Seeing the Sky Visualization & Astronomers

Alyssa A. Goodman Harvard Smithsonian Center for Astrophysics & Radcliffe Institute for Advanced Study @aagie WorldWide Telescope

Gm1m2

F= gluemultidimensional data exploration R2 Cognition “Paper of the Future” Language* Data Pictures

Communication

*“Language” includes words & math Why Galileo is my Hero Explore-Explain-Explore

Notes for & re-productions of Siderius Nuncius 1610 WorldWide Telescope

Galileo’s New Order, A WorldWide Telescope Tour by Goodman, Wong & Udomprasert 2010 WWT Software Wong (inventor, MS Research), Fay (architect, MS Reseearch), et al., now open source, hosted by AAS, Phil Rosenﬁeld, Director see wwtambassadors.org for more on WWT Outreach WorldWide Telescope

Communication

*“Language” includes words & math enabled by d3.js (javascript) outputs

d3po

Cognition

Communication

[demo] [video]

Many thanks to Alberto Pepe, Josh Peek, Chris Beaumont, Tom Robitaille, Adrian Price-Whelan, Elizabeth Newton, Michelle Borkin & Matteo Cantiello for making this posible. 1610 4 Centuries from Galileo to Galileo

1665 1895 2009

2015 WorldWide Telescope

Gm1m2

F= gluemultidimensional data exploration R2 WorldWide Telescope

gluemultidimensional data exploration WorldWide Telescope

gluemultidimensional data exploration

Data, Dimensions, Display

1D: Columns = “Spectra”, “SEDs” or “Time Series” 2D: Faces or Slices = “Images” 3D: Volumes = “3D Renderings”, “2D Movies” 4D:4D Time Series of Volumes = “3D Movies” Data, Dimensions, Display

Spectral Line Observations

Loss of 1 dimension

Mountain Range No loss of information

Data, Dimensions, Display mm peak (Enoch et al. 2006) sub-mm peak (Hatchell et al. 2005, Kirk et al. 2006) 13CO (Ridge et al. 2006) mid-IR IRAC composite from c2d data (Foster, Laakso, Ridge, et al.)

Optical image (Barnard 1927) 3D Viz made with VolView AstronomicalMedicine@ NATURE Vol 457 1 January 2009 LETTERSLETTERS NATURE | Vol 457 | 1| January 2009|

abab usingusing 2D maps 2D of maps column of column density. density. With this With early this 2D work early as 2D inspira- work as inspira- tion,tion, we have we developed have developed a structure-identification a structure-identification algorithm algorithm that that abstractsabstracts the hierarchical the hierarchical structure structure of a 3D (ofp– ap– 3Dv) data (p–p cube–v) data into cube into an easilyan easily visualized visualized representation representation called a ‘dendrogram’ called a ‘dendrogram’10. Although10. Although 11,12 well developedwell developed in other in data-intensive other data-intensive fields , fields it is curious11,12, it is that curious the that the applicationapplication of treeof methodologies tree methodologies so far in astrophysicsso far in astrophysics has been rare, has been rare, and almostand almost exclusively exclusively within thewithin area the of galaxy area of evolution, galaxy evolution, where where ‘merger trees’ are being used with increasing frequency13. ‘merger trees’ are being used with increasing frequency13.

(dec.) (dec.) Figure 3 and its legend explain the construction of dendrograms v vz z y y

(dec.) (dec.) Figure 3 and its legend explain the construction of dendrograms v vz z y y schematically. The dendrogram quantifies how and where local max- x (RA) x (RA) ima ofschematically. emission merge The with dendrogram each other, quantifies and its implementation how and where is local max- x (RA) x (RA) explainedima inof Supplementary emission merge Methods. with each Critically, other, the and dendrogram its implementation is is Click to rotate determinedexplained almost in Supplementary entirely by the data Methods. itself, and Critically, it has negligible the dendrogram is Click to rotate sensitivitydetermined to algorithm almost parameters. entirely To by make the data graphical itself, presentation and it has negligible c 8 Self-gravitating Self-gravitating All structure leaves structures possiblesensitivity on paper to and algorithm 2D screens, parameters. we ‘flatten’ Tothe makedendrograms graphical of 3D presentation c 8 Self-gravitating Self-gravitating All structure data (see Fig. 3 and its legend), by sorting their ‘branches’ to not 6 leaves structures possible on paper and 2D screens, we ‘flatten’ the dendrograms of 3D cross, which eliminates dimensional information on the x axis while (K) 6 data (see Fig. 3 and its legend), by sorting their ‘branches’ to not

