Sunteți pe pagina 1din 93

UNCLASSIFIED

Final Report Technical Annex


AGIS Task 5 Visualising Uncertainty

© Envitia Ltd. (2013)

MOD Contract No: DSTLX-1000063699

Report Number: AGIS_SD014-02

Authors: Dr. Ed Figura (Envitia), Dr. Adam Rousell (Nottingham


University), Mr Adam Chilton (Envitia), Dr. Gobe Hobona
(Envitia), Prof. Mike Jackson (Nottingham University)

Date of Publication: 21 June 2013

CONDITIONS OF SUPPLY – FULL RIGHTS

This document is supplied in confidence to MOD in accordance with Contract No DSTLX1000063699. The
document comprises information proprietary to Envitia Ltd and whose unauthorised disclosure would cause
damage to the interests of Envitia Ltd.

The document is supplied to MOD as a FULL RIGHTS VERSION under the terms of DEFCON 705 (Edn 11/02)
and, except with the prior written permission of Envitia Ltd, MOD’s rights of use and dissemination in the
document are limited to those set out in Clause 12 of DEFCON 705 (Edn 11/02) for the use of Full Rights
Versions of Technical Deliverables.

Requests for permission for wider use or dissemination should be made to Envitia Ltd.

THE INVESTIGATION WHICH IS THE SUBJECT OF THIS REPORT WAS INITIATED BY THE
PROGRAMME LEADER – ISTAR, UNDER THE ISTAR AND SENSORS DOMAIN, DSTL AND
WAS CARRIED OUT UNDER THE TERMS OF CONTRACT NO DSTLX-1000063699

UNCLASSIFIED
UNCLASSIFIED Amendment Record

Amendment Record
Issue Pages Description of Change Date

01 93 Initial versions 31/05/2013

02 Changes after review 21/06/2013

ALL RECIPIENTS OF THIS REPORT ARE ADVISED THAT IT MUST NOT BE COPIED IN
WHOLE OR IN PART OR BE GIVEN FURTHER DISTRIBUTION OUTSIDE THE AUTHORITY
WITHOUT THE WRITTEN APPROVAL OF THE PROGRAMME LEADER – ISTAR, UNDER THE
ISTAR AND SENSORS DOMAIN, DSTL RESEARCH PROGRAMME DSTL, PORTON DOWN,
SALISBURY WILTS SP4 0JQ

AGIS_SD014-02 UNCLASSIFIED Page 2 of 93


UNCLASSIFIED List of Contents

List of Contents
A. Appendix A: Military Uncertainty Visualisations........................................................... 8
A.1 APP-6................................................................................................................................ 8
A.2 Canadian Recognised Maritime Picture (RMP) Uncertainty Visualisations ........................ 8
A.2.1 Lego Design ...................................................................................................................... 8
A.2.2 Rectangle Design .............................................................................................................. 9
B. Appendix B: Analysis of Visualisation Techniques .................................................... 10
B.1 Colour ............................................................................................................................. 10
B.1.1 Colour Hue ...................................................................................................................... 10
B.1.2 Intensity........................................................................................................................... 10
B.1.3 Saturation........................................................................................................................ 11
B.1.4 Traffic Light Colours ........................................................................................................ 12
B.2 Visual Variables .............................................................................................................. 13
B.2.1 Transparency .................................................................................................................. 13
B.2.2 Line Width ....................................................................................................................... 13
B.2.3 Size ................................................................................................................................. 14
B.2.4 Shapes ............................................................................................................................ 14
B.2.5 Location .......................................................................................................................... 15
B.2.6 Feature Fuzziness / Focus .............................................................................................. 15
B.2.7 Grain ............................................................................................................................... 16
B.2.8 Orientation ...................................................................................................................... 16
B.2.9 X-Hatch Density .............................................................................................................. 17
B.2.10 Sketchiness ..................................................................................................................... 17
B.2.11 Resolution ....................................................................................................................... 18
B.2.12 Jitter ................................................................................................................................ 18
B.3 Additions ......................................................................................................................... 19
B.3.1 Buffers............................................................................................................................. 19
B.3.2 Confidence Buffers .......................................................................................................... 19
B.3.3 Stroke Buffers ................................................................................................................. 19
B.3.4 Fuzzy Buffers .................................................................................................................. 20
B.3.5 Analyst-Defined Area ...................................................................................................... 21
B.3.6 Fuzzy Area ...................................................................................................................... 21
B.3.7 Star Bursts ...................................................................................................................... 22
B.3.8 Displacement .................................................................................................................. 22
B.3.9 Bubbles ........................................................................................................................... 23
B.4 Temporal ......................................................................................................................... 23
B.4.1 Fading Features .............................................................................................................. 23
B.4.2 Expanding Features ........................................................................................................ 24
B.4.3 Flashing .......................................................................................................................... 24
B.4.4 Colour Legend................................................................................................................. 24
B.4.5 Moving Objects ............................................................................................................... 25
B.5 Stereotypes ..................................................................................................................... 25
B.5.1 Traffic Lights.................................................................................................................... 26
B.5.2 Emoticons ....................................................................................................................... 29
B.6 Labels ............................................................................................................................. 30
B.6.1 Textual Labels ................................................................................................................. 30
B.6.2 GetFeatureInfo ................................................................................................................ 30
B.6.3 Pop-Ups .......................................................................................................................... 30
B.7 Visualisations requiring GIS ............................................................................................ 30
B.7.1 3D Extrusion.................................................................................................................... 30
B.7.2 Heat Maps ....................................................................................................................... 31
B.7.3 Clusters ........................................................................................................................... 31
B.7.4 Hexagon Grids ................................................................................................................ 32

AGIS_SD014-02 UNCLASSIFIED Page 3 of 93


UNCLASSIFIED List of Tables

B.7.5 Contouring / Isopleth Maps.............................................................................................. 33


B.7.6 Graphs ............................................................................................................................ 33
B.7.7 3D Threat Domes ............................................................................................................ 34
C. Appendix C: SLD Structure .......................................................................................... 35
C.1 SLD Header/Namespaces ............................................................................................... 35
C.2 SLD Information .............................................................................................................. 35
C.3 Styling Information........................................................................................................... 36
C.3.1 Basic Styling.................................................................................................................... 36
C.3.2 Advanced SLD styling ..................................................................................................... 37
C.3.3 Closing Elements of the SLD file ..................................................................................... 45
D. Appendix D: Example SLD ........................................................................................... 46
E. Appendix E: Uncertainty Information for KML ............................................................ 48
F. Appendix F: Metrication of Visualisation Techniques ................................................ 52
F.1 Introduction ..................................................................................................................... 52
F.2 Methodology.................................................................................................................... 52
F.2.1 Identification of Uncertainty Types ................................................................................... 54
F.3 Credibility ........................................................................................................................ 56
F.3.1 Identification and Creation of Visualisations .................................................................... 57
F.3.2 Experimental Design ....................................................................................................... 62
F.3.3 DGC Assessment Workshop ........................................................................................... 62
F.3.5 Test Results .................................................................................................................... 69
F.3.6 Summary of Results ........................................................................................................ 72
F.4 Summary & Conclusions ................................................................................................. 73
F.5 Further Investigations ...................................................................................................... 76
G. Appendix G: Statistical Testing.................................................................................... 77
H. Appendix H: Experiment Data and Analysis ............................................................... 80
H.1 Friedman Test Results .................................................................................................... 81
H.2 Mann-Whitney Test Results ............................................................................................ 81
H.2.1 Accuracy in Point Features.............................................................................................. 82
H.2.2 Accuracy in Line Features ............................................................................................... 83
H.2.3 Accuracy in Polygon Features ......................................................................................... 84
H.2.4 Precision in Line Features ............................................................................................... 85
H.2.5 Precision in Polygon Features ......................................................................................... 86
H.2.6 Credibility in Point Features............................................................................................. 87
H.2.7 Credibility in Polygon Features ........................................................................................ 88
H.2.8 Vagueness in Point Features .......................................................................................... 89
H.2.9 Vagueness in Polygon Features ...................................................................................... 90
I. Appendix I: Colour Ramp Statistics ............................................................................. 91
I.1 Friedman Test ................................................................................................................. 91
I.2 Mann-Whitney Test ......................................................................................................... 91
J. List of References ......................................................................................................... 92
K. Abbreviations and Acronyms ....................................................................................... 93

List of Tables
Table 1: Visualisation generation tools .......................................................................................... 58
Table 2: Topologically consistent and inconsistent representations .............................................. 59
Table 3: Examples of features using different visualisation methods ............................................. 62

AGIS_SD014-02 UNCLASSIFIED Page 4 of 93


UNCLASSIFIED List of Figures

Table 4: Visualisations identified in the DGC workshop ................................................................ 63


Table 5: Visualisation count against feature type .......................................................................... 64
Table 6: Visualisation count against uncertainty type .................................................................... 65
Table 7: Participant organisations ................................................................................................. 67
Table 8: Response ranks by uncertainty and feature type ............................................................. 68
Table 9: Results of experiment...................................................................................................... 73
Table 10: Summary of preferences ............................................................................................... 74
Table 11: Raw data from Visual Variables experiment .................................................................. 80
Table 12: Visual variables Friedman test results ........................................................................... 81
Table 13: Mann-Whitney test results for techniques showing accuracy in point features ............... 82
Table 14: Mann-Whitney test results for techniques showing accuracy in line features ................. 83
Table 15: Mann-Whitney test results for techniques showing accuracy in polygon features .......... 84
Table 16: Mann-Whitney test results for techniques showing precision in line features ................. 85
Table 17: Mann-Whitney test results for techniques showing precision in polygon features .......... 86
Table 18: Mann-Whitney test results for techniques showing credibility in point features .............. 87
Table 19: Mann-Whitney test results for techniques showing credibility in polygon features ......... 88
Table 20: Mann-Whitney test results for techniques showing vagueness in point features ............ 89
Table 21: Mann-Whitney test results for techniques showing vagueness in polygon features ....... 90
Table 22: Mann-Whitney test results for colour ramp preference .................................................. 91

List of Figures
Figure 1: APP-6 symbols for uncertainty ......................................................................................... 8
Figure 2: RMP uncertainty visualisations ........................................................................................ 9
Figure 3: Colour hue change for showing uncertainty as a map overlay ........................................ 10
Figure 4: Intensity change for showing uncertainty as a map overlay ............................................ 10
Figure 5: Two saturation examples showing an uncertainty range for map overlays ..................... 11
Figure 6: Uncertainty visualisation for line features using the saturation technique ....................... 11
Figure 7: Traffic light colours to represent uncertainty for point features ....................................... 12
Figure 8: Transparency change for showing uncertainty as a map overlay ................................... 13
Figure 9: Line width visualisation to portraying uncertainty in line features .................................... 14
Figure 10: Changes in shape to portray different uncertainty values ............................................. 14
Figure 11: Location iconographic symbology for uncertainty portrayal of a point feature ............... 15
Figure 12: Fuzziness / Focus visualisation for uncertainty portrayal of polygon features ............... 15
Figure 13: Grain fill; density changes for showing uncertainty as a map overlay ........................... 16
Figure 14: Orientation fill; map overlay where line direction indicates uncertainty values .............. 17
Figure 15: X-hatch density fill; density changes for showing uncertainty as a map overlay ........... 17
Figure 16: Sketchiness visualisation for uncertainty portrayal of polygons .................................... 18
Figure 17: Resolution visualisation for uncertainty portrayal of polygons ....................................... 18
Figure 18: Jitter visualisation for uncertainty portrayal of polygons ................................................ 19
Figure 19: Stroke buffers visualisation for uncertainty portrayal of point features .......................... 20
Figure 20: Fuzzy buffers visualisation for uncertainty portrayal of point features ........................... 20
Figure 21: Analyst-defined area for indicating uncertainty of a point location ................................ 21
Figure 22: Fuzzy area for indicating uncertainty of a point location ............................................... 21
Figure 23: Star bursts visualisation to show the extent and most uncertain locations .................... 22
Figure 24: Displacement visualisation to show uncertainty through the change in location ........... 22
Figure 25: Bubbles experimental visualisation to imply uncertainty as a map overlay ................... 23
Figure 26: Fading features to portray increasing uncertainty of overlain features .......................... 23
Figure 27: Diagram showing how uncertainty increases into the future ......................................... 24
Figure 28: Expanding features for prediction into the future .......................................................... 24
Figure 29: Star rating visualisation to indicate levels of uncertainty in point features ..................... 25
Figure 30: Alternative single star symbology ................................................................................. 26
Figure 31: Tabular star metaphor for visualising a datasets total uncertainty ................................ 26
Figure 32: Traffic light symbology in food packaging labels ........................................................... 27

AGIS_SD014-02 UNCLASSIFIED Page 5 of 93


UNCLASSIFIED List of Figures

Figure 33: Traffic light visualisations to indicate levels of uncertainty in point features .................. 27
Figure 34: Traffic light metaphor for visualising a datasets total uncertainty .................................. 28
Figure 35: Alternative five point traffic light scale........................................................................... 28
Figure 36: Emoticon visualisation indicating uncertainty in point features ..................................... 29
Figure 37: GetFeatureInfo portraying feature information including uncertainty values ................. 30
Figure 38: Code required for 3D extrusion using GeoServer & Google Earth ................................ 30
Figure 39: 3D extrusion, where increasing height indicates greater uncertainty ............................ 31
Figure 40: Heat map where the red (hotter) colours indicate higher uncertainty ............................ 31
Figure 41: Cluster technique showing an overview of a high density point dataset........................ 32
Figure 42: Hexagon grid technique where higher opaqueness represents higher uncertainty ....... 32
Figure 43: Isopleth map as an overlay for contouring uncertainty values. ..................................... 33
Figure 44: Graphs within the polygon to show uncertainty values graphically ............................... 33
Figure 45: ESRI ArcGIS chart symbology ..................................................................................... 33
Figure 46: 3D threat dome utilising COLLADA .............................................................................. 34
Figure 47: SLD header used in every SLD .................................................................................... 35
Figure 48: SLD information section ............................................................................................... 35
Figure 49: Basic point SLD styling rule .......................................................................................... 36
Figure 50: Basic polygon SLD styling rule ..................................................................................... 36
Figure 51: Basic line SLD styling rule ............................................................................................ 37
Figure 52: Basic raster SLD styling ............................................................................................... 37
Figure 53: Transparency SLD parameters .................................................................................... 37
Figure 54: Dash array for creating dashed lines in SLDs............................................................... 37
Figure 55: SLD text symbology options ......................................................................................... 38
Figure 56: SLD links to image files ................................................................................................ 39
Figure 57: SLD point shapes ......................................................................................................... 39
Figure 58: X-hatch styling (suitable for X-hatch fills)...................................................................... 39
Figure 59: Grain styling (suitable for hatch fills) ............................................................................. 40
Figure 60: Orientation styling (suitable for polygon fills) ................................................................ 40
Figure 61: Equal To SLD filter function.......................................................................................... 40
Figure 62: Greater Than SLD filter function ................................................................................... 41
Figure 63: Less Than SLD filter function ....................................................................................... 41
Figure 64: Greater Than Or Equal To SLD filter function ............................................................... 41
Figure 65: Less Than Or Equal To SLD filter function ................................................................... 41
Figure 66: Multiple SLD filter functions .......................................................................................... 42
Figure 67: SLD scale filtering maximum and minimum scale denomination .................................. 42
Figure 68: SLD interpolation active function - colour ..................................................................... 43
Figure 69: SLD Interpolation active function – numeric ................................................................. 43
Figure 70: SLD recode active function .......................................................................................... 44
Figure 71: SLD categorize active function ..................................................................................... 44
Figure 72: Arithmetic operators within SLD styling ........................................................................ 45
Figure 73: Specifying drawing units for scale specific SLD styling ................................................. 45
Figure 74: Closing elements of SLD .............................................................................................. 45
Figure 75: Basic method for exposing uncertainty information in KML .......................................... 48
Figure 76: Using text styling to improve the uncertainty information visualisation .......................... 49
Figure 77: KML schema included with uncertainty information ...................................................... 50
Figure 78: Uncertainty information accompanying the feature but not processed by Google Earth 51
Figure 79: Feature datasets .......................................................................................................... 53
Figure 80: GPS trace .................................................................................................................... 54
Figure 81: Spatial precision ........................................................................................................... 55
Figure 82: Spatial vagueness; feature extent (top) and overlapping features (bottom) .................. 56
Figure 83: Topologically consistent and inconsistent representations ........................................... 59
Figure 84: Continuous vs. discrete classification ........................................................................... 60
Figure 85: Program flow ................................................................................................................ 66
Figure 86: Main assessment window ............................................................................................ 66
Figure 87: Equality of preference example .................................................................................... 70

AGIS_SD014-02 UNCLASSIFIED Page 6 of 93


UNCLASSIFIED List of Figures

Figure 88: Accuracy in line features preference ............................................................................ 70


Figure 89: Accuracy in polygon features preference ..................................................................... 71
Figure 90: Precision in line features preference ............................................................................ 71
Figure 91: Precision in polygon features preference ..................................................................... 71
Figure 92: Vagueness in point features preference ....................................................................... 71
Figure 93: Vagueness in polygon features preference .................................................................. 72
Figure 94: Credibility in polygon features preference .................................................................... 72
Figure 95: Consistently ranked values .......................................................................................... 78
Figure 96: Randomly ranked values .............................................................................................. 78
Figure 97: Box plots for techniques showing accuracy in point features ........................................ 82
Figure 98: Box plots for techniques showing accuracy in line features .......................................... 83
Figure 99: Box plots for techniques showing accuracy in polygon features ................................... 84
Figure 100: Box plots for techniques showing precision in line features ........................................ 85
Figure 101: Box plots for techniques showing precision in polygon features ................................. 86
Figure 102: Box plots for techniques showing credibility in point features ..................................... 87
Figure 103: Box plots for techniques showing credibility in polygon features................................. 88
Figure 104: Box plots for techniques showing vagueness in point features ................................... 89
Figure 105: Box plots for techniques showing vagueness in polygon features .............................. 90
Figure 106: Box plots of the colour ramps against ranking by survey participants ......................... 91

AGIS_SD014-02 UNCLASSIFIED Page 7 of 93


UNCLASSIFIED Technical Annex

A. Appendix A: Military Uncertainty Visualisations


A.1 APP-6

In the current APP-6 symbology set there are a number of dynamic symbols for showing
uncertainty; the next revision currently in draft does not make changes to these symbols.

a. b. c. d.

Figure 1: APP-6 symbols for uncertainty1

The four symbols shown in Figure 1 are Area of Uncertainty (AOU) indicators. These represent the
area where an object is most likely to be, based upon the last report of the object’s location and the
reporting accuracy of the sensor which detected the object. AOU indicators can be displayed as an
ellipse, bounding box or a line of bearing.

The centre of the ellipse and bounding box AOU indicator (Figure 1 a. & b.) is the last reported
position for the object. The symbol for the object is shown in the centre and the ellipse or bounding
box is depicted as a solid line whose parameters are based upon the attribute of the sensor
detecting the object.

The line of bearing AOU (Figure 1 c. & d.) is a solid line with the rotation of the line representing
the bearing of the object and the length of the line represents the range estimate. The indicator can
be shown as a single solid centre line as in Figure 1 c. but can also be drawn with dotted bearing
error or ‘V’ lines determined from the bearing error as shown in Figure 1 d.

A.2 Canadian Recognised Maritime Picture (RMP) Uncertainty Visualisations

The visualisations described in this section are those developed by the Defence Research and
Development Canada (DRDC). The document produced by Matthews et al. (2008) examined two
visualisations for uncertainty in the maritime domain shown in Figure 2.

A.2.1 Lego Design

The Lego® design represents information along three axes: identity of an object, spatial
uncertainty and how temporally relevant the data is. This is represented by the horizontal and
vertical rectangles filled with a maximum of three squares as well as the colour of those squares.

1
Source: Department of Defence (2008)

AGIS_SD014-02 UNCLASSIFIED Page 8 of 93


UNCLASSIFIED Technical Annex

Spatial uncertainty is represented along the horizontal axis with an increasing number of squares
indicating a greater level of uncertainty. A single square indicates a precise position, two squares
indicate an approximate position and three squares indicate an unknown position.

Temporal information is represented along the vertical axis with uncertainty increasing with the
number of squares. One square represents current information, two squares represent old
information and three squares represent out-of-date information.

