Ten Challenges of Producing an Astronomical Gigapixel Image Best Practices Practices Best

Home , FITS Liberator, Paranal Observatory, VISTA (telescope)

Mathias Jäger Lars Lindberg Christensen Keywords ESA/Hubble Public Information Officer Head of ESO education and Public Outreach Image processing, astronomical images, [email protected] Department (ePOD) VISTA, VVV survey [email protected]

Summary

Public outreach involves developing new methods, testing new technologies and integrating new ideas. Sometimes, the craft of outreach even leads into completely unknown territory. This is the story of a project that led into astronomical and technological terra incognita. It is about the production of a mosaic of the central parts of the Milky Way made with ESO’s VISTA telescope as part of the VVV survey. The outreach system at ESO was tested to its limits, and beyond, by the production of what is still likely to be the largest astronomical image in the world. Several significant challenges had to be over- come, extensive hardware and software upgrades were undertaken and compromises had to be made to produce this stunning image for the public.

VISTA and the VVV survey source of stunning images to be used for public outreach team quickly realised that public outreach. As of March 2015 ESO the huge mosaic of the central Milky Way VISTA — the Visible and Infrared Survey has published 17 press releases based on could be the main focus of a release. When Telescope for Astronomy — is part of results delivered by VISTA, seven of which the work on the image began the team was ESO’s Paranal Observatory and is the larg- were directly related to the VVV survey. confronted with ten major problems which est survey telescope in existence. It is also had to be solved. the most powerful near-infrared survey telescope ever built. The telescope has a The final image 4.1-metre primary mirror and is dedicated Problem 1: Getting the astronomical to conducting wide-angle surveys of the The image which tested the limits of ESO’s data skies with its 67-megapixel digital camera. system, and its team members, covers Simply moving the raw data from place to about 315 square degrees (20.4 × 15.4 place turned out to be a great challenge. VISTA’s observing time is entirely devoted degrees) and shows the centre of the Early on it was decided — on an excep- to mapping the sky systematically and six Milky Way. The observations were carried tional basis and only because of the sheer huge public surveys will take up the major- out using three different near-infrared (JHK) amount of data — that the data had to be ity of the telescope’s first years of opera- filters and the resulting image is monu- down-sampled by a factor of almost two. tion. One of these surveys is the VVV — or mental. It is 9 gigapixels in size, measur- Ordinarily ESO always releases images VISTA Variables in the Vía Láctea — survey, ing 108 199 by 81 503 pixels1. The image pixel by pixel as they are observed, to which started in 2010 and was granted a is so large that, if printed with the resolution make sure that the images the end-user total of 1929 hours of observing time over of a typical book, it would be nine metres receives are optimal. With the support of a five-year period. The survey is scanning across and seven metres tall. This makes ESO’s helpdesk the available storage on the southern plane and bulge of our gal- it likely the largest astronomical image in the ESO FTP server was extended and the axy — 520 square degrees in total — in the world. vast quantities of data — 166 gigabytes five near-infrared filters. of FITS3 files — could be moved back The gigantic dataset contains about and forth between science team member When the survey is completed in 2016, the 173 million objects, out of which about Ignacio Toledo (Atacama Large Millimeter/ outcome will be a catalogue with around 84 million have been confirmed as stars, submillimeter Array [ALMA]) and ESO over one billion point sources, including about which is ten times as many stars as in any several iterations. a million variable objects. This will be used previous study. It is a major step forward to create a three-dimensional map of the in our understanding of our home galaxy. bulge of our galaxy, the Milky Way, to cal- Problem 2: Distortions over the large culate the ages of stellar populations and field study the evolution of globular clusters. Ten problems to solve The next problem occurred during the astronomical data processing. Due to the In addition to its scientific purpose, VISTA’s Work on the image began in 2012 with the large field of view, the image happened superb image quality and the wide field preparation of a science release2 based to have significant projection and distor- of view make the telescope an excellent on a paper by Saito et al. (2012). The ESO tion effects. Most of these problems were

CAPjournal, No. 17, June 2015 29 Ten Challenges of Producing an Astronomical Gigapixel Image

Figure 1. This striking view of the central part of the Milky Way was obtained with the VISTA survey telescope at ESO’s Paranal Observatory in Chile. This huge picture is 108 200 by 81 500 pixels and contains nearly nine billion pixels. It was created by combining thousands of individual images from VISTA. Credit: ESO/VVV Survey/D. Minniti

30 CAPjournal, No. 17, June 2015 CAPjournal, No. 17, June 2015

Ten Challenges of Producing an Astronomical Gigapixel Image 31 Ten Challenges of Producing an Astronomical Gigapixel Image

