more recent geologic activity, such as cryovolcanism capable of resurfacing most of the moon (e.g., Croft et al. 1995), must be driven by an alternative energy source, such as obliquity tides (Nimmo & Spencer 2015).
Unfortunately, our understanding of Triton, including the existence and potential habitability of an internal ocean, is limited by a sparsity of available data.Although ground-based observations have provided invaluable information on Triton's composition and volatile transport (e.g., Cruikshank & Apt 1984;Cruikshank et al. 1984Cruikshank et al. , 1993;;Bauer et al. 2010;Grundy et al. 2010;Buratti et al. 2011;Merlin et al. 2018;Hicks et al. 2022), most of what we know about Triton and its geologic history comes from a single flyby of NASA's Voyager 2 spacecraft.Voyager 2 launched on 1977 August 20 and, after rendezvousing with Jupiter, Saturn, and Uranus, flew past Neptune almost exactly 12 years later on 1989 August 25 (closest approach).The spacecraft also made a close Original content from this work may be used under the terms of the Creative Commons Attribution 4.0 licence.Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.approach to Triton at that time, acquiring 43 images with pixel scale below 2 km pixel −1 , including 10 images at pixel scales <420 m pixel −1 .These few images of Triton are the only data set that permits examination of the geology of this potential ocean world.It is therefore vital that the scientific community can access and utilize these data with as little overhead as possible.

The Challenges of the Voyager 2 Imaging Data Set, Existing Triton Data Products, and the Goal of This Work
The raw Voyager 2 Triton images are challenging to use for modern scientific analysis.The images are publicly available from NASA's Planetary Data System (PDS) Imaging and Cartography Node as compressed raw image files (.imq format), which is the format in which Voyager image files were originally archived. 4The images have not been geometrically or radiometrically calibrated, are unprojected, and contain reseaux and corner markers used for geometric calibration.Uncompressed versions (.img format) of the images are available from the PDS Ring-Moon System Nodes in VICAR-formatted binary (VICAR, or Video Image Communication And Retrieval, is the Jet Propulsion Laboratory's image processing software5 ).These images have already been processed through radiometric and geometric calibration and reseau removal; however, the binary files are not compatible with the commonly used USGS ISIS software package for planetary image processing (Laura et al. 2022) (ISIS applications such as pds2isis and vicar2isis do not attach the metadata necessary for further processing), nor is the VICAR format easily used by common geospatial software (e.g., a Geographical Information System or GIS).The individual processed images of Triton are thus not currently available to the scientific community in a widely used, spatially enabled image format such as GeoTIFF.
Many members of the planetary science community use the USGS's ISIS software to ingest raw PDS image data, generate images in ISIS cube format, and either use that with compatible geospatial tools or convert it to a GeoTIFF using ISIS routines (e.g., isis2std) or the Geospatial Data Abstract Library (GDAL).Unfortunately, the images of Triton processed using ISIS still have limited utility.Because of uncertainty in the location and orientation (position and pointing) of the Voyager 2 spacecraft during the Triton flyby, images are poorly located on the surface.That is, the reconstructed NAIF SPICE kernels (Navigation and Ancillary Information Facility, Spacecraft, Planet, Instrument, C-matrix (pointing) and Events; Acton et al. 2017) that help define the latitude and longitudes of features on the surface are inaccurate, resulting in features that are misplaced by up to 200 km (Figure 2).Because of this, two or more images cannot easily be used together unless their locations on the surface are first corrected.