Colour represents identity information where green is good, yellow indicates poor identity and
orange is no identity information about the object.

A.2.2 Rectangle Design

The rectangle design represents the same three axes as the Lego design but represent these
through the colour of the rectangle, the size of the circle and the position of the circle.

Spatial uncertainty is represented by a black filled circle, where the size of the circle represents the
precision of the information available for the object. The smallest circle represents a precise
location, the intermediate size represents an approximate position and the largest circle indicates
an unknown spatial position.

Temporal information is represented by the position of the circle within the rectangle. Current
information is presented as the circle being in the leftmost position, old information is shown
through a circle positioned in the middle and out-of-date information is shown through by circle
being in the rightmost position.

Colour represents identity information where green is good, yellow indicates poor identity and
orange is no identity information about the object.

Figure 2: RMP uncertainty visualisations2

2
Source: Matthews et al. (2008)

AGIS_SD014-02 UNCLASSIFIED Page 9 of 93


UNCLASSIFIED Technical Annex

B. Appendix B: Analysis of Visualisation Techniques


A number of the visualisations presented in this section are created using SLD technology. For
these cases, the accompanying makes reference to aspects of SLD styling – a discussion of the
technical details of SLDs can be found in Section C.

B.1 Colour

B.1.1 Colour Hue

Colour hue represents a changing of colour. In the Task 5 research, several colour ramps of
changing colour hue were presented (Figure 3). To create this visualisation using SLD technology,
either categorisation (SLD code shown in Figure 71) or multiple filters (SLD code shown in Figure
66) were used to create a categorized view by choosing the colours for each category.
Alternatively, interpolation (SLD code shown in Figure 68) can be used to create a continuous
scale setting colours against values at either end of the colour ramp, for example, green = 0 (no
uncertainty) and red = 100 (complete uncertainty).

Figure 3: Colour hue change for showing uncertainty as a map overlay

B.1.2 Intensity

Intensity is much like colour hue except that one end of the colour ramp is fixed as white and
therefore the colour becomes lighter and lighter with increasing uncertainty. To create this
visualisation using SLD technology, either categorisation (SLD code shown in Figure 71) or
multiple filters (SLD code shown in Figure 66) were used to create a categorized view by choosing
the colours for each category and changing the RGB values towards the white end of the
spectrum. Alternatively for a continuous scale, one can specify the colours at either end of the
scale, for example, white = 100 (high uncertainty) and green = 0 (low uncertainty) and the SLD
system will perform the interpolation (SLD code shown in Figure 68).

Figure 4: Intensity change for showing uncertainty as a map overlay

AGIS_SD014-02 UNCLASSIFIED Page 10 of 93


UNCLASSIFIED Technical Annex

B.1.3 Saturation

Colour saturation refers to the intensity of a colour where high saturation produces the purest
colour and low saturation produces a dark or grey colour. The more vivid or clearer the colour the
more certain or less uncertain the information being portrayed is. In Figure 5 two examples of this
are shown and the most vivid red and blue areas are regions of low uncertainty.

Figure 5: Two saturation examples showing an uncertainty range for map overlays

Saturation can also be applied to lines and points. However, it is difficult to distinguish the changes
in the saturation when using line features on a complex and fully saturated map background as
shown in Figure 6.

Figure 6: Uncertainty visualisation for line features using the saturation technique

To create this visualisation using SLD technology, either categorisation (SLD code shown in Figure
71) or multiple filters (SLD code shown in Figure 66) were used to create a categorized view by
choosing the colours for each category and changing the RGB values towards the black end of the
spectrum. Alternatively for a continuous scale, one can specify the colours at either end of the
scale, for example, black = 100 (high uncertainty) and blue = 0 (low uncertainty) and the SLD
system will perform the interpolation (SLD code shown in Figure 68).

AGIS_SD014-02 UNCLASSIFIED Page 11 of 93


UNCLASSIFIED Technical Annex

B.1.4 Traffic Light Colours

Traffic light colours are based upon the traffic light stereotype explained in Section B.5.1. This
involves colouring points, lines, polygons or raster features by categorising the data into three
uncertainty levels: high (red), medium (amber) and low (green).

Figure 7: Traffic light colours to represent uncertainty for point features

To create this visualisation using SLD technology, either use the active function called ‘categories’
(SLD code shown in Figure 71) on the ‘fill’ attribute of the styling or else use multiple filters (SLD
code shown in Figure 66). Then specify the colour codes for red, amber and green per category or
filter.

AGIS_SD014-02 UNCLASSIFIED Page 12 of 93


UNCLASSIFIED Technical Annex

B.2 Visual Variables

This section describes the visual variables visualisations presented in Section 5.2.2 of the Task 5
Final Report. An advantage of these techniques is that they can be combined with colour variables,
for example to show a thematic attribute, perhaps the attribute about which the uncertainty is being
shown.

B.2.1 Transparency

Transparency refers to how much of the background can be seen through the feature. If a feature
is highly transparent, then much of the background can be seen and the feature itself becomes
more difficult to make out. If a feature has low transparency then it tends to become opaque and so
the textures and features behind it cannot be seen. In computer graphics, this is often referred to
as the ‘alpha’ component of the colour.

Figure 8: Transparency change for showing uncertainty as a map overlay3

To create the transparency visualisation using SLD technology, either categorisation (SLD code
shown in Figure 71) or multiple filters (SLD code shown in Figure 66) were used to create a
categorized view choosing a ‘fill-opacity’ value for each category. Alternatively, use interpolation
(SLD code shown in Figure 69) for a continuous scale by setting the values of the ‘fill-opacity’
attribute based on uncertainty values. For example, for low uncertainty, a value of 0 would be used
(opaque); for high uncertainty a value of 1 would be used (or vice-versa if one wanted to highlight
low uncertainty).

B.2.2 Line Width

The original paradigm for line width was to use wider lines to represent high uncertainty in spatial,
temporal or thematic attributes. However, after discussion with stakeholders, the conclusion was
that, because wider lines stand out more clearly against typical map backgrounds, these were
more representative of certainty, rather than uncertainty, of features.

3
In this example a red hexagon grid was used as a map overlay, portraying uncertainty.

AGIS_SD014-02 UNCLASSIFIED Page 13 of 93


UNCLASSIFIED Technical Annex

Figure 9: Line width visualisation to portraying uncertainty in line features

To create the line width visualisation using SLD technology, either categorisation (SLD code shown
in Figure 71) or multiple filters (SLD code shown in Figure 66) were used to form a categorised
styling range by choosing the ‘stroke-width’ values for each category or filter. Alternatively for a
continuous scale, interpolation (SLD code shown in Figure 69) was used by setting a value for
‘stroke-width’ against values at both ends of the uncertainty scale. This is a numerical scale and
therefore for high uncertainty, a value of 1 was used to render a thin line/stroke and for low
uncertainty, a value of 100 was used to render thick lines. The SLD engine automatically linearly
interpolates the uncertainty to produce appropriate values for the ‘stroke-width’ attribute and hence
rendered line thickness.

B.2.3 Size

When first developed, smaller sized objects were considered to have lowest uncertainty. However,
after discussion with stakeholders and comments from some participants in the contextual
composite experiment, this was not always deemed intuitive, with disagreements over whether
large or small should be high uncertainty.

To create the size visualisation for points using SLD technology, either categorisation (SLD code
shown in Figure 71) or multiple filters (SLD code shown in Figure 66) were used to form a
categorized styling range by choosing the ‘size’ values for each category or filter. Alternatively for a
continuous scale, interpolation (SLD code shown in Figure 69) was used by setting a value for
‘size’ for values at both ends of the uncertainty scale. For example, for low uncertainty, a value of 1
(small size) would be used for the ‘size’ attribute; for high uncertainty, a larger value (e.g. 5) would
be used. For polygons and line features, the ‘stroke-width’ rather than ‘size’ attribute should be
changed.

B.2.4 Shapes

In this study the shape variable has only been applied to point features. Different classifications of
uncertainty have been depicted as a different shape (e.g. square, circle or triangle).

Figure 10: Changes in shape to portray different uncertainty values

AGIS_SD014-02 UNCLASSIFIED Page 14 of 93


UNCLASSIFIED Technical Annex

To create the shape visualisation using SLD technology use multiple filters (SLD code shown in
Figure 66) combined with shape symbols (SLD code shown in Figure 57) or linked images (SLD
code shown in Figure 56) for the point symbol.

B.2.5 Location

Location is a symbol-based visualisation technique founded in research by MacEachren at al.


(2012) where a dot in the centre of the diagram is moved off-centre by varying amounts (see
Figure 11.

Figure 11: Location iconographic symbology for uncertainty portrayal of a point feature4

This is an immature technique which could be used in one of two ways: (a) the placement of the
dot could be purely symbolic in that the distance of it from the centre of the cross-hairs is
proportional to uncertainty but its placement has no meaning; (b) the placement represents
information about the directionality of the uncertainty.

Currently there is not an SLD function to create this symbology. However, the symbols could be
created in other software and exported as PNG images to be used in an SLD with multiple filters
(SLD code shown in Figure 66) in conjunction with the linked images function (SLD code shown in
Figure 56).

B.2.6 Feature Fuzziness / Focus

The fuzziness (or focus) refers to how sharp the feature appears. If a feature has a high amount of
uncertainty it appears blurred and unfocussed. On the other hand, if it has a low amount of
uncertainty then it appears sharp and well defined. The visualisation presented in Figure 12 was
created using photo editing software5,6 and the software called Processing7. The point symbols
however, could be used with SLD technology through using PNG images of the fuzzy points and
linking these into the SLD using the code shown in Figure 56. To create the categories, multiple
rules and filter functions would be required as shown in Figure 66.

Figure 12: Fuzziness / Focus visualisation for uncertainty portrayal of polygon features

4
Image source: MacEachren et al. (2012)
5
http://www.getpaint.net/index.html
6
http://www.gimp.org/
7
http://www.processing.org/

AGIS_SD014-02 UNCLASSIFIED Page 15 of 93


UNCLASSIFIED Technical Annex

B.2.7 Grain

Grain in this project refers to either dashed lines or hatch filled polygons. In the case of lines,
higher uncertainty values are represented by drawing those lines using dots with large gaps whilst
low uncertainty values are drawn as solid lines. High uncertainty polygons are drawn with a hatch
where the lines are far apart and low uncertainty polygons have the lines much closer together.

To create the polygon grain visualisation using SLD technology, use the grain styling shown in
Figure 59 or any one of the variables shown in Figure 60. Then use either categorisation (Figure
71) or multiple filters (Figure 66) for a categorized view, manually choosing the ‘size’ values for
each category. Alternatively for a continuous scale, interpolation (SLD code shown in Figure 69)
was used by setting a value for ‘size’ for values at both ends of the uncertainty scale. For example,
for low uncertainty, a value of 1 (small size) would be used; for high uncertainty, a larger value
(e.g. 10) would be used.

To create the line grain visualisation using SLD technology use a dash-array shown in (Figure 54)
conjunction with multiple filters (Figure 66) to present a categorised view.

Figure 13: Grain fill; density changes for showing uncertainty as a map overlay

B.2.8 Orientation

Orientation relates to the direction of the filler lines within a polygon. This technique was sourced
from MacEachren et al. (2012). To create this visualisation using SLDs a technique involving the
use of multiple filters (SLD example shown in Figure 66) and the same technique for the polygon
fill as explained for Grain (Section B.2.7 with SLD code shown in Figure 59). The grain styling is
then altered using the SLD code variations shown in Figure 60. Unfortunately, only four
orientations are possible. This visualisation was not used in the experiments as it is not intuitive
and would always require a legend to enable a judgement to be made.

AGIS_SD014-02 UNCLASSIFIED Page 16 of 93


UNCLASSIFIED Technical Annex

Figure 14: Orientation fill; map overlay where line direction indicates uncertainty values

B.2.9 X-Hatch Density

X-hatch density is a variation on the grain visualisation using cross-hatching to increase the
amount of colour shown in areas of low uncertainty. This aims to create a feeling of completeness
(and therefore certainty) through seeing almost block colour in areas of low uncertainty compared
to the largely white areas for high uncertainty features or regions.

Figure 15: X-hatch density fill; density changes for showing uncertainty as a map overlay

To create the X-hatch density visualisation using SLD technology, use the X-hatch styling shown in
Figure 58. Then either the active function called categorisation (SLD code shown in Figure 71) or
multiple filters (SLD code shown in Figure 66) were used to create a categorized view by selecting
the ‘size’ values for each category or filter. Alternatively, for a continuous scale of X-hatch density
the active function called ‘interpolation’ can be used (SLD code shown in Figure 69), setting the
‘size’ parameter based on the values at both ends of the uncertainty scale. For example, for low
uncertainty, a value of 5 for ‘size’ produces tightly hatched fills and for high uncertainty, a value
such as 55 produces low density hatching.

B.2.10 Sketchiness

Sketchiness is an abstract concept that refers to how ‘neatly’ a feature is drawn and has been
presented in studies by Boukhelifa et al. (2012) and Wood et al. (2012). If a feature has a high
level of uncertainty, then the features are drawn in a highly sketched manner as if they were
created quickly by a child. On the other hand, a more certain feature would be drawn with a lower
level of sketchiness, similar to that of a drawing made by a professional artist. In a similar way to
feature fuzziness/focus, the point images can be saved as PNG files and used in an SLD via the

AGIS_SD014-02 UNCLASSIFIED Page 17 of 93


UNCLASSIFIED Technical Annex

linked images SLD code shown in Figure 56. For more extensive drawing capabilities, download
the Handy library8.

Figure 16: Sketchiness visualisation for uncertainty portrayal of polygons

B.2.11 Resolution

The resolution of a feature refers to how large the pixels are that are used to draw it. Features with
a high value of uncertainty will be drawn coarser (the pixels will be larger) than those that have a
low level of uncertainty.

Figure 17: Resolution visualisation for uncertainty portrayal of polygons

This visualisation was created using photo editing software5, 6 in combination with the processing14
software. The point symbols however, can be exported as PNG images to be used in an SLD with
multiple filters (SLD code shown in Figure 66) and linked into the SLD through the SLD code
shown in Figure 56).

B.2.12 Jitter

The term Jitter refers to a distortion placed on a feature that looks similar to a ripple effect. It shifts
pixels horizontally by a certain amount so that features with a high level of uncertainty appear more
distorted than those with a low level.

8
http://gicentre.org/handy/

AGIS_SD014-02 UNCLASSIFIED Page 18 of 93


UNCLASSIFIED Technical Annex

Figure 18: Jitter visualisation for uncertainty portrayal of polygons

This visualisation was created using photo editing software5, 6 in combination with the processing14
software. The point symbols however, can be exported as PNG images to be used in an SLD with
multiple filters (SLD code shown in Figure 66) and linked into the SLD through the SLD code
shown in Figure 56).

B.3 Additions

B.3.1 Buffers

The simple buffer refers to expanding the area around a feature and involves a single colour.
Buffering is a common function in most GIS and is available in desktop GIS software.

To create a buffer using SLD technology one of the attributes of the feature would need to contain
the distances to be buffered. With these distance figures, the active function ‘interpolation’ can be
used (SLD code shown in Figure 69) for the ‘size’ attribute with the distance value and the ‘size’
attribute being equal. To create a buffer that uses map scale values, insert the code shown in
Figure 73 into the point, line or polygon symbolizer element.

B.3.2 Confidence Buffers

The confidence buffer is similar to the normal buffer except that there is additional information
regarding the confidence intervals of the data. These could be, for example, the mean and
standard deviation or a 95% confidence interval. An example of this visualisation was not
calculated and created as part of this research but could be created using the buffering function in
desktop GIS software.

To create a buffer using SLD technology one of the attributes of the feature would need to contain
the distances to be buffered. With these distance figures, the active function ‘interpolation’ can be
used (SLD code shown in Figure 69) for the ‘size’ attribute with the distance value and the ‘size’
attribute being equal. To create a buffer that uses map scale values, insert the code shown in
Figure 73 into the point, line or polygon symbolizer element. To show the different confidence
intervals, the GetMap request would need to include multiple instances of the same layer with
different styling for each of the different confidence intervals.

B.3.3 Stroke Buffers

Stroke buffers were developed specifically for spatial precision. The reason only the outlining
stroke feature is varied is because precision is a function of the data grouping. Therefore a stroke
buffer could be combined with another variable to present the point or polygon feature type with the
stroke buffer used to portray spatial precision.

AGIS_SD014-02 UNCLASSIFIED Page 19 of 93


UNCLASSIFIED Technical Annex

Figure 19: Stroke buffers visualisation for uncertainty portrayal of point features

To create a stroke buffer using SLD technology one of the attributes of the feature would need to
contain the distances to be buffered would be required. With these distance figures, the active
function ‘interpolation’ can be used (SLD code shown in Figure 69) for the ‘stroke-width’ attribute
with the distance value and the ‘stroke-width’ attribute being equal. To create a buffer that uses
map scale values, insert the code shown in Figure 73 into the point, line or polygon symbolizer
element.

B.3.4 Fuzzy Buffers

Fuzzy buffers work along the same principle of the normal buffer; however, towards the outer
edges of the buffer, the buffer becomes fuzzier. The reasoning behind the increasing fuzziness is
that in a normal distribution the majority of results revolve around the central area of the
distribution. The fuzzy buffer applies the same theory to a spatial representation. Although
represented as a point in Figure 20, the fuzzy buffer technique could be used to present a different
distribution of values as well as buffering around the physical map features.

This research could not find a method of creating a true fuzzy buffer; instead, layers of transparent
buffers were stacked to simulate a fuzzy effect.

Figure 20: Fuzzy buffers visualisation for uncertainty portrayal of point features

AGIS_SD014-02 UNCLASSIFIED Page 20 of 93


UNCLASSIFIED Technical Annex

B.3.5 Analyst-Defined Area

An analyst-defined area is used to represent spatial vagueness. For example, in Figure 21, an
analyst is required to use their judgement of the vagueness of spatial location information about
the black point and extrapolate outwards. This would require the use of a desktop GIS to draw the
polygon which could then be styled using basic SLD styling (SLD code shown in Figure 50).

Figure 21: Analyst-defined area for indicating uncertainty of a point location

B.3.6 Fuzzy Area

Fuzzy Area extends the analyst-defined area (Section B.3.5) to include fuzzy symbology, in effect
showing the focus of the spatial location while indicating that the spatial location is vague.

Figure 22: Fuzzy area for indicating uncertainty of a point location

AGIS_SD014-02 UNCLASSIFIED Page 21 of 93


UNCLASSIFIED Technical Annex

B.3.7 Star Bursts

Star bursts were developed by Reuschel & Hurni (2011) to show uncertainty in locations for
mapping locations in fictional literature. The visualisation defines a polygon region and uses lines,
like spokes on a wheel spreading outwards, while becoming increasingly transparent towards the
edge of the polygon. This introduces a focus to the area, at the centre of the spokes. This
visualisation cannot be made using SLDs or common GIS functions.

Figure 23: Star bursts visualisation to show the extent and most uncertain locations

B.3.8 Displacement

Displacement presents a pair of points, lines or polygons, one being the feature itself shown in
black and the other a red feature (the displacement) connected together by thin blue lines. The
distance between the feature and the displacement (shown in red, Figure 24) is dependent on the
value of the spatial uncertainty. Those features that are highly uncertain are drawn further away
from the black feature than those that are fairly certain. For point and polygon features, the
direction for displacement is random and determined at the time of rendering. The linear
displacement is made parallel to the line that the displaced line is associated with.

Figure 24: Displacement visualisation to show uncertainty through the change in location

This visualisation was made using software called Processing7.

AGIS_SD014-02 UNCLASSIFIED Page 22 of 93


UNCLASSIFIED Technical Annex

B.3.9 Bubbles

The bubble technique was developed based on Nature; the idea is that the size and density of
bubbles acts as a representation of uncertainty that can be overlaid on maps (Figure 25). The
bubbles can be combined with many of the other visual variables including, size, density, opacity
and colour while still maintaining the structure of a bubble and thus the presence of uncertainty.
Currently, there are no means by which this visualisation can be produced automatically.
However, a stylized version of the bubble technique is that of hexagon grids, discussed in Section
B.7.4.

Figure 25: Bubbles experimental visualisation to imply uncertainty as a map overlay

B.4 Temporal

B.4.1 Fading Features

The fading features technique was developed for temporal uncertainties whereby a feature fades
towards complete transparency as its relevance decreases over time. This technique is not a
capability of SLDs or desktop GIS. However, Task 2 developed software which used fading
features to indicate the age of the locational information about friendly and enemy positions and
routes.

