interactive visual analysis of multi-faceted scientific data · (multiple data attributes, e.g.,...

Interactive Visual Analysis of Multi-faceted Scientific Data

Johannes Kehrer 1,2,3

1 Institute of Computer Graphics and Algorithms, Vienna University of Technology

2 Department of Informatics, University of Bergen3 VRVis Research Center, Vienna

J. Kehrer Visual Analysis of Multi-faceted Scientific Data

Increasing amounts of scientific data

Hard to analyze and understand

Motivation

time-dependent3D datamedical scanner computational simulation

2


“The purpose of visualizationis insight, not pictures”

[Shneiderman ’99]

Different application areas

Visualization

[Burns et al., 2007] [Laramee et al., 2003] [SequoiaView]

3

J. Kehrer Visual Analysis of Multi-faceted Scientific DataJohannes Kehrer4

Typical Visualization Tasks

Visualization is good for visual exploration

find unknown/unexpected

generate new hypothesis

visual analysis (confirmative vis.) verify or reject hypotheses

information drill-down

presentation show/communicate results


Spatiotemporal data

Multi-variate/multi-field data(multiple data attributes, e.g., temperature or pressure)

Multi-modal data(CT, MRI, large-scale measurements, simulations, etc.)

Multi-run/ensemble simulations (repeated with varied parameter settings)

Multi-model scenarios(e.g., coupled climate model)

Multi-faceted Scientific Data

multi-run distribution per cell

3D time-dependent simulation data

5


[ Böt

tinge

r, Cl

imaV

is08

]

Land

Multi-faceted Scientific Data

Coupled climate models


Literature review of 200+ papers on scientific data

How are vis., interaction, and comput. analysis combined?

Categorization

[compare to Keim et al. 2009; Bertine & Lalanne 2009]

what are main characteristics /

features

data abstraction& aggregation

how to represent the data

visual data fusion

visual mapping comput. analysis

relation &comparison

navigation

focus+context &overview+detail interactive

feature spec.

interaction concepts(linking & brushing, zooming,

view reconfiguration, etc.)

interactive visual analysis

7


Visual vs. Computational Analysis

Interactive Visual Analysis+ user-guided analysis possible

+ detect interesting features without looking for them

+ understand results in context

+ uses power of human visual system

− human involvement not always possible or desirable (expensive!)

− limited dimensionality

− often only qualitative results

− (still) often unfamiliar

Automated Data Analysis- needs precise definition of goals

- limited tolerance of data artifacts

- result without explanation

- computationally expensive

+ hardly any interaction required (after setup)

+ scales better w.r.t. many dimensions

+ precise results

+ long history (mostly statistics)

8


Fusion within a single visualization common frame of reference

layering techniques (e.g., glyphs, color, transparencey)

multi-volume rendering (coregistration, segmentation)

Helix glyphs [Tominski et al. 05] Layering [Kirby et al. 99] Multi-volume rendering [Beyer et al. 07]

visual mapping comput. analysisinteractive visual analysis

visual data fusion


focus+context &overview+detail

navigationinteractive

feature spec.


spatiotemporal multi-modalmulti-variate

9


Layering techniques [Wong ‘02]

opacity modulation

filigreed

colormap enhancement

2D heightmap

colormap + square wave modulation


visual data fusion




feature spec.



Textures and Colors [Healey & Enns 02]

temperature color

wind speed coverage

pressure size

precipitation orientation


visual data fusion




feature spec.


11


Variations of super ellipse-based glyphs

LowHigh

High

Pressure

Vapor

Glyph color: Normalized Temperature Glyph Rotation (-45°, 45°): Flow Velocity

Low


visual data fusion




feature spec.


Glyph-basedvisualization[Lie et al. 09]


Fusion of multiple simulation runs spaghetti plots [Diggle et al. 02]

summary statistics (box plots and glyphs)

EnsembleVis [Potter et al. 09]


visual data fusion




feature spec.


Glyph-based overview [Kehrer et al. 11]

multi-run multi-run

isocontours

13


Fusion of multiple simulation runs spaghetti plots [Diggle et al. 02]

summary statistics (box plots and glyphs)

EnsembleVis [Potter et al. 09]


visual data fusion




feature spec.