In practice, the limitation of poorly positioned images is often overcome by using an existing image mosaic provided by a mission team or data scientist.Such mosaics include multiple images that have been projected to a common coordinate system and pixel scale and therefore provide continuous coverage of the surface.At least three "global" Triton image mosaics are now publicly available (excluding several airbrush "pictorial maps" available only as PDFs).Here we use the term "global" to describe mosaics that include all of the portions of Triton observed by Voyager 2 at moderate or high resolution, despite the fact that actual global image coverage was not achieved (images were primarily acquired in a single hemisphere).
At the time of writing, a global color mosaic in orthographic projection at 600 m pixel −1 is available through the PDS Image Annex. 6The mosaic was created by the USGS and originally released in NASA's "Photojournal"7 in JPEG and standard TIFF (not GeoTiff) format in 1998.In 2016, the mosaic was converted to a GeoTIFF from an unreleased, now-obsolete ISIS2-formatted cube and rereleased to the PDS Imaging Node Annex by the USGS.Unfortunately, most of the metadata have been lost in the almost two decades since the mosaic was originally created.At the time of writing, the PDS Imaging Node Annex also includes a Triton color mosaic at 600 m pixel −1 in cylindrical projection that was created independently by Dr. Paul Schenk of the Lunar and Planetary Institute. 8Unfortunately, the metadata available in the Image Annex incorrectly attribute this mosaic to the USGS.Like the orthographic mosaic, this cylindrical mosaic was first released in NASA's "Photojournal"9 in JPEG and standard TIFF (not GeoTiff) formats in 2014, and then it was added to the PDS Image Annex in 2016.
More recently, new mosaics and digital elevation models (DEMs) of Triton in cylindrical projection were generated by Paul Schenk and released by the Lunar and Planetary Institute. 10The products are described in detail by Schenk et al. (2023) and include three DEMs and the associated mosaic.The DEMs were released in cube format, which can be used in a GIS environment; however, the reference image mosaics were only released as JPEGs.
Although desirable for many applications (especially geologic mapping, where a consistent pixel scale is required; e.g., Martin et al. 2023), such mosaics are limited in that they only contain a subset of the images obtained by the spacecraft and image pixel scale is usually either up-sampled or downsampled.Thus, images are not available at their native resolution, which in some cases is substantially greater (i.e., "better") than that of the mosaic.For example, existing Triton mosaics are projected to a scale of 600 m pixel −1 ; however, roughly one-quarter on the Voyager 2 Triton images (10 of 41 included here) have pixel scale smaller than 600 m pixel −1 , with several <350 m pixel −1 (Table 1).Additionally, such image mosaics do not include every image acquired by the spacecraft, and these images are therefore frequently not included in geologic analysis.Put another way, Voyager 2, which is the only spacecraft to have visited Triton and required 12 years to get there, acquired only several dozen images from which geologic detail can be discerned, and the community cannot easily use all of them at their full resolution.
Here we describe our efforts to process and improve the locations of Triton images such that users have the option of using either an established mosaic or individual images at their native resolution.We seek to maximize the usability and accessibility of this truly unique data set by minimizing the technical overhead required to use the data and releasing the data in well-known repositories.This work therefore follows the recent practice of releasing analysis-ready data (ARD) that are intended to "allow analysis with a minimum of additional user effort" (Dwyer et al. 2018).A critical component of ARD is thorough documentation (and metadata) describing how the data were processed so that users understand the provenance of the data sets they use.This paper describes such processing and follows the standards set by similar work for Enceladus (Bland et al. 2018) and Europa (Bland et al. 2021).These efforts are also in line with the findings of NASA's Planetary Data Ecosystem Independent Review Board,11 which emphasized the development of more ARD products (findings 60 and 61, recommendation 49) and the creation of Planetary Spatial Data Infrastructures (PSDIs; Laura et al. 2017Laura et al. 2018).