Ageing/fading
features

Figure 26: Fading features to portray increasing uncertainty of overlain features

AGIS_SD014-02 UNCLASSIFIED Page 23 of 93


UNCLASSIFIED Technical Annex

B.4.2 Expanding Features

Expanding features work on the theory that the larger the period of time since the last known
information about an attribute, the greater the uncertainty about that information. Therefore, using
animation, the longer the period of time since an object was last detected by a sensor, the larger
that object will grow.

Figure 27: Diagram showing how uncertainty increases into the future

Another use for expanding features is presenting future possibilities. The further into the future that
a prediction is being made, the larger the cone becomes as shown in Figure 28.

Figure 28: Expanding features for prediction into the future

B.4.3 Flashing

Flashing objects require animation or a time slider which is a currently capability in Quantum GIS
(QGIS) as well as ESRI ArcGIS 10.0 and Google Earth. Flashing can be used to signify when a
feature is not temporally relevant by the speed at which that feature flashes.

Flashing can also be used for spatial uncertainties and can work in a number of ways:

 Flashing between extremes of spatial location when including uncertainty


 Flashing between the expected location and the extreme extent of the spatial uncertainty
when including uncertainty
 Flashing periodically along values between the extremes repeatedly

It can also be used to signify uncertain thematic attributes by flashing at different tempos, slow to
no flashing to signify low uncertainty and fast flashing to signify high uncertainty.

B.4.4 Colour Legend

The colour legend technique is used to represent when a feature is temporally relevant and is
shown through a colour scale. This could be combined with another visual variable to indicate the
uncertainty of this temporal period.

AGIS_SD014-02 UNCLASSIFIED Page 24 of 93


UNCLASSIFIED Technical Annex

To colour features by time using SLD technology, one of the attributes of the feature would need to
contain the time that the attribute is relevant for. To colour the features categorically either the
active function categorisation (SLD code shown in Figure 71) or multiple filters are used. The
colour at each time period is then set for each time period. Alternatively to create a continuous
scale the active function ‘interpolation’ can be used (SLD code shown in Figure 69) for the ‘fill’
attribute. This requires time values to be set against colours as described in Figure 69.

B.4.5 Moving Objects

The moving objects technique is like flashing (Section B.4.3), but instead of periodic flashes the
object appears to shift its location within the localised area to the starting point, increasing in
distance with time. The movement could also be given a direction if an estimated direction is
known. This technique requires animation or a time slider, which is a capability currently available
in Quantum GIS, ESRI ArcGIS 10 and Google Earth.

B.5 Stereotypes

Stars are regularly used by critics in their rating of films, books and food. The CGTS report by
Team Sparta (2010) also highlighted stars as a technique to signify data quality. Despite
uncertainty making up only one component of data quality, the theory behind this visualisation is
that stars might also be able to convey uncertainty. In keeping with common convention, the best
or most certain data is attributed with five stars and the most uncertain data is attributed with just
the one star.

Figure 29: Star rating visualisation to indicate levels of uncertainty in point features

This technique is most suited to digital mapping through the use of an uncertainty layer or as an
icon which is then exposed when the mouse is hovered over the object on the map.

The Task 6 workshop produced an improvement to the visualisation, through the use of empty
stars to make the map user aware of the maximum scale available. At the DGC workshop another
improvement was proposed: to use the five limbs of the star to represent the uncertainty and where
the centre of the star is used to indicate the spatial location of the object (Figure 30). Unfortunately,
these proposals came too late to be included within the metrication survey.

AGIS_SD014-02 UNCLASSIFIED Page 25 of 93


UNCLASSIFIED Technical Annex

Figure 30: Alternative single star symbology

To create the star visualisation using an SLD, the symbols were created using photo editing
software5 and saved in PNG format. Multiple SLD filter functions were used (shown in Figure 66)
linking the PNG symbology to the point feature through the code shown in Figure 56.

A more traditional visualisation using the star metaphor would be to rate uncertainty out of five in a
tabular format. This can be used as a guide when selecting data from a database or downloading
data from a web portal as shown in Figure 31. Again this is more regually used to signify quality but
could be used to the dataset’s total uncertainty.

Figure 31: Tabular star metaphor for visualising a datasets total uncertainty9

B.5.1 Traffic Lights

The traffic light visualisations rely upon the everyday signalling of traffic lights on roads. Officially,
according to direct.gov.uk, a red light means ‘stop and wait’, a yellow light also means ‘stop’ unless
it is too late to do so and a green light means to ‘proceed if it is safe to do so’. However, more
colloquially, a yellow light is often understood to mean ‘prepare to stop’ or ‘proceed if safe’ and
therefore acts as more of an intermediary.

9
This particular example used the paradigm that more stars mean more certainty – this is supported by the
results from the contextual composite metrication experiment which indicate that users prefer more stars to
mean more certainty. However, this also highlights the fact that users are more comfortable with the use of
the term ‘certainty’ rather than ‘uncertainty’ and this should perhaps be reflected in the labelling.

AGIS_SD014-02 UNCLASSIFIED Page 26 of 93


UNCLASSIFIED Technical Annex

Traffic light colouring has recently become commonplace on food products to signify how healthy
the levels of fat and salt, etc. are in the product. With regards to food packaging, Kelly et al. (2009)
conducted a consumer test with 790 participants to compare the effectiveness of a traffic light
colour system against presenting the consumer with percentage figures both colour coded and
monochrome. This has a strong relation to showing the uncertainty of a product or dataset
because there are subcategories in the different forms of uncertainty (e.g. accuracy, precision,
etc.) as there are in food (e.g. fats and salt, etc.). Kelly et al. (2009) found that despite consumers
preferring to see the percentages they actually made 3x better decisions when using a traffic light
scale compared to the colour coded scale and 5x better decisions than consumers using a
monochrome scale; results that backed up an earlier study by Conquest Research in 2006 (see
Figure 32 for food labelling symbology). This would require similar studies in a military context but
is a good basis for including a traffic light visualisation technique.

Figure 32: Traffic light symbology in food packaging labels

Figure 33: Traffic light visualisations to indicate levels of uncertainty in point features

AGIS_SD014-02 UNCLASSIFIED Page 27 of 93


UNCLASSIFIED Technical Annex

The icon-based traffic light would be best used as a digital overlay which can be switched on and
off when analysing data. A potential issue with this stereotype is that the red light is a metaphor for
stopping but uncertainty does not mean that the object being portrayed is wrong or dangerous,
unlike driving through a red light at a crossroads.

To create the traffic light visualisation using an SLD, the symbols were created using photo editing
software5 and saved in PNG format. Multiple SLD filter functions were used (shown in Figure 66)
linking the PNG symbology to the point feature through the code shown in Figure 56.

Traffic light symbology could also be used as a representation of the overall uncertainty of a
datasets (Figure 34), in a similar manner to the earlier star example (Figure 31). In this context, the
green light is a clear signal that the data is good for use, yellow acts as a warning and red implies
‘do not use’.

Figure 34: Traffic light metaphor for visualising a datasets total uncertainty

Figure 34 only shows a traffic light with a three point scale (red, yellow & green) but this could have
been expanded to a five point scale using the colour combinations shown in Figure 35.

Figure 35: Alternative five point traffic light scale

AGIS_SD014-02 UNCLASSIFIED Page 28 of 93


UNCLASSIFIED Technical Annex

B.5.2 Emoticons

Emoticons have become a common expression of emotional feeling when sending textual
information via mobile phones or email. To portray uncertainty through emoticons smiling, unsure
and sad faces were chosen with the point feature being shown by the location of the face’s nose in
the middle of the icon.

Low Uncertainty Low Uncertainty

Medium Medium
Uncertainty Uncertainty

Figure 36: Emoticon visualisation indicating uncertainty in point features10

Although fairly clear when given time to examine Figure 36 (left), the monochrome emoticons do
not work particularly well at a glance. To improve the original emoticons visualisation it was
combined with traffic light colours which seem to enhance the visualisation, especially when
viewed quickly. These visualisations would work best as a digital overlay or as an icon which only
appears when the mouse is hovered over the object on the map.

To create the emoticon visualisation using an SLD, the symbols were created using photo editing
software5 and saved in PNG format. Multiple SLD filter functions were used (shown in Figure 66)
linking the PNG symbology to the point feature through the code shown in Figure 56.

10
left: monochrome, right: traffic light colours

AGIS_SD014-02 UNCLASSIFIED Page 29 of 93


UNCLASSIFIED Technical Annex

B.6 Labels

B.6.1 Textual Labels

Textual labels are a very simple method for portraying any information. It would be possible to set
up an SLD which shows the uncertainty percentage as a label using the SLD code in Figure 55.

B.6.2 GetFeatureInfo

GetFeatureInfo is a WMS request that allows information about features and their attributes to be
displayed. In Figure 37, the Kandahar polygon has been clicked and all of the attribute information
has been exposed. This example presents the results in a tabular format but this is application
specific – the GetFeatureInfo response is text-based and can be displayed in many ways.

Figure 37: GetFeatureInfo portraying feature information including uncertainty values

B.6.3 Pop-Ups

Pop-ups require the user to click on the feature when more information is required. Primarily this
example has been shown in Google Earth and KML. Appendix E. shows four techniques for
including uncertainty information within KML.

B.7 Visualisations requiring GIS

B.7.1 3D Extrusion

Figure 39 shows the 3D extrusion visualisation technique. This is displayed using Google Earth
using its ‘network link’ functionality to request WMS data from a server (in this case, GeoServer).
To extrude the GeoServer WMS by uncertainty attribute, an additional file called height.ftl
needs to be created and saved in the GeoServer workspace11. This file contains the following
code:

${Uncert.value?number}

Alternately: change the extrusion with a mathematical function


${Uncert.value?number * 100}

Figure 38: Code required for 3D extrusion using GeoServer & Google Earth

11
See http://docs.geoserver.org/stable/en/user/googleearth/tutorials/heights/heights.html for details.

AGIS_SD014-02 UNCLASSIFIED Page 30 of 93


UNCLASSIFIED Technical Annex

Figure 39: 3D extrusion, where increasing height indicates greater uncertainty

B.7.2 Heat Maps

To create a heat map image (Figure 40) requires a desktop GIS application. This example was
made using ESRI ArcGIS with the Geospatial Analyst add-in. Heat maps can also be made using
the open source Quantum GIS (QGIS) heat map function. Once created, the raster layer can be
coloured using SLDs through the use of the basic raster symbology (SLD code shown in Figure
52).

Figure 40: Heat map where the red (hotter) colours indicate higher uncertainty

B.7.3 Clusters

Clusters can be created using desktop GIS software. These could be used to provide an overview
of the level of uncertainty of a dataset by displaying such information in a manner which
discourages the user from concentrating on specific points. The size of the symbol, the indicator

AGIS_SD014-02 UNCLASSIFIED Page 31 of 93


UNCLASSIFIED Technical Annex

around the outside and the number within the symbol can all be altered to reflect some aspect of
uncertainty of the information within the cluster.

Figure 41: Cluster technique showing an overview of a high density point dataset12

B.7.4 Hexagon Grids

Hexagon grids can be used to amalgamate point data, producing an overview depicting the
uncertainty related to the information portrayed.
To create the hexagon grids, the open source GIS, Quantum GIS (QGIS) was used with the open
source plug-in mmgis. The mmgis plug-in has an option for creating a grid layer with cells in the
shape of a rectangle, diamond or hexagon. Hexagons were chosen because they represented a
stylised version of the bubble visualisation and can also be fully tessellated. This grid was then
overlaid over point data and a QGIS point count function was executed. This counted the number
of points within the bounds of each polygon in the grid layer. The grid was then coloured based on
the point count value using an SLD and one of the colour or visual variables techniques.

Figure 42: Hexagon grid technique where higher opaqueness represents higher uncertainty

12
Source: www.police.co.uk

AGIS_SD014-02 UNCLASSIFIED Page 32 of 93


UNCLASSIFIED Technical Annex

B.7.5 Contouring / Isopleth Maps

Isopleth maps are like the heat maps but shown as contours. These require GIS software to extract
the contours from the data or to be manually drawn using the raw data. To extract contours from
the data, ESRI ArcGIS Geospatial Analyst add-in was used. Once the contours have been created
they can be visualised using one of the colour or visual variables techniques.

70%

Figure 43: Isopleth map as an overlay for contouring uncertainty values.

B.7.6 Graphs

To add graphs to a polygon or point a desktop GIS is required. ESRI ArcGIS was used to create
the graph visualisation. This option can be found under the ‘Layer Properties’, ‘Symbology’ tab
shown in Figure 45. The three chart options available are pie charts, bar charts and stacked bar
charts.

Figure 44: Graphs within the polygon to show uncertainty values graphically

Figure 45: ESRI ArcGIS chart symbology

AGIS_SD014-02 UNCLASSIFIED Page 33 of 93


UNCLASSIFIED Technical Annex

B.7.7 3D Threat Domes

The theory behind the 3D threat domes is that the certainty of a threat decreases with distance
from the source, therefore the larger the dome the greater the uncertainty of the threat. This
visualisation was created using a COLLADA 3D model and altering the colours for the varying
uncertainty levels. This model was then referenced by a KML file which scaled the COLLADA
dome model to create the different sized domes. This was an experimental technique but proves
the capability within Google Earth.

Figure 46: 3D threat dome utilising COLLADA

AGIS_SD014-02 UNCLASSIFIED Page 34 of 93


UNCLASSIFIED Technical Annex

C. Appendix C: SLD Structure


This appendix describes the structure and components to create the SLDs used in the uncertainty
visualisations as well as the basics of SLD styling.

An SLD can be separated out into four main sections; these are header/namespaces, SLD
information, styling and closing elements. Throughout this SLD structure appendix the SLD
snippets will be placed against line numbers or letters as can be seen on the left of Figure 47
through to Figure 74 these will be referenced within the text and also show how the SLD sections
slot together. Figures which include the same line numbers are an either/or styling option, these
only occur in Section C.3. Sections with letters depicting each line show styling options over and
above the basic styling.

Within each figure there is this colour scheme:

Blue = XML information


Green = Comments
Red = User entered/editable information
Purple = Rule brackets

C.1 SLD Header/Namespaces

These seven lines began every SLD used to create the Task 5 SLD visualisations. Line 1 tells
applications that the code is xml based. Line 2 defines the file as a Styled Layer Descriptor (SLD)
and the version. Lines 3-7 are namespaces which define the parsing of the prefixes such ‘ogc:’
when these are used within the SLD.

1 <?xml version="1.0" encoding="ISO-8859-1"?>


2 <StyledLayerDescriptor version="1.0.0"
3 xsi:schemaLocation="http://www.opengis.net/sld StyledLayerDescriptor.xsd"
4 xmlns="http://www.opengis.net/sld"
5 xmlns:ogc="http://www.opengis.net/ogc"
6 xmlns:xlink="http://www.w3.org/1999/xlink"
7 xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance">

Figure 47: SLD header used in every SLD

C.2 SLD Information

This next section contains the name of the SLD, the title and abstract before the opening bracket
for the main body of the SLD styling information.

8 <!-- a Named Layer is the basic building block of an SLD document -->
9 <NamedLayer>
10 <!-- Styles can have names, titles and abstracts -->
11 <Name>SLD_NAME</Name>
12 <UserStyle>
13 <Title>SLD_TITLE</Title>
14 <Abstract>Enter Styling Abstract Here</Abstract>
15 <!-- A FeatureTypeStyle for rendering polygons -->
16 <FeatureTypeStyle>

Figure 48: SLD information section

AGIS_SD014-02 UNCLASSIFIED Page 35 of 93


UNCLASSIFIED Technical Annex

C.3 Styling Information

The main body of the SLD is defined by one or many rules which contain the information that
defines the styling of text, raster, point, line and polygon features. Within each <Rule> is an
optional a name, title and abstract then a <PointSybolizer>, <PolygonSymbolizer>, <LineSymbolizer>
or <RasterSymbolizer> function optionally used in conjunction with a <TextSymbolizer> function. Basic
SLD styling styles all points, polygons, lines and rasters by one form of symbology. Whereas
advanced styling involves the use of multiple rules, filters and active functions to define the styling
by attributes and attribute values.

C.3.1 Basic Styling

This form of styling is of most use when used in conjunction with a scale filter (see C.3.2) to only
show simple symbology when viewed at a small scale avoiding overcrowding the map. If used with
a scale filter this symbology would then switch to the uncertainty symbology when zoomed into the
area of interest on the map.

17 <Rule>
18 <Name>POINT_RULE</Name>
19 <Title>Black_Circle</Title>
20 <Abstract>
21 A 6 pixel circle with a black
22 fill and no stroke (outline)
23 </Abstract>
24 <PointSymbolizer>
25 <Graphic>
26 <Mark>
27 <WellKnownName>circle</WellKnownName>
28 <Fill>
29 <CssParameter name="fill">#000000</CssParameter>
30 </Fill>
31 </Mark>
32 <Size>6</Size>
33 </Graphic>
34 </PointSymbolizer>
35 </Rule>

Figure 49: Basic point SLD styling rule

17 <Rule>
18 <Name>POLYGON_RULE</Name>
19 <Title>Red Polygon with Black Outline</Title>
20 <Abstract>
21 A polygon with a Red fill and
22 a 1 pixel Black outline
23 </Abstract>
24 <PolygonSymbolizer>
25 <Fill>
26 <CssParameter name="fill">#FF0000</CssParameter>
27 </Fill>
28 <Stroke>
29 <CssParameter name="stroke">#000000</CssParameter>
30 <CssParameter name="stroke-width">1</CssParameter>
31 </Stroke>
32 </PolygonSymbolizer>
33 </Rule>

Figure 50: Basic polygon SLD styling rule

AGIS_SD014-02 UNCLASSIFIED Page 36 of 93


UNCLASSIFIED Technical Annex

17 <Rule>
18 <Name>LINE_RULE</Name>
19 <Title>Blue_Line</Title>
20 <Abstract>A solid blue line with a 1 pixel width</Abstract>
21 <LineSymbolizer>
22 <Stroke>
23 <CssParameter name="stroke">#0000FF</CssParameter>
24 </Stroke>
25 </LineSymbolizer>
26 </Rule>

Figure 51: Basic line SLD styling rule

17 <Rule>
18 <Name>RASTER_RULE</Name>
19 <Title>Opaque Greyscale Raster</Title>
20 <Abstract>A raster with 100% opacity and greyscale</Abstract>
21 <RasterSymbolizer>
22 <Opacity>1.0</Opacity>
23 </RasterSymbolizer>
24 </Rule>

Figure 52: Basic raster SLD styling

C.3.2 Advanced SLD styling

Advanced SLD styling involves many of the components met in the basic section and some
additional ones. These are opacity, dashed lines, text, image files, filters, active functions,
arithmetic operators and specifying the drawing units.