mb 4 preserving all information about connectivity and hierarchy. T cross, which eliminates dimensional information on the x axis while (K) Numbered ‘billiard ball’ labels in the figures let the reader match

mb 42 preserving all information about connectivity and hierarchy. T features between a 2D map (Fig. 1), an interactive 3D map (Fig. 2a online)Numbered and a sorted ‘billiard dendrogram ball’ labels (Fig. 2c). in the figures let the reader match 20 features between a 2D map (Fig. 1), an interactive 3D map (Fig. 2a d CLUMPFIND segmentation A dendrogram of a spectral-line data cube allows for the estimation 8 online) and a sorted dendrogram (Fig. 2c). 2009 0 of key physical properties associated with volumes bounded by iso- d 86 CLUMPFIND segmentation surfaces,A such dendrogram as radius (ofR), a velocityspectral-line dispersion data cube (sv) and allows luminosity for the estimation (L). Theof keyvolumes physical can have properties any shape, associated and in other with work volumes14 we focus bounded on by iso- (K)

mb 64 the significancesurfaces, such of the as especially radius (R elongated), velocity features dispersion seen (s inv) L1448 and luminosity T (Fig.( 2a).L). The The volumes luminosity can is have an approximate any shape, and proxy in otherfor mass, work such14 we focus on

(K) 20 2 21 21 2 that Mlum 5 X13COL13CO, where X13CO 5 8.0 3 10 cm K km s

mb 4 the significance of the especially elongated features seen in L1448 T (ref. 15;(Fig. see 2a). Supplementary The luminosity Methods is an and approximate Supplementary proxy Fig. for 2). mass, such 0 The derived values for size, mass and velocity dispersion can20 then2 be 21 21 2 that Mlum 5 X13COL13CO, where X13CO 5 8.0 3 10 cm K km s 3D PDF Figure 2 | Comparison of the ‘dendrogram’ and ‘CLUMPFIND’ feature- used to estimate the role of self-gravity at each point in the hierarchy, (ref. 15; see Supplementary Methods and Supplementary Fig. 2). 13 via calculation of an ‘observed’ virial parameter, a 5 5s 2R/GM . identification0 algorithms as applied to CO emission from the L1448 obs v lum region of Perseus. a, 3D visualization of the surfaces indicated by colours in In principle,The derived extended values portions for size, of the mass tree (Fig. and 2,velocity yellow highlighting)dispersion can then be used to estimate the role of self-gravity at each point in the hierarchy, Figurethe dendrogram 2 | Comparison shown ofin c the. Purple ‘dendrogram’ illustrates theand smallest ‘CLUMPFIND’ scale self- feature- where aobs , 2 (where gravitational energy is comparable to or larger 13 2 High-Dimensional identificationgravitating structures algorithms in the as region applied corresponding to CO emission to the leaves from of the the L1448 than kineticvia calculation energy) correspond of an ‘observed’ to regions virial of parameter,p–p–v spacea whereobs 5 self-5sv R/GMlum. dendrogram; pink shows the smallest surfaces that contain distinct self- region of Perseus. a, 3D visualization of the surfaces indicated by colours ingravityIn is principle, significant. extended As aobs only portions represents of the the tree ratio (Fig. of kinetic 2, yellow energy highlighting) thegravitating dendrogram leaves shown within inthem;c. Purple and green illustrates corresponds the smallest to the surface scale self-in the to gravitationalwhere a energy, 2 (where at one gravitational point in time, energy and does is comparable not explicitly to or larger data cube containing all the significant emission. Dendrogram branches obs gravitating structures in the region corresponding to the leaves of the 16 corresponding to self-gravitating objects have been highlighted in yellow capturethan external kinetic over-pressure energy) correspond and/or magnetic to regions fields of ,p its–p measured–v space where self- dendrogram; pink shows the smallest surfaces that contain distinct self- over the range of T (main-beam temperature) test-level values for which valuegravity should only is significant. be used as aAs guidea toonly therepresents longevity (boundedness) the ratio of kinetic of energy data in a gravitating leaves withinmb them; and green corresponds to the surface in the obs the virial parameter is less than 2. The x–y locations of the four ‘self- any particularto gravitational feature. energy at one point in time, and does not explicitly data cube containing all the significant emission. Dendrogram branches gravitating’ leaves labelled with billiard balls are the same as those shown in capture external over-pressure and/or magnetic fields16, its measured correspondingFig. 1. The 3D visualizations to self-gravitating show position–position–velocity objects have been highlighted (p–p–v) in space. yellow value should only be used as a guide to the longevity (boundedness) of overRA, theright range ascension; of Tmb dec.,(main-beam declination. temperature) For comparison test-level with the values ability for of which Local max any particular feature. “Paper” thedendrograms virial parameter (c) to track is less hierarchical than 2. The structure,x–y locationsd shows of a pseudo- the four ‘self- Test level gravitating’dendrogram leaves of the labelled CLUMPFIND with billiard segmentation balls are (b), the with same the assame those four shown in labels used in Fig. 1 and in a. As ‘clumps’ are not allowed to belong to larger Local max