Image Processing Methodology Overview
We aim to enable science users to completely bypass the initial processing steps described herein.However, to ensure that users understand the provenance of the data and to enable reproducible data processing, we document our methodology in detail.
We processed and improved the locations of Voyager 2 images (described in Section 3.1) of Triton using principles of photogrammetry and the USGS's publicly available ISIS software (predominately version 7.0.0;Laura et al. 2022).A list of the 41 images we included is provided in Table 1.We first download the data from the PDS and minimally process the images (Section 3.2).We then improved the locations of images in two stages.In the first stage, we performed only a rough alignment of the images (Section 3.3.1).This stage was necessitated by the poor initial image locations, which made accurate calculation of image overlaps impossible.In the second stage, we used the roughly aligned images to develop a more robust control network to further refine image locations (Section 3.3.2).With the image geometry corrected, additional processing was performed to remove reseaux, corner flags, and dropped lines (Section 3.4).

The Voyager Imaging Science System at Triton
The Voyager 2 spacecraft carried an Imaging Science Subsystem (ISS) consisting of both narrow-and wide-angle slow-scan vidicon cameras.The narrow-angle camera had a focal length of 1500 mm (actual calibrated focal length was 1503.49mm), a 7.5 × 7.5 mrad field of view, and a 9.25 μrad angle subtended by each scan line.The wide-angle camera had a 200 mm focal length, a 55.6 × 55.6 mrad field of view, and a 69.4 μrad angle subtended by each scan line (Smith et al. 1977).The ISS used 800 scan lines per frame and 800 picture elements per line, resulting in images with 640,000 pixels per frame.The images include fiducial markers (reseaux) and corner markers used for making calibration measurements (e.g., focal length and camera distortion).The camera also included an eight-position filter wheel assembly, enabling color imaging to be generated when images obtained with different filters are combined.
Voyager 2ʼs closest approach to Triton occurred on the morning of 1989 August 25 (UTC); however, Voyager 2 began acquiring images of Neptune and Triton that improved on ground-based observations nearly 1 yr in advance of the flyby (Chapman & Cruikshank 1995), with imaging of Triton itself beginning as early as 1989 June.Throughout the extended encounter period (from 1989 late June to early September), a total of 660 images of Triton were acquired and archived in the PDS; however, the vast majority of these have pixel scales >10 km pixel −1 (e.g., images acquired 48 hr before closest approach have pixel scales of ∼21 km pixel −1 ), and in many, Triton is just a few pixels across.Beginning roughly 8 hr before closest approach, Voyager 2 captured a sequence of images in multiple color filters, enabling the creation of low-resolution (∼4 km pixel −1 ) color imaging.Because they are of lower resolution, these images were not included in our processing, but see, e.g., Schenk et al. (2023).Images in which surface features could be well resolved (<2 km pixel −1 ; Table 1) were acquired starting roughly 2 hr before closest approach.Image acquisition at Triton was extremely successful; however, the low light levels at Neptune necessitated long exposure times, which, despite slewing the instrument to compensate for spacecraft motion, resulted in significant smear in the highestresolution images.A set of unsmeared images exist, but as of our writing they are not publicly available (Schenk et al. 2023), and we have not corrected the smear in this work.Image acquisition continued after closest approach, but only a crescent of Triton was illuminated.Note.
a Illumination statistics were calculated after image locations were corrected.
Values listed are averages over the image.

Initial Image Processing
We acquired images from the PDS Cartography and Imaging Node at the USGS (see Section 2) in compressed format (.imq) and ingested them into ISIS using the voy2isis application.After ingestion, preliminary geographic information was attached to each image in the form of NAIF SPICE kernels.12These kernels provide the location and orientation of the Sun, target (in this case Triton), and camera at the time each image was acquired (along with other supplementary information) such that, when combined with an appropriate camera model (we use the standard Voyager 2 camera model included in the USGS's ISIS software), the latitude and longitude of every pixel can be determined.We began with the default reconstructed SPICE kernels, which were attached to each image using ISIS's spiceinit application.Specifically, the default kernels included the pck0009.tpcplanetary constants kernel (PCK) for target information such as body radius, the de430.bspand nep081.bspfor Sun and target ephemeris (SPKs), the vgr2_nep081.bspspacecraft ephemeris kernel (SPK), the vg2_nep_version1_type2_iss_sedr.bccamera pointing kernel (CK), the vg2_v02.tfframes kernel (FK), the vg2_issna_v02.tiinstrument kernel (IK), the voyagerAdden-dum004.ti (IAK), and the vg200044.tscspacecraft clock kernel (SCLK).
Once geometric information was attached, we radiometrically calibrated the images using ISIS's voycal application to convert pixel values ("data number" or DN) to a quantity proportional to radiance (I/F).Vidicon detectors, such as those used by Voyager 2, build up signal (dark current) even when the camera shutter is closed.The amplitude of this signal varies by line and sample on the detector.Additionally, the sensitivity of the detector depends on the signal intensity, which is a function of the scene (i.e., like the dark current, it varies by line and sample) and the distance from the Sun.Voycal uses supplementary dark current and gain files, as well as the distance between the target and the Sun, to correct each image and enable calculation of a "radiance factor," for which a calibrated DN of 10,000 is equivalent to an I/F of 1.The calibration is directly analogous to VICAR's FICOR77 program.
In addition to the radiometric quirks of the vidicon detectors described directly above, this type of detector also produces geometric distortion that varies with time and the distribution of signal on the detector.To correct for this distortion, the Voyager cameras included a grid of reseau markings on the camera faceplate that can be used to geometrically correct each image.Every ISIS-formatted image label (attached to the image) includes a table of reseau locations (line/sample) that the camera model uses to correct the distortion when displaying, projecting, or querying the image (e.g., calculating illumination angles).The geometric correction was improved by using ISIS's findrx application, which uses area-based matching to refine the line/sample locations of each reseau in the image.Using findrx overwrites the reseau table within the ISIS label and therefore has the effect of improving the distortion correction based on these updated locations.The practical effect is that the latitudes and longitudes of a pixel will differ (in some cases substantially) before and after running findrx because the geometric distortion has been mitigated.This processing step is analogous to the use of VICAR's RESLOC and GEOMA programs.

