A Critical Review of Scientific Visualization in Astronomy Christopher Fluke & Amr Hassan Those not properly initiated into the mysteries of visualization research often seek to understand the images rather than appreciate their beauty. Globus & Raible (1994) CRICOS provider 00111D
Astronomy Data Astronomy is a data intensive science By 2010, 3 Petabytes of data (Brunner et al. 2002) Doubles every 1.5 years (Szalay & Gray 2001) Knowledge is not growing at same rate (Djorgovski 2005) Observation: Optical: LSST, SkyMapper, Radio: ALMA, ASKAP, MeerKat, SKA Simulation: Cosmology: N > 10 10 particles CSIRO Springel et al.
Petascale Astronomy Era Analysis, storage, access? escience paradigm Automated data mining will be essential HOWEVER Still need to look at data Understanding Interpretation Verification (of automated results)
Human Visual System: Ultimate Pattern Matcher! http://img.dailymail.co.uk/
Visuals in Astronomy: A Long History Credits: Rappenglück (1999); Wikimedia Commons; Linda Hill Library of Science, Engineering & Technology; NASA; C Fluke
Much of astronomy is making (pretty) images and graphs Images from the Hubble Space Telescope
The Role of Visualization in Astronomy Planning Observation/monitoring Quality control Knowledge discovery Analysis Publication Formal Education Informal outreach Data V isualization System Knowledge discovery Much of the data is >3D Science
Scientific Visualization = SciVis SciVis is the process of turning N > 3D data into images Inherent geometric structure Interactive examination Understanding not just presentation Fundamental enabling technology for knowledge discovery McCormick, De Fanti & Brown (1987) Multidisciplinary
unc.renci.org http://ensight.com http://medgadget.com http://ensight.com Application Areas Architecture Engineering Molecular modelling Medical imaging
SciVis Research Techniques for displaying data Surface Rendering Volume Rendering Streamlines Efficient, scalable implementations Effective HPC use Parallel rendering Computational steering
Review: What we hope to achieve Identified papers addressing SciVis issues in astronomy Not an account of projects that have used SciVis No similar review exists c.f. Palomino (2003); Leech & Jenness (2005); Dubinski (2008); Kapferer & Riser (2008); Li et al. (2008) Hassan (PhD thesis) and Fluke & Hassan (in prep) Open questions How has SciVis advanced in astronomy? Are there areas for improvement? What future work is required? (Petascale Astronomy Era)
A Bit of Perspective Astronomical SciVis Medical SciVis Number of active researchers < 50 Number of active researchers > 1000 Beneficiaries O(10 4 ) astronomers Beneficiaries O(10 9 ) people!! However O(billion dollar) investments in astronomy data production hardware in next decade probably warrants some additional effort
Literature: Breakdown Large-N particle Simulations AMR Simulation Data Other Simulation Data Custom Visualization Software Multi-wavelength/ Spectral data Distributed/Remote Visualization Collaborative Visualization General Visualization Software Public Outreach Outcomes Will talk about Won t talk about
Case Study: Large-Scale Visualization Numerical simulations N-body gravity N > 10 9 Outputs (x,y,z) positions (v x,v y,v z ) velocity Temperature, density, particle type, Multiple time steps Custom file formats Scattered Points
Brodbeck et al. (1998) Comparative visualization competing Dark Matter models Provided animations on videotape
Navratil et al. (2007) Comparison between techniques Consider impact of visualization on understanding Direct visualization (Partiview) Direct visualization (ParaView) Isosurfaces (ParaView) Spatial relationships and extent of structures visible
Dubinski (2008) Spiral Metamorphosis MYRIAD library Visualization outputs as simulation proceeds Ability to map to different display formats Andromeda and Milky Way Δt = 170 million years
Szalay et al. (2008) 10 10 particles? Brute force approach does not scale Hierarchical structure (spatial octree) GPU = 10 fps on TB of data Selecting points of interest
Case Study: Adaptive Mesh Refinement simulations Hydrodynamical technique Berger & Colella (1989) Multi-scale hierarchy Resolution where required Reduce memory CPU overhead lowered Astronomy use Cosmology Structure formation Gas dynamics First stars