Glyph-based overview [Kehrer et al. 11]

multi-run multi-run

q1

q2

q3

isocontours

14


Taxonomy [Gleicher et al. 2011]

side-by-side comparison

overlay in same coordinate system

explicit encoding of differences / correlations


visual data fusion


navigation


feature spec.


2-tone coloring [Saito et al. 05] Nested surfaces [Buskin et al. 11]

spatiotemporal multi-modal

side-by-side comp.

explicit encodingof differences

overlay

15


Mon Tue

Thu Fri

Wed

Sat

Sun

average traffic

Difference Views [Daae Lampe et al. 2010]


visual data fusion


navigation


feature spec.


spatiotemporal

16


3D transition between 2 scatterplots

scatterplotmatrix

visual mapping comput. analysisinteractive visual analysisrelation &

comparisonnavigation


feature spec.


visual data fusion

multi-variate

Interactive search, zooming, and panning

Scatterplot Matrix Navigation [Elmqvist et al. 2008]

17


[Vio

la e

t al.

06]

segmented volume data

Ranking/quality metrics [Bertini et al. 2011]

clustering, correlations, outliers, image quality, etc.

Automated viewpoint selection information-theoretic measures




feature spec.


visual data fusion

[Joha

nsso

n &

Joha

nsso

n 09

]multi-variate

18





feature spec.


visual data fusion

variationsfocal point

input output

variations

Parameter space navigation (multi-run data)

[Berger et al. 11]

focal point

19


Focus+context visualization different graphical resources (space, opacity, color, etc.)

focus specification (e.g., by pointing, brushing or querying)

Clustering & outlierpreservation


visual data fusion


navigation


feature spec.


Outlier-preserving focus+context [Novontný & Hauser 06]

20



visual data fusion


navigation


feature spec.


Overview+detail representation of multi-run data

Brushing statistical moments [Kehrer et al. 10]

multi-run datasummary statistics

quantile plot

21


Brushing in multiple linked views

SimVis [Doleisch et al. 03, 04]

attribute views

3D view


visual data fusion


navigation


feature spec.data abstraction& aggregation

22


Select function graphs based on similarity [Muigg et al. 2008]

pattern sketched by user

similarity evaluated on gradients (1st derivative)


visual data fusion


navigation




Tight integration with supervised machine learning


visual data fusion


navigation



Visu

al h

uman

+mac

hine

le

arni

ng [F

uchs

et a

l. 09

]multi-variate

alternative hypotheses

24


Fluid-structure interactions (multi-model data) heat exchange between fluid structure

feature specification/transfer across data parts [Kehrer et al. 11]


visual data fusion


navigation



feature transfer

cooler aluminum foam

feature

25


Algorithmic extraction of values & patterns dimensionality reduction (PCA, SOM, MDS)

aggregation, summary statistics

clustering, outliers, etc.


visual data fusion


navigation


feature spec.


clustering of multi-run simulations [Bruckner & Möller 10] [Andrienko & Andrienko 11]

multi-run

spatiotemporal

26


Cluster Calendar View [vanWijk & van Selow ’99]

Time series clusteredby similarity (K-means)


visual data fusion


navigation


feature spec.



Categorization of approaches

28


Open Issues

How to deal with data heterogeneity? most approaches only address one or two data facets

coordinated multiple views with linking & brushing

investigation of features across views, data facets, levels of abstraction, and data sets

fusion of heterogeneous data at feature/semantic level

Combination of vis., interaction, and comput. analysis analytical methods can controll steps in visualization pipeline

(e.g., visualization mapping or quality metrics)

interactive feature specification + machine learning

29


Conclusions

Scientific data are becomming multi-faceted

Categorization based on common visualization, interaction, and comput. analysis methods

Promising data facets, e.g., multi-run & multi-model data


visual data fusion




feature spec.


30


Acknowledgements

H. Hauser, H. Schumann, M. Chen, T. Nocke,VisGroup in Bergen, H. Piringer, M.E. Gröller

Supported in part by the Austrian Funding Agency (FFG)

Interactive Visual Analysis (IVA)Enables visual dialogue between user and data

directly interact with data (zoom, filter, select, analyze)drill-down into information(“overview first, zoom and filter, then details on demand” [Shneiderman])interpret complex datafind relations (“read betw. the lines”)detect features / patterns that are difficult to describeintegrate expert knowledge

Johannes Kehrer 0

SimVis Framework for IVAcoordinated, multiple views

linking & brushing

focus+context vis.

degree-of-interest(DOI ∈ [0, 1])on-the-fly data derivation

interactivity, etc.