Improving Image Locations: Photogrammetric Control
Correcting the locations of images on Triton's surface relative to one another requires identifying points in common between images, which we refer to as tie points.Each tie point must include at least two images but may include numerous images where many images cover the same region.We refer to each image associated with a tie point as an image measure.Two image measures are required to solve for a tie point's 3D location (e.g., latitude, longitude, and radius, which is the height as measured from the moon's center of figure), and three image measures are required to solve for the location uncertainty.In general, the more image measures that are included, the better the location solution (i.e., with the assumption of no systematic errors, uncertainty is proportional to N , where N is the number of measurements).
The first step in creating tie points is determining which images overlap each other.To do this, we used ISIS's findimageoverlaps application, which uses the SPICE (i.e., geometry) information attached to each image to calculate regions of overlap.Once image overlaps were determined, we placed tie points within overlaps using ISIS's autoseed application, which distributes tie point candidates based on a given line/sample spacing.Unfortunately, because the initial locations of the Triton images are so poor (Figure 2), the automated determination of overlaps was highly inaccurate and included regions of fictitious overlap (i.e., the images do not actually overlap), as well as regions of unaccounted-for overlap (i.e., images do overlap but were not included).We therefore determined that the locations of each image had to be improved before a more robust control network could be created to further refine image locations.

Rough Initial Image Alignment
For the rough initial alignment, we began with the set of tie points generated by findimageoverlaps and autoseed.High emission angles result in excessive foreshortening and poor image matching, and we therefore required that all tie point locations had an emission angle <70°.0. To prevent tie points from being placed in poorly illuminated regions, we also required tie point locations to have an incidence angle <89°.0.Finally, because image overlaps were uncertain and it ensures that a sufficient number of surrounding pixels were available for matching, we required tie points to be at least 8 pixels from the edge of the image.Once points were placed, we performed automated image matching using ISIS's pointreg, which uses an area-based maximum correlation algorithm with a goodnessof-fit threshold and subpixel refinement.Because initial image geometries are in some cases extremely inaccurate, we used a large 501 × 501 pixel search area (i.e., a substantial fraction of the 800 × 800 pixel image).This set of tie points was supplemented with manually created tie points (using ISIS's qnet application in conjunction with pointreg) in regions where point density was low, including regions of high emission where necessary.
Once a fully connected network of images and tie points was established, we used ISIS's least-squares bundle adjustment application, jigsaw (Edmundson et al. 2012), to solve for the camera pointing (three rotation angles) and ground coordinates (latitude, longitude, and radius) of every tie point in the network.Given our lack of constraints and considering that spacecraft position is highly correlated with camera pointing for the camera's narrow field of view, for this initial control network the spacecraft position was held fixed.Camera pointing was constrained to ±4°, and ground coordinates were given large uncertainties of 20 km in latitude and longitude and 2 km in radius.The large initial uncertainty is necessitated by the highly inaccurate initial locations of some images based on the a priori geometry from the default SPICE kernels.Running jigsaw on the control network provided residual information for each tie point.This enabled us to identify and examine points with the highest residuals, which often result from incorrect tie point matching.These high-residual points were manually assessed using ISIS's qnet application and either corrected (properly matched) or removed from the network.The bundle adjustment was then run again to identify any additional points with high residuals.Thus, jigsaw was used iteratively until the overall network residuals fall below requirements (a solution variance σ 0 = 0.5 is desired).After removing inaccurate tie points, our initial network consisted of 402 points and 1470 image measures (Figure 3(a)), which was sufficient to tie all 41 images together but was limited in point density relative to many modern control networks (e.g., Bland et al. 2018Bland et al. , 2021)).Bundle adjustment yielded σ 0 = 1.265, with median point residual of −0.007 pixels, first and third quartile residuals of −0.321 and +0.328 pixels, respectively, and minimum and maximum residuals of −62.16 and +20.81 pixels, respectively.Although these residuals are higher than typically achieved (see Section 3.3.2),the solution was sufficient to meet its objective of "moving" the images closer to their correct location so that more accurate overlaps could be calculated.
With an adequate initial solution achieved, we updated the pointing information for all 41 images and generated new pointing kernels (CK) from them using ISIS's ckwriter application.These CKs provide the foundation for a more robust calculation of the image geometries (Section 3.3.2).

