PostGIS Raster : Extending PostgreSQL for The Support of ... - CoDE

Faculté de Sciences Appliquées
Service Ingénierie Informatique
et de la Décision (CoDE)
PostGIS Raster :
Extending PostgreSQL for The Support of
Continuous Fields
Thi Ngoc Uyen DANG
Directeur : Prof. Esteban Zimányi
Mémoire présentée en vue de l’obtention
du diplôme d’Ingénieur Informaticien
Année académique 2011–2012

Acknowledgements
I would like to thank all people for helping me in this work and also for those who gave me advice
and ideas when i faced problems. Especially, i would like to thank my promoter Esteban Zimányi for
the help, support and responsibility of which he has shown throughout the year. I also would like to
thank Alejandro Vaisman, who worked with me from start to finish.
The last thanks will go to my family, for allowing me to study, to progress and realize my own
mistakes.
i

Abstract
Nowadays, in many application domains, complex analysis tasks often require to take geographical
information into account. However, there are very few attempts to support continuous fields, a phenomena
that are perceived as having a value at each point in space and/or time. Examples of such
phenomena include temperature, altitude or land use. This thesis is intended to introduce PostGIS
Raster, a new raster data type supporting continuous fields in PostgreSQL database. Later, based
on this primitive data type, a new composite type is build, along with a set of functions to take into
account the temporal aspect. Finally, an application that allows user to exploit the temporal feature
of such raster data is also provided.
ii

Contents
Acknowledgements i
Abstract ii
Contents vi
List of Figures 1
1 Introduction 2
1.1 Context . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Goals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Organisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2 Geographic Information Systems 4
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 Elements of GIS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2.1 Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2.2 Attributes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2.3 Imagery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2.4 Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3 Geographic Data Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.3.1 Vector Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.3.2 Raster Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.4 Geographic Information Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.4.1 Map Layers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.4.2 Data Theme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.5 Georeference and Coordinate Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.5.1 Georeference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.5.2 What Is a Coordinate System? . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.6 GIS Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.6.1 Crime Mapping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.6.2 GIS and Hydrogeology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.6.3 Traditional Knowledge GIS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3 Why PostGIS Raster ? 19
3.1 Goals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.1.1 Simplicity and Complementarity . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.1.2 Transparent Integration with Vector . . . . . . . . . . . . . . . . . . . . . . . . 19
3.1.3 Storage Flexibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.1.4 Interoperability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2 PostGIS Raster Versus ORACLE GEORaster . . . . . . . . . . . . . . . . . . . . . . . 20
3.2.1 Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.2.2 Physical Storage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.2.3 Indexing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
iii

3.2.4 Overlapping Raster and Vector Layers . . . . . . . . . . . . . . . . . . . . . . . 24
3.2.5 Loading Raster Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.2.6 Intersecting Raster and Vector Layers . . . . . . . . . . . . . . . . . . . . . . . 25
4 PostGIS Raster 27
4.1 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.2 Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.3 Spatial Reference Identifier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.4 Raster Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.4.1 Usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.4.2 Cell Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.4.3 Zones and Regions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.4.4 Representing Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.5 Spatial Resolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.6 Spatial Resolution Versus Scale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.7 Bands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.8 Pixel Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.9 Nodata Values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.10 Blocks or Tiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.11 Pyramids or Overviews . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.11.1 Concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.11.2 Creating Pyramids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.12 Masks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.13 Arrangements of Raster Layers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.14 Data Formats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.15 Physical Storage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.15.1 Storing In-database Raster Versus Storing Out-database Raster . . . . . . . . . 49
4.15.2 Registering Out-database Raster . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.15.3 Storing In-database Raster . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.16 Indexing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.17 Retrieving . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.17.1 In-database Raster Retrieving . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.17.2 Out-database Raster Retrieving . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.18 Conversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.18.1 Raster to vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.18.2 Vector to raster . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.19 Intersection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.20 Temporal Raster . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.20.1 Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.20.2 Operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5 User Guide 62
5.1 Store and Manage Rasters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
5.1.1 Import Rasters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
5.1.2 Get and Set The Raster Properties . . . . . . . . . . . . . . . . . . . . . . . . . 62
5.1.3 Vector to Raster Conversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
5.1.4 Raster to Vector Conversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
5.2 Exporting Rasters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.3 Get Raster Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.4 Display Rasters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.5 Edit and Compute Rasters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
5.6 Convert Rasters to GDAL Format . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
5.6.1 Raster to TIFF File . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
5.6.2 Raster to JPEG File . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
iv

5.6.3 Raster to PNG File . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
5.7 Intersect Rasters with Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
5.7.1 Intersect Rasters with Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
5.7.2 Intersect Rasters with Lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
5.7.3 Intersect Rasters with Polygons . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
5.8 Aggregate Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.9 Create a High Resolution Analysis Grid . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.10 Create a Specialised Web or Desktop GIS Application . . . . . . . . . . . . . . . . . . 70
6 Application 72
6.1 Quantum GIS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
6.1.1 Advantages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
6.1.2 Utilisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
6.2 Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
6.2.1 Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
6.2.2 Color Legend . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
6.3 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
6.3.1 Time Travel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
6.3.2 Color Legend . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
7 Conclusion 83
Bibliography 84
Annex 87
A PostGIS Raster Utilisation 88
A.1 Dependencies Installation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
A.2 Installing and Configuring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
A.2.1 Linux . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
A.2.2 Windows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
A.3 Creating Database . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
B Python Plugin in QGIS 89
B.1 Necessary Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
B.2 Python Code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
B.2.1 __init__.py . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
B.2.2 Plugin.py . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
B.2.3 Resources.qrc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
C Code Implementation 92
C.1 Temporal Raster Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
C.2 Temporal Raster Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
D Raster Sources 95
D.1 Remote Sensing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
D.1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
D.1.2 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
D.2 Photogrammetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
D.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
D.2.2 Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
D.3 Cartography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
D.4 Digital Image Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
D.4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
D.4.2 Digital Camera Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
v

D.5 Geology, Geophysics and Geochemistry . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
E Raster Resolution 101
F Continuous Surfaces 102
vi