C.3.2.1 Transparency/Opacity

In the basic styling section there was no use of transparency. To use transparency include one of
the lines that are included within Figure 53 either b, d or f, for the relevant feature type. The input
value for transparency can range between 0 and 1, where 0 is fully transparent and 1 is opaque.
Transparency was used in many of the visualisations to provide the ability to view the underlying
map through the visualisation. It is also used as the changing parameter in the transparency
visualisation (see Section B.2.1).

a <!-- Polygon & point transparency -->


b <CssParameter =name “fill-opacity”>value between 0 and 1</CssParameter>
c <!-- Line or stroke transparency -->
d <CssParameter =name “stroke-opacity”>value between 0 and 1</CssParameter>
e <!-- Raster transparency -->
f <Opacity>value between 0 and 1</Opacity>

Figure 53: Transparency SLD parameters

C.3.2.2 Dashed lines

Dashed lines can be used for the grain visualisation (see Section B.2.7) to create this using an
SLD, use the dash array function (see Figure 54).

a <!-- To create the dashed line enter a series of numbers separated by spaces
b odd-indexed numbers (first, third etc.) define the pixel length of the dash, even-indexed
c numbers (second, fourth etc.) define the space between dashes -->
d <CssParameter =name “stroke-dasharray”>5 2 3 2</CssParameter>

Figure 54: Dash array for creating dashed lines in SLDs

AGIS_SD014-02 UNCLASSIFIED Page 37 of 93


UNCLASSIFIED Technical Annex

C.3.2.3 Text

To include text in the styling of a layer the <TextSymbolizer> function is required. Within this function
it is possible to define all aspects of the font as well as how that text is presented with regards to
features on the map.

a <TextSymbolizer>
b <!--Label to be written, CDATA causes direct formatting e.g. a carriage return (below)-->
c <Label>
d <ogc:PropertyName>Property1</ogc:PropertyName><![CDATA[
e ]]>(<ogc:PropertyName>Property2</ogc:PropertyName>)
f </Label>
g <!-- Font parameters -->
h <Font>
i <CssParameter name="font-family">Times New Roman</CssParameter>
j <CssParameter name="font-size">11</CssParameter>
k <CssParameter name="font-style">normal</CssParameter>
l <CssParameter name="font-weight">bold</CssParameter>
m </Font>
n <!-- Text colour -->
o <Fill>
p <CssParameter name="fill">#000000</CssParameter>
q </Fill>
r <!-- Halo around the text, parameters for radius and colour of the halo -->
s <Halo>
t <Radius>2</Radius>
u <Fill>
v <CssParameter name="fill">#FFFFFF</CssParameter>
w </Fill>
x </Halo>
y
z <LabelPlacement>
aa <!-- Places the label on a point or <!-- Places the label parallel to a line
ab in the centre of a polygon --> feature -->
ac <PointPlacement> <LinePlacement>
ad <!-- Point location within label --> <!-- displacement from line feature -->
ae <AnchorPoint> <Perpendicularoffset>0</Perpendicularoffset>
af <AnchorPointX>0.5</AnchorPointX> </LinePlacement>
ag <AnchorPointY>0.5</AnchorPointY>
ah </AnchorPoint>
ai <!-- displacement from central point --> Or
aj <Displacement>
ak <DisplacementX>0</DisplacementX>
al <DisplacementY>25</DisplacementY>
am </Displacement>
an <!-- Angle of rotation -->
ao <Rotation>-45</Rotation>
ap
aq </PointPlacement>
ar </LabelPlacement>
as
at </TextSymbolizer>
au </LabelPlacement>
av
aw </TextSymbolizer>

Figure 55: SLD text symbology options

AGIS_SD014-02 UNCLASSIFIED Page 38 of 93


UNCLASSIFIED Technical Annex

C.3.2.4 Image, shape & texture styling

Along with standard styling it is possible to link to image files this is used in the stars, traffic lights
and emoticon visualisations (Figure 56). It is also possible to show a point as a series of shapes
(Figure 57) and fill polygons by a number of different textures (Figure 58 to Figure 60).

a <PointSymbolizer>
b <Graphic>
c <ExternalGraphic>
d <OnlineResource xlink:type="simple"
e xlink:href="file://folder/picture.png" />
f <Format>image.png</Format>
g </ExternalGraphic>
h </Graphic>
i </PointSymbolizer>

Figure 56: SLD links to image files

a <!-- Circular Point -->


b <WellKnownName>circle</WellKnownName>
c
d <!-- Square Point -->
e <WellKnownName>square</WellKnownName>
f
g <!-- Triangular Point -->
h <WellKnownName>triangle</WellKnownName>
i
j <!-- 5 pointed star -->
k <WellKnownName>star</WellKnownName>
l
m <!-- + cross Point -->
n <WellKnownName>cross</WellKnownName>
o
p <!-- x cross point -->
q <WellKnownName>x</WellKnownName>
r
s <!-- For other Webdings symbols, ttf://webdings then include the symbol hex code -->
t <!-- This symbol is a shield -->
u <WellKnownName>ttf://Webdings#0x0064</WellKnownName>

Figure 57: SLD point shapes

a <Fill>
b <GraphicFill>
c <Graphic>
d <Mark>
e <!-- An 'X' symbol suitable for X-hatch fills -->
f <WellKnownName>shape://times</WellKnownName>
g <Stroke>
h <!-- Colour and width of the X-hatch -->
i <CssParameter name="stroke">#000000</CssParameter>
j <CssParameter name="stroke-width">1</CssParameter>
k </Stroke>
l </Mark>
m <!-- Size of the 'X' change this attribute to change the density of the X-hatching -->
n <Size>50</Size>
o </Graphic>
p </GraphicFill>
q </Fill>

Figure 58: X-hatch styling (suitable for X-hatch fills)

AGIS_SD014-02 UNCLASSIFIED Page 39 of 93


UNCLASSIFIED Technical Annex

a <Fill>
b <GraphicFill>
c <Graphic>
d <Mark>
e <!-- An '|' symbol suitable for vertical hatch fills -->
f <WellKnownName>shape://horline</WellKnownName>
g <Stroke>
h <!-- Colour and width of the hatch -->
i <CssParameter name="stroke">#000000</CssParameter>
j <CssParameter name="stroke-width">1</CssParameter>
k </Stroke>
l </Mark>
m <!-- Size of the '|' change this attribute to change the grain width -->
n <Size>50</Size>
o </Graphic>
p </GraphicFill>
q </Fill>

Figure 59: Grain styling (suitable for hatch fills)

Grain is the density of straight lines in the same direction but orientation is the varying in the
direction of the fill so in Figure 60 lines ‘e’ and ‘f’ are repeated from Figure 59 but with all the
different orientations possible using SLD code.

e <!-- An '|' symbol suitable for vertical hatch fills -->


f <WellKnownName>shape://horline</WellKnownName>

e <!-- An '-' symbol suitable for vertical hatch fills -->


f <WellKnownName>shape://vertline</WellKnownName>

e <!-- An '/' symbol suitable for vertical hatch fills -->


f <WellKnownName>shape://slash</WellKnownName>

e <!-- An '\' symbol suitable for vertical hatch fills -->


f <WellKnownName>shape://backslash</WellKnownName>

Figure 60: Orientation styling (suitable for polygon fills)

C.3.2.5 Filters

There are several mathematical functions which are used to filter against the dataset’s attributes to
apply the rule’s symbology only to those features which satisfy the function (see Figure 61-Figure
65). The filters are best applied in conjunction with multiple rules and can be used to show
symbology for classified sets of uncertainty values.

a <ogc:Filter>
b <!-- Filters the attribute ‘Property_Name’ for the value or text shown between the literal
c brackets. Only those that are equal to that value are applied with the symbology -->
d <ogc:PropertyIsEqualTo>
e <ogc:PropertyName>Property_Name</ogc:PropertyName>
f <ogc:Literal>Value_Numeric_or_Textual</ogc:Literal>
g </ogc:PropertyIsEqualTo>
h
h </ogc:Filter>

Figure 61: Equal To SLD filter function

AGIS_SD014-02 UNCLASSIFIED Page 40 of 93


UNCLASSIFIED Technical Annex

a <ogc:Filter>
b <!-- Filters the attribute ‘Property_Name’ for the value or text shown between the literal
c brackets. Only those that are greater than the value are applied with the symbology -->
d <ogc:PropertyIsGreaterThan>
e <ogc:PropertyName>Property_Name</ogc:PropertyName>
f <ogc:Literal>Value_Numeric_or_Textual</ogc:Literal>
g </ogc:PropertyIsGreaterThan >
h
h </ogc:Filter>

Figure 62: Greater Than SLD filter function


a <ogc:Filter>
b <!-- Filters the attribute ‘Property_Name’ for the value or text shown between the literal
c brackets. Only those that are less than the value are applied with the symbology -->
d <ogc:PropertyIsLessThan>
e <ogc:PropertyName>Property_Name</ogc:PropertyName>
f <ogc:Literal>Value_Numeric_or_Textual</ogc:Literal>
g </ogc:PropertyIsLessThan >
h
h </ogc:Filter>

Figure 63: Less Than SLD filter function


a <ogc:Filter>
b <!-- Filters the attribute ‘Property_Name’ for the value or text shown between the literal
c brackets. Only those that are greater than or equal to the value are applied with the
d symbology -->
e <ogc:PropertyIsGreaterThanOrEqualTo>
f <ogc:PropertyName>Property_Name</ogc:PropertyName>
g <ogc:Literal>Value_Numeric_or_Textual</ogc:Literal>
h </ogc:PropertyIsGreaterThanOrEqualTo >
i
h </ogc:Filter>

Figure 64: Greater Than Or Equal To SLD filter function


a <ogc:Filter>
b <!-- Filters the attribute ‘Property_Name’ for the value or text shown between the literal
c brackets. Only those that are less than or equal to the value are applied with the
d symbology -->
e <ogc:PropertyIsLessThanOrEqualTo>
f <ogc:PropertyName>Property_Name</ogc:PropertyName>
g <ogc:Literal>Value_Numeric_or_Textual</ogc:Literal>
h </ogc:PropertyIsLessThanOrEqualTo >
i
h </ogc:Filter>

Figure 65: Less Than Or Equal To SLD filter function

These filters can be used in conjunction with one another as well as referencing multiple attributes
of the dataset an example of this is shown in Figure 66. More complex filtering can be
accomplished through adding more filter functions in conjunction.

AGIS_SD014-02 UNCLASSIFIED Page 41 of 93


UNCLASSIFIED Technical Annex

a <ogc:Filter>
b <!-- 'And', 'Or' & 'Not' function allows for the use of multiple filters -->
c <ogc:And>
d <!-- Filter 1 -->
e <ogc:PropertyIsEqualTo>
f <ogc:PropertyName>Property_1</ogc:PropertyName>
g <ogc:Literal>Value_Numeric_or_Textual</ogc:Literal>
h <!-- Filter 2, this can reference the same property or an entirely new property -->
i </ogc:PropertyIsEqualTo>
j <ogc:PropertyIsLessThanOrEqualTo>
k <ogc:PropertyName>Property_2</ogc:PropertyName>
l <ogc:Literal>Value_Numeric_or_Textual</ogc:Literal>
m </ogc:PropertyIsLessThanOrEqualTo>
n <!-- Additional filters as required can be inserted here -->
o </ogc:And>
p </ogc:Filter>

Figure 66: Multiple SLD filter functions

Another form of filter is the scale filter which does not act upon the data but upon the map when
used only showing the styling within the rule up to, between or below the specified scales. Figure
67 shows the two scale filters which can be included within a <Rule> bracket, these can be used
standalone or used in conjunction to set an upper and lower bounds.

a <!-- The minimum scale at which a graphic or symbol is to be visualised -->


b <MinScaleDenominator>80000</MinScaleDenominator>
c
d <!-- The maximum scale at which a graphic or symbol is to be visualised -->
e <MaxScaleDenominator>120000</MaxScaleDenominator>

Figure 67: SLD scale filtering maximum and minimum scale denomination

AGIS_SD014-02 UNCLASSIFIED Page 42 of 93


UNCLASSIFIED Technical Annex

C.3.2.6 Active functions

Active functions can be used in a similar way to using multiple rule statements but, act within one
rule. These functions include interpolation (Figure 68 & Figure 69), recode (Figure 70) and
categorize (Figure 71). Interpolation allows for custom creation of colour ramps as well as numeric
ramps; for scales which are numeric such as size or width. The interpolation method can also be
configured to be linear, cubic or sinusoidal. Recode uses input to output pairs and is equivalent to
using multiple ‘equal to’ filters within separate rules each defining different colours. The categorize
method sets colours according against numeric values, with the order of the list defining which
colour is shown above and below the numeric value.

a <!--Parameter to be interpolated: fill, fill-opacity, stroke, size or width-->


b <CssParameter name="fill">
c <ogc:Function name="Interpolate">
d <!-- Property to transform -->
e <ogc:PropertyName>Property_1</ogc:PropertyName>
f
g <!-- Mapping curve definition pairs (input, output), minimum of 2 pairs -->
h <ogc:Literal>Lowest_Value</ogc:Literal>
i <ogc:Literal>#fefeee</ogc:Literal>
j
k <ogc:Literal>Inbetween_Value</ogc:Literal>
l <ogc:Literal>#00ff00</ogc:Literal>
m
n <ogc:Literal>Highest_Value</ogc:Literal>
o <ogc:Literal>#ff0000</ogc:Literal>
p
q <!-- Interpolation method - numeric (default) or colour -->
r <ogc:Literal>color</ogc:Literal>
s <ogc:Literal>linear</ogc:Literal>
t <!-- Interpolation mode - linear (default), cubic or cosine-->
u </ogc:Function>
v </CssParameter>

Figure 68: SLD interpolation active function - colour

a <!--Parameter to be interpolated: fill, fill-opacity, stroke, size or width-->


b <CssParameter name=”fill-opacity">
c <ogc:Function name="Interpolate">
d <!-- Property to transform -->
e <ogc:PropertyName>Property_1</ogc:PropertyName>
f
g <!-- Mapping curve definition pairs (input, output), minimum of 2 pairs -->
h <ogc:Literal>Lowest_Value</ogc:Literal>
i <ogc:Literal>0.1</ogc:Literal>
j
k <ogc:Literal>Inbetween_Value</ogc:Literal>
l <ogc:Literal>0.7</ogc:Literal>
m
n <ogc:Literal>Highest_Value</ogc:Literal>
o <ogc:Literal>0.9</ogc:Literal>
p
q <!-- Interpolation method - numeric (default) or colour -->
r <ogc:Literal>numeric</ogc:Literal>
s <ogc:Literal>linear</ogc:Literal>
t <!-- Interpolation mode - linear (default), cubic or cosine-->
u </ogc:Function>
v </CssParameter>

Figure 69: SLD Interpolation active function – numeric

AGIS_SD014-02 UNCLASSIFIED Page 43 of 93


UNCLASSIFIED Technical Annex

a <!--Parameter to be recoded: fill, fill-opacity, stroke, size or width-->


b <CssParameter name="fill">
c <ogc:Function name="Recode">
d <!-- Value to transform -->
e <ogc:Function name="strTrim">
f <ogc:PropertyName>Value_To_Transform</ogc:PropertyName>
g </ogc:Function>
h
i <!-- Map of input to output value pairs (output can be colour or numeric) -->
j <ogc:Literal>Attribute_1</ogc:Literal>
k <ogc:Literal>#6495ED</ogc:Literal>
l
m <ogc:Literal>Attribute_2</ogc:Literal>
n <ogc:Literal>#B0C4DE</ogc:Literal>
o
p <ogc:Literal>Attribute_3</ogc:Literal>
q <ogc:Literal>#00FFFF</ogc:Literal>
r
s <ogc:Literal>Attribute_4</ogc:Literal>
t <ogc:Literal>#9ACD32</ogc:Literal>
u <!-- As many pairs as required -->
v
w </ogc:Function>
x </CssParameter>

Figure 70: SLD recode active function

a <!-- Parameter to be categorise styling for: fill, fill-opacity, stroke, size or width -->
b <CssParameter name="fill">
c <ogc:Function name="Categorize">
d
e <!-- Value to transform -->
f <ogc:PropertyName>Property_1</ogc:PropertyName>
g
h <!-- Output values and thresholds (example below colours
i by categories defined as [<20],[20-100] & [>100] -->
j <ogc:Literal>#87CEEB</ogc:Literal>
k <ogc:Literal>20</ogc:Literal>
l <ogc:Literal>#FFFACD</ogc:Literal>
m <ogc:Literal>100</ogc:Literal>
n <ogc:Literal>#F08080</ogc:Literal>
o
p </ogc:Function>
q </CssParameter>

Figure 71: SLD categorize active function

AGIS_SD014-02 UNCLASSIFIED Page 44 of 93


UNCLASSIFIED Technical Annex

C.3.2.7 Arithmetic operators

Arithmetic operators allow for the combination of numeric attributes to style by the result of the
function. This can be done by entering the values of by entering the name of the attribute
containing the values. The arithmetic operators include ‘add’, ‘sub’, ‘mul’ & ‘div’ and must have the
prefix of ‘ogc:’.

a <!-- Arithmetic Operators include: Add, Sub, Mul & Div -->
b <ogc:Add>
c <!--Minimum of two expressions required as sub-elements -->
d <ogc:PropertyName>Property_or_numeric_value</ogc:PropertyName>
e <ogc:PropertyName>Property_or_numeric_value</ogc:PropertyName>
f </ogc:Add>

Figure 72: Arithmetic operators within SLD styling

C.3.2.8 Specifying the drawing units – for scale specific styling

When defining the size of a point or width of a stroke for line or polygon feature types the styling
can be defined in terms of meters, feet or pixels through the addition of a line of code shown in
Figure 73 included within the <PointSymbolizer>, <LineSymbolizer> & <PolygonSymbolizer> element
respectively. This is more effective for printing maps as the symbology remains ‘n’ meters/feet wide
at all scales. This was used for buffering to the correct value spatial uncertainty value in the Task 5
visualisations.

a <!-- Unit of Measure = meters -->


b <PointSymbolizer uom="http://www.opengeospatial.org/se/units/metre">
c <!-- Contents of PointSymbolizer -->
d </PointSymbolizer>
e
f <!-- Unit of Measure = feet -->
g <LineSymbolizer uom="http://www.opengeospatial.org/se/units/foot">
h <!-- Contents of LineSymbolizer -->
i </LineSymbolizer>
j
k <!-- Unit of Measure = pixels -->
l <PolygonSymbolizer uom="http://www.opengeospatial.org/se/units/pixel">
m <!-- Contents of PolygonSymbolizer -->
n </PolygonSymbolizer>

Figure 73: Specifying drawing units for scale specific SLD styling

C.3.3 Closing Elements of the SLD file

After the header, and the one or many rules containing filters and other functions come the closing
elements of the SLD. These elements close all those that were opened in the header of the SLD;
the line numbers continue from Figure 49 but the SLD document could be much longer when using
the advanced styling.

36 </FeatureTypeStyle>
37 </UserStyle>
38 </NamedLayer>
39 </StyledLayerDescriptor>

Figure 74: Closing elements of SLD

AGIS_SD014-02 UNCLASSIFIED Page 45 of 93


UNCLASSIFIED Technical Annex

D. Appendix D: Example SLD


1 <?xml version="1.0" encoding="ISO-8859-1"?>
2 <StyledLayerDescriptor version="1.0.0"
3 xsi:schemaLocation="http://www.opengis.net/sld StyledLayerDescriptor.xsd"
4 xmlns="http://www.opengis.net/sld"
5 xmlns:ogc="http://www.opengis.net/ogc"
6 xmlns:xlink="http://www.w3.org/1999/xlink"
7 xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance">
8 <NamedLayer>
9 <Name>SLD_Example</Name>
10 <UserStyle>
11 <Title>SLD_Example</Title>
12 <Abstract>Colours polygons by attribute with categorised text drawn with a white halo</Abstract>
13 <FeatureTypeStyle>
14 <Rule>
15 <Name>SLD_name</Name>
16 <Title>SLD_title</Title>
17 <Abstract>Description of the SLD styling</Abstract>
18 <TextSymbolizer>
19 <Label>
20 <ogc:PropertyName>Property_Name_for_example ‘UNCERT’</ogc:PropertyName>
21 </Label>
22 <Font>
23 <CssParameter name="font-family">Arial</CssParameter>
24 <CssParameter name="font-size">15</CssParameter>
25 <CssParameter name="font-style">normal</CssParameter>
26 <CssParameter name="font-weight">bold</CssParameter>
27 </Font>
28 <LabelPlacement>
29 <PointPlacement>
30 <AnchorPoint>
31 <AnchorPointX>0.5</AnchorPointX>
32 <AnchorPointY>0.5</AnchorPointY>
33 </AnchorPoint>
34 </PointPlacement>
35 </LabelPlacement>
36 <Fill>
37 <CssParameter name="fill">#000000</CssParameter>
38 </Fill>
39 <VendorOption name="autoWrap">60</VendorOption>
40 <VendorOption name="maxDisplacement">150</VendorOption>
41 <Halo>
42 <Radius>2</Radius>
43 <Fill>
44 <CssParameter name="fill">#FFFFFF</CssParameter>
45 </Fill>
46 </Halo>
47 </TextSymbolizer>
48 </Rule>
49
50 <Rule>
51 <Name>0% to &lt;=20%</Name>
52 <Title>0% to &lt;=20%</Title>
53 <ogc:Filter>
54 <ogc:PropertyIsLessThanOrEqualTo>
55 <ogc:PropertyName>UNCERT</ogc:PropertyName>
56 <ogc:Literal>20</ogc:Literal>
57 </ogc:PropertyIsLessThanOrEqualTo>
58 </ogc:Filter>
59 <PolygonSymbolizer>
60 <Fill>
61 <CssParameter name="fill">#FFFF00</CssParameter>
62 </Fill>
63 <Stroke>
64 <CssParameter name="stroke">#000000</CssParameter>
65 <CssParameter name="stroke-width">1</CssParameter>
66 </Stroke>
67 </PolygonSymbolizer>
68 </Rule>
69
70 <Rule>
71 <Name>&gt;40% to &lt;=60%</Name>
72 <Title>&gt;40% to &lt;=60%</Title>

AGIS_SD014-02 UNCLASSIFIED Page 46 of 93


UNCLASSIFIED Technical Annex

73 <ogc:Filter>
74 <ogc:And>
75 <ogc:PropertyIsGreaterThan>
76 <ogc:PropertyName>UNCERT</ogc:PropertyName>
77 <ogc:Literal>40</ogc:Literal>
78 </ogc:PropertyIsGreaterThan>
79 <ogc:PropertyIsLessThanOrEqualTo>
80 <ogc:PropertyName>Certainty</ogc:PropertyName>
81 <ogc:Literal>60</ogc:Literal>
82 </ogc:PropertyIsLessThanOrEqualTo>
83 </ogc:And>
84 </ogc:Filter>
85 <PolygonSymbolizer>
86 <Fill>
87 <CssParameter name="fill">#FF8000</CssParameter>
88 </Fill>
89 <Stroke>
90 <CssParameter name="stroke">#000000</CssParameter>
91 <CssParameter name="stroke-width">1</CssParameter>
92 </Stroke>
93 </PolygonSymbolizer>
94 </Rule>
95 <!--This Rule is then repeated but with values 60 and 80, then greater than 80 -->
96 </FeatureTypeStyle>
97 </UserStyle>
98 </NamedLayer>
99 </StyledLayerDescriptor>

AGIS_SD014-02 UNCLASSIFIED Page 47 of 93


UNCLASSIFIED Technical Annex

E. Appendix E: Uncertainty Information for KML


KML (Keyhole Markup Language) originally developed by Google, for presenting data in Google
Maps and Google Earth, has recently been accepted as an OGC standard. KML has many
similarities with GML but also includes the symbology styling within the one file. This is beneficial
when using standardised symbology or a single dataset as the display loads with symbology
already specified. This can be problematic however, when using KML files from a variety of
sources as colours and symbols could be repeated.