Final Image Location Refinement
The control network described in Section 3.3.1 was necessary to bring the images into closer alignment but insufficient to provide a robust solution for the locations of Voyager 2 Triton images (e.g., σ 0 > 1, and maximum residuals were still >60 pixels).As noted in Section 3.3, the initial geometries prevented adequate matching of images, both because the correct pixel location was often far from a priori estimates and because the image-overlap information was incorrect.The number of tie points used and the number of image measures associated with each tie point were therefore limited.
In order to further refine image locations, we created a secondgeneration control network that enabled further refinement of Triton image locations.To do this, we began with a fresh set of Voyager 2 images, which were redownloaded from the PDS and ingested into ISIS using voy2isis.We attached image geometries using spiceinit, but in this case we specified the CKs generated from our rough alignment (Section 3.3.1).All other kernels were identical to those described in Section 3.2.The data were calibrated using voycal and findrx, and reseaux were removed (but not "filled in") using remrx (see Section 3.4).We again began the control process by determining image overlaps using findimageoverlaps, but with the images now in closer alignment, the calculation was much more robust, enabling the majority of points to be placed automatically using autoseed.We again used the constraint that tie points must have an emission angle <70°.0, have an incidence angle <89°.0, and be at least 8 pixels from the edge of the image.Manually created points were again added using qnet to ensure that the network was fully connected.Automated, area-based matching was again performed using pointreg, but because the initial geometry was already improved, a much smaller search area (51 × 51 pixels) could be used (i.e., tie points were closer to their correct location).ISIS's jigsaw bundle adjustment application was again used iteratively to identify incorrect matches, which were subsequently corrected or removed using qnet.The final control network included 41 images, 958 points, and 3910 measures (Figure 3(b)).Bundle adjustment with 2 km uncertainty in latitude, longitude, and radius yielded a σ 0 = 0.385, with a median point residual of −0.008 pixels (i.e., errors are nearly centered on 0), first and third quartile residuals of −0.302 and +0.301 pixels, respectively, and minimum and maximum residuals of −4.21 and +3.20 pixels, respectively.The total rms uncertainty was 0.50, 0.52, and 0.51 pixels in latitude, longitude, and radius, respectively.

Reseau Removal and Cosmetic Improvements
The Voyager cameras included reseau marks on the camera faceplates to aid in geometric calibration (Section 3.1).Many users prefer that these reseaux be removed to provide a more visually appealing image and avoid misinterpretations.In our work, this processing step was performed after the photogrammetric solution was completed because it introduces interpolated data that can reduce the quality of image matching.We removed image reseaux by first replacing them with a 7 × 7 block of null pixels using ISIS's remrx application (Figure 4(b)) and then applying a low-pass filter only to null pixels (ISIS's lowpass).The low-pass filter effectively replaces the null pixels with a value interpolated from the surrounding pixels.The low-pass filter also removed dropped lines that occur in several images (e.g., Figures 4(a) and (b)).This  process is analogous to the use of VICAR's VGRFILLIN and RESSAR77 programs.
As a final step, we performed cosmetic improvements on a very small number of images (c1139350, c1139521, c1139627).This "cleanup" involved manually replacing obviously bad pixels (i.e., pixels that do not contain data) with nulls and then using the low-pass filter to replace those null pixels with interpolated values.We intentionally kept these improvements to a minimum in order to provide users with a data set that is relatively unaltered from the original (see the discussion of "partial processed" data in Section 5).To that end, no effort was made to remove other sources of noise, such as from radiation hits.