List of Figures
2.1 Point feature [27]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Line feature [27]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.3 Polygon feature representing parcels [27]. . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.4 Grid of cells in raster data [27]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.5 Raster data [27]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.6 Multiband raster data [27]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.7 Surface expressed by contour lines [27]. . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.8 Surface expressed by contour bands [27]. . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.9 Surface expressed by raster data [27]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.10 Surface expressed by TIN layers [27]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.11 Vector image at various scales. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.12 The original vector image is on the left side. The upper-right image illustrates magnification
of 7x as a vector image. The lower-right image illustrates the same magnification
as a raster image [29]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.13 Raster image in zooming [28]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.14 Multiple layers geographic representation. . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.15 Themes for a area [30]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.16 Georeferencing a geographic object [26]. . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.17 The location of the measured point (in red) is latitude plus 40 degrees and longitude
plus 50 degrees [26]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.18 Parallels lines of latitude and meridians line of longitude [26]. . . . . . . . . . . . . . . 14
2.19 2D Cartesian coordinate system. [26]. . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.20 3D Cartesian coordinate system [26]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.21 Different methods for map projection [26]. . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.22 Distorsion in map projection [26]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.23 Mapping of homicides in Washington D.C. [18] . . . . . . . . . . . . . . . . . . . . . . 17
2.24 Groundwater level change of the High Plains Aquifer, 1980-95 [31]. . . . . . . . . . . . 17
3.1 GeoRaster data architecture [3]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.2 PostGIS Raster data architecture. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.3 Physical storage of GeoRaster data [3]. . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.4 Physical storage of PostGIS Raster data. . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.5 Oracle GeoRaster georeference approach [14]. . . . . . . . . . . . . . . . . . . . . . . . 23
3.6 PostGIS Raster georeference approach [14]. . . . . . . . . . . . . . . . . . . . . . . . . 23
4.1 PostGIS Raster implementation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.2 Information representation using raster data [16]. . . . . . . . . . . . . . . . . . . . . . 30
4.3 Basemap raster for road data [16]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.4 Surface map raster [16]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.5 Thematic map raster [16]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.6 Tree attribute [16]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.7 On the left side: cell values applied at the center point. On the right side: cell values
applied for the whole square. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.8 Raster cell values [16]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
vii

4.9 Raster cell [16]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.10 Raster resolution [16]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.11 Pixel or cell location. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.12 Raster extent [16]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.13 Raster zones and regions [32]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.14 Discrete data represented in raster format [33]. . . . . . . . . . . . . . . . . . . . . . . 35
4.15 Raster point representation [34]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.16 Raster line representation [34]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.17 Raster polygon representation [34]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.18 Raster elevation data for a part of the province of Quebec [8]. . . . . . . . . . . . . . . 37
4.19 Spatial resolution. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.20 Raster in different spatial resolutions [11]. . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.21 Rasters at different scales [11]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.22 Same scaled rasters at different spatial resolution [11]. . . . . . . . . . . . . . . . . . . 39
4.23 Raster bands. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.24 Three main ways to display single-band raster [35]. . . . . . . . . . . . . . . . . . . . . 40
4.25 Multi-band raster [35]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.26 RGB composite from a three-band raster [35]. . . . . . . . . . . . . . . . . . . . . . . . 41
4.27 Multi-band raster image [35]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.28 Raster tiles [23]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.29 Pyramids (or Overviews) in PostGIS Raster [24]. . . . . . . . . . . . . . . . . . . . . . 43
4.30 Pyramids at different levels and scales [36]. . . . . . . . . . . . . . . . . . . . . . . . . 44
4.31 In the right side: The mask. In the left side: the mask with the transparency set to
70% overlays the digital elevation model raster [10]. . . . . . . . . . . . . . . . . . . . 45
4.32 Image warehouse of untiled and unrelated images (4 images) [6]. . . . . . . . . . . . . 45
4.33 Image warehouse of cars [22]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.34 Irregular tiled raster coverage (36 tiles))[6]. . . . . . . . . . . . . . . . . . . . . . . . . 46
4.35 Regular tiled raster coverage (36 tiles) [6]. . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.36 Rectangular regular tiled raster coverage (54 tiles) [6]. . . . . . . . . . . . . . . . . . . 46
4.37 Tiled images (2 tables of 54 tiles) [6]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.38 Raster object coverage (9 raster objects corresponding 9 rows in a raster table) [6]. . . 47
4.39 Vector to raster conversion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.40 Raster to vector conversion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.41 Image formed by the raster [17]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.42 Web application with out-database raster [23]. . . . . . . . . . . . . . . . . . . . . . . 50
4.43 Storing in-database raster. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.44 In-database raster. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.45 Conversion from raster to vector [22]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.46 Conversion from vector to raster [22]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.47 Vector and vector intersection with a resulting vector layer [22]. . . . . . . . . . . . . . 56
4.48 Raster intersection paradigm [22]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.49 Vector and raster intersection with a resulting raster layer [22]. . . . . . . . . . . . . . 57
4.50 Vector and raster intersection with a resulting vector layer [22]. . . . . . . . . . . . . . 57
4.51 Raster and raster intersection with a resulting vector layer [22]. . . . . . . . . . . . . . 58
4.52 Mutual exclusive polygons intersection with a resulting vector layer [22]. . . . . . . . . 58
4.53 Mutual exclusive vector-raster intersection with a resulting raster layer [22]. . . . . . . 58
4.54 Mutual exclusive vector-raster intersection with a resulting vector layer [22]. . . . . . . 58
4.55 Mutual exclusive raster-raster intersection with a resulting raster layer [22]. . . . . . . 59
5.1 Raster circle [17]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
5.2 Clipping raster effect [17]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
5.3 Algebra operation on rasters [23]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
5.4 Intersection of raster and geometric points [23]. . . . . . . . . . . . . . . . . . . . . . . 68
5.5 Intersection of raster and roads [23]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
viii