To utilise KML as a format for displaying uncertainty information the <ExtendedData> element can
be used, adding custom data to a point, line or polygon. There are three methods to assign
uncertainty information to a feature in a KML file. They differ on their attribution of a data type i.e.
specifying that the data comes in the form of an integer or a string or not specifying at all.

To include information on data types then the schema method is required. This is important
because stipulating the data type within the file or not affects the interoperability of the data when
used in other software other than Google’s products. Google regards all values as data type string
and will show values contrary to the type specified in the KML schema.

The first and simplest method uses name/value pairs which will automatically be shown as a table
in the pop up balloons used in Google Earth & Google Maps (Figure 75).

1 <?xml version="1.0" encoding="UTF-8"?>


2 <kml xmlns="http://www.opengis.net/kml/2.2">
3 <Placemark>
4 <name>POI</name>
5 <ExtendedData>
6 <Data name="Positional_Accuracy_x">
7 <value>+/- 1 m</value>
8 </Data>
9 <Data name="Positional_Accuracy_y">
10 <value>+/- 0.5 m</value>
11 </Data>
12 </ExtendedData>
13 <Point>
14 <coordinates>-4.109864,50.371102,0</coordinates>
15 </Point>
16 </Placemark>
17 </kml>

Figure 75: Basic method for exposing uncertainty information in KML

Using the same name/value pairs a custom style can be applied to the pop up to improve
visualisation in which the name is used as a reference for the value in order to insert that value in a
descriptive section of text (Figure 76). In both of these examples the name value pairs are
Positional_Accuracy_x & Positional_Accuracy_y with their respective values.

AGIS_SD014-02 UNCLASSIFIED Page 48 of 93


UNCLASSIFIED Technical Annex

1 <?xml version="1.0" encoding="UTF-8"?>


2 <kml xmlns="http://www.opengis.net/kml/2.2">
3 <Document>
4 <name>Positional Accuracy</name>
5 <Style id="Positional Accuracy">
6 <BalloonStyle>
7 <text>
8 <![CDATA[
9 <h2>Positional Accuracy</h2></br>
10 Positional Accuracy <b>x</b> = +/- <b>$[Positional_Accuracy_x] m</b>
11 Positional Accuracy <b>y</b> = +/- <b>$[Positional_Accuracy_y] m</b>
12 ]]>
13 </text>
14 </BalloonStyle>
15 </Style>
16 <Placemark>
17 <name>POI 1</name>
18 <styleUrl>#Positional Accuracy</styleUrl>
19 <ExtendedData>
20 <Data name="Positional_Accuracy_x">
21 <value>1</value>
22 </Data>
23 <Data name="Positional_Accuracy_y">
24 <value>0.5</value>
25 </Data>
26 </ExtendedData>
27 <Point>
28 <coordinates>-4.109864,50.371102,0</coordinates>
29 </Point>
30 </Placemark>
31 </Document>
32 </kml>

Figure 76: Using text styling to improve the uncertainty information visualisation

In Figure 76 any descriptive text can be used as a standardised style with the value being input
using the $[Positional_Accuracy_x] notation.

To include information about data type in KML a schema is required the method for this is shown in
Figure 77 lines 19-30. This changes the naming convention in the styling of the pop up but allows
for a display name, as well as a schema name (Figure 77 lines 12-14 & 22, 25 & 28). This allows
for a free text name such as ‘positional accuracy’ to be shown on screen, whereas within the code
the value is stated as ‘x’, ‘y’ and ‘z’; which might be the output from an automated process. The
main benefit of including a schema is that this specifies the data type of each value for use in other
software. Google Earth and Google Maps do not use the schema, just display the values.

AGIS_SD014-02 UNCLASSIFIED Page 49 of 93


UNCLASSIFIED Technical Annex

1 <kml xmlns="http://www.opengis.net/kml/2.2">
2 <<Document>
3 <name>Positional_Uncert_Schema</name>
4 <open>1</open>
5
6 <Style id="Positional_Uncertainty">
7 <BalloonStyle> <!-- Style of the pop up -->
8 <text>
9 <![CDATA[
10 <h2>Positional Uncertainty</h2>
11 <br/>
12 $[PositionalAccuracy/x/displayName] = +/- <b> $[PositionalAccuracy/x] </b> <br/>
13 $[PositionalAccuracy/y/displayName] = +/- <b> $[PositionalAccuracy/y] </b> <br/>
14 $[PositionalAccuracy/z/displayName] = +/- <b> $[PositionalAccuracy/z] </b>
15 ]]>
16 </text>
17 </BalloonStyle>
18 </Style>
19 <!-- Schema for Positional Accuracy -->
20 <Schema name="PositionalAccuracy" id="PositionalAccuracyID">
21 <SimpleField type="double" name="x">
22 <displayName><![CDATA[Positional Accuracy x]]></displayName>
23 </SimpleField>
24 <SimpleField type="double" name="y">
25 <displayName><![CDATA[Positional Accuracy y]]></displayName>
26 </SimpleField>
27 <SimpleField type="double" name="z">
28 <displayName><![CDATA[Positional Accuracy z]]></displayName>
29 </SimpleField>
30 </Schema>
31 <Placemark>
32 <name>POI 1</name>
33 <styleUrl>#ISO_19115_Positional_Uncertainty</styleUrl>
34 <ExtendedData>
35 <SchemaData schemaUrl="#PositionalAccuracyID">
36 <!-- Placemark's positional accuracy values -->
37 <SimpleData name="x">0.5</SimpleData>
38 <SimpleData name="y">1.3</SimpleData>
39 <SimpleData name="z">0.0</SimpleData>
40 </SchemaData>
41 </ExtendedData>
42 <Point>
43 <coordinates>-4.109864,50.371102,0</coordinates>
44 </Point>
45 </Placemark>
46 <Placemark>
47 <name>POI 2</name>
48 <styleUrl>#ISO_19115</styleUrl>
49 <ExtendedData>
50 <SchemaData schemaUrl="#PositionalAccuracyID">
51 <SimpleData name="x">2.5</SimpleData>
52 <SimpleData name="y">0.3</SimpleData>
53 <SimpleData name="z">0.0</SimpleData>
54 </SchemaData>
55 </ExtendedData>
56 <Point>
57 <coordinates>-4.108864,50.372102,0</coordinates>
58 </Point>
59 </Placemark>
60 </Document>
61 </kml>

Figure 77: KML schema included with uncertainty information

AGIS_SD014-02 UNCLASSIFIED Page 50 of 93


UNCLASSIFIED Technical Annex

The final method allows for uncertainty data to accompany the feature in the KML file but Google
Earth does not process it and only displays the location of the feature (Figure 78).

1 <kml xmlns="http://www.opengis.net/kml/2.2">
2 <Document>
3 <name>Positional Uncertainty</name>
4 <Placemark>
5 <name>POI</name>
6 <!-- Imported schema requires use of namespace prefix -->
7 <ExtendedData xmlns:uncert="http://uncert_schema.com">
8 < uncert:positionalAccuracy_x>1.4</uncert:positionalAccuracy_x>
9 < uncert:positionalAccuracy_y>2.0</uncert:positionalAccuracy_y>
10 < uncert:positionalAccuracy_z>0.0</uncert:positionalAccuracy_z>
11 </ExtendedData>
12 <Point>
13 <coordinates>-4.109864,50.371102,0</coordinates>
14 </Point>
15 </Placemark>
16 </Document>
17 </kml>

Figure 78: Uncertainty information accompanying the feature but not processed by Google
Earth

For the best interoperability, including a schema as shown in Figure 77 is the best method,
however, this does increase the file length. Whereas, Figure 75 or Figure 76 displays the same
information in Google Earth, using fewer lines of code. The simpler methods might be acceptable
for the basic user. The difference in file size is very small and would only become noticeable when
dealing with a very large number of files. It is therefore the recommendation of this report, to
include a schema as this maintains data type for use in other software and also encompasses the
ability to decode value headers that might be produced in automatic processes.

AGIS_SD014-02 UNCLASSIFIED Page 51 of 93


UNCLASSIFIED Technical Annex

F. Appendix F: Metrication of Visualisation Techniques


The following three sections of this Technical Annex present the academic work on the metrication
of visualisation techniques. They provide a detailed discussion and analysis of the visual variables
experiment and include the raw experimental data together with an explanation of the statistical
techniques used by the analysis.

F.1 Introduction

One of the primary objectives of the Task 5 work package was to obtain a set of metricated results
relating to the best methods for portraying uncertainty information. Due to the size of the research
topic as a whole, an experiment was developed which focussed on one particular aspect of
uncertainty. This experiment aimed at providing options to a user for possible visualisation
techniques and tests user preference for displaying spatial data uncertainty. The primary outcome
of the experiment and its development is the overall method used to assess preference, which
could be implemented in further investigations within the visualisation of uncertainty context.

The primary basis study for this experiment was that by MacEachren et al. (2012) whereby the
changing of visual variables to portray uncertainty information was investigated. In addition to the
standard visual variables, a number of iconic representations were also included. Overall, it was
found that these iconic representations were deemed to be more intuitive, but the abstract visual
variable methods lead to quicker judgements. One important aspect of the MacEachren et al.
(2012) study was that they did not investigate all the uncertainty types individually, but instead
used three higher level types: accuracy, precision, and trustworthiness. The reasoning behind this
was to make the experiment practical, as investigating the nine uncertainty types they identified
would have resulted in a much more complex experiment. These three uncertainty types were
assessed over the three data components: spatial, temporal, and attribute.

There are a number of possible methods for assessing aspects relating to visualisation and
uncertainty. Such methods include the use of online trials (Boukhelifa et al. (2012), Kubíček and
Šašinka (2011) and Idris et al. (2011)), direct user trials (Karnick et al. (2010) and Edwards and
Nelson (2001)), and workshops and focus groups (Tomaszewski and MacEachren et al. (2012)
and Pappenberger et al. (2013)). Each of these methods has strengths and weaknesses including
the type and depth of data that can be collected and the possible reach of the experiment in terms
of participants. The wide variety of research methods highlights that the concept of uncertainty
spans not only domains but also the types and implementations of data. As such, consideration
needs to be placed on a number of aspects before implementing an experiment including the
sample that will be used to represent the population and constraints in terms of time for collection
and analysis of data.

A key document with respect to the analysis of data captured from experiments that aim to assess
the preferences of people is the ISO 8587 Standard (ISO, 2006). This standard details a
methodology for analysing ranked data based on sensory perception and suggests a number of
methods for analysis based on the number of ‘treatments’ and the aim of the test. For the
comparison where the order of products is unknown (the products are compared as opposed to the
assessors) and where more than two treatments are present then the use of the Friedman test is
recommended.

F.2 Methodology

The methods used for the development and implementation of the experiment and the analysis of
data were identified through evaluation of previous studies and recommendations made by defined
standards. The process involved for this experiment consisted of a number of stages:

1. Identification of uncertainty types


2. Identification of visual variables to be altered

AGIS_SD014-02 UNCLASSIFIED Page 52 of 93


UNCLASSIFIED Technical Annex

3. Development of a visualisation portfolio of all visualisations


4. Undertake assessment to select the most suitable visualisations for portrayal in the
experiment
5. Design of experimental procedure
6. Development of the experimental system
7. Performance of experiment and collection of data
8. Analysis of results and presentation of findings.

The experiment itself presents users with different visualisations depicting uncertainty in geospatial
features. All uncertainty values contained within these features are artificially generated. Three
datasets were created for use within the experiment:

 Trees (point features)


 Roads (line features)
 Parking areas (polygon features)

The tree and parking datasets were developed through digitisation of aerial imagery (from Google
Maps), and the road dataset was acquired from the Ordnance Survey’s ITN network. The
background datasets depicting the land and sea were also digitised from Google Map’s Street
dataset. These datasets can be seen in Figure 79. Methods were implemented to attempt to make
the data more realistic by basing uncertainty values on the amount of uncertainty in the previous
feature of the dataset, which is likely to be similar to real data. The values produced were fixed to
be between 0 and 100 to represent the percentage of ‘certainty’ in the particular uncertainty item.
For example, a value of 90 in the accuracy attribute would indicate that the feature was fairly
accurate, whereas a value of 10 in credibility would show that the feature was not very credible. In
real data, the uncertainty information may not be a percentage value, and in the case of accuracy
the uncertainty is often represented using mean and standard deviation values. Methods could be
used to convert these raw uncertainty values into the 0-100 scale as to make understanding and
representation simpler, although the raw uncertainty information should be maintained for more in
depth analysis.

Figure 79: Feature datasets

AGIS_SD014-02 UNCLASSIFIED Page 53 of 93


UNCLASSIFIED Technical Annex

F.2.1 Identification of Uncertainty Types

Due to a number of restrictions (in terms of time and resources) it was decided that a selection of
uncertainty types should be included in the experiment as opposed to all of them. In the study by
MacEachren et al. (2012) the accuracy, precision and trustworthiness uncertainty types were
investigated and considered to be higher level aspects that covered a number of lower level
uncertainty types. In the study undertaken here, the accuracy and precision types are also
implemented. The trustworthiness type has been replaced with credibility (seen in the MacEachren
et al. study as part of trustworthiness) as this is a concept that is easier to understand and is
already widely used, and the vagueness uncertainty type has been introduced. Each of these types
will now be discussed in greater detail.

F.2.1.1 Accuracy

The spatial accuracy of a feature can be seen as how much error is introduced into the position
given by the measuring tool as influenced by effects that vary over space and time. A classic
example of spatial accuracy is that of GPS locations. If there are a number of satellites visible
which are spread out over the sky then the GPS fix of the locating device will be fairly accurate. If
on the other hand there are few satellites visible that are ‘clustered’ together, the accuracy will be
particularly low. This means that any position stored by the device at this location will likely not be
the actual location the device is at. This is often seen in track logs where the user enters a building
and a ‘spiky ball’ feature appears in the resulting visualisation (Figure 80). This is due to the logger
storing a position that is inaccurate and as such does not represent the actual location of the
device at that point in time.

Figure 80: GPS trace

F.2.1.2 Precision

The precision of a spatial feature refers to how much certainty can be placed in its position based
on measuring parameters of where the feature was digitised from, or where the data is stored. If
the feature was created from orthographic imagery that had a resolution of one metre, then the
precision of the feature would be at best one metre. This is because the person digitising the data
would need to draw the feature across the pixels and so decide as to whether the pixel represents
the feature or not. In some cases, the boundary of the pixel may actually match the boundary of
the feature and so the drawing would match well. In most instances however the border will lie

AGIS_SD014-02 UNCLASSIFIED Page 54 of 93


UNCLASSIFIED Technical Annex

somewhere in the pixel and so if the feature is drawn through the centre, the actual boundary is
likely to be to one side of the feature (±0.5m).

As an example, Figure 81 shows a building polygon being created from an orthographic image. In
the left hand image, the outlines are clear as the image is of high resolution, meaning that the
polygon can be drawn relatively precisely. On the right image however, the resolution is much
courser and so the boundary more difficult to identify. This means that if a vertex was placed at the
centre of each pixel where the perceived boundary changed, the resulting shape can be seen as
being ‘out’ by a certain amount (generally the size of each pixel). If you were to draw a buffer
around the feature at this distance, it is likely that most of the actual vertices would be found within
it (as can be seen in the comparison in the image).

Figure 81: Spatial precision

In addition to the idea of resolution of an image used for digitisation, the actual storage of data has
an impact on the level of precision. For example, if a database stored positional information in
Eastings and Northings, but only to an integer value, then the precision of the value would likely be
to the nearest metre. If the latitude and longitude values were stored, then the precision would be
to the nearest degree (which could be as large as 111,000 metres). If this was the case, then the
precision of the value could be up to 111,000 metres which is obviously a very large amount of
uncertainty. The most likely case where precision is altered based on the data structure is where
the number of decimal points is altered (i.e. a number is rounded up to two decimal places).

AGIS_SD014-02 UNCLASSIFIED Page 55 of 93


UNCLASSIFIED Technical Annex

F.2.1.3 Vagueness

In attribute data, vagueness refers to the uncertainty relating to whether a feature is correctly
classified. For example, if a building consists of apartments above some shops and it is classified
as commercial it could be said that there is some vagueness present when this classification is
used for analysis. For spatial data, the same principle applies for identifying whether a particular
location in space can be seen as being part of (or located on/within) the feature. If a park is
recorded and depicted as a point feature and then a pin placed on the map, you could only be sure
that that location would be in the park if it landed exactly on the point. If it was represented as a
polygon whose edges matched a fence surrounding the park, then if the pin landed inside this
polygon we could say that it was in the park, else it would be outside of it (Figure 82, top). Also, if
polygons overlap then vagueness increases as it is unclear as to which of the polygons the point is
associated to. This can be seen in Figure 82 (bottom) where a subset of the park has been marked
a children’s play area. If the play area is simply drawn over the park then it is unclear whether a
point inside that area is part of the park or part of the play area. To negate this, the polygons would
be drawn so that they do not overlap (i.e. the park is clipped) so then if the point is in the play area
feature, it is not also within the park feature. Sometimes however the first representation may still
be needed as in reality, the play area is part of the park and so the point is actually still within the
park as a whole.

Figure 82: Spatial vagueness; feature extent (top) and overlapping features (bottom)

F.3 Credibility

Credibility refers to how much surety should be placed that the data is correct and a true
representation of reality. This is not to say that credibility is based on the amount of other forms of
uncertainty in the data. On the contrary, credibility can be seen in a way as the uncertainty relating

AGIS_SD014-02 UNCLASSIFIED Page 56 of 93


UNCLASSIFIED Technical Annex

to the uncertainty information. As an example, if two datasets were acquired and one described as
having a high level of uncertainty and one described a low level of uncertainty, intuition tells us that
the data with the least uncertainty should be used. However, if the data stated as being uncertain
is from a highly credible source (they are trusted to provide correct information) and the one with
low uncertainty is from a low credibility source (one that we cannot be sure provides correct
information) then the choice of dataset becomes more difficult. On one hand, we can be fairly sure
that the uncertainty relating to the credible source is accurate and so can be accounted for in
analysis. On the other hand, the uncertainty information from the low credibility source may itself
be subject to a large amount of uncertainty and so highly erroneous, as well as the possibility of
miss/dis-information.