The Improved Voyager 2 Images: Illustrations of Use Cases
Photogrammetric control of all of the Voyager 2 Triton images with pixel scale less than 2 km pixel −1 enables easier, and thus more extensive, use of these data by the scientific community.Figure 5 shows a mosaic of the hemisphere of Triton observed by Voyager 2 and demonstrates the quality of the image alignment after image locations have been improved (another example is shown in Figure 6).Mosaics with a consistent pixel scale, such as that shown in Figure 5, are necessary for geologic mapping (e.g., Martin et al. 2023).This new mosaic is similar in appearance to the 1998 mosaic described in Section 2 but lacks color (see discussion below).However, it provides an advantage over the older product in that all of the Triton images are aligned to it.That is, only 16 images are included in the mosaic itself (Table 2), but the other 25 images, which have different illumination and, in Figure 6.Example of the improvement in image locations provided by our photogrammetric solution.The images are the same as those shown in Figure 2. The red and white arrows indicate features that were misaligned when using the default reconstructed SPICE information (same features as in Figure 2) but have now been brought into alignment.The region shown is roughly 1300 km east to west and is centered at approximately 343°E and 12°S (north is up).The projection is identical to that used in Figure 5. Notes.Images are listed in the order shown in the mosaic, with the first image listed on the "bottom" and the final image listed on the "top." a Illumination statistics were calculated after image locations were corrected.
Values listed are averages over the image.some cases, higher resolution, are spatially aligned with it and can be used as supplementary data (Figures 7 and 8).All of these images (those included in the mosaic and those not) have been released (Section 5) and can be used independently from the mosaic itself.
Visual comparison of our mosaic and NAIF's SPICEenhanced Cosmographia software13 indicates that our mosaic is aligned to Triton's longitude system by better than 1°, and we take 1°as the accuracy to which the absolute locations of features on Triton are known.Direct comparison between our new mosaic and the 1998 USGS mosaic shows a longitudinal shift in feature locations that is minimized in the western portion of the mosaic but increases to the east.This shift is 1°.2 in the center of the mosaic.Our new mosaic also does not align with the 2014 mosaic produced by Paul Schenk, although the offset of features is smaller (0°.The same region as shown in panel (a), but as seen in image c1139627 at 343 m pixel −1 , which was not included in the 1998 USGS mosaic and was previously only available as an uncontrolled image (unaligned to the lower-resolution images) from the PDS.This image has a resolution twice that of the underlying data and was acquired at very high phase (115°; Table 1).Tie pointing is therefore challenging, and the image is still slightly misaligned.The image is centered at 349°E and 12°N (north is up) and uses the same projection as Figure 5.
These misalignments (no two products are aligned to each other) accentuate the need for a cartographic definition of Triton's longitude system, which currently does not exist.For the terrestrial planets and large moons of Jupiter and Saturn, the longitude system is fixed cartographically by defining a surface feature (often a crater) to be at a specific meridian.Although such definitions are somewhat arbitrary, they permit control network solutions to have at least one point fixed at a specific longitude to help ensure consistency among data products (see, e.g., the approach used in Bland et al. 2018Bland et al. , 2021, which ensured that the crater Salih on Enceladus and Cilix on Europa were at 5°W and 182°W (178°E), respectively).Triton's longitude system, in contrast, is defined only mathematically (Davies et al. 1992Archinal et al. 2018) and is therefore difficult to use as a reference.To address this, we propose to define Triton's longitude system by fixing the location of a currently unnamed crater (name proposal is pending with the IAU) at 40°E longitude (because Triton rotates retrograde, east longitudes are used; Archinal et al. 2018; see Figure 9).The crater was chosen because it is located near Triton's equator (1°S), it is easily identifiable relative to currently named features (Figure 9), it is readily visible in existing Triton data (the crater is visible in 5 of 41 images), and it is located at precisely 40°.0 E in our mosaic.A proposal to the IAU working Group on Cartographic Coordinates and Rotational Elements to formalize the cartographic longitude definition is pending; however, the feature should not be used to define Triton's coordinate system unless (and until) approved by the IAU.
We also note that the 1998 USGS mosaic used a radius of 1350 km, whereas our work uses the currently accepted IAU radius of 1352.6 ± 2.4 km (Davies et al. 1991Archinal et al. 2018).Independently solving for the best-fit mean radius of Triton using our new control network yields a similar value and error, and we do not propose any change to the IAU-accepted radius.
We emphasize that the value of our work is not in the production of a new mosaic, which mostly shows the same data included in the 1998 USGS mosaic and the 2014 Schenk mosaic, but the ability to use individual images that have high-quality metadata (e.g., processing information) in spatial analysis software (e.g., a GIS; Figure 6) at their native resolution and in coordination with a global mosaic that provides context.Figures 7 and 8 provide two example use cases.In Figure 7(a) we show a distinctive region of cantaloupe terrain as it appears in the 1998 USGS mosaic.Here the mosaic uses Voyager 2 image c1139350, which has been up-sampled from its original pixel scale of 1262 m pixel −1 (Table 1) to 600 m pixel −1 .Two higherresolution (smaller pixel scale) images of this region exist but were not included in the USGS 1998 mosaic: c1139627 and c1139629 at 343 and 337 m pixel −1 , respectively (Table 1).These images have almost two times better resolution than the mosaic (Figure 7(b)); however, they have been largely unavailable for scientific analysis except by the those with the expertise to process Voyager 2 images from raw data and rigorously align it to the surrounding lower-resolution data.We further illustrate this in Figure 8, which shows a sequence of six high-resolution images of Triton (pixel scale < 367 m pixel −1 ) that are not included in the 1998 USGS mosaic, the 2014 Schenk mosaic, or our new mosaic.Four other images of similar pixel scale illuminate regions farther east (not shown).In sum, these images constitute roughly onequarter of the Voyager 2 Triton images usable for geologic analysis, yet until now they have remained unavailable to all but a small fraction of the planetary science community.These images can now simply be downloaded (see Section 5) and utilized in any GIS software without any additional processing, opening the full potential of the Voyager 2 Triton data set to the scientific community.