5.6 Intersection of raster polygons [23]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
5.7 Different temperature values in a polygon [23]. . . . . . . . . . . . . . . . . . . . . . . 70
5.8 Gird of cells of USA map [23]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
5.9 Web GIS application [23]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
6.1 QGIS interface. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
6.2 QGIS toolbars. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
6.3 PostGIS connection. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
6.4 Temporal raster window. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
6.5 Time travel interface. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
6.6 Band area. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
6.7 The temperature color legend. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
6.8 The color legend interface. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
6.9 The loading map sequential approach. . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
6.10 The loading map parallel approach. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
6.11 Maps representing world climate [7] are loaded once time. Their visibility is handled
by turning on/off the check boxes list. . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
6.12 Media area. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
6.13 Single band rasters. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
6.14 Maps representing world climate are loaded as single band rasters. They are turned
into color maps by loading a created color palette "worldclim.txt". . . . . . . . . . . . 80
6.15 Map is loaded as multi band raster. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
6.16 Visualisation of band number 2 of a multi band map. . . . . . . . . . . . . . . . . . . . 81
6.17 Maps with a color legend. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
D.1 Synthetic aperture radar image of Death Valley [37]. . . . . . . . . . . . . . . . . . . . 96
D.2 Relief map Sierra Nevada (Spain) [38]. . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
D.3 William Smith’s geologic map of England,Wales, and southern Scotland. Completed in
1815, it was the first national-scale geologic map and by far the most accurate of its
time [39]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
D.4 Age of the sea floor. Much of the dating information comes from magnetic anomalies
[40]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
E.1 Raster data represented at different resolution or cell sizes [11]. . . . . . . . . . . . . . 101
F.1 Elevation represented in raster format [33]. . . . . . . . . . . . . . . . . . . . . . . . . 102
F.2 Source-concentration surface represented in raster format [33]. . . . . . . . . . . . . . . 103
1

Chapter 1
Introduction
1.1 Context
Our daily life is full of situations that require us to make a choice among several alternative
solutions, so-called decision process. For example, we have from simple situations such as shopping,
voting to complicated situations that happen in government and business like resource management,
urban planning and regional planning. But on what do we base to make the best choice ?
Due to the complexity of many decisions in government and business that require us to think of
different elements like stakeholders and categories, it leads to the need of using software for decision
making process. Examples of such applications include environmental impact-assessment, geographic
history, population and demographic studies, criminology and military planning.
Among the elements that influence decision making, geographical information represents a relevant
source to decision makers. Whether the issue is the location of a new public or the development of
a new project, decision makers should consider such geographic information as location, character of
environment and landscapes.
Furthermore, in the private sectors, a large group of companies use geographic information as a
tool for their locational decision making. These include retail marketing chains (for example, Dayton-
Hudson, a major retail firm headquartered in Minneapolis), railroads (for example, Southern Pacific
Railroad’s land division), electric power and gas utilities, international import-export firms, transportation
and travel service organizations, publishing firms and real estate planners and investors
[19].
1.2 Goals
Not all sources of geographic information are reliable. The precision of geographic information can
be influenced indirectly through scholarly publications. These publications tend to influence decision
makers’s understanding about climate issues. So it is important to have a useful tool, such as a
geographic information system, that helps user to handle directly geographic information and thus
guarantees the correctness of information.
Most of the current geographic information systems provide a vector type for manipulating geographic
information. However, there are very few attempts to support a data type that describes the
distribution of physical phenomena that change continuously in time and space, so-called continuous
fields. Example of such phenomena are temperature, pressure, precipitation, land elevation, land use,
population density and etc.
The main contribution of this thesis is to introduce a new data type PostGIS Raster that supports
continuous fields in PostgreSQL database. In addition, it also describes an application that allows
user to exploit the temporal feature of geographic data using PostGIS Raster.
2

1.3 Organisation
This thesis is organized as follows : Section 2 starts with GIS and its applications. Section 3 describes
the goals of PostGIS Raster, along with some comparisons with the existed plugin supporting raster
Oracle GeoRaster to give users more motivations. The following section describes in detail PostGIS
Raster structure and how this one is stored, indexed, retrieved and overlapped. Section 5 provides
guides for users about utilisation of some PostGIS Raster functions. Section 6 extends PostGIS
Raster by adding temporal feature and describes a set of operators related to aspect temporal. An
implemented application that enhances Quantum Geographic Information System (QGIS) to exploit
the temporal feature of the PostGIS Raster data will be shown in the next section. Finally, Section 8
concludes our work and point out to future perspectives.
3

Chapter 2
Geographic Information Systems
Geographic data is stored, manipulated, analyzed by a Geographic Information System (GIS). This
section describes what is a GIS and different types of geographic data such as raster and vector. Then
there are a short introduction to some interesting GIS applications.
2.1 Introduction
Geographic data is much more than electronic pictures. The geographic data describes not only
real objects and relations in space but contains also spatial reference, geometric and thematic information.
Nevertheless, to fully exploit these characteristics, an additional tool is needed. A geographic
information system is an information system designed to integrates, stores, edits, analyzes, shares and
displays geographic information for decision makers. GIS applications are tools that allow users to
create interactive queries, analyze spatial information, edit data in maps and present the results of all
these operations.
A GIS is the merging of cartography, statistical analysis and database technology. It is customdesigned
for an organization. A GIS developed for jurisdiction or enterprise purposes may not be
necessarily interoperable or compatible with a GIS that has been developed for some other applications.
2.2 Elements of GIS
All GISs that provide rich behaviors for representing and managing geographic information are based
on four fundamental elements:
• Features
• Attributes
• Imagery
• Surfaces
2.2.1 Features
A feature is used to describe an entity in space-time. Besides a number of additional features,
common geographic features are collections of points, lines and polygons. They represent discrete
things that occur naturally (such as rivers and vegetation) to those of constructions (such as roads,
pipelines, wells and buildings) or subdivisions of land (such as counties, political divisions and land
parcels).
Points
Points define discrete things that are too small to be depicted as lines or areas like locations,
telephone poles and stream gauges. Points can also represent address locations, Global Positioning
System (GPS) coordinates or mountain peaks.
4

Lines
Figure 2.1: Point feature [27].
Lines represent the shape of geographic objects that are too narrow to be depicted as areas (such
as street centerlines and streams). Lines are also used to represent features that have length but no
area such as contour lines and administrative boundaries.
Polygons
Figure 2.2: Line feature [27].
Polygons are enclosed areas that represent the shape of large objects such as states, counties,
parcels, soil types and landuse zones.
2.2.2 Attributes
Geographic data transmits descriptive information through symbols, colors, and labels. For example:
• Roads are displayed based on road class. For instance, line symbols represent divided highways,
main streets, residential streets, unpaved roads and trails.
• Blue is used to indicate water and streams.
• City streets are labeled with their name and often some address.
• Special point and line symbols denote specific features such as rail lines, airports, schools, hospitals
and special facilities.
5