Fig. 1. The 3D visualizations show position–position–velocity (p–p–v) space. Leaf RA,structures, right ascension; each pseudo-branch dec., declination. in d is simply For acomparison series of lines with connecting the ability the of Local max Merge dendrogramsmaximum emission (c) to value track in hierarchical each clump structure,to the thresholdd shows value. a pseudo-A very large Leaf number of clumps appears in b because of the sensitivity of CLUMPFIND to Test level on its way dendrogram of the CLUMPFIND segmentation (b), with the same four Local max noise and small-scale structure in the data. In the online PDF version, the 3D labels used in Fig. 1 and in a. As ‘clumps’ are not allowed to belong to larger

Intensity level Local max Leaf

cubes (a and b) can be rotated to any orientation, and surfaces can be turned Branch structures, each pseudo-branch in d is simply a series of lines connecting the on and off (interaction requires Adobe Acrobat version 7.0.8 or higher). In Leaf Merge Merge maximumthe printed emission version, thevalue front in eachface of clump each 3D tothe cube threshold (the ‘home’ value. view A in very the large Leaf numberinteractive of clumps online version) appears corresponds in b because exactly of the to sensitivity the patch of sky CLUMPFIND shown in to Local max

to the Future Trunk noiseFig. 1, and and small-scale velocity with structure respect to in the the Local data. Standard In the online of Rest PDF increases version, from the 3D 21 21

2 Intensity level cubesfront ( (a and0.5 kmb) cans ) be to rotated back (8 km to any s ). orientation, and surfaces can be turned Branch

on and off (interaction requires Adobe Acrobat version 7.0.8 or higher). InFigure 3 Merge| Schematic illustration ofLeaf the dendrogram process. Shown is the the printed version, the front face of each 3D cube (the ‘home’ view in theconstruction of a dendrogram from a hypothetical one-dimensional interactivedata, CLUMPFIND online version) typically corresponds finds features exactly on a to limited the patch range of of sky scales, shown inemission profile (black). The dendrogram (blue) can be constructed by