The Data Release
We have released the Voyager 2 Triton images (Bland 2023) described above as ARD at the USGS's ScienceBase Data portal,14 which is a Trusted Digital Repository.Here we define ARD as data (in this case images) that are geometrically and radiatively calibrated and that are rigorously tied to a consistent geospatial reference frame (to the extent permitted by the data type, quality, and availability of such references).This implies that the data are orthorectified where topographic information is available, or projected onto a sphere where it is not, and that latitudes and longitudes are accurate and consistent with the IAU defined coordinate system.The data must have thorough qualitative and quantitative metadata; be provided in a standard, widely used format; and be easily discoverable.Put more pragmatically, ARD are data that a scientist can easily find and then simply drop into GIS software and begin their investigation.Our data release includes the ARD products described below.

Individual Calibrated and Projected, Photogrammetrically Controlled Voyager 2 Images
We have released each individual Voyager 2 image of Triton included in this work (41 images; Table 1) as cloud-optimized GeoTIFFs, which are simply GeoTIFFs that have been internally formatted for efficient usage in the cloud.These images have been geometrically and radiometrically calibrated, had reseaux and corner marks removed and replaced with interpolated data, and had their locations updated as described in Section 3 (Figure 10(a)).Some images have had minor additional cosmetic improvements (Section 3.4).The images have been projected to a common orthographic projection centered at 15°E and 18°N, but each retains its original pixel scale.The center of the projection was chosen for consistency with previous mosaics.No photometric correction has been applied.These images provide the user with a set of "clean" images that can be easily utilized in a GIS for many scientific applications.This new image set shows Triton's surface at a greater range of pixel scales and illumination conditions than are available in the existing Triton mosaics (including the new one described here).However, because some of the data have been interpolated, users should be careful to assess whether small-scale features are "real" or artifacts of the low-pass filtering used to interpolate the small gaps left from reseau removal (see description of partially processed images next).

Individual Partially Processed Photogrammetrically Controlled Voyager 2 Images
To provide end users with a better understanding of how the images were processed, we also have released versions of the images that have not had reseaux and corner marks removed (Figure 10(b)).These images have been geometrically and radiometrically calibrated and have identical locations and utilize the same projection (orthographic centered at 15°E and 18°N) as the more aesthetically pleasing images described immediately above.The intention is to provide the user with the ability to determine whether small-scale features are associated with reseaux or other cosmetic improvements.By using the two sets of images together, the end user can be more confident that their scientific analysis is robust.

Associated Triton Mosaics
As described in Section 2, several Triton mosaics have previously been released.In general, our mosaic does not include data that were not included in previous versions (although there may be some minor differences in image selection and how images were ordered or trimmed).However, because the three mosaics (1998 USGS, 2014 Schenk, and our new mosaic) are not aligned (Section 4), we believe that it will be useful to provide users with a version of the mosaic that aligns with the individual images included in this data release.We have therefore released three versions of the mosaic, each using the same set of images in the same mosaic order (Table 2), with the same orthographic projection as the individual images described above (centered at 15°E and 18°N ) and a pixel scale of 600 m pixel −1 .One version of the mosaic uses the fully processed level 2 images described at the top of this section.No photometric correction or modification of the pixel DNs has been performed (they are only radiometrically calibrated).A second version of the mosaic uses the same set of images but with a high-pass filter applied to sharpen the images (51 × 51 pixel window with 100% albedo add-back).This is the mosaic that is shown in Figure 5.A final version of the mosaic uses the "partially processed" images described above.This mosaic lets the user see where reseaux or corner marks occur and can be used with the first two mosaics to evaluate data quality.

Summary
Much of what we know about the geology of Triton has been inferred through the analysis of images returned by the Voyager 2 spacecraft, the only mission to have visited that potential ocean world.At the time of writing, no follow-up mission to the Neptune system is planned for decades to come.The Voyager 2 images are therefore an extraordinarily unique data set.Unfortunately, those images are difficult for the community to use.They are archived in compressed format within NASA's PDS and require extensive processing to ) has numerous small plume deposits (Hansen et al. 1990;Hofgartner et al. 2022;Martin et al. 2023), and distinguishing real features from reseau marks is critical.calibrate them, remove reseaux and corner marks, and properly locate them on the surface.Although several image mosaics are now publicly available, those mosaics do not include all of the images returned by Voyager 2 (they specifically exclude the highest-resolution data) and are projected at a common pixel scale.We have improved the usability of Voyager 2 images of Triton by processing them, removing reseaux and corner marks, correcting their locations on Triton's surface, and releasing each image in a standard and commonly used format (cloud-optimized GeoTIFF).Our objective is to provide the scientific community with a restored, corrected, and optimized set of ARD that will enable new investigations of Triton and increase accessibility to this unique data set.