2.2.3 Imagery
Figure 2.3: Polygon feature representing parcels [27].
Imagery is organized as a raster data type composed of cells organized in a grid of rows and
columns. In addition to cell values, a raster includes also its cell size and a reference coordinate (the
upper left or lower left corner of the grid). These properties enable a raster to be described by a series
of cell values starting in the upper left row. Each cell location can be automatically located using the
reference coordinate, the cell size and the number of rows and columns.
Figure 2.4: Grid of cells in raster data [27].
Aerial imagery is a raster data structure obtained from various sensors carried in satellites and
aircraft. Digital orthophotography is a typical imagery sources that comes from cameras.
Figure 2.5: Raster data [27].
Imagery is also used to collect data in both the visible and nonvisible portions of the electromagnetic
spectrum. One system is the multispectral scanner carried in landsat satellites that records imagery
in seven bands along with the electromagnetic spectrum. The measures for each band are recorded in
a separate grid. The stack of seven grids makes up a multiband image.
6

2.2.4 Surfaces
Figure 2.6: Multiband raster data [27].
A surface describes a phenomenon that has a value for each point in space. For example, surface
elevation is a continuous area such that every spatial location has a value for ground elevation above
sea level. Other examples include surface rainfall, surface pollution concentration and etc.
The major problem in surface representation is that it is impossible to represent all values for all
locations of a area. Various alternatives exist for representing such surfaces using either features or
rasters. Here are some alternative solutions for surface representation:
Contour Lines
Contour lines: Isolines represent a set of points having an equal value, such as elevation contours.
Contour Bands
Figure 2.7: Surface expressed by contour lines [27].
Contour bands: Each band describes a specified range of values. An area, where each point value
within it belong to a band, will be represented by a corresponding color. Example of such band
consists of all average annual rainfall between 25 cm and 50 cm per year.
Raster Data
Raster data: A matrix of cells where each cell value represents a measure of each point in space.
For example, digital elevation models (DEMs) represented by grids of squares, are frequently used to
describe surface elevation.
Triangulated Irregular Network
A triangulated irregular network (TIN): A data structure represents surfaces as a connected network
of triangles. Each triangle node has x,y coordinates and a surface value z.
The raster and TIN representations can be used to estimate the surface value for any location
using interpolation.
7

Figure 2.8: Surface expressed by contour bands [27].
Figure 2.9: Surface expressed by raster data [27].
Figure 2.10: Surface expressed by TIN layers [27].
8

2.3 Geographic Data Representation
Traditionally, real objects in geographic data can be stored into two forms: discrete objects (such
as roads, houses, trees) using vector and continuous fields (such as rainfall amount, elevations) using
raster. A new hybrid method of storing data identifies point clouds, combines three-dimensional points
with RGB information at each point and returns a 3D color image. However, the following Section
focus only on vector and raster representation.
2.3.1 Vector Images
Vector images use geometrical primitives like points, lines, curves and polygons, which are all based
on mathematical expressions to represent images in computer graphics.
Each geometric primitive is presented by lines and surfaces. All lines, individually, are defined by
characteristic points that define its equation. These characteristic points form a vector. Thus, the
vector representation is expressed in terms of characters of line and color of the surface. This leads
to the notion of drawing multiple layers where planes curves can be overlaid. This phenomenon can
not happen in raster representation since there is only one layer, then each new point overwrites the
previous point.
The advantage of a vector representation is the delicacy and precision of the results. Every line
can be edited very quickly and all at once. Indeed, this representation involves a limited number of
graphical objects, so these graphics files are very light.
Resolution
As vector images are defined by mathematical expressions, they can be zoomed in or out without
any loss of quality. Concretely, the mathematical description of the object is simply multiplied by a
zooming factor. For example a 1 inch square object will be multiplied by a factor of 2 in order to
double in size. The mathematical expression is thus recalculated to produce an object twice the size of
the original. So vector images can be output at any resolution that a printer is capable of producing.
Unlike raster images, quality of vector images is not limited by scanning resolution. This is a big
reason that vector images are so popular for clip art.
Color
Figure 2.11: Vector image at various scales.
Since vector images are composed of objects, coloring vector objects is similar to coloring with crayons
in a coloring book. Using drawing program, users can easily change the color of individual objects
by clicking inside this object and define its color, along with defining width of lines. Coloring vector
images is much easier than coloring raster images.
File Size
Storing vector images need to keep only mathematical descriptions. For this reason vector files are
very small in file size. A 2-inch by 4-inch logo based vector will be the same files size as a 2-foot by
9

4-foot logo. The file size is the same because the only difference in file is one number defining the size
of the file. Most vector-based logos are going to be under 100 KB. For this reason, vector files are
ideally suited for transfer over the Internet.
File Formats
Common vector formats include EPS (Encapsulated PostScript), WMF (Windows Metafile), AI
(Adobe Illustrator), CDR (CorelDraw), DXF (AutoCAD), SVG (Scalable Vector Graphics) and PLT
(Hewlett Packard Graphics Language Plot File).
Modern displays and printers are raster devices, vector formats have to be converted to raster format
before they can be displayed or printed. Or in graphic work, devices such as cameras and scanners
produce essentially continuous-tone raster graphics that are impractical to convert into vectors. So
an image editor will operate on the pixels rather than on drawing objects defined by mathematical
expressions.
2.3.2 Raster Images
A raster image (or bitmap) is a 2-dimensional array data structure representing a rectangular grid
of pixels, or dots or points of color. A raster image is characterized by its width and height in pixels
and by the number of bits per pixel. This number determines the number of colors can be represented
in the image.
Resolution
The resolution of a raster image is expressed in terms of the dots per inch or dpi. Printer resolution
is also measured in dots per inch. Typical desktop laser printers print at 300 - 600 dpi. Printers with
higher dpi are capable of producing smoother and cleaner output. Nevertheless, the quality of a raster
image depends on its resolution (dpi) and the capabilities of the printing technology [9]. For example,
a 300 dpi raster image will output at the same quality on a 300 dpi laser printer as on a 2,500 dpi
printer. By increasing size of a 300 dpi raster image, the tiny pixel squares will become bigger and
create jaggy edges as no additional information is added. In this case, the raster image losses quality.
Exammple in Figure 2.12 shows the jaggy edges of raster image through zooming effect. In contrast,
When decreasing the size, the squares get smaller and the image retains its original edge without
jaggies.
Figure 2.12: The original vector image is on the left side. The upper-right image illustrates magnification
of 7x as a vector image. The lower-right image illustrates the same magnification as a raster
image [29].
Color
As raster images are reproduction of real world, a large number of colors will be required to render
raster images accurate as their original sources. For example, if scanner works at 24-bit color (16
million colors), most human eyes could not distinct the difference between the original image and
the scanned raster image. In contrast, if a smaller color palette with 256 colors is used, it would be
10