Norman et al. (1999) Lack of suitable tools Not first class data type Re-sampled to structured grid Not scalable Custom VTK software
Kähler et al. (2003; 2006) Volume rendering the first stars Large range of length-scales 1000 light years to a fraction of an AU (10 19 to 10 11 m) Use GPU for real-time interaction
Kähler et al. (2006) Comparison of volume rendering techniques Refinements Texturebased rendering GPU assisted ray-casting
Astronomy data challenges Observation Data volume (Petabytes ) (Low) Signal-to-noise (High) Dynamic range Incomplete or sparse sampling Astronomy-specific coordinate systems (RA, Dec, z) Simulation Number of particles Mesh resolution Range of length scales (solar to cosmological) Data formats!! (FITS, VOTable, Gadget-2, HDF5, ASCII )
SciVis Software Options Commercial SciVis software IDL VisIt TecPlot Amira Free SciVis software Drishti ParaView Write your own code VTK, OpenGL S2PLOT
SciVis Software Options Adapt from another discipline Astronomical Medicine Project (IIC/Harvard) Osirix 3D Slicer
SciVis Software Options Custom astronomy code AstroMD (Becciani et al. 2003) DVR (Beeson et al. 2003) Glnemo (Lambert) Hubble in a Bottle (Kuehne 2005) IFRIT (Gnedin) Karma (Gooch 1996) Partiview (Levy) SPLASH (Price 2007) StarSplatter (Welling) Tipsy (Katz & Quinn) TOPCAT (Taylor) VisIVO (Comparato et al. 2007) VOPlot3D (VO-India)
Why do Astronomers need their own SciVis software? General visualization software can be challenging to use Steep learning curve? Too much functionality? Not quite right functionality? (e.g. quantitative analysis) Lack of native data format support Costly file conversion Inflated file sizes Units! RA, Dec, z (x, y, z) Does not integrate directly with work process Need to convert data to another format
Preliminary Findings Visualization is not Science for astronomers Too many short-term projects (situation improving ) Funding sources? Visualization does not do the Science Interpretation of results Adding meaning is still a task Qualitative vs. Quantitative Emphasis still on pretty pictures /public outreach? Few packages give astronomers statistical tools they need
Preliminary Findings Usability and Interoperability Applications with focused functionality had wider use Being cross platform is very important Lifetime of software? Documentation? Support? Adjustable parameters and configurations Deep understanding required for both visualization and data Only a small portion of work targeted general data
Preliminary Findings Publication and citation Does astronomy software receive enough credit? E.g. Weiner et al. 2009 Hard to track number of software users Where to publish astronomy SciVis papers? MNRAS, A&A, ApJ, PASA, New Astronomy key words: Methods: data analysis Methods: numerical Techniques: image processing What about Methods: visualization or Software:??? Not all astronomy SciVis papers are in ADS
Opportunities Increased use of GPU For graphics and computation [Hassan talk Wed] Comparison between observation and models Enhanced comprehension and cognition Colour maps Shaders/Transfer functions Alternative interaction devices Spaceball Gooch (1996) Programmable handheld devices; Game controllers Software development Modular craftsmanship (Goodman 2009) [Fluke talk Wed]
Collaborative Environments Nakasone et al. (2009) Second Life Djorgovski et al. (2009) Google Wave
Conclusion Pretty pictures are important BUT Aim to maximize scientific return from data through advances in scientific visualization Many current astronomy visualization approaches will be seriously challenged by, or completely incompatible with, the Petascale Data Era The time for new advancements in SciVis for astronomy is now!
Types of Astronomical Data Brunner et al. (2002): Imaging data: 2D, narrow Δλ, fixed epoch Catalogs: secondary parameters determined from processing (coordinates, fluxes, sizes, etc). Spectroscopic data and products (e.g. redshifts, chemical composition, etc). Studies in the time domain: moving objects, variable and transient sources Numerical simulations
Common Data Formats Structured Grid Unstructured Grid Scattered Points Adaptive Mesh Refinement Points + Connectivity Points Only
Palomino (2003)