Fig.above 1, and but velocity close to thewith physical respect resolution to the Local of Standard the data, and of Rest its results increases can from‘dropping’ a test constant emission level (purple)Trunk from above in tiny steps frontbe overly (20.5 dependent km s21) to on back input (8 km parameters. s21). By tuning CLUMPFIND’s (exaggerated in size here, light lines) until all the local maxima and mergers two free parameters, the same molecular-line data set8 can be used to are found, and connected as shown. The intersection of a test level with the emissionFigure is a set 3 | ofSchematic points (for illustration example the of light the purpledendrogram dots) in process. one Shown is the show either that the frequency distribution of clump mass is the same construction of a dendrogram from a hypothetical one-dimensional as the initial mass function of stars or that it follows the much shal- dimension, a planar curve in two dimensions, and an isosurface in three data, CLUMPFIND typically finds features on a limited range of scales,dimensions.emission The profile dendrogram (black). of 3D The data dendrogram shown in Fig. (blue) 2c is can the be direct constructed by abovelower but mass close function to the physical associated resolution with large-scale of the data, molecular and its clouds results cananalogue‘dropping’ of the tree a test shown constant here, only emission constructed level (purple)from ‘isosurface’ from above rather in tiny steps (Supplementary Fig. 1). than ‘point’(exaggerated intersections. in size It here,has been light sorted lines) and until flattened all the for local representation maxima and mergers be overly dependent on input parameters. By tuning CLUMPFIND’s9 twoFour free parameters,years before theadvent same molecular-line of CLUMPFIND, data ‘structure set8 can trees’be used toon a flatare page, found, as fully and connectedrepresenting as dendrograms shown. The for intersection 3D data cubes of a would test level with the were proposed as a way to characterize clouds’ hierarchical structure require four dimensions. [demo/video] show either that the frequency distribution of clump mass is the same emission is a set of points (for example the light purple dots) in one as64 the initial mass function of stars or that it follows the much shal- dimension, a planar curve in two dimensions, and an isosurface in three © 2 0 0 9 Macmillan Publis hers Limited.dimensions. All rights res e Therved dendrogram of 3D data shown in Fig. 2c is the direct lower mass function associated with large-scale molecular clouds analogue of the tree shown here, only constructed from ‘isosurface’ rather (Supplementary Fig. 1). than ‘point’ intersections. It has been sorted and flattened for representation Four years before the advent of CLUMPFIND, ‘structure trees’9 on a flat page, as fully representing dendrograms for 3D data cubes would were proposed as a way to characterize clouds’ hierarchical structure require four dimensions. Goodman et al. 2009, Nature, 64 © 2 0 0 9 Macmillan Publis hers Limited. All rights res erved cf: Fluke et al. 2009

Why AstronomicalMedicine?

“Keith” “Perseus”

“z” is depth into head “z” is line-of-sight velocity Why AstronomicalMedicine?

Composite X-ray CT/MRI CT composite

Astronomy & Medicine both rely on Optical high-dimensional,Radio big, wide,MRI data for insight.SPECT

chandra.harvard.edu/photo/2014/m106/ Chang, et al. 2011, brain.oxfordjournals.org/content/134/12/3632 WorldWide Telescope

gluemultidimensional data exploration mm peak (Enoch et al. 2006) Wide Data sub-mm peak (Hatchell et al. 2005, Kirk et al. 2006)

13CO (Ridge et al. 2006) mid-IR IRAC composite from c2d data (Foster, Laakso, Ridge, et al.)

Optical image (Barnard 1927) Wide Data

Temperature Foreground amplitudes from Commander, Planck Data [Feb 2015] WorldWide Telescope Big DAta BIG DATA and ”Human-Aided Computing”

mark bubbles

machine- learning algorithm (Brut)

example here from: Beaumont, Goodman, Kendrew, Williams & Simpson 2014; based on Milky Way Project catalog (Simpson et al. 2013), which came from Spitzer/GLIMPSE (Churchwell et al. 2009, Benjamin et al. 2003), cf. Shenoy & Tan 2008 for discussion of HAC; astroml.org for machine learning advice/tools BIG DATA and ”Human-Aided Computing”

mark neurons

machine- learning algorithm (RF+CRF)

example here from: Kaynig...Lichtman...Pﬁster et al. 2013, “Large-Scale Automatic Reconstruction of Neuronal Processes from Electron Microscopy Images”; cf. Shenoy & Tan 2008 for discussion of HAC; astroml.org for machine learning advice/tools (Note: RF=Random Forest; CRF=Conditional Random Fields.) Big and Wide Data

Movie: Volker Springel, formation of a cluster of galaxies. Millenium Simulation requires 25TB for output. WorldWide Telescope

gluemultidimensional data exploration LETTERS NATURE | Vol 457 | 1 January 2009

ab using 2D maps of column density. With this early 2D work as inspira- tion, we have developed a structure-identification algorithm that abstracts the hierarchical structure of a 3D (p–p–v) data cube into an easily visualized representation called a ‘dendrogram’10. Although well developed in other data-intensive fields11,12, it is curious that the application of tree methodologies so far in astrophysics has been rare, and almost exclusively within the area of galaxy evolution, where ‘merger trees’ are being used with increasing frequency13.