Figure 1 .
Figure 1.Examples of Triton's surface morphology, which is dissimilar from the other ice-covered moons in the outer solar system.(a) Broad double ridges, similar to those on Europa but of much greater size, cross portions of the "cantaloupe terrain."(Image center 342°E, 25°N.)(b) Darker smooth terrain transitions to bright frostcovered terrain near the southern pole.Several macula (dark spots) in the bright terrain mark the locations of plume deposits.(Image center 27°E, 17°S.)(c) Extensive smooth planitia (plains) disrupt the cantaloupe terrain.A few small craters are notable in their sparsity.(Image center 23°E, 21°N.)Each panel is a portion of our Triton mosaic, which is in orthographic projection at a scale of 600 m pixel −1 .

Figure 2 .
Figure 2. Example of Voyager 2 Triton images with incorrect location and orientation.These images (image numbers as shown) were downloaded from the PDS Imaging and Cartography Node, had default NAIF SPICE kernels (i.e., position and pointing) attached, and warped to a common projection and pixel scale (image reseaux were also removed to aid visualization).The red arrows highlight the same crater appearing in c1139503 and c1139445, which is misaligned by 200 km.The white arrows highlight a sinuous ridge in c1139509 and c1139445, which is misaligned by 130 km.Such misalignment is the "rule" rather than the exception for Triton data.

Figure 3 .
Figure 3. Visualizations of the tie point locations used in the (a) preliminary network and (b) final control network.Blue polygons show image footprints, and green crosses show tie point locations.The orientation is identical to that shown in Figure 5 below.

Figure 4 .
Figure 4. Example of our processing of "level 1" (unprojected) including reseau removal.(a) The raw image after ingestion into ISIS.(b) Reseaux have been replaced by null pixels.(c) Null pixels, corner marks, and dropped lines are filled in with a low-pass filter.The unprojected cube shown is c1139527.

Figure 5 .
Figure 5.The new Triton mosaic.The mosaic uses an orthographic projection centered on 15°E and 18°N (the 15°E meridian is vertical with north up) and at a scale of 600 m pixel −1 .The images have been high-pass filtered with a 51 × 51 pixel window and 100% albedo add-back (i.e., a sharpening filter) to accentuate surface features.No photometric correction has been applied.Images included in this mosaic are listed in Table2.
7) and within the estimated absolute accuracy of our mosaic of 1°.Notably, the 1998 USGS mosaic and the 2014 Schenk mosaic are also unaligned by almost 2°, and the recent DEMs released by Schenk are also unaligned with the 2014 Schenk mosaic by 0°.3.

Figure 7 .
Figure 7.Comparison of the existing Triton mosaic with available high-resolution data.(a) A portion of the 1998 USGS global color mosaic in orthographic projection.This region is covered by image c1139350 at a scale of 1262 m pixel −1 , which has been up-sampled to the 600 m pixel −1 scale of the mosaic.(b)The same region as shown in panel (a), but as seen in image c1139627 at 343 m pixel −1 , which was not included in the 1998 USGS mosaic and was previously only available as an uncontrolled image (unaligned to the lower-resolution images) from the PDS.This image has a resolution twice that of the underlying data and was acquired at very high phase (115°; Table1).Tie pointing is therefore challenging, and the image is still slightly misaligned.The image is centered at 349°E and 12°N (north is up) and uses the same projection as Figure5.

Figure 8 .
Figure8.Sequence of six high-resolution images overlaid on our new mosaic.The high-resolution images are c1139619, c1139621, c1139623, c1139625, c1139627, and c1139529, all at pixel scale < 367 m pixel −1 .This mosaic is a screenshot from a GIS in which the individual images have simply been added as layers on top of the underlying mosaic (orthographic projection at 600 m pixel −1 ).

Figure 9 .
Figure9.Location of impact crater (circled in red) proposed to define Triton's longitude system.The crater is located precisely at 40°. 0 E (1°. 1 S) in our mosaic and is readily identifiable in 5 (of 41) Voyager 2 images.A name for the crater is pending IAU approval.

Figure 10 .
Figure 10.Examples of ARD released for Triton.(a) An individual level 2 (calibrated and projected) photogrammetrically controlled Voyager 2 image (c1139445).(b)A partially processed version of the image shown in panel (a) that is calibrated and projected but still includes reseaux and corner marks.This region of Triton (identical to that shown in Figure1(b)) has numerous small plume deposits(Hansen et al. 1990;Hofgartner et al. 2022;Martin et al. 2023), and distinguishing real features from reseau marks is critical.

Table 1
List of the 41 Voyager 2 Images Included in the Triton Control Network

Table 2
List of Images Included in the Global Mosaics