Johannes Kehrer 1

brushing: mark interesting data subset

updated linked views

move/alter/extend brush

explore data relations

Interactive Brushing

–pressure–»

–TK

E–»

–vel.–»

color: temp.

(SimVis)

[Doleisch et al., ’03]

3 Kehrer et al.

Function Graphs View [Muigg et al. 08]

Visualize large amounts of function graphs (e.g., 500K)Transfer functions [Johansson et al. 05] map line count to pixel‘s luminance

Basic edge enhancement, F+C coloring

Data aggregation (frequency binmaps) [Novotný & Hauser 06]

Visual Exploration of Climate Data

Hypothesis Generation search for potential sensitive & robust indicators for climate changecharacteristic climate signals thatdeviate from natural variabilityuseful to monitor atmospheric change

ECHAM5, B1, temp.

Johannes Kehrer 4

Usual Workflow

Set research focus

Acquire data

Iterateexplore / investigate dataformulate particular hypothesisevaluate with statistics

Challenging to come up with new hypothesesGoal: accelerate process (fast interactive visualization,

more informed partner more directed search)

Johannes Kehrer 5

Integrated data derivation linear trends & signal to noise ratios (SNR)

Interactive visual exploration for quick and flexible data investigation (“preview on statistics”)

Generated hypotheses evaluated using statistics trend testing [Lackner et al. 08]

Narrow down parameters

Our Visual Exploration Process

Johannes Kehrer 6

Localize robust indicatorsareas with high significance

smooth specification

northsouth

Focus on Expressive Data

excludelow |SNR|

strato-sphere

tropo-sphere

+– 7

Explore Trend Variation over Time

+

–

strato-sphere

tropo-sphere

+–

Johannes Kehrer 8

Robust cooling trends

Up to now: investigation in one direction

check relation in other direction

Analyze Relations between Dimensions

similarity based brushing

high SNR

SN

R

northsouth

Pres

sure

leve

ls

function graph

Kehrer et al.

Generated Hypothesis / ECHAM5 temp.

Promising indicator region inlower stratosphere at northern latitudes & tropics.Cooling trend considered robust over investigated time span.

strato-sphere

tropo-sphere

strato-sphere

tropo-sphere

northsouth

northsouth +–

+

–

hypothesis handed over to statistics

10

Hypothesis Generation with Visual ExplorationKehrer et al. Hypothesis generation in climate research with interactive visual data exploration. IEEE TVCG, 14(6):1579–1586, 2008.

Ladstädter et al. SimVis: an interactive visual field exploration tool applied to climate research. In New Horizons in Occultation Research, pages 235–245. Springer, 2009.

Ladstädter et al. Exploration of climate data using interactive visualization. Journal of Atmospheric and Oceanic Technology, 27(4):667–679, 2010.

Johannes Kehrer 11

Single-part scenariosgiven in a coherent form

(spatio-temporal / multi-variate)

Multi-part scenarios2 or more related data partsdifferent simulation modelsdifferent data sourcesvarious data gridsdifferent dimensionality (2D/3D data)

Johannes Kehrer 12

IVA across two Parts of Scientific Data

time-dependent 3D data

coupled climate model

Multi-physics Simulations

Johannes Kehrer 13

cooler aluminum foam

heat exchange

Example: fluid-structure interactions (FSIs)movable or deformable structure fluidflexible roofs, bridges, blood flow in arteries, etc.[Bungartz & Schäfer 2006]

How to relate features across different data parts?

Sample Analysis of Heat Transfer

Johannes Kehrer 14

How is the heating (solid) influenced by surrounding vortical flow?