Figure 2. VISTA is the largest telescope in the world dedicated to surveying the sky. Its primary mirror is 4.1 metres in diameter and it has a huge camera that is sensitive to near-infrared wavelengths. Credit: ESO corrected by Ignacio Toledo. However, a format, failed as the ESO/ESA/NASA FITS Problem 4: Reading BigTIFF much smaller residual misalignment of Liberator programme4 was unable, at the Not only did the available version of the only a few pixels between the three differ- time, to create tiff files above 2 gigabytes. FITS Liberator prove to be inadequate, ent exposures (JHK filters) had to be cor- Since the dynamic range in the dataset was but Photoshop 5, which was used by ESO rected manually in Photoshop later on. not extremely large, the team reverted to at the time, also fell short as it could not using a less interactive method, using the read BigTIFF files properly. Thankfully, software STIFF5 which can create BigTIFF6 and as a matter of pure luck, Photoshop Problem 3: Dynamic range compression files of almost unlimited size. Meanwhile, 6 was released only a few weeks before The normal dynamic range compression, the FITS Liberator programme has now the start of the project and this version was which converts files from FITS format to tiff been updated to write BigTIFF files as well. able to read BigTIFF files. With the new

32 CAPjournal, No. 17, June 2015 CAPjournal, No. 17, June 2015

Figure 3. This view compares the huge mosaic in infrared light from the VISTA survey telescope and a visible-light mosaic view of the same region taken with a small telescope. Credit: ESO/VVV Survey/D. Minniti/Serge Brunier. Acknowledgement: Ignacio Toledo, Martin Kornmesser

software installed, the images could finally Problem 7: Cosmetic cleaning Problem 9: Zoomable version be opened in Photoshop; and due to their The next step was a cosmetic cleaning Assuming that very few users would be size had to be reduced to eight-bit colour stage to remove instrumental artefacts able to actually handle the full 9-gigapixel depth immediately — normally astronom- and other blemishes of a non-cosmic ori- image, ESO wanted to offer a zoomable ical images have 16-bit colour depth and gin. ESO normally uses an outsourcing image10 as the main vehicle for deliver- hence have twice the size. This is some- company, but in this case the transport of ing the experience to the public. However thing which is otherwise only done at the the individual layers was impractical and Zoomify11, the usual tool for creating zoom- last stage in the production of an astro- the clean itself would have been too costly able products at ESO, did not work with PSB nomical colour composite. as it is paid for per megapixel. Therefore it files. The tools Krpano12 and Panotour Pro was decided to perform a more modest, from Kolor13 were finally used and proved but sufficient, in-house cleaning. This led themselves to be very good alternatives. Problem 5: Swapping 600 gigabytes to some interesting feedback from the pub- As soon as the team started to work on lic later on, which will be discussed in the the image in Photoshop the next challenge lessons learned. Problem 10: Web serving emerged. Working on three 9-gigapixel JHK The final challenge that the team had layers with corresponding adjustment lay- to face was to actually serve the large ers in Photoshop used unforeseen amounts Problem 8: Distribution file format individual files and the panorama to the of memory — up to 600 gigabytes of mem- It was clear that the final image should be public. The news of the image spread like ory/swap space. To perform the swapping accessible to as many people as possi- wildfire. The image is the most successful a state-of-the-art solid-state disk (SSD) ble, so the team had to look for the most ESO release to date with many more than was quickly procured to give a workable appropriate and viewable final file format. one million visitors to the press release on solution by providing a pseudo-memory Unfortunately Photoshop 6 does not write the ESO page alone. At the time of publica- delivering up to 1000 megabits per sec- BigTIFF, so an alternative format was tion this success nearly melted the servers ond in real throughput. needed. In the end only Photoshop’s pro- at ESO Headquarters during the peak load prietary PSB format proved itself suitable — they were sluggish for days and at times for formats above 65 000 pixels. Since this the available slots for download were all Problem 6: Graphics card is a much less used format than formats filled, which undoubtedly left some visitors Even with the SSD onscreen interac- like TIFF, this was naturally a compromise. disappointed. Therefore the large 24.6 GB tion with the image for the clean-up in The final PSB file was 24.6 gigabytes in PSB file first had to be moved to another Photoshop, the image was impossible size; but smaller intermediate size formats server and finally to peer-to-peer distri- to handle. Besides the new SSD, a mod- of 407, 258 and 109 thousand pixels were bution in Bittorrent as it took up too many ern graphics card was added to the sys- also created and made available in TIFF slots for too long. After a couple of weeks tem to provide the necessary speed and format for download. the normal distribution system was rein- onboard memory for the onscreen naviga- stated, but the image is still downloaded tion in Photoshop. many times per day. Since the release the ESO server system has been upgraded to a more resilient system.