F.3.1 Identification and Creation of Visualisations

A number of different visualisation techniques have been identified in previous studies. As in the
study by MacEachren et al. (2012), this study focuses on the alteration of visual variables. As
identified from the literature, these include:

 Colour hue
 Colour saturation
 Colour intensity
 Transparency
 Size
 Grain (hatch and dotted line)
 Location (displacement)
 Sketchiness
 Shape (point only)
 Fuzziness (focus)
 Resolution

In addition to these items, three additional representations were added based on availability in
current tools and an initial thesis that they could be used to portray uncertainty:

 Star rating (point only)


 Traffic light symbol (point only)
 Jitter

In addition, two forms of colour hue were implemented: one where the colours are random and
discrete and one where the colour used is based on a red-yellow-green colour ramp. Again, this
was implemented due to method availability and knowledge that the red to green colour ramp is
often used in a number of implementations and so may be more familiar.

To create all of the visualisations discussed, three software packages were used:

 Quantum GIS13 (GIS software)


 Processing14 (scripting based graphical renderer)
 Paint.Net 15 (graphical production and editing software)

Quantum GIS (QGIS) is an open source GIS that provides a huge amount of functionality, with the
addition of being able to create custom scripts using the Python language. In this project, the
visualisations were created using standard representations and through the creation of custom

13
www.qgis.org/
14
www.processing.org/
15
www.getpaint.net/index.html

AGIS_SD014-02 UNCLASSIFIED Page 57 of 93


UNCLASSIFIED Technical Annex

icons (for the traffic light and star rating representations). QGIS could not (at the time of conducting
this study) be used to create the visualisations where the structure of the feature itself changed,
such as the sketchy rendering and changing of resolution. For the simpler of these representations
(fuzziness, resolution and jitter) Paint.Net was used to manipulate the individual classifications of
features using various filters and adjustment tools. This was done by exporting each classification
of features (i.e. all lines with an accuracy of 60-80) using SQL and QGIS export facility. These
were then imported into new layers of the Paint.Net canvas and the varying levels of modifications
applied. The final implementation method used the Processing graphical scripting tool which is
based on the Java programming language. By writing a custom GML parser, the system was able
to read any dataset exported from QGIS as a GML file and then have the special adjustments
placed on the representations. For the sketchy rendering representation, the Handy Renderer
library16 was implemented which recursively draws geometric primitives in a way that resembles
human drawing. The location method was accomplished by using mathematical algorithms to
determine a distance and direction for translation of the feature based on the level of uncertainty,
and then drawing this displaced feature along with the location of the feature in the dataset. At this
current stage, there is no one software package that can implement all of these representations.
Therefore, if any of these non-standard GIS representations are deemed to be useful, then it may
be of benefit to develop custom scripts within the software used to render them. Table 1 shows
which system was used to create each of the visual variable implementations.

GIS Processing Paint.Net


Colour hue Location Fuzziness
Colour saturation Sketchiness Resolution
Colour intensity Jitter
Transparency
Size
Grain
Shape
Star rating*
Traffic lights*
* Methods required the creation of custom icons for use in the GIS
package
Table 1: Visualisation generation tools

The visual variables can generally be split into two groups based on whether the topology of the
feature is maintained or altered as part of the visualisation. When topology is maintained
(topologically consistent), the feature appears to be in the same location and has the same
dimensions whether uncertainty is visualised or not. In the case of non-maintenance of topology
(topologically inconsistent) the location and/or dimensions of the feature changes when uncertainty
is portrayed. This difference can be seen in Figure 83 where features are depicted without
uncertainty portrayal, and when intensity (consistent) and sketchiness (inconsistent) are
implemented to depict this information. If a point was to be placed on the map in the same location
of each representation, on the sketchy features it could be mistaken that the point is on a very
erratically shaped feature in the real world. In the intensity representation however the point is
clearly not on the feature. Although not an explicit aspect investigated in this study, the concept
should be considered when representations are used for depicting features as it may lead to
unexpected decision making. A list of the types of representations and whether they are
topologically consistent or inconsistent can be seen in Table 2.

16
http://gicentre.org/handy/

AGIS_SD014-02 UNCLASSIFIED Page 58 of 93


UNCLASSIFIED Technical Annex

Figure 1: Topologically consistent and inconsistent representations


Figure 83: Topologically consistent and inconsistent representations

Topologically consistent Topologically inconsistent

Hue Size

Saturation Location

Intensity Sketchiness

Transparency Shape

Grain Fuzziness

Traffic light symbology Resolution

Star rating symbology Jitter

Table 2: Topologically consistent and inconsistent representations

F.3.1.1 Uncertainty classification

Based on feedback from the AGIS Task 6 Workshop, it was decided to use a classified
representation scale as opposed to a continuous one when determining how to represent the
varying levels of uncertainty. This means that rather than presenting each feature based on the
exact uncertainty value, the features are classified into a group (based on the uncertainty value)

AGIS_SD014-02 UNCLASSIFIED Page 59 of 93


UNCLASSIFIED Technical Annex

whose representation is fixed for all features contained within it (see Figure 84). The method has
some advantages and disadvantages. The main disadvantages are that the selection of where to
place boundaries could have an effect on the representation and as such there are a number of
different classification methods including fixed interval (as used in this study), natural breaks, and
standard deviations. Also, two features that actually have very similar uncertainty values may be
represented quite differently if they happen to fall either side of a classification boundary. The main
advantages however are that it may be easier to determine a level of uncertainty of the feature as
there are a limited number of representations, and examples of all of these can be portrayed in the
legend. Also, using this method may stop users attempting to put an exact uncertainty value to the
feature which in some cases is beneficial (i.e. all that is important is a general interpretation) or
detrimental (it adds additional uncertainty in the interpretation).

0 100

Continuous

0 20 40 60 80 100

Discrete

Figure 84: Continuous vs. discrete classification

To enable consistent representations, the uncertainty values used are in the form of percentages,
whereby 0% indicates high amounts of uncertainty and 100% is absolute certainty in the data. In a
real world implementation, the representations of uncertainty would vary between types and
implementations. For example, it would likely be the case that accuracy would be stored in the
form of mean and standard deviation values, and precision as a direct indication of measurement
precision (i.e. 0.5 metres). However, for a generic visualisation approach these values should be
standardised by pre-defined methods (i.e. the standard deviation as a percentage of the data
value) to give the single 0-100 value that can be easily represented.

F.3.1.2 Representations

Based on the visual variable and the feature type, a number of representations have been
developed. Each representation format has merits and disadvantages which can often be
dependent on other aspects of the data and visualisation as a whole. For example, although the
red-green colour ramp is often used for portraying risk and danger information (and so may be
inherently easier to interpret), it means that colour cannot be used to represent other aspects of the
data (such as road classification). With transparency, the method of making features transparent is
seldom used as a method of portraying information and so its usage would likely not interfere with
the portrayal of other information. However, as a feature becomes more transparent it is more
difficult to distinguish such information, so this could be seen as a disadvantage (though in some
instances this could be a merit to the representation as it makes the uncertain features more
difficult to interpret). An example of the visualisations can be seen in Table 3.

AGIS_SD014-02 UNCLASSIFIED Page 60 of 93


UNCLASSIFIED Technical Annex

Visualisation Uncertainty Level (%)


Method 100-80 80-60 60-40 40-20 20-0

Colour Hue

Colour Hue
(Red – Green)

Colour
Intensity

Colour
Saturation

Transparency

Grain

Size

Shape

Fuzziness /
Blur / Focus

Resolution

Jitter

Sketchiness

AGIS_SD014-02 UNCLASSIFIED Page 61 of 93


UNCLASSIFIED Technical Annex

Visualisation Uncertainty Level (%)


Method 100-80 80-60 60-40 40-20 20-0

Location

Traffic Light

Star Rating

Table 3: Examples of features using different visualisation methods

F.3.2 Experimental Design

To collect information regarding preference of visualisations, an experiment was conducted at the


AGIS Task 5 & 6 demonstration at JAGO HQ (Dennison Barracks, Hermitage) on the 25 th April,
2013. This was deemed the most appropriate setting for the collection of data due to the large
presence of military personnel who would ultimately be the end beneficiaries of recommendations
generated by this work package. However, this posed a number of restrictions on the type of study
that can be conducted due to:

 A limited number of possible test subjects


 Restrictions in terms of time available for both the collection and analysis of results
 A full representation of the ‘type’ of user may not be present

Many of the experiments conducted in previous studies have had a large number of test
participants and a relatively large amount of time for collection and analysis, but in this experiment
both items are restricted. This means that a number of the methods used in those studies could not
be directly applied to this experiment. As such, it was decided that the experiment should take the
form of a low-level preference investigation in that users would be asked to identify which of the
visualisations they prefer over the others for particular uncertainty and feature types. Based on the
ISO 8587 standard for sensory testing (ISO, 2006), it was identified that a ranking approach would
be the most suitable whereby the users would be presented with a number of visualisations and
then asked to rank them in order of preference. To reduce the amount of visualisations that each
user would be asked to assess, an initial assessment was carried out with the aim of selecting the
four most useful visualisation techniques for depicting uncertainty from each feature-uncertainty
type combination. This assessment was carried out at a half-day assessment session at the DGC.

F.3.3 DGC Assessment Workshop

Due to the limited amount of time and the nature of the ranking experiment, it was decided that the
full list of visualisations should be reduced to a level that would allow for an appropriate statistical
assessment. For analysis purposes, a selection of four visualisations for each uncertainty and
feature type was deemed suitable. To determine which visualisations would be used in the
experiment a preliminary assessment was undertaken on all possible visualisations against all
combinations of feature and uncertainty type. Some methods were not applicable to all features
(i.e. the shape variable could only be applied to point features) and so these were not implemented
across all combinations. In total, 164 visualisations were produced for the initial assessment stage

AGIS_SD014-02 UNCLASSIFIED Page 62 of 93


UNCLASSIFIED Technical Annex

and collated into a single portfolio. This assessment was conducted with the help of the DGC via a
half day assessment session. In this session the complete portfolio of the maps developed was
presented to the attendees. The maps were presented in groups based on the uncertainty type and
feature type and participants of the assessment event decided as a group which four should be
deemed the most useful. The visualisation techniques selected for use in the experiment can be
seen in Table 4.

Uncertainty Type & Form

Accuracy Precision Credibility Vagueness

Polygon

Polygon

Polygon

Polygon
Point

Point

Point

Point
Line

Line

Line

Line
Visualisation
Colour hue (green to red)

Focus

Grain

Intensity

Jitter

Resolution

Size (large to small)

Size (small to large)

Transparency

Table 4: Visualisations identified in the DGC workshop

The maps generated all used the same datasets between groups (i.e. all visual representations in
the precision-point groups used the same uncertainty values as other representations in the
precision-point group), with the exception of the iconic representations (traffic lights and stars).
These representations required that the number of features presented was reduced as the large
area covered by each icon would obscure other close by features. This in itself could be a problem
for real-life implementation and so if these methods were to be used, it may be of benefit to provide
a more interactive interface whereby they appear when the feature is clicked. From this
assessment however, neither of these representations were selected as being the most useful and
so they do not appear in the experimental phase.

During the assessment, a number of observations were made and feedback generated. One key
observation was that the more abstract visualisation methods (sketchy and displacement) were
often almost immediately rejected. The decision to reject these methods was often made even
before viewing the other maps and so suggests that in general, these methods are not liked.
Feedback as to why these were rejected often suggested that the methods looked unprofessional,
were difficult to distinguish between the levels of uncertainty, and in the case of the displacement
visualisation it was simply too confusing or implied that the displaced feature was the actual
location.

AGIS_SD014-02 UNCLASSIFIED Page 63 of 93


UNCLASSIFIED Technical Annex

A number of additional pieces of feedback were provided revolving around the general
implementation of the visualisations. One aspect that was mentioned was how the design of the
map itself can have an impact on the usefulness of the visualisations. One example provided was
in relation to the colour scale variable whereby presenting the visual on a map that used a classic
green depiction for land against a blue one for water meant that the uncertain features stood out
much more than the more certain ones (the green features blended in with the green background).
In the same way, it was also highlighted that for the intensity variable, the white features (and so
the uncertain ones) stood out more than the more ‘intense’ features which depicted a higher level
of certainty. In addition, where colour is used there may be issues with people who are colour blind
and so the actual user needs to be considered.

With regards to the grain visualisation for the line feature type (dotted lines), this was often
highlighted as being a more traditional visualisation technique that was already in use. As such it
was often identified as being particularly useful. In addition, the colouring method which used the
red-yellow-green (r-y-g) colour ramp was widely identified as being useful, as was the transparency
method.

Overall it is clear that although the most useful visualisations were selected from the portfolio, there
are several aspects that could alter how effective these actually are. Some issues arise from how
the variables interact with other features and the background map, and others from issues such as
the scale of the map and how many features are in the environment. The assessment has also
highlighted that although some visualisations are suitable in one instance, they may not be as
useful for different uncertainty of feature types. However, from the counts presented in Table 5 and
Table 6, it can be said that at this initial assessment stage, the variables of colour ramp (r-y-g),
focus, grain and transparency appear to be the most transferrable between the different
uncertainty and feature types. This may suggest that these variables provide a more general
impression of uncertainty which could be advantageous for future implementation.

Variable Point Line Polygon Total


Colour (green to red) 4 4 2 10
Focus 4 3 3 10
Transparency 2 3 3 8
Grain - 4 3 7
Resolution 1 0 2 3
Size (small to large) 1 1 1 3
Size (large to small) 2 0 1 3
Intensity 2 0 0 2
Jitter 0 1 1 2
Table 5: Visualisation count against feature type

Variable Accuracy Precision Vagueness Credibility Total

Colour (green to red) 2 3 2 3 10

Focus 2 2 3 3 10
Transparency 3 0 2 3 8

Grain 2 2 2 1 7
Resolution 0 1 2 0 3

Size (small to large) 1 0 0 2 3

AGIS_SD014-02 UNCLASSIFIED Page 64 of 93


UNCLASSIFIED Technical Annex

Variable Accuracy Precision Vagueness Credibility Total

Size (large to small) 0 2 1 0 3


Intensity 1 1 0 0 2

Jitter 1 1 0 0 2

Table 6: Visualisation count against uncertainty type

F.3.3.1 Experimental System

A bespoke system was developed to present the visualizations to the user and to collect their
preferences. This system was developed in C# with all results being stored in a standalone
database. After the initial development a pilot study was conducted using a number of participants
(staff and students of the NGI in Nottingham University) with three primary objectives:

1. To ensure stability of the system and collection of data


2. To obtain feedback from users as to possible improvements to the interface
3. To collect test data that could be used to verify the suitability of statistical tests

Preliminary runs from the pilot study indicated that the system was stable and collected data as
expected. Feedback from users identified a few minor issues with text, with the addition of a ‘Back’
button being suggested by a number of participants. These modifications were implemented in an
improved system which was tested to ensure that the data was still being collected and system
stability remained intact.

The main flow of the system can be seen in Figure 85. When the system begins, a new trial can be
started by clicking on a button. This then opens some explanatory text for the subject detailing
what the trial is about and providing some very brief information about the types of uncertainty that
will be assessed. The screen also provides the ability of presenting more in depth descriptions
about the uncertainty which opens in a new window. The same extended uncertainty information is
available during the trials by clicking on the name of the uncertainty type on the assessment
screen. Once the user is happy with the explanations, they proceed with the trial and a screen to
collect some basic information is opened. This screen collects information about the organization
the user is associated with, their role within the organization, and what level of geospatial
information user they consider themselves to be. Once entered, the assessment part of the
experiment begins. To reduce any learning effects that appear in the data due to extended use of
the visualisations, the order that the uncertainty types, feature types, and visual variable are
randomised. However, the feature types will always be grouped to the same uncertainty type. For
example, in one assessment the order may be Accuracy (Point, Line, Polygon), Precision (Line,
Polygon Point), Vagueness (Point, Line, Polygon) and finally Credibility (Line, Point, Polygon),
whereas in a second it may be Vagueness (Polygon, Point, Line), Accuracy (Point, Polygon, Line),
Credibility (Polygon, Line, Point) and finally Precision (Point, Line, Polygon). Within each of these,
the position of the four visual variables is randomised (the same items in A, B, C and D for one
group in a trial may appear in the positions of D, B, A and C in another trial). The assessment
process itself presents the user with the four visualisations in a small format (Figure 86), but by
clicking on the smaller images a larger window is opened containing the more detailed
representation along with a legend. The user then ranks the visualisations based on how effective
they are at representing the specific uncertainty for the feature type. A visualisation cannot be
ranked the same as another in the same group, and all visualisations must be given a rank before
the user can proceed to the next assessment. Upon completing the final assessment a thank you
message is displayed along with two buttons which allow the user to end the trial or to return to the
last assessment in case of user error. When the trial is ended, the start screen is displayed again
ready for the next trial to begin.

AGIS_SD014-02 UNCLASSIFIED Page 65 of 93


UNCLASSIFIED Technical Annex

Start

Starting
Map popup
screen

Introductory User Thank you


Assessment
text information screen

Uncertainty
type
information

Figure 85:1:
Figure Program flow
Program flow

Figure 86: Main assessment window

AGIS_SD014-02 UNCLASSIFIED Page 66 of 93


UNCLASSIFIED Technical Annex

F.3.4 Results

In total, 16 participants took part in the experiment. Based on feedback regarding user level this
group consisted of ten geospatial specialists, two advanced users, and four basic users.
Participants were from a number of organisations, as can be seen in Table 7.

Organisation Count

JFIG 3
RSMS 3

DGC 2

UKHO 2

ICSP 1

DSTL 1

Met Office 1

University 1

NCHQ 1

GeoCore 1

Table 7: Participant organisations

AGIS_SD014-02 UNCLASSIFIED Page 67 of 93


UNCLASSIFIED Technical Annex

Uncertainty Type

Accuracy Precision Credibility Vagueness


Point
Feature Type
Line
Polygon

Table 8: Response ranks by uncertainty and feature type

AGIS_SD014-02 UNCLASSIFIED Page 68 of 93


UNCLASSIFIED Technical Annex

The ISO 8487 Standard (ISO, 2006) describes that when the order of products is unknown (it is the
treatments that are being compared and not the subjects) and there are more than two products in
the comparison, a Friedman test should be used to determine any preferences between the
treatments. It further suggests the use of the Least Significant Difference (LSD) test as the post-
hoc testing method for identifying between which treatments the differences are found if the
Friedman test has indicated as such. However, as discussed in Meier (2006) and Williams and
Abdi (2010), the LSD test can introduce a number of type I statistical errors (false positives) as it
does not explicitly correct for multiple observations. Therefore, individual Mann-Whitney U tests
were performed between the treatments (visualisation techniques) with a Holm-Bonferroni
correction procedure implemented to account for the multiple observations.

F.3.5 Test Results

The data collected for all uncertainty-feature type groups were analysed using the above
mentioned methods whereby initial testing was performed using the Friedman test, and then post-
hoc analysis using the Mann-Whitney test with the Holm-Bonferroni correction applied. The
hypotheses for the Friedman testing are:

H0 = There are no differences in preference between the visualisation methods


implemented for portraying uncertainty information;

H1 = There are preferences between at least two of the visualisation methods implemented
for presenting uncertainty information.

The hypotheses for all Mann-Whitney tests are:

H0 = There are no differences in preference between visualisation method A and


visualisation method B for portraying uncertainty information;

H1 = There are preferences between visualisation method A and visualisation method B for
portraying uncertainty information.

All Friedman tests were performed using an alpha level of 5%, with 16 blocks and 4 treatments,
with 3 degrees of freedom. Mann-Whitney tests were performed at the alpha level of 1% using a
two-tailed test with 16 observations for each group. These Mann-Whitney tests were all corrected
for multiple observations using the Holm-Bonferroni correction. As with all statistical testing, the
aim is to identify if there is sufficient evidence to reject the null hypothesis (H 0) in favour of the
alternative hypothesis (H1). If a difference was found using the Mann-Whitney test, then the order
of preference (i.e. A is preferred over B or B is preferred over A) is determined based on visual
analysis of boxplots representing the data. All data and boxplots can be found in Appendix H.