impossible to reproduce the original colors. To get around this, scanners use a process called dithering
to approximate colors that don’t occur in the current color palette.
Figure 2.13: Raster image in zooming [28].
The biggest disadvantage when using raster images locates at editing and manipulating image
colors. As vector images are object-oriented while raster images are pixel oriented, in order to change
pixel colors, a specific color or range of colors must be isolated from the rest to be changed. This work
is quite a challenge for even experienced users.
File Size
For each raster image file, a large amount of information, including the exact location and color of
each pixel in the grid, needs to be kept track to reproduce the image. This results in large file sizes
for raster images. Higher resolutions (dpi) and greater color depths (the number of bits per pixel)
produce bigger file sizes. A typical 2" by 3" 150 dpi black and white raster image logo will be less than
70 KB in file size. The same file saved as a 300 dpi 24-bit (millions of colors) raster image logo might
be 100 times larger (over 7 MB). When creating and scanning raster images, raster file size becomes
a real issue, as big files tend to make computer processor and hard drive work overtime. Transferring
big files (over 1 MB) over the Internet requires a high speed Internet connection on both ends for
timely uploads and downloads.
In despite of these disadvantages, raster images are more practically than vector images as they
are viewable via monitors or other display medium and it is impossible in practice to obtain vector
images from a photo.
File Formats
Common raster image formats include BMP (Windows Bitmap), PCX (Paintbrush), TIFF (Tag
Interleave Format), JPEG (Joint Photographics Expert Group), GIF (Graphics Interchange Format)
, PNG (Portable Network Graphic), PSD (Adobe PhotoShop) and CPT (Corel Photo PAINT).
Perfect graphical tools will combine images from vector and raster sources and provide editing
tools for both, since some parts of an image could come from a camera source and others could have
been drawn using vector tools.
2.4 Geographic Information Organization
2.4.1 Map Layers
GIS organizes geographic information in terms of layers. A layer is a representation of geographic
data on a map. For example, vector layers are collections of simple geographic elements of same type
such as a road network (represented by lines), parcel boundaries (polygons), soil types (polygons),
well locations (points) and so on. Examples of raster layers include an elevation surface, satellite
imagery, land use and etc. Within the data frame, each layer that represents a specific geographic
data is displayed and overlaid by the other layers to form a significant geographic representation in
real world [5].
11

2.4.2 Data Theme
Figure 2.14: Multiple layers geographic representation.
Based on the data themes concept, geographic information in GIS could be partitioned into a series
of logical information layers rather than a random collection of objects. Hence, users can organize
information in various data themes that described the distribution of a phenomenon and how each
one should be represented across a geographic extent.
Many data themes are best represented by a single layer such as soil types or well locations. Other
themes like a transportation framework, are represented by multiple layers, where each feature (such
as streets, intersections, bridges, highway ramps, railroads and so on) is represented by a separate
layer.
Figure 2.15: Themes for a area [30].
Each GIS can contain multiple themes for a common geographic area. The collection of themes
acts as a stack of layers. Each theme can be managed as an information set independent of other
themes. Each one has its own representation and could be a collection of points, lines, polygons,
surfaces, rasters and so on. Because layers are spatially referenced, maybe they overlap one another
and can be displayed in a common map display. So GIS analysis operations like polygon intersection
can combine information between layers to discover and work with the derived spatial relationships.
The concept of layer-based data themes implies that:
• All layers must be georeferenced to a place on the Earth so that they can be combined together.
The georeferencing process is accomplished by defining the geographic coordinate system for
each dataset.
• Layers can be combined in many ways such as following the ordered layers in a map or by
employing operators or commands. Geographic operators can work with the relationships both
12

within and between layers.
• GIS layers can be combined from many sources and many users. In fact, most users are dependent
on one another for portions of the data they want to use. So interoperability becomes a
fundamental feature in GIS.
2.5 Georeference and Coordinate Systems
2.5.1 Georeference
All elements in a map have a specific geographic location and extent that enable them to be
located on the Earth’s surface. So to georeference something means to describe its existence on the
Earth’s surface. The term is used when establishing the relation between raster or vector images and
coordinates or when determining the spatial location of other geographical features. Examples would
include establishing the correct position of an aerial photograph within a map or finding the geographical
coordinates of street address. The georeferencing procedure is thus imperative for modeling
geographical data in the field of geographic information systems.
2.5.2 What Is a Coordinate System?
Figure 2.16: Georeferencing a geographic object [26].
A coordinate system is a system that uses coordinates to uniquely determine the location of a point
or other geographic element on the Earth. There are two common types of coordinate system used in
GIS:
• A global or spherical coordinate system such as latitude-longitude. This system is often referred
to as geographic coordinate system.
• A projected coordinate system based on map projection. There exist different map projections
that provides various mechanisms to project maps of the Earth’s spherical surface onto a twodimensional
Cartesian coordinate plane. Projected coordinate systems are also referred to as
map projections.
Geographic Coordinate Systems
A geographic coordinate system is a coordinate system that enables every location on the Earth
to be specified by a set of numbers. The coordinates are often chosen such that one number represents
vertical position and two or three numbers represent horizontal position. A common choice of
coordinates is latitude and longitude.
Latitude and Longitude are the angles from the center of the Earth to a measured point on the
Earth’s surface.
Latitude measures are angles in a north-south direction. The equator is at an angle of 0. Often,
the northern hemisphere has positive measures of latitude and the southern hemisphere has negative
13

