Jan Brus - User study for representing the spatial data uncertainty in land cover maps with use of intrinsic and extrinsic methods

www.geoinformatics.upol.cz

User study for representing the spatial data uncertainty in land cover maps with use of

intrinsic and extrinsic methods

Jan BRUS


Question about quality of outputs

• Are the map true?• How about the quality of presented data?• What about the subjectivity?• Will reader know about the positional and other errors

caused by data manipulation?

• Study focused on uncertainty visualisations intuitiveness


Sources of Uncertainty – data quality elements

• Lineage (description of the source material from which the data were derived and the methods of derivation)

• Positional accuracy (resolution of the measurement)• Attribute accuracy (both measurement accuracy and class assignment

accuracy)• Logical consistency (describing the fidelity of relationships inside data

structure)• Completeness (relationship between the objects represented and the

abstract universe)• Currency (time currency, time relevance)• Credibility (reliability of information source, experiences)• Subjectivity (amount of human judgments in the information)• Interrelatedness (source independence)


• Isn’t better to provide geoinformation with some kind of uncertainty?

• Isn‘t maps (geovisualizations) with information about data uncertainty confusing?

• What‘s the right/good way of uncertainty visualization?

• What‘s better in a real decision process?

Is uncertainty visualisation necessary?


Approach in uncertainty visualizations

Examples

Usability testing




• Presenting the data sets with different associated uncertainty

• positional accuracy • attribute accuracy• subjectivity

• How best to represent the data?• How to best reflect reflect the uncertainty?

Representation of Uncertainty


Laboratory setup

• SMI RED 250 eye-tracker with 120 Hz sampling rate• SMI Experiment Center - design of experiment• SMI BeGaze, OGAMA, R software - data analyses

• remote eye tracker most practical method of ET• illuminator/eye camera module placed below line of

sight• all participants were recorded and have to speak during

testing• evaluation of right and wrong answers was based on

post processing of recorded video


Study and experiments

• the aim of our study was to evaluate the effect of uncertainty visualisations on eye movements and performance in map-related tasks

• the study involved decision making questions where the participants were presented with several uncertainty visualisation methods based on intrinsic and extrinsic methods

• finding areas with the least or most uncertainty of selected land cover class – based on intuitiveness

• additive factor of the study also compared user performance with and without the use of the legend


Testing details

• user perception of uncertainty visualizations derived from photointerpretation of land cover classes

• maps without legend – intuitiveness of uncertainty methods• 14 participants – 8 uncertainty methods as stimuli• dependent variables were represented by following metrics

derived from the analysis of eye-tracking data:– fixation duration– number of fixation– fixation count, saccade count– and more


• Land Facet Corridor Tools for ArcGIS

Delineation of uncertainty – entropy approach

• can be used for each map layer

• combination of entropies

• showing most uncertain

• map algebra

(Wellmann and Regenauer-Lieb, 2012


Entropy calculation

• concept of entropy was applied to landcover classes


Methods exampleshillshade - positivehillshade - negativeglyphstransparent dots - sizetransparent gridgrid - width of linetransparencyquadtree


Stimuli setup


Stimuli setup


Results

hillshad

e - positi

ve

hillshad

e - neg

ative

glyphs

transp

arent d

ots

transp

arent g

rid

grid - w

idth of line

transp

arency

quadtre

e

8

6

4

9

12

7

98

6

8

10

5

2

7

56

right answer wrong answer


Results

hillshad

e - positi

ve

hillshad

e - neg

ative

glyphs

transp

arent d

ots

transp

arent g

rid

grid - w

idth of line

transp

arency

quadtre

e0

50

100

150

200

250

300

350

Fixation Lenght

MeanMedian


Results

hillshade - positivehillshade - negative glyphs transparent dots transparent grid grid - width of line transparency quadtree0

10

20

30

40

50

60

70

Number of Fixations


Results


Results• As a top rated when compared to all methods and metrics

have been examined methods:• transparent grid • transparent circles

• problem with implementation these methods• quantification of uncertainty based on blur or transparency• Semantic Depth of Field (Kosara, 2011)

• partly method grid - width of line• and quadtree method


Problems

• small amount of respondents • respondents not domain experts• very specific task – can be domain depended• difficult visualisation methods• same area (rotated and fliped)• target group mostly cartographers and geoinformatics

professionals• not statistically proved• long interpretation of results from recorded video


Conclusion and future

• in our study we try to capture the uncertainty visualisation connected with land cover classes

• study focused more on uncertainty visualisations methods• this should bring more adequate results to uncertainty

visualisation community• it is clear that uncertainty visualizations will have great

importance in optimization of cartographic products and presenting geographic data in the future

• comparison of different uncertainty visualization methods• proofing and confirming results from the past research


Conclusion and future

• same method for different studies• more respondents• combination of different spatial quality components in one

visualisation


Thank you for your

attention…

[email protected]

The presentation has been completed within the project CZ.1.07/2.2.00/28.0078 “InDOG” which is co-financed from European Social Fund and State financial resources of the Czech Republic.

Jan Brus - User study for representing the spatial data uncertainty in land cover maps with use of intrinsic and extrinsic methods

Education

stimuli setup

uncertainty

data

www

methods

results

quality

information