(dec.) (dec.) Figure 3 and its legend explain the construction of dendrograms v vz z y y schematically. The dendrogram quantifies how and where local max- x (RA) x (RA) ima of emission merge with each other, and its implementation is explained in Supplementary Methods. Critically, the dendrogram is Click to rotate determinedThese almost entirely by theare data itself, and it has negligible sensitivity to algorithm parameters. To make graphical presentation c 8 Self-gravitating Self-gravitating All structure leaves structures possible on paper and 2D screens, we ‘flatten’ the dendrograms of 3D 6 data (see Fig.“dead” 3 and its legend), by sorting their ‘branches’ to not cross, which eliminates dimensional information on the x axis while (K)

mb 4 preserving all information about connectivity and hierarchy. T Numberedpanels! ‘billiard ball’ labels in the figures let the reader match 2 features between a 2D map (Fig. 1), an interactive 3D map (Fig. 2a online) and a sorted dendrogram (Fig. 2c). 2009 0 A dendrogram of a spectral-line data cube allows for the estimation d 8 CLUMPFIND segmentation That’s not of key physical properties associated with volumes bounded by iso- 6 surfaces, such as radius (R), velocity dispersion (sv) and luminosity (L). The volumes can have any shape, and in other work14 we focus on

(K) good

mb 4 the significance of the especially elongated features seen in L1448 T (Fig. 2a). The luminosity is an approximate proxy for mass, such 20 2 21 21 3D PDF 2 that Menough.lum 5 X13COL13CO, where X13CO 5 8.0 3 10 cm K km s (ref. 15; see Supplementary Methods and Supplementary Fig. 2). 0 The derived values for size, mass and velocity dispersion can then be Figure 2 | Comparison of the ‘dendrogram’ and ‘CLUMPFIND’ feature- used to estimate the role of self-gravity at each point in the hierarchy, 13 2 High-Dimensional identification algorithms as applied to CO emission from the L1448 via calculation of an ‘observed’ virial parameter, aobs 5 5sv R/GMlum. region of Perseus. a, 3D visualization of the surfaces indicated by colours in In principle, extended portions of the tree (Fig. 2, yellow highlighting) the dendrogram shown in c. Purple illustrates the smallest scale self- where aobs , 2 (where gravitational energy is comparable to or larger gravitating structures in the region corresponding to the leaves of the than kinetic energy) correspond to regions of p–p–v space where self- dendrogram; pink shows the smallest surfaces that contain distinct self- gravity is significant. As a only represents the ratio of kinetic energy data in a gravitating leaves within them; and green corresponds to the surface in the obs to gravitational energy at one point in time, and does not explicitly data cube containing all the significant emission. Dendrogram branches 16 corresponding to self-gravitating objects have been highlighted in yellow capture external over-pressure and/or magnetic fields , its measured value should only be used as a guide to the longevity (boundedness) of over the range of Tmb (main-beam temperature) test-level values for which “Paper” the virial parameter is less than 2. The x–y locations of the four ‘self- any particular feature. gravitating’ leaves labelled with billiard balls are the same as those shown in Fig. 1. The 3D visualizations show position–position–velocity (p–p–v) space. RA, right ascension; dec., declination. For comparison with the ability of Local max dendrograms (c) to track hierarchical structure, d shows a pseudo- Test level on its way dendrogram of the CLUMPFIND segmentation (b), with the same four labels used in Fig. 1 and in a. As ‘clumps’ are not allowed to belong to larger Local max structures, each pseudo-branch in d is simply a series of lines connecting the Leaf Merge maximum emission value in each clump to the threshold value. A very large Leaf number of clumps appears in b because of the sensitivity of CLUMPFIND to to the Future Local max noise and small-scale structure in the data. In the online PDF version, the 3D Intensity level

cubes (a and b) can be rotated to any orientation, and surfaces can be turned Branch

on and off (interaction requires Adobe Acrobat version 7.0.8 or higher). In Merge Leaf the printed version, the front face of each 3D cube (the ‘home’ view in the interactive online version) corresponds exactly to the patch of sky shown in