measures of latitude. Longitude measures are angles in an east-west direction. Longitude measures
are traditionally based on the Prime Meridian, which is an imaginary line running from the North
Pole through Greenwich, England to the South Pole. This line has longitude as 0. West of the Prime
Meridian is often recorded as negative longitude and east is recorded as positive.
Figure 2.17: The location of the measured point (in red) is latitude plus 40 degrees and longitude plus
50 degrees [26].
Although longitude and latitude can locate exactly any positions on the surface of the globe, they
are not uniform in units. Only along the equator, the distance represented by one degree of longitude
approximate the distance represented by one degree of latitude. This is because the equator is the
only parallel line of latitude that is as large as a meridian line of longitude.
Figure 2.18: Parallels lines of latitude and meridians line of longitude [26].
Above and below the equator, the circles defining the parallel lines of latitude get gradually smaller
until they become a single point at the North and South Poles. As the meridians converge toward
the poles, the distance represented by one degree of longitude decreases to zero. For example, one
degree of longitude at the equator equals 111.321 km, while at 60 ◦ latitude, it is only 55.802 km. Since
degrees of latitude and longitude don’t have a standard length, distances or areas can not be measured
accurately or display the data easily on a flat map. Efficient GIS analysis and mapping applications
require a more stable coordinate framework, which is provided by projected coordinate systems.
Projected Coordinate Systems using Cartesian coordinates
Projected coordinate systems are any coordinate system designed for a flat surface. 2D and
3D Cartesian coordinate systems provide the mechanism for describing the geographic location of
geographic objects using x, y and z values. The x,y measurements represent the position on the
Earth’s surface and the z represents the height above or below sea level.
2D Cartesian coordinate system is a coordinate system that uses two axes: one horizontal x
representing east-west and one vertical y representing north-south. The point at which these axes
intersect is the origin. Locations of objects are defined relatively to the origin using the notation
14

(x,y), where x refers to the distance along the horizontal axis and y refers to the distance along the
vertical axis. The origin is defined as (0,0). For example, in Figure 2.19, the notation (4, 3) gives a
point that is four units over in x and three units up in y from the origin.
Figure 2.19: 2D Cartesian coordinate system. [26].
3D Cartesian coordinate system is a coordinate system that uses also a z value to measure
elevation above or below sea level. The notation (2, 3, 4) in Figure 2.20 records a point that is two
units over in x and three units in y from the origin and whose elevation is 4 units above the Earth’s
surface such as 4 meters above sea level.
Figure 2.20: 3D Cartesian coordinate system [26].
Since the Earth is spherical, a challenge faced by cartographers is how to represent the real world
using a flat coordinate system. The process of flattening the Earth is called projection, hence the term
map projection is resulted. Examples of various methods for map projection are illustrated in Figure
2.21.
Unlike a geographic coordinate system, a projected coordinate system has constant lengths, angles
and areas across the two dimensions. However, all map projections representing the Earth’s surface
create distortions in some aspect of distance, area, shape, scale or direction (Figure 2.22). Depend on
user application, many map projections are designed for specific purposes. One map projection might
be used for preserving shape while another might be used for preserving the area.
These properties of the map projection are important parameters in the definition of the coordinate
system for each GIS dataset and each map. With these properties, the geographic locations of dataset
elements can be reprojected and transformed into an appropriate coordinate system. As a result, it’s
possible to integrate and combine information from multiple GIS layers. This is a fundamental GIS
capability as accurate location forms the basis for almost all GIS operations.
15

2.6 GIS Applications
2.6.1 Crime Mapping
Figure 2.21: Different methods for map projection [26].
Figure 2.22: Distorsion in map projection [26].
Crime mapping is an application used by analysts in law enforcement agencies to map, visualize and
analyse crime incident. Using Geographic Information Systems, crime analysts can overlay different
datasets (such as census demographics, locations of pawn shops, schools and etc.) to identify crime
hot spots, to better understand the underlying causes of crime and thus to devise strategies dealing
with the problem, along with other trends and patterns [18].
2.6.2 GIS and Hydrogeology
Nowadays, with increasing demand on surface water resources and 98% of the world’s available
fresh water is groundwater, it implies that the demand for groundwater will also increase. In fact, in
some places, this resource has already been seriously exploited and even mismanaged (Figure 2.24).
Hence, more knowledge about its disposition is quickly required.
Hydrogeology, a subset of hydrology, is concerned with the way in which groundwater is stored and
its availability for use. Not as apparent as surface water flows, the description of groundwater is a more
complex task than that of surface water. Groundwater is water located under the Earth’s surface.
Groundwater moves much more slowly than surface water and is 3-dimensional flow (2-dimensional in
the case of surface water flow). Groundwater flow is a function of geology and head. That means it
flows from higher head to lower head and its flow path can be predicted by geology.
Head values, geology and groundwater flow direction are the features which may be presented in
GIS. Then hydrogeology is especially suited to GIS because groundwater can be characterized spatially
in GIS and analyzed by scientists. For example, in a simple application of GIS, the effect of a new
feature can be studied on the groundwater. The results of such study then can be used by decision
makers to determine whether or not to proceed with drilling.
16

Figure 2.23: Mapping of homicides in Washington D.C. [18]
Figure 2.24: Groundwater level change of the High Plains Aquifer, 1980-95 [31].
17

2.6.3 Traditional Knowledge GIS
Traditional knowledge GIS is designed to document and utilize local knowledge of communities
around the world. Traditional knowledge includes the experiences of a particular culture or society.
Traditional knowledge GIS is richer than ordinary maps in the sense that they express environmental
and spiritual relationships among real and conceptual entities.
Cultural preservation is the principal application of a traditional knowledge GIS. Its central feature
is language revitalization. Bilingual visual and audible maps are used to describe oral traditions and
historical information of significant cultures on various scales.
18