Transfer vortex feature to solid

feature transfer

Relate grid cells across data parts

Transfer features(DOI values) in both directions

Keep feature specification up to date

IVA across an Interface

Johannes Kehrer 15

data part1 data part2

feature transfer

“Scientific” data:some data values f(p) (e.g., temperature, pressure values)measured/simulated wrt. a domain p(e.g., 2D/3D space, time, simulation input parameters)

If dimensionality of p > 3, then traditional visual analysis is hard

Reducing the data dimensionality can help (e.g., computing stat. aggregates)

Johannes Kehrer 16

Higher-dimensional Scientific Data

3D time-dependent multi-run simulation data

data distribution per cell

Reducing the Data Dimensionality

Statistics: assess distributional characteristics along an independent dimension (e.g., time, spatial axes)

Integrate into IVA through attribute derivation

[from IPCC AR #4, 2007]

average temp. in ten years

17Johannes Kehrer

Case Study: Multi-run climate Data

size: IQR

Cooling event (8200 yrs ago)ocean simulation (2D sections)10 x 10 = 100 runstime-dependent (250 time steps)

Compute statistics wrt. the multiple runs

median, quartiles, etc.billboard glyphs[Lie et al. 2009]

year 100

Johannes Kehrer

Case Study: Multi-run climate Data

size: IQR

Cooling event (8200 yrs ago)ocean simulation (2D sections)10 x 10 = 100 runstime-dependent (250 time steps)

Compute statistics wrt. the multiple runs

median, quartiles, etc.billboard glyphs[Lie et al. 2009]

year 100

q1

q2

q3

Johannes Kehrer

Moment-based Visual Analysis

Get big picture (data trends & outliers)

Multitude of choices, e.g,statistical moments(mean, std. deviation, skewness, kurtosis)traditional and 2 robust estimatescompute relation (e.g., differences, ratio) change scale(e.g., data normalization, log. scaling,measure of “outlyingness”)

How to deal with this “management challenge”?

4×3

×2

×3

= 72 possible configurations per axis

20Johannes Kehrer

right skewed

peakedvs.flat

Moment-based Visual Analysis

quantile plot(focus+context)

multi-run data aggregated data

Iterative view transformationsalter axis/attribute configuration(construct a multitude of informative views)maintain mental model of views

classification of moment-based views

Relate multi-run data aggregated data

Johannes Kehrer

meanmedian

Robust Statistics

Outlier influence traditional estimates

Johannes Kehrer 22

Iterative View TransformationsChange axis/attribute configuration of view

change order of moment robustify moment

compute relation (e.g., difference or ratio)change scale(e.g., normalize, z-standardization)

23

traditionalmed/MAD-based octile-based

Closer related to data tranformations

Johannes Kehrer

change order of moment

study relationsbetw. moments

investigate basiccharacteristicsof distributions

Basic View Setup: Opposing Different Moments

multi-run data aggregated data

quantile plot(focus+context)

1st vs. 2nd moment 3rd vs. 2nd moment

3rd vs. 4th moment

24Johannes Kehrer

Views: Opposing Different Momentsrobustify moment assess influence

of outliers

25

traditionalestimates

robustestimates

Johannes Kehrer

Other View Transformationscompute relation (e.g., difference or ratio)

change scale(e.g., z-standardization,normalize to [0,1])

z-score (measure of outlyingness)

distance to median

26

quantiles of original data

Johannes Kehrer

IVA across two Parts of Scientific DataJ. Kehrer, P. Muigg, H. Doleisch, and H. Hauser. Interactive visual analysis of heterogeneous scientific data across an interface. IEEE TVCG, 17(7):934–946, 2011.

Johannes Kehrer 27

Moment-based Visual AnalysisJ. Kehrer, P. Filzmoser, and H. Hauser. Brushing moments in interactive visual analysis. CGF, 29(3):813–822, 2010.

Johannes Kehrer 28

Conclusions

Johannes Kehrer 29

IVA across 2 data partsrelating multi-run data aggregated statisticsanalyst can work with both parts (e.g., check validity)

Integration of statistical momentstraditional vs. robust statistics, outliersiterative view transformationsinteractive statistical plots (linking & brushing)

Workflow for hypothesis generation

Johannes Kehrer 30

Acknowledgements

Helwig Hauser, VisGroup @ UiB

Helmut Doleisch, Philipp Muigg, Wolfgang Freiler

Florian Ladstädter, Andrea Steiner,Bettina Lackner, Barbara Pirscher,Gottfried Kirchengast

Peter Filzmoser, Andreas Lie, Ove Daae Lampe

Thomas Nocke, Michael Flechsig

Armin Pobitzer, C. Turkay, Stian Eikeland

many others

interactive visual analysis of multi-faceted scientific data · (multiple data attributes, e.g.,...

Documents