Fig. 1, and velocity with respect to the Local Standard of Rest increases from Trunk front (20.5 km s21) to back (8 km s21). Figure 3 | Schematic illustration of the dendrogram process. Shown is the construction of a dendrogram from a hypothetical one-dimensional data, CLUMPFIND typically finds features on a limited range of scales, emission profile (black). The dendrogram (blue) can be constructed by above but close to the physical resolution of the data, and its results can ‘dropping’ a test constant emission level (purple) from above in tiny steps be overly dependent on input parameters. By tuning CLUMPFIND’s (exaggerated in size here, light lines) until all the local maxima and mergers two free parameters, the same molecular-line data set8 can be used to are found, and connected as shown. The intersection of a test level with the show either that the frequency distribution of clump mass is the same emission is a set of points (for example the light purple dots) in one as the initial mass function of stars or that it follows the much shal- dimension, a planar curve in two dimensions, and an isosurface in three dimensions. The dendrogram of 3D data shown in Fig. 2c is the direct lower mass function associated with large-scale molecular clouds analogue of the tree shown here, only constructed from ‘isosurface’ rather (Supplementary Fig. 1). than ‘point’ intersections. It has been sorted and flattened for representation Four years before the advent of CLUMPFIND, ‘structure trees’9 on a flat page, as fully representing dendrograms for 3D data cubes would were proposed as a way to characterize clouds’ hierarchical structure require four dimensions. Goodman et al. 2009, Nature, 64 © 2 0 0 9 Macmillan Publis hers Limited. All rights res erved cf: Fluke et al. 2009 Linked Views of High-dimensional Data

John Tukey

3D Data Abstraction

Statistics 100 75 50 25 0 figure, by M. Borkin, reproduced from Goodman 2012, “Principles of High-Dimensional Data Visualization in Astronomy” John Tukey’s Legacy

PRIM-9 PRIM-H

XGobi GGobi RGGobi

1970 1980 1990 2000 2010 Linked Views of High-dimensional Data (in Python) Glue gluemultidimensional data exploration

video by Tom Robitaille, lead glue developer glue created by: C. Beaumont, M. Borkin, P. Qian, T. Robitaille, and A. Goodman, PI Linked Views of High-dimensional Data (in Python) Glue gluemultidimensional data exploration

video by Chris Beaumont, glue developer glue created by: C. Beaumont, M. Borkin, P. Qian, T. Robitaille, and A. Goodman, PI your handout

Data File 1 deﬁne new variables, standard data import/export insights, standard loaders interactive plots for the web, 1D, 2D & 3D

built-in save state, all from GUI Data plots File 2 link data ﬁles’ highlight live or attributes algorithmic selections

custom data … with Boolean logic custom loaders custom buttons, features plots plug-in Data File N ?

user conﬁg.py ﬁle access to all matplotlib functions run & interact with glue from (loaders, colors, plot types, +) through built-in IPython terminal Jupyter notebook & other tools +options

glueviz.org What is visualization (and all this software) for?

INSIGHT

CONTEXT PATTERN RECOGNITION EVALUATION Spatial Ideas Algorithms Non-Spatial Outliers Errors glueing together the Milky Way

gluemultidimensional data exploration Logan airport (and my FBI file)

gluemultidimensional data exploration JupYter lab: Glue in the browser

Video courtesy of Maarten Breddels, consulting developer

in the browser gluemultidimensional data exploration Yes, more on Nah, this maybe visualization later… challenge! a visualization saga…

WorldWide Telescope

gluemultidimensional data exploration The challenge of 3D Selection

A state-of-the-art 3D model of the stars & gas near the Orion nebula, created at Orion (un)plugged, Vienna, 2015. Expert builders (~20 total) include: Joao Alves, John Bally, Alyssa Goodman & Eddie Schlafly. (cf. “Image & Meaning” workshops by Felice Frankel) YouTube video explanation; WWT Tour The challenge of 3D Selection The challenge of 3D Selection WorldWide Telescope