Chapter 3
Why PostGIS Raster ?
ORACLE GEORaster is a component that supports continuous fields in ORACLE. Nevertheless,
it sounds complex and cumbersome for inexperienced users. PostGIS Raster is thus designed to
overcome these limitations. This fact will be demonstrated through PostGIS Raster goals and how
Oracle GeoRaster and PostGIS Raster are different in architecture, physical storage, indexing and
overlapping, along with some principal operations when working with raster data.
3.1 Goals
PostGIS Raster, an extension of PostGIS, implements a new raster data type that adapts to
diversity of applications by achieving four objectives [6]:
3.1.1 Simplicity and Complementarity
Seen a raster type is designed to adapt to diversity of applications, then the first objective of
PostGIS Raster is to provide a simple loader to import rasters into database. The loader allows users
to load a single raster or a set of rasters once. This new data type complements the existing PostGIS
vector type in the sense that it works in the same way as the one of vector type, but instead of
association with geometric elements, raster relates to a matricial data structure.
3.1.2 Transparent Integration with Vector
PostGIS Raster provides a set of similar operators and functions with those available for vector
type to support raster. In this way, users can use their knowledge about vector type to expect similar
behaviors when using these operator and functions. Thus it enhances user’s interaction with geospatial
applications.
About developers, a single SQL paradigm for both raster and vector types will help them write
better GIS applications. Developers will build a unique graphical user interface for both raster and
vector data.
3.1.3 Storage Flexibility
In addition to traditional storage (in-database storage), PostGIS Raster allows user to simply register
only metadata of raster images stored in the filesystem (out-database storage). When retrieving these
out-database rasters from database, they are accessed directly from the filesystem.
3.1.4 Interoperability
PostGIS Raster uses Geospatial Data Abstraction Library (GDAL) [25] to load rasters into the
database and to work with out-database rasters. With GDAL, PostGIS Raster can work with nearly
a hundred file formats 1 ).
1 http://www.gdal.org/formats_list.html.
19

These additional functionalities make PostGIS Raster more than just a new raster format but also
a necessary complement to PostGIS.
3.2 PostGIS Raster Versus ORACLE GEORaster
3.2.1 Architecture
General Architecture
In general, tools supporting raster data include five components [3]:
Characteristics Oracle GeoRASTER PostGIS Raster
SQL API Standard SQL A single set of SQL
functions for vector and
raster data
Accessing to Java, Oracle Call Interface (OCI), Oracle C++ Call PostGIS SQL API
raster data Interface (OCCI), and GeoRaster SQL API
Viewing tools Oracle Fusion Middleware MapViewer, Open- QGIS and gvGIS
Import/Export
data
JUMP
Extract, transform and load (ETL) GDAL allows loading a
single raster or a set of
Data formats Six standard image formats: GeoTIFF, DEM,
PNG, BMP, GIF, and JPEG
Data Architecture
rasters
As many image formats
as GDAL does [21]
In Oracle GeoRASTER, each image is stored as a single object of type SDO_GEORASTER. A GeoRaster
table is a table which has at least one data column of type SDO_GEORASTER. SDO_GEORASTER objects
include metadata and information about how to retrieve GeoRaster data which is stored in another
table, called a Raster Data Table (RDT). GeoRaster data is an object of type SDO_RASTER which
includes a Binary Larger Object (BLOB) column called RASTERBLOCK storing the raster blocks (The
core of raster data can be partioned into small blocks for optimal storage and retrieval). A block is
also called a tile [3].
Figure 3.1: GeoRaster data architecture [3].
Other information associated with the GeoRaster objects can be stored in separate columns or
tables, such as a Value Attribute Table (VAT).
More simply, PostGIS Raster uses only one type raster to store a raster image. Like the geometry
type, the raster type is a complex type, embedding information about raster itself with its georeference.
In PostGIS Raster, there is no table for storing the raster data (like Oracle Spatial SDO_RASTER)
and another table for storing the georeference and the metadata (like Oracle Spatial SDO_GEORASTER).
20

Figure 3.2: PostGIS Raster data architecture.
Everything is stored in a single attribute and raster attributes composing a table are not necessarily
related to each other to form a significant coverage.
PostGIS Raster uses in addition a raster_columns table to contain information about which
tables have a raster column and the metadata (pixeltype, block size, nodata value and etc.) of the
rasters in these columns.
3.2.2 Physical Storage
GeoRaster data consists of a multidimensional matrix of cells and the GeoRaster metadata. The
multidimensional matrix is blocked into small blocks. A block (or tile) is stored as a BLOB and a
geometry object of type SDO\_GEOMETRY is used to define the spatial extent (footprint) of the block.
Each block and information related to that block are stored in one row. Thus for storing an image
having size greater than block size, at least two rows are necessary to be used in Oracle database table
[3].
To do this, GeoRaster requires two types of object:
• The SDO_GEORASTER type contains spatial extent geometry and metadata.
• The SDO_RASTER contains a BOLB storing a block and the block information.
Each SDO_GEOMETRY data has a pair of attributes (rasterDataTable, rasterID) that identify the
RDT and the rows within the RDT that are used to store the raster data.
For Example, Figure 3.3 shows the storage of GeoRaster object used for an image of Boston,
Massachusetts, where:
• Each row in the table of city images contains information about the image for a specific city
(such as Boston), including an SDO_GEORASTER object.
• The SDO_GEORASTER object includes the spatial extent geometry covering the entire area of the
image, the metadata, the raster ID and the name of the raster data table associated with this
image.
• Each row in the raster data table contains information about a block (or tile) of the image,
including the block’s minimum bounding rectangle (MBR) and image data (stored as a BLOB).
A MBR is defined as a single rectangle that minimally encloses a geometry or a collection of
geometries.
In PostGIS Raster, the raster type is a complex type composed of many attributes like attribute,
georeference information, band information and band data. Thus a single raster attribute can contain
all information about raster itself (width, height, number of bands, pixel type for each band, nodata
21

Figure 3.3: Physical storage of GeoRaster data [3].
value and data values for each band) with its georeference (pixel type, upper left pixel center, rotation
and spatial reference system identifier (SRID)).
A single raster attribute may be composed of many bands having a common size, pixel size and
georeference. The storage used for raster bands is band sequential (BSQ) in which each band has its
pixel type and its nodata value.
One table with a column of type raster is called raster coverage. One table row with a column of
type raster is called tile. Thus there are no differences between rasters and tiles. A tile is a raster and
a raster is a tile in which each tile has all that are necessary to do basic raster operations.
Figure 3.4: Physical storage of PostGIS Raster data.
Another important advantage of PostGIS Raster over Oracle GeoRaster is the georeferencing
approach [14]. As Oracle GeoRaster uses a one-georeference-by-layer georeference schema instead of
one-georeference-by-raster one, so non-continuous raster data cannot be stored in the same raster data
table. That means that Oracle GeoRaster requires one raster data table for each raster file loaded
when these raster files are from irregular raster coverage. In contrast, PostGIS Raster allows only one
table for all kinds of raster coverages, regular or not.
In addition to this traditional storing method (storing in-database raster), PostGIS Raster enables
user to simply register basic metadata (path to the actual file, pixeltype, with, height, georeference)
of images stored in the file system, without having to load their data values into the database. This is
22