Ten Challenges of Producing an Astronomical Gigapixel Image 33 Ten Challenges of Producing an Astronomical Gigapixel Image

Figure 4. Central part of the big VVV image with some prominent objects labelled. Credit: ESO/VVV Survey/D. Minniti. Acknowledgement: Ignacio Toledo, Martin Kornmesser

Lessons learned The first and most obvious lesson learned It is also of the upmost importance not was the fact that the data volumes of the to underestimate the needs of the target The completion of the project — despite new survey telescopes are a new paradigm groups. In far more cases than one might the setbacks — and its great success for the creation of astronomical images. expect the public prefers the largest files in made everyone involved proud of the A significant investment in the design of the highest resolution. This also indicates result. The aftermath of the release was new outreach pipelines and in modern that the users have the capabilities to han- also a good time to summarise the les- hard- and software is needed to handle, dle and use these files. sons learned during the project; and there edit and publish the images produced by are almost as many as there were chal- these telescopes. The overloaded ESO servers showed that lenges. alternative ways of content delivery have to

34 CAPjournal, No. 17, June 2015 CAPjournal, No. 17, June 2015

cosmetic cleaning stage. The in-house 7 Download the 40 000 pixel VVV image: cleaning left some artefacts, such as black http://www.eso.org/public/archives/images/ spots in the centres of saturated stars, in publicationtiff40k/eso1242a.tif the image. As a consequence, ESO is still 8 Download the 25 000 pixel VVV image: receiving feedback from hopeful people, http://www.eso.org/public/archives/images/ who report spotting possible new transit- publicationtiff25k/eso1242a.tif ing planets around a few of the bright stars. 9 Download the 10 000 pixel VVV image: http://www.eso.org/public/archives/images/ The last and final lesson learned was prob- publicationtiff10k/eso1242a.tif ably the most important one: teamwork. 10 The zoomable VVV image: Without an amazing team of experts in the http://www.eso.org/public/images/ areas of astronomy, technology and graph- eso1242a/zoomable/ ics it would have been impossible to over- 11 The Zoomify tool: http://www.zoomify.com/ come the individual challenges. In particu- 12 Zoom tool Krpano: http://krpano.com/ lar this project was made possible due to 13 Zoom tool Panotour Pro from Kolor: a collaboration between Olivier Hainaut http://www.kolor.com/ (project lead, PhD astronomer), Martin Kornmesser (graphic designer), Richard Hook (astronomer, press officer), Mathias Andre (developer), Davide de Martin (engi- neer and amateur astronomer), Kaspar Nielsen (PhD physicist and developer), Luis Calçada (graphic designer), Georgia Bladon (science communicator) and Lars Lindberg Christensen (astronomer, head).

Producing colour composites from astronomical data is an important way of illus- trating astronomical progress to the public. Sometimes, work such as this brings us to the forefront of existing technology and stretches the imaginations of the science communicators. The VVV image is such a case, but in turn it has also managed to make a significant impact on the public.

References

Saito et al. 2012, Astronomy and Astrophysics, Biographies 544.

Lars Lindberg Christensen is a science communication specialist and head of the Notes ESO education and Public Outreach Depart- ment (ePOD) in Munich, Germany. He leads 1 Download the full VVV image: public outreach and education for the La Silla http://www.eso.org/public/archives/images/ Paranal Observatory, for ESO’s part of ALMA original/eso1242a.psb and APEX, for the European Extremely Large Telescope, for ESA’s part of the Hubble Space 2 ESO science press release: Telescope and for the IAU Press Office. http://www.eso.org/public/news/eso1242/ be found and used. Peer-to-peer networks 3 FITS is a specialised format used in the Mathias Jäger is a science communicator for like Bittorrent turned out to be an excellent astronomical community. astronomy from Austria. He obtained a PhD in astronomy from the University of Heidelberg, 4 way to carry out distributed content deliv- ESO/ESA/NASA FITS Liberator: then worked for the Haus der Astronomie ery. Content distribution networks may also http://www.spacetelescope.org/projects/ before spending six months as an intern at be an interesting solution to take the load fits_liberator/ the European Southern Observatory (ESO) in off the central server. 5 The software STIFF: the education and Public Outreach Depart- ment (ePOD). Since the beginning of 2015 he http://www.astromatic.net/software/stiff has been working as a science communica- As already mentioned above, the team 6 More about bigTIFF files: http://bigtiff.org/ tion freelancer for organisations including ESO learned that it is necessary to remove and ESA/Hubble. every single artefact from the image at the

Ten Challenges of Producing an Astronomical Gigapixel Image 35