Gm1m2

F= gluemultidimensional data exploration R2 Literature as (a filter for) Data

Many thanks to Alberto Pepe, August Muench, Thomas Boch, Jonathan Fay, Michael Kurtz, Alberto Accomazzi, Julie Steffen, Laura Trouille, David Hogg, Dustin Lang, Christopher Stumm, Chris Beaumont & Phil Rosenfield for making this all work! Bringing “Dead” Data Back to Life ADS All-Sky Survey & Astronomy Rewind your handout ”putting articles and images (back) on the Sky” 1. Images Extracted from Journal Articles 0. ADS All-Sky Survey offers (ﬁltered) article density layer on the Sky 2. Missing coordinate metadata added back to images, either…

3. “Solved” images returned to ADS & Astronomy Image Explorer …automatically, applying astronometry.net to “historical” wide-ﬁeld optical images, or images

4. New button in Astronomy Image Explorer offers image-in- context, using AAS’ WorldWide Telescope in the browser via “Astronomy Rewind” Zooniverse Citizen Science Project

“recent” images

click entries on the ADS timeline to WorldWide Telescope Zooniverse Astrometry.net ADS All Sky Survey Astronomy Image Explorer Astronomy Rewind 1992 try out 2008 2009 2011 2014 2014 2017 services ≡

ASTRONOMY REWIND ABOUT CLASSIFY TALK COLLECT

FEEDBACK

Thanks all for helping with our 2nd beta! Beta is now complete. We'll be revising the site Who, How, and Who’s Paying? yourbased on yourhandout feedback. Stay tuned for our full launch! The ADS All Sky Survey was ﬁrst funded via a 2012 grant Help us create a database of astro- referenced old Astronomy images. from the NASA ADAP program

to Seamless Astronomy, in Learn more Get started collaboration with CDS, Astrometry.net and Microsoft Research.

Articles-on-the-Sky was ﬁrst deployed in 2014, using 1 person is talking about Astronomy Rewind right now. APIs from WWT (Microsoft Join in Research, now AAS)

and CDS (Aladin) ASTRONOMY REWIND STATISTICS

Images-on-the-Sky 100% Complete relies on the astrometry.net, 201 1,987 360 360 Volunteers Classiﬁcations Subjects Completed Subjects Zooniverse, IOP/AAS Astronomy

Image Explorer and WorldWide WORDS FROM THE RESEARCHER Telescope platforms, and it is "Your contributions unlock the information from old astronomy journals. Thank you and enjoy the funded by the American images!" Astronomical Society, in addition to the NASA ADAP ABOUT ASTRONOMY REWIND This project is part of an ongoing NASA-funded eﬀort aimed at turning the SAO/NASA Astrophysics grant. Data System (ADS) into a data resource. The result will be a database of astro-referenced images, i.e., images of the sky for which coordinates, orientation, and pixel scale will be publicly available through NASA data archives, the Astronomy Image Explorer, and World Wide Telescope, thanks to your help! These projects rely on open source sofware, primarily hosted on GitHub.

PI to contact for more information

Alyssa Goodman, Harvard Projects About Us Zooniverse Talk ! " + Collections Education Daily Zooniverse [email protected] Build a Project Our Team Blog How to Build Publications Project Policies Acknowledgements Contact Us

Alyssa A. Goodman Harvard Smithsonian Center for Astrophysics & Radcliffe Institute for Advanced Study @aagie To continue the conversation…

Creativity & Collaboration: Revisiting Cybernetic Serendipity Natonal Academy of Sciences Sackler Colloquium, March 13-14, 2018, Washington, DC

Role/Play: Collaborative Creativity and Creative Collaborations Natonal Academy of Sciences Sackler Student Fellows Symposium, March 12, 2018, Washington DC

www.nasonline.org/Sackler-Creatvity-Collaboraton

Creativity & Collaboration at NAS March 2018 with Ben Shneiderman, Maneesh Agrawala, Roger Malina, 10qviz.org with Arzu Çöltekin (beta 2017, release 2018) Youngmoo Kim & Donna Cox