F.3.5.1 Initial Testing of Groups

The initial testing on the groups to identify preferences was conducted using the Friedman test.
Results indicated that at the 5% alpha level there was sufficient evidence to reject the null
hypothesis for the groups:

 Accuracy – point, line and polygon


 Precision – line and polygon
 Vagueness – point and polygon
 Credibility – point and polygon

F.3.5.2 In-Depth Analysis between Treatments

After the identification of the groups that showed difference in rankings, further post-hoc analysis
was performed using the Mann-Whitney test to identify which visualisation methods were ranked

AGIS_SD014-02 UNCLASSIFIED Page 69 of 93


UNCLASSIFIED Technical Annex

differently. These preferences are depicted using a series of charts which group the visualisations
based on whether preferences were identified in the Mann-Whitney tests. If a box spans multiple
visualisations, this indicates that they can be seen as being equally preferable to each other. As an
example, Figure 87 shows such a diagram. In this, A can be seen as equal in preference to C (the
blue box spans both A and C columns), but more preferable than B or D. C on the other hand is
equally preferable to B as well as A, and is more preferable than D. B can only be seen as being
equally preferable to C, and is less preferable than A but more preferable than D. Finally, D is not
equal in preference to any other item (the box only spans the column for D), and is less preferable
than A, B and C.

A C B D

Figure 87: Equality of preference example

For the groups of accuracy and credibility in point features, although the Friedman test identified
that there was some difference in the rankings, when further analysis was performed no
differences were found. The difference in rankings for the other groups can be seen in the following
diagrams.

Accuracy in Line Features

When portraying accuracy of line features, it was found that the use of the red-green colour ramp
was preferable over the use of transparency. No other preferences were found.

Colour (red to green) Size (small to large) Grain Transparency

Colour (red to green)

Size (small to large)

Grain

Transparency

Figure 88: Accuracy in line features preference

Accuracy in Polygon Features

For the portrayal of accuracy in polygon features, the use of transparency was found to be
preferred over all other methods. Jitter was found to be the least preferable with all other methods
being ranked higher.

Transparency Focus Grain Jitter

Transparency

Focus

AGIS_SD014-02 UNCLASSIFIED Page 70 of 93


UNCLASSIFIED Technical Annex

Grain

Jitter

Figure 89: Accuracy in polygon features preference

Precision in Line Features

In the portrayal of precision for line feature group, results indicated that both the red to green
colour ramp and grain methods were preferred over the use of jitter.

Colour (red to green) Grain Focus Jitter

Colour (red to green)

Grain

Focus

Jitter

Figure 90: Precision in line features preference

Precision in Polygon Features

When portraying the precision in polygon features, the only significant result was that the use of
the red-green colour ramp is preferable to using the large-small size method.

Colour (red to green) Grain Resolution Size (large to small)

Colour (red to green)

Grain

Resolution

Size (large to small)

Figure 91: Precision in polygon features preference

Vagueness in Point Features

For portraying Vagueness in point features, it was found that the use of the red-green colour ramp
and large-small size are preferable to the use of resolution. No other preferences were identified in
the data.

Colour (red to green) Size (large to small) Focus Resolution

Colour (red to green)

Size (large to small)

Focus

Resolution

Figure 92: Vagueness in point features preference

AGIS_SD014-02 UNCLASSIFIED Page 71 of 93


UNCLASSIFIED Technical Annex

Vagueness in Polygon Features

When portraying the vagueness of polygon features, the use of transparency was found to be
preferable over all other visualisation methods assessed. In addition, all visualisation methods
were found to be more preferable than the use of resolution for depicting the vagueness
information.

Transparency Focus Grain Resolution

Transparency

Focus

Grain

Resolution

Figure 93: Vagueness in polygon features preference

Credibility in Polygon Features

The final group with preferences identified is the credibility of polygon features. In this group, it was
found that the use of the red-green colour ramp is preferable over the focus and small to large size
methods. No other preferences were found.

Colour (red to green) Transparency Focus Size (small to large)

Colour (red to green)

Transparency

Focus

Size (small to large)

Figure 94: Credibility in polygon features preference

F.3.6 Summary of Results

Overall, the results from the statistical analysis of data collected from the experiment can be seen
in Table 9. In this table there are three columns representing the order of choice for which
visualisation is preferred. Where more than one visualisation method is in a cell, then any of those
methods are equally preferable at that choice level. In some instances a visualisation method will
appear in multiple cells for the same uncertainty-feature type combination. This is due it being seen
equally preferable to one method which was seen as a different preference to another. In effect,
the next level of choice should only be entered if all of the options in the higher level are not
available. The items could be presented in a way that the methods are not duplicated between
cells, but this could imply that the item in the lower level choice is less preferable to all others,
which may not be the case.

It should be noted that the experiment and analysis does not aim at assessing whether a
visualisation method is good or bad. What is investigated is as to whether there are any
preferences to be found between the visualisations presented. Therefore, if no significant
differences were found between the methods this does not mean that all methods are good or bad.
What it does indicate is that there is no general preference between the usage of the visualisations
for depicting that type of uncertainty in the specific feature type. In effect, the determination of

AGIS_SD014-02 UNCLASSIFIED Page 72 of 93


UNCLASSIFIED Technical Annex

‘good’ visualisation methods conducted using the initial DGC assessment, and this experiment
aimed at identifying if there were preferences between those ‘good’ visualisations.

1st choice 2nd Choice 3rd Choice


Colour Hue (green-red)
Focus
Point
Intensity
Accuracy

Transparency
Colour Hue (green-red) Grain
Line Grain Size (small-large)
Size (small-large) Transparency
Grain
Polygon Transparency Jitter
Focus
Colour Hue (green-red)
Focus
Point
Intensity
Precision

Size (large-small)
Colour Hue (green-red)
Focus
Line Focus
Jitter
Grain
Colour Hue (green-red) Grain
Polygon Grain Resolution
Resolution Size (small-large)
Colour Hue (green-red)
Focus
Point Focus
Resolution
Vagueness

Size (large-small)
Colour Hue (green-red)
Focus
Line
Grain
Transparency
Grain
Polygon Transparency Resolution
Focus
Colour Hue (green-red)
Focus
Point
Size (small-large)
Transparency
Credibility

Colour Hue (green-red)


Focus
Line
Grain
Transparency
Focus
Colour Hue (green-red)
Polygon Size (small-large)
Transparency
Transparency

Table 9: Results of experiment

F.4 Summary & Conclusions

The aim of this investigation was to identify whether there were preferences as to which
visualisation method should be used for portraying uncertainty information in a spatial context. To
accomplish this, an experiment was performed that collected data from users in the form of

AGIS_SD014-02 UNCLASSIFIED Page 73 of 93


UNCLASSIFIED Technical Annex

preference rankings for a number of visualisations against uncertainty type and feature type
groups.

The first stage of the study was to identify the uncertainty types and visualisation techniques that
would be assessed. A decision was made to use low level visualisation in the form of altering the
visual variables that can be used to portray information. The list used assimilated classical types
(i.e. colour, saturation and focus) along with more novel approaches such as sketchiness and jitter.
The uncertainty types selected were accuracy, precision, vagueness and credibility. Accuracy and
precision were selected due to their use in the MacEachren et al. (2012) study, along with
credibility which can be seen as a sub-component of trustworthiness which was also implemented
in that study. The type of vagueness was included as this appears to be a major component in
decision making in terms of knowing whether a particular location belongs to one feature (or area)
or another.

After identification of the uncertainty types and visualisation methods, a portfolio of 164 maps
depicting uncertainty was generated. These were taken to a workshop at the DGC where the top
four were selected for each uncertainty-feature type group. These maps were those included in the
final experiment conducted. As well as determining which representations would be assessed, a
number of additional findings were obtained from the meeting. These included the observation that
often the more abstract and novel visualisation were rejected in favour of more classical
representations. Other comments included the importance of the map as a whole for determining
the most suitable representation (the red-green colour ramp did not work very well for linear
features drawn on a green background) and the amount of other data in the visualisation (the star
rating method would quickly become too cluttered). Recommendations were also obtained with
regards to changes that could be made to the representations such as drawing borders around
points that use transparency, and the possible inversion of the saturation. Also, consideration
regarding the number of classifications applied and how the boundaries for these are decided was
also highlighted, although such aspects are outside of the scope of this study.

The experiment itself was conducted at the AGIS Task 5 and 6 demonstration day. Data was
collected from the attendees and analysed using a series of statistical tests to identify whether
users preferred one visualisation technique over another. Results showed that in some cases no
preference was found, in a number of them some preferences were found in that some methods
were preferable over another, and in two the results indicated that there was an overall preferred
method. A general summary of these findings is shown in Table 10:

Uncertainty Feature
Findings
type type
Accuracy Line Red-green colour ramp is preferable to the transparency
method
Polygon Transparency is the most preferable, followed by grain and
focus (joint), with jitter being the least preferable
Precision Line Red-green colour ramp and grain are more preferable than
jitter
Polygon Red-green colour ramp is preferable to size (large-small)

Vagueness Point Red-green colour ramp and size (large-small) are


preferable to resolution
Polygon Transparency is the most preferable, followed by focus and
grain (joint) with resolution being the least preferable
Credibility Polygon Red-green colour ramp is preferable over focus and size
(large-small)
Table 10: Summary of preferences

AGIS_SD014-02 UNCLASSIFIED Page 74 of 93


UNCLASSIFIED Technical Annex

A number of observations can also be made about the preferences identified. Interestingly, in the
two instances where one method was found preferable over all others, this method was
transparency. Also, in both those instances the use of the red-green colour ramp did not feature in
the maps presented. When this method was presented and preferences identified in the data, in all
instances the colour ramp was ranked preferable over at least one other visualisation method.
Additionally, in all instances where the jitter method was presented, it was out-ranked by at least
one other method.

A major concept that should be noted about this study is that the information gathered is in terms
of user preference and does not reflect whether the techniques are actually better at allowing the
user to make the correct decision about the data presented. As such, any significant findings
simply suggest that of the visualisation methods used, a difference has been found in the number
of people who preferred one visualisation over another. So even though it can be said that for
portraying vagueness in polygon data the use of transparency is preferable, this does not imply
that it should be used without further understanding as to whether it actually is of benefit in the
interpretation of the information.

In addition, from information collected about the background of the users and their perceived level
of geospatial data user, it was found that over half of the sample population were specialist users.
By specialist user it is meant that their role focuses heavily on the use and interpretation of
geospatial data and as such may have a different outlook as to what is preferable in a
representation. There is not enough data from this study to suggest whether this is indeed the
case, but it should certainly be considered before any implementation of findings.

Overall, although this study has been conducted to identify preferences in methods for portraying
uncertainty, the primary contribution is that of the methods used to identify these preferences. As
discussed, an early decision was made to investigate the low level visual variable aspect for
portraying information. Obviously there are higher level aspects such as the presentation of text
and graphs, as well as 3D representations and other sensory inputs such as audio. The study
could have compared such high level aspects to obtain information regarding whether people
preferred the use of graphs over audio, but there would be a huge number of variables that could
affect such decisions. For example, if graphs were identified as being more preferable than audio,
is this because of the type of graph used, the colours used, the sounds used for the audio or the
method of audio delivery. Without thorough investigation it would be highly possible that in fact, the
audio method overall is more preferable, but that specific implementation is not suitable. The
methods used in this investigation aim at assessing what alters the perception of each individual
high level aspect. Indeed, the visual variables used to represent a feature are but one low level
aspect that contributes to the overall effectiveness of a map. Other aspects include the levels of
classification, delivery method, use of a legend, and the scale of the map to name a few. Only by
investigating what methods in each of these make a preferable representation can the higher level
aspects be fully assessed.

Due to the nature of the experiment, the results obtained should not be seen as a fully indicative
evaluation of the best methods for portraying uncertainty. As such, there are several caveats that
should be imposed on these results:

 Analysis has only been performed on the spatial data component, and so the findings
cannot be applied to the attribute and temporal components;
 Only four uncertainty types out of the full list have been investigated, meaning that the
findings cannot be transferred to all uncertainty cases;
 The sample used for the study was relatively small and so may not be fully indicative of the
full population, as well as limiting the power of statistical testing performed;
 A large proportion of test subjects can be seen as specialist users, and so the results may
not transfer fully to instances where basic or advanced users are using the visualisations;

AGIS_SD014-02 UNCLASSIFIED Page 75 of 93


UNCLASSIFIED Technical Annex

 As a fairly low level study of visual variables, there are a number of higher level delivery
methods that may be more beneficial which have not been investigated and so the method
of altering the visual variable may not be the most effective;
 Data collected is in the form of personal preferences and not in performance, meaning that
although a visualisation method may be identified as preferable, it may not be effective in
terms of successfully calculating uncertainty.

F.5 Further Investigations

This study can be seen as a small contribution to a much larger area of investigation that should be
investigated before any decisions are made in terms of implementation. As discussed, this
experiment has investigated but one low level aspect out of many that could have an impact on
how effectively a user can interpret uncertainty information from a portrayal. Although the results
have indicated that there are indeed preferences in terms of which visual variable should be
altered in some instances to portray uncertainty information, this experiment did not have the
resources in terms of time and access to test subjects to complete a thorough investigation.
However, the methods used can easily be implemented in other similar studies that would
contribute to the visualising uncertainty topic as a whole, such as what type of graphs are preferred
if used, does the delivery method affect perception of uncertainty in data, and is a particular style of
textual information preferred over others. By investigating such aspects, a more thorough ‘picture’
can be developed.

It is recommended that to identify what components can affect the usefulness of a visualisation
(scale, delivery, graphs etc.) a series of workshops should be conducted to identify the high level
aspects and then to identify the lower level aspects that contribute to each of these. Once these
have been identified, similar experiments to those presented in this report can be conducted to
determine the optimal representations.

In addition to the further investigations with regards to high and low level aspect identification, the
results from this experiment can be extended by performing an experiment to identify whether the
methods do in fact alter performance. Such experiments would require a larger number of
participants and more time to collect and analyse results but would be highly beneficial in terms of
any findings.

From general feedback it has also been found that although a low-level aspect has been
investigated, there are still a number of other factors that could alter how useful each of the
methods presented are. For example, only one representation of each visual variable was
implemented (i.e. only one colour hue for the intensity and saturation variables) but it may well be
the case that there are a number of sub factors (i.e. would using a different colour alter the
perceived usefulness of the intensity and saturation variables). Also, the visualisation as a whole
may impact on the usefulness of each variable and so should there be a pre-defined set of base
representations that can be accounted for when determining what visual variable to use? Finally,
the users themselves may have an impact on the usefulness of the visualisation. These differences
could be cultural (the classic example being red where in a number of cultures it portrays danger
and is negative, yet in others it portrays luck and is positive) or physical (i.e. colour blindness) but
should be considered before any final implementation of the visualisation.

On a higher level, this experiment has only investigated a limited number of uncertainty types for
the spatial data component. As such, the other types of uncertainty should also be investigated as
well as the temporal and attribute data components to determine whether there are different
preferences in these to those investigated in this experiment.

AGIS_SD014-02 UNCLASSIFIED Page 76 of 93


UNCLASSIFIED Technical Annex

G. Appendix G: Statistical Testing


As mentioned, two statistical tests were used in the analysis of the results obtained from the
experiment. These are the Friedman Ranked Sum and the Mann-Whitney U test.

Before discussing these tests it is important to clarify a number of terms relating to statistical
testing. Firstly, the term alpha (α) is used to denote the probability value (P) that is deemed
significant. The P-value is basically a probability representation on the chances that the values
analysed could have occurred at random, or where two samples are compared, it represents the
probability that the two samples are from the same population. The alpha value is the cut off point
for where it is said that the chances of the values occurring at random or are from the same
population is not possible. So if an alpha value of 1% (or 0.01) is declared, only those test statistics
that have a probability value of less than 1% are deemed as significant.

G.1.1.1 Friedman Ranked Sum

The Friedman Ranked Sum test is a non-parametric test (it does not assume that the data is from
a specific distribution) which works by ranking results from ‘blocks’ against ‘treatments’. In the case
of this experiment, the ‘blocks’ are the individual assessments and the ‘treatments’ are the
visualisation methods. As the test works by using ranks as opposed to the raw data, this test is
ideal for the results from this experiment as the data itself is already in a ranked format. Also, due
to the data being ordinal as opposed to continuous, a non-parametric test should be used.

The test statistic for the Friedman test (normally referred to as S) is obtained by squaring the sum
of the ranks in each treatment and then modifying this value based on the number of blocks and
treatments. If there are no differences between the ranks then the resultant test statistic will be
small due to the squaring of sums being balanced. If there are differences, then the statistic will be
higher due to the difference.

For example, if 15 groups of 3 treatments were analysed, and the ranking of treatments was the
same between all groups (i.e. one treatment always ranked first, one always second and one
always third), then the sum of the ranks in the treatments would be 15, 30 and 45. Squaring these
values gives 225, 900 and 2025, with an overall sum of 3150. If these rankings were completely
random however, we may find that the rankings are completely balanced (equal number of each
rank in each treatment) which would give sums of 30, 30 and 30. These values when squared give
900, 900 and 900 which when summed provides a value of 2700. When the rest of the algorithm is
applied to these values, we would get 30.105 and 0.09 for our test statistics which are obviously
considerably different. To determine statistical significance, these values are compared to those
provided in the Friedman Test Statistic tables depending on the number of blocks and treatments,
and on the alpha level that we wish to perform the test at. If we say that (as in the case of the
analysis performed in the main experiment) the alpha value is 5%, then for analysis of data with
three treatments and 15 blocks, we would require an S value of greater than 6.400. Obviously, the
first test statistic (30.105) is larger than this and so can be seen as being significant, and the
second value (0.09) is smaller and so cannot be seen as significant. Again, the term significance
refers to whether it can be said that the values obtained could have occurred at random or not –
significant results come from those where it is highly unlikely that the values occurred at random
and some ‘structure’ can be seen in the data.

The Friedman test however only tests to see if there is some difference between the treatments – it
does not indicate which treatments the differences occur in. Therefore, if the Friedman test
identifies such differences, further post-hoc testing is required. In the case of this experiment, this
is done using the Mann-Whitney U test (also known as a Wilcoxon Ranked Sum test).

AGIS_SD014-02 UNCLASSIFIED Page 77 of 93


UNCLASSIFIED Technical Annex

G.1.1.2 Mann-Whitney / Wilcoxon Ranked Sum

The Mann-Whitney (also known as the Wilcoxon Ranked Sum) test is used to compare two groups
of data to see if one group is from a different population than the other. It is again non-parametric
and so does not assume an underlying distribution to the data. This makes the test ideal for this
experiment as it can be performed on ordinal as opposed to continuous data.

For the Mann-Whitney test, the test statistic (generally referred to as U) is determined by putting
the data from both groups together and ranking each item. In the case of tied items (the
occurrence of which is inevitable in the main experiment of this report) the average rank for all
items with the same value is used. For example, if the combined data as 1, 2, 3, 3, 3, 3, 4 then the
resultant ranking values would be 1, 2, 4.5, 4.5, 4.5, 4.5, 7. These values are then put back into the
original groups and the values summed. The smaller of the two summed values is the W value
used for the test.

For this example, consider two groups that comprise of ranked values.

A 1 1 1 1 1 1 1 1

B 2 2 2 2 2 2 2 2

Figure 95: Consistently ranked values

In this example, when the two groups are put together and ranked, A has a rank sum of 36 and B
has a rank sum of 100. Therefore, the W value is 36. When compared with the statistical tables two
considerations are needed. Firstly, whether the test is two-tailed or one tailed is important. A one
tailed test is one where the alternative hypothesis assumes a direction of difference i.e. group B
has higher ranks than group A. A two tailed test on the other hand considers both directions i.e.
group A has either higher or lower ranks than group A. For this test (and in the main experiment) a
two tailed test is performed with an alpha value of 1% (P=0.01). Using the tables, when there are 8
observations in both groups, a significant W at the 1% alpha is 43, meaning that as the lowest
value from our ranked sums (36) is lower than this we can say that there is significant differences
between the groups.

A 1 2 1 2 1 2 1 2

B 2 1 2 1 2 1 2 1

Figure 96: Randomly ranked values

If the values were randomly assigned and balanced, the rank sums for both groups would be 68.
Therefore, using the same significant value from the table (W=43) we can conclude that there is
insufficient evidence to say that there is a difference between the groups as our smallest rank sum
is not smaller than the critical W value.

G.1.1.3 Holm-Bonferroni Correction

One problem with the statistical testing of the data collected is that if a significant result is obtained
from the Friedman test, each of the treatments is then compared with multiple other treatments.
Due to these multiple comparisons, it is more likely that Type I errors (false-positives) will occur.
This is because when testing is performed at (for example) the 5% alpha level, it would be
expected that one out of every 20 tests comprising of random data would be identified as
significant (a false positive). As each treatment will be tested three times there is more opportunity
for these false positives to occur for that treatment. As such, a process known as Holm-Bonferroni
Correction can be applied to the data. This works by modifying the alpha value used in a test

AGIS_SD014-02 UNCLASSIFIED Page 78 of 93


UNCLASSIFIED Technical Annex

based on the number of multiple observations present. In the case of this experiment, when the
Friedman test shows significant results there will be six Mann-Whitney tests performed. The test
itself is performed as normal and the W-value does not change. In the analysis performed on the
experiment, the base alpha value used is 1% (P=0.01). The Holm-Bonferroni approach involves
ordering the test values from the smallest P-value to the largest. The Holm-Bonferroni corrected
alpha value is then said to be the base alpha divided by the position of the test when ordered by
the P-value. In the case of the smallest P-value test result in the group (within this experiment), the
corrected alpha value would be 0.0017 (0.01 / 6). The next smallest would have a corrected alpha
of 0.002 (0.01 / 5) and only the largest P-value result would be tested to the original alpha value of
0.01.

It should be noted however that although by correcting the alpha values the number of type I errors
(false positives) are reduced, there is the chance that this is at the cost of some type II errors (false
negatives). This means that although there will be less cases where a preference is found where it
does not exist, there may be some cases where preferences that do exist are not identified. It is
generally preferable however to have this type of scenario rather than providing recommendations
based on preferences that do not actually exist.

AGIS_SD014-02 UNCLASSIFIED Page 79 of 93


UNCLASSIFIED Technical Annex

H. Appendix H: Experiment Data and Analysis


The data collected from the experiment is:

Rank (count)
Uncertainty Feature Variable 1 2 3 4
Colour (red to green) 9 3 3 1
Focus 5 0 1 10
Point
Intensity 0 9 3 4
Transparency 2 4 9 1
Colour (red to green) 10 3 1 2
Grain 1 7 2 6
Accuracy Line
Size (small to large) 5 4 1 6
Transparency 0 2 12 2
Focus 2 4 10 0
Grain 3 7 3 3
Polygon
Jitter 0 1 3 12
Transparency 11 4 0 1
Colour (red to green) 7 2 3 4
Focus 2 3 2 9
Point
Intensity 1 9 4 2
Size (large to small) 6 2 7 1
Colour (red to green) 10 3 2 1
Focus 2 4 5 5
Precision Line
Grain 3 9 2 2
Jitter 1 0 7 8
Colour (red to green) 10 3 2 1
Grain 2 6 7 1
Polygon
Resolution 3 4 2 7
Size (large to small) 1 3 5 7
Colour (red to green) 7 5 1 3
Focus 4 3 5 4
Point
Resolution 0 1 8 7
Size (large to small) 5 7 2 2
Colour (red to green) 8 4 0 4
Focus 3 4 2 7
Vagueness Line
Grain 3 2 10 1
Transparency 2 6 4 4
Focus 2 3 10 1
Grain 3 7 4 2
Polygon
Resolution 0 2 2 12
Transparency 11 4 0 1
Colour (red to green) 8 4 1 3
Focus 3 1 2 10
Point
Size (small to large) 4 7 3 2
Transparency 1 4 10 1
Colour (red to green) 7 4 2 3
Focus 4 2 2 8
Credibility Line
Grain 1 4 7 4
Transparency 4 6 5 1
Colour (red to green) 10 5 0 1
Focus 1 3 5 7
Polygon
Size (small to large) 2 2 4 8
Transparency 3 6 7 0

Table 11: Raw data from Visual Variables experiment

AGIS_SD014-02 UNCLASSIFIED Page 80 of 93


UNCLASSIFIED Technical Annex

H.1 Friedman Test Results

The Friedman testing was performed with an alpha level of 5% (α=0.05) with 16 blocks (n=16) and
four treatments (k=4). With these parameters, the significant S value is 7.800 (any value obtained
from the test which is larger than this is deemed significant).

The notation used in the Friedman test result tables is:

 S – the Friedman test statistic value


 P – the probability value (P-value) that the data within all of the treatments comes from the
same population (there is no difference in the rankings between visualisation methods)

All significant results are highlighted in bold.

Feature type

Point Line Polygon

Accuracy S=8.18 S=9.69 S=24.52


Uncertainty type

P=0.043 P=0.022 P=0.000


Precision S=5.33 S=16.57 S=12.07
P=0.149 P=0.001 P=0.007
Vagueness S=11.63 S=3.52 S=23.48
P=0.009 P=0.318 P=0.000
Credibility S=8.32 S=5.47 S=17.70
P=0.040 P=0.140 P=0.001
Table 12: Visual variables Friedman test results

H.2 Mann-Whitney Test Results

All significance testing is applied to the 1% alpha level (α=0.01), although the Holm-Bonferroni
correction has applied which is what the comparison is made to (corrected α). Therefore in the
case of accuracy in point features the P-value for Colour (red to green) against Intensity resulted in
a test statistic with a p-value of 0.007, although without correction (α=0.01) this value is significant,
with correction (α=0.0017) it is not significant.

In the tables, the following notation is used:

 α – the alpha value that identifies significance


 W – the Mann-Whitney test statistic value
 P – the probability value (P-value) of the samples being from the same population (no
difference between the rankings)

All significant results are highlighted in bold within the tables. Again, significance testing is
performed against the corrected alpha value (Corrected α column in the tables).

AGIS_SD014-02 UNCLASSIFIED Page 81 of 93


UNCLASSIFIED Technical Annex

H.2.1 Accuracy in Point Features

Comparisons Test values Corrected

Colour (red to green) Intensity W=195 0.0017


P=0.0070
Colour (red to green) Focus W=201 0.002
P=0.0117
Colour (red to green) Transparency W=204 0.0025
P=0.0179
Transparency Focus W=306.5 0.0033
P=0.0984
Focus Intensity W=286.5 0.005
P=0.3792
Transparency Intensity W=265.5 0.01
P=0.9679

Table 13: Mann-Whitney test results for techniques showing accuracy in point features

Accuracy in Point Features

Colour (red to green)

Focus

Intensity

Transparency

1 2 3 4
Rank

Figure 97: Box plots for techniques showing accuracy in point features

AGIS_SD014-02 UNCLASSIFIED Page 82 of 93


UNCLASSIFIED Technical Annex

H.2.2 Accuracy in Line Features

Comparisons Test values Corrected

Colour (red to green) Transparency W=177 0.0017


P=0.0006
Colour (red to green) Grain W=189.5 0.002
P=0.0035
Colour (red to green) Size (small to large) W=217.5 0.0025
P=0.0633
Transparency Size (small to large) W=238 0.0033
P=0.3131
Grain Size (small to large) W=284.5 0.005
P=0.4269
Grain Transparency W=238 0.01
P=0.5600

Table 14: Mann-Whitney test results for techniques showing accuracy in line features

Accuracy in Line Features

Colour (red to green)

Grain

Size (small to large)

Transparency

1 2 3 4
Rank

Figure 98: Box plots for techniques showing accuracy in line features

AGIS_SD014-02 UNCLASSIFIED Page 83 of 93


UNCLASSIFIED Technical Annex

H.2.3 Accuracy in Polygon Features

Comparisons Test values Corrected

Transparency Jitter W=380 0.0017


P=0.0000
Focus Jitter W=163 0.002
P=0.0001
Grain Jitter W=177 0.0025
P=0.0005
Focus Transparency W=349 0.0033
P=0.0007
Grain Transparency W=335 0.005
P=0.0045
Focus Grain W=280 0.01
P=0.5324

Table 15: Mann-Whitney test results for techniques showing accuracy in polygon features

Accuracy in Polygon Features

Focus

Grain

Jitter

Transparency

1 2 3 4
Rank

Figure 99: Box plots for techniques showing accuracy in polygon features

AGIS_SD014-02 UNCLASSIFIED Page 84 of 93


UNCLASSIFIED Technical Annex

H.2.4 Precision in Line Features

Comparisons Test values Corrected

Colour (red to green) Jitter W=165 0.0017


P=0.0001
Grain Jitter W=179.5 0.002
P=0.0010
Colour (red to green) Focus W=188.5 0.0025
P=0.0032
Colour (red to green) Grain W=214.5 0.0033
P=0.0488
Focus Grain W=309 0.005
P=0.0786
Focus Jitter W=223.5 0.01
P=0.1077

Table 16: Mann-Whitney test results for techniques showing precision in line features

Precision in Line Features

Colour (red to green)

Focus

Grain

Jitter

1 2 3 4
Rank

Figure 100: Box plots for techniques showing precision in line features

AGIS_SD014-02 UNCLASSIFIED Page 85 of 93


UNCLASSIFIED Technical Annex

H.2.5 Precision in Polygon Features

Comparisons Test values Corrected

Colour (red to green) Size (large to small) W=174 0.0017


P=0.0005
Colour (red to green) Resolution W=194.5 0.002
P=0.0063
Colour (red to green) Grain W=199.5 0.0025
P=0.0112
Grain Size (large to small) W=210 0.0033
P=0.0345
Grain Resolution W=237.5 0.005
P=0.3092
Resolution Size (large to small) W=248 0.01
P=0.5367

Table 17: Mann-Whitney test results for techniques showing precision in polygon features

Precision in Polygon Features

Colour (red to green)

Grain

Resolution

Size (large to small)

1 2 3 4
Rank

Figure 101: Box plots for techniques showing precision in polygon features

AGIS_SD014-02 UNCLASSIFIED Page 86 of 93


UNCLASSIFIED Technical Annex

H.2.6 Credibility in Point Features

Comparisons Test values Corrected

Colour (red to green) Focus W=200 0.0017


P=0.0111
Focus Size (small to large) W=324.5 0.002
P=0.0183
Colour (red to green) Transparency W=208.5 0.0025
P=0.0307
Focus Transparency W=315.5 0.0033
P=0.0430
Size (small to large) Transparency W=220 0.005
P=0.0820
Size (small to large) Colour (red to green) W=239.5 0.01
P=0.3408

Table 18: Mann-Whitney test results for techniques showing credibility in point features

Credibility in Point Features

Colour (red to green)

Focus

Size (small to large)

Transparency

1 2 3 4
Rank

Figure 102: Box plots for techniques showing credibility in point features

AGIS_SD014-02 UNCLASSIFIED Page 87 of 93


UNCLASSIFIED Technical Annex

H.2.7 Credibility in Polygon Features

Comparisons Test values Corrected

Colour (red to green) Focus W=166 0.0017


P=0.0001
Colour (red to green) Size (small to large) W=173 0.002
P=0.0004
Colour (red to green) Transparency W=197 0.0025
P=0.0076
Transparency Focus W=330 0.0033
P=0.0098
Size (small to large) Transparency W=329 0.005
P=0.0115
Focus Size (small to large) W=260 0.01
P=0.8876

Table 19: Mann-Whitney test results for techniques showing credibility in polygon features

Credibility in Polygon Features

Colour (red to green)

Focus

Size (small to large)

Transparency

1 2 3 4
Rank

Figure 103: Box plots for techniques showing credibility in polygon features

AGIS_SD014-02 UNCLASSIFIED Page 88 of 93


UNCLASSIFIED Technical Annex

H.2.8 Vagueness in Point Features

Comparisons Test values Corrected

Size (large to small) Resolution W=353.5 0.0017


P=0.0005
Colour (red to green) Resolution W=181 0.002
P=0.0013
Focus Resolution W=212.5 0.0025
P=0.0415
Colour (red to green) Focus W=229 0.0033
P=0.1770
Focus Size (large to small) W=296.5 0.005
P=0.2113
Colour (red to green) Size (large to small) W=254 0.01
P=0.7044

Table 20: Mann-Whitney test results for techniques showing vagueness in point features

Figure 104: Box plots for techniques showing vagueness in point features

AGIS_SD014-02 UNCLASSIFIED Page 89 of 93


UNCLASSIFIED Technical Annex

H.2.9 Vagueness in Polygon Features

Comparisons Test values Corrected

Resolution Transparency W=378 0.0017


P=0.0000
Resolution Grain W=175 0.002
P=0.0004
Focus Transparency W=351.5 0.0025
P=0.0005
Focus Resolution W=179 0.0033
P=0.0007
Transparency Grain W=334.5 0.005
P=0.0048
Focus Grain W=293.5 0.01
P=0.2445

Table 21: Mann-Whitney test results for techniques showing vagueness in polygon features

Figure 105: Box plots for techniques showing vagueness in polygon features

AGIS_SD014-02 UNCLASSIFIED Page 90 of 93


UNCLASSIFIED Technical Annex

I. Appendix I: Colour Ramp Statistics


I.1 Friedman Test

The Friedman test provides a test statistic of 19.46. At the 95% confidence interval (5% alpha
level) with n=21, a critical value is 7.800. Therefore, at the 5% alpha level it can be said that there
is sufficient evidence to reject the null hypothesis and accept the alternative hypothesis (there is a
difference in the rankings).

Preference of Colour Ramps

Blues

Blue to Red

Traffic Light

Greyscale

1 2 3 4
Rank

Figure 106: Box plots of the colour ramps against ranking by survey participants

I.2 Mann-Whitney Test

Due to the Friedman test signifying that there was some difference in rankings, Mann-Whitney
tests were conducted between the colour ramp types. Using the Holm-Benferroni correction
method for multiple comparisons, significant preference was found in all cases for the traffic light
implementation but not for any other colour ramp. This means that the traffic light method was
preferred over all other methods, but there is no statistical difference present in the survey results
between the other three implementations.

Comparisons Test values Corrected


Blue to Red Traffic Light W=627.5 0.0017
P=0.0000
Traffic Light Greyscale W=310.0 0.002
P=0.0002
Blues Traffic Light W=585.5 0.0025
P=0.0004
Blue to Red Greyscale W=517.0 0.0033
P=0.0848
Blues Blue to Red W=409.5 0.005
P=0.2731
Blues Greyscale W=475.0 0.01
P=0.5465
Table 22: Mann-Whitney test results for colour ramp preference

AGIS_SD014-02 UNCLASSIFIED Page 91 of 93


UNCLASSIFIED Technical Annex

J. List of References
1 AKLI-ASTOUATI, K., GUEBAILI, R. & MOKHTARI, A. 2011. Geographic metadata: Data quality, uncertainty
and imprecision. International Journal of Reasoning-based Intelligent Systems, 3, 164-172.
2 BOUKHELIFA, N., BEZERIANOS, A., ISENBERG, T. & FEKETE, J. D. 2012. Evaluating sketchiness as a
visual variable for the depiction of qualitative uncertainty. IEEE Transactions on Visualization and Computer
Graphics, 18, 2769-2778.
3 CASTILLA, G. & HAY, G. J. 2007. Uncertainties in land use data. Hydrology and Earth System Sciences, 11,
1857-1868.
4 DEPARTMENT OF DEFENCE. 2008. Common Warfighting Symbology. MIL-STD-2525C
5 DUCKHAM, M. & WORBOYS, M. 2005. An algebraic approach to automated geospatial information fusion.
International Journal of Geographical Information Science, 19, 537-557.
6 EDWARDS, L. D. & NELSON, E. S. 2001. Visualizing data certainty: A case study using graduated circle
maps. Cartographic Perspectives, 38, 19-36.
7 HARDY, D., FREW, J. & GOODCHILD, M. F. 2012. Volunteered geographic information production as a
spatial process. International Journal of Geographical Information Science, 26, 1191-1212.
8 HEUVELINK, G. B. M., BROWN, J. D. & VAN LOON, E. E. 2007. A probabilistic framework for representing
and simulating uncertain environmental variables. International Journal of Geographical Information Science,
21, 497-513.
9 IDRIS, N. H., JACKSON, M. J. & ABRAHART, R. J. 2011. Credibility Labelling in Map Mash-up DesignL
Influence on Users' Judgement. EuroSDR/ISPRS Workshop on Web Cartography. Lund, Sweden.
10 ISO 2006. 8587 Sensory analysis - Methodology - Ranking. Switzerland.
11 KARNICK, P., CLINE, D., JESCHKE, S., RAZDAN, A. & WONKA, P. 2010. Route visualization using detail
lenses. IEEE Transactions on Visualization and Computer Graphics, 16, 235-247.
12 KUBÍČEK, P. & ŠAŠINKA, C. 2011. Thematic uncertainty visualization usability - Comparison of basic
methods. Annals of GIS, 17, 253-263.
13 LASKEY, K. B., WRIGHT, E. J. & DA COSTA, P. C. G. 2010. Envisioning uncertainty in geospatial
information. International Journal of Approximate Reasoning, 51, 209-223.
14 MACEACHREN, A. M., ROTH, R. E., O'BRIEN, J., LI, B., SWINGLEY, D. & GAHEGAN, M. 2012. Visual
semiotics & uncertainty visualization: An empirical study. IEEE Transactions on Visualization and Computer
Graphics, 18, 2496-2505.
15 MEIER, U. 2006. A note on the power of Fisher's least significant difference procedure. Pharmaceutical
Statistics, 5, 253-263.
16 MOITRA, A., BARNETT, B., CRAPO, A. & DILL, S. J. 2010. Addressing uncertainty and conflicts in cross-
domain data provenance. In, 2010 San Jose, CA. 912-917.
17 PAPPENBERGER, F., STEPHENS, E., THIELEN, J., SALAMON, P., DEMERITT, D., VAN ANDEL, S. J.,
WETTERHALL, F. & ALFIERI, L. 2013. Visualizing probabilistic flood forecast information: Expert
preferences and perceptions of best practice in uncertainty communication. Hydrological Processes, 27,
132-146.
18 REUSCHEL. A-K. HURNI. L. 2011. Mapping literature: Visualisation of spatial uncertainty in fiction. The
Cartographic Journal, 48 no.4, 293-308
19 SKARLATIDOU, A., HAKLAY, M. & CHENG, T. 2011. Trust in Web GIS: The role of the trustee attributes in
the design of trustworthy Web GIS applications. International Journal of Geographical Information Science,
25, 1913-1930.
20 TOMASZEWSKI, B. & MACEACHREN, A. 2012. Geovisual analytics to support crisis management:
Information foraging for geo-historical context. Information Visualization, 11, 339-359.
21 WILLIAMS, L. J. & ABDI, H. 2010. Fisher’s Least Significant Difference (LSD) Test. Encyclopedia of
Research Design, Sage Publications, Thousand Oaks, CA.
22 WOOD, J., ISENBERG, P., ISENBERG, T., DYKES, J., BOUKHELIFA, N. & SLINGSBY, A. 2012. Sketchy
rendering for information visualization. IEEE Transactions on Visualization and Computer Graphics, 18,
2749-2758.

AGIS_SD014-02 UNCLASSIFIED Page 92 of 93


UNCLASSIFIED Technical Annex

K. Abbreviations and Acronyms

AGIS Advanced Geospatial Information and Intelligence Services


C2BL Command and Control Battle Lab
CGTS Common Geospatial Tool Set
DGI Defence Geospatial Information
DGIWG Defence Geospatial Information Working Group
DI ICSP Defence Intelligence – Intelligence Capability Strategy & Policy
GEOINF Geospatial Information
GEOINT Geospatial Intelligence
GI2 Geospatial Information and Intelligence
GIS Geographic Information Systems
GUI Graphical User Interface
HCI Human Computer Interaction
HUMINT Human Intelligence
ISO International Organization for Standardization
JISC Joint Information Systems Committee
KML Keyhole Markup Language
MGMP MOD Geospatial Metadata Profile
NATO North Atlantic Treaty Organisation
NCHQ Navy Command Head Quarters
NEO Non-combatant Evacuation Operation
NGA National Geospatial Intelligence Agency
OGC Open Geospatial Consortium
RSMS Royal School of Military Survey
SDI Spatial Data Infrastructure
SIE Single Intelligence Environment
SIGINT Signals Intelligence
SKOS Simple Knowledge Organization System
SLD Styled Layer Descriptor
TTP Tactics, Techniques & Procedures
WMS Web Map Service
WFS Web Feature Service

AGIS_SD014-02 UNCLASSIFIED Page 93 of 93

S-ar putea să vă placă și