Figure 3.5: Oracle GeoRaster georeference approach [14].
Figure 3.6: PostGIS Raster georeference approach [14].
referred to as storing out-database raster. With the second method, web or desktop applications may
access out-database rasters directly from the filesystem (such as JPEG files) and use most transparently
PostGIS Raster operators and functions on those rasters.
3.2.3 Indexing
The most important index is created on a GeoRaster object is a spatial index on the spatial extent
geometry of the GeoRaster object. Oracle GeoRaster creates spatial indexes over the footprint of raster
data while PostGIS Raster creates GiST indexes over the raster data itself. In Oracle GeoRaster,
spatial operations (like intersection) can only be done with the MBR of the data as the indexes are
created over the spatial extent. In PostGIS Raster, user can do really complex operations with raster
data like getting a set of geometry and geometry value from a given raster band (ST_DumAsPolygons()
function). Then, creating indexes over these raster data has more sense [15].
PostGIS Raster uses GDAL driver to calculate the georeferenced coordinates for upper and lower
corners (the spatial extent or footprint) of a raster. In this case, its spatial extent is the enclosing
geometry. Nervertheless, this is not necessarily in this way for Oracle GeoRaster.
In GeoRaster, one of the prerequisites for creating a spatial index over a geometry column is
that the USER_SDO_GEOM_METADATA view must contain an entry with the dimensions and coordinate
boundary information for the table column to be spatially indexed. So the following code will create
spatial index on the spatial extent geometry of the GeoRaster objects stored in table spain_images:
DELETE FROM user_sdo_geom_metadata
WHERE table_name = ’spain_images’
AND column_name = ’IMAGE.SPATIALEXTENT’;
INSERT INTO user_sdo_geom_metadata
VALUES (’spain_images’,’IMAGE.SPATIALEXTENT’,
SDO_DIM_ARRAY(SDO_DIM_ELEMENT(’X’, -180, 180, .00000005),
SDO_DIM_ELEMENT(’Y’, -90, 90, .00000005)), 4326);
23

DROP INDEX spain_images_idx;
CREATE INDEX spain_images_idx
ON spain_images(image.spatialExtent)
INDEXTYPE IS mdsys.spatial_index;
In contrast, PostGIS Raster loads raster data and creates spatial index using simply these two
commands:
python raster2pgsql.py -r *.tif -t spain_images -s 4326 -k 50x50 -I -o spain.sql
psql -d postgis_db -f spain.sql
The first command uses Python loader to import a set of raster images into raster table spain_images
(option -t). Option -I is used to create indexes over these rasters. All necessary code is created and
registered in a SQL file spain.sql (option -o). Then, the second command will execute spain.sql
in PostGIS database.
3.2.4 Overlapping Raster and Vector Layers
Each spatial instance has a spatial reference identifier (SRID). However, the result of any spatial
method from two spatial instances is valid only if those instances that have the same SRID. It is the
same in the case of overlapping operation between vector and raster data.
To reset SRID of a geometry object, GeoRaster uses sdo_cs.transform() function [14]:
SELECT sdo_cs.transform(geom, srid_number) geom FROM geom_table ;
In the same way in PostGIS Raster:
SELECT ST_Transform(the_geom, srid_number)) FROM geom_table;
3.2.5 Loading Raster Data
Loading Using Oracle GeoRaster
To load raster image into GeoRaster, firstly users need to create tables to store the raster data.
One table stores the metadata and another one stores the data [14]:
CREATE TABLE spain_images (image_id NUMBER PRIMARY KEY,
image_description VARCHAR2(50), image SDO_GEORASTER);
CREATE TABLE spain_images_rdt OF SDO_RASTER
(PRIMARY KEY (rasterID, pyramidLevel, bandBlockNumber, rowBlockNumber,
columnBlockNumber))
TABLESPACE users LOB(rasterBlock) STORE AS SECUREFILE lobseg (NOCACHE);
Then, the image is loaded using PL/SQL code:
DECLARE
geor SDO_GEORASTER;
BEGIN
-- Initialize an empty GeoRaster object into which the external image
-- is to be imported.
INSERT INTO spain_images
VALUES( 1, ’Spain_TIFF_1’, sdo_geor.init(’spain_images_rdt’) );
-- Import the TIFF image.
SELECT image INTO geor FROM spain_images
WHERE image_id = 1 FOR UPDATE;
24

sdo_geor.importFrom(geor, ’blocksize=(256,256)’, ’TIFF’, ’file’, ’srtm_new.tif’);
UPDATE spain_images SET image = geor WHERE image_id = 1;
END;
The same operation is repeated for every image file. If these images form a non-continuous coverage,
two new tables are needed for each image file.
Loading Using PostGIS Raster
When loading raster data by using GDAL driver, the loader creates all that are necessary to store
the raster data into a SQL file. It consists of creating the needed tables, the indexes (if it is specified)
and inserting raster data into the tables. A single or a set of raster images can be loaded all only once.
Example:
python raster2pgsql.py -r path\*.tif -t srtm_tiled -s 4326 -k 50x50 -I -M
-o srtm.sql
Next step is to execute the srtm.sql file using psql:
psql -d postgis_db -f srtm.sql -U user -W
3.2.6 Intersecting Raster and Vector Layers
Intersecting in Oracle GeoRaster
It is impossible to intersect directly raster and vector data in Oracle GeoRaster. Instead, what user
can do is to select a sampling window to clip an area of the raster and get raster data over this area.
The sampling windows must be rectangular or the round windows cannot be used (As described in
SDO_GEOR.generateStatistics documentation 2 ). So the MBR is used as solution.
In the following example, the sampling windows are the geometry objects stored in sampling_win table.
The request will calculate the intersection between the sampling windows and the spatial extent of
the GeoRaster objects [15]:
DECLARE
cellCoordinate mdsys.sdo_geometry;
ret varchar2(256);
gr sdo_georaster;
BEGIN
-- Intersect spatial extent of the rasters with the sampling windows
(SELECT t.id, t.geom, r.image_id as rid
FROM spain_images r, sampling_win t
WHERE sdo_geom.sdo_intersection(r.image.spatialExtent, t.geom, 0.005) is not null)
END;
Intersecting in PostGIS Raster
PostGIS Raster uses st_intersection() and st_intersects() functions, that operate directly
on raster and vector data, to calculate the intersection. Before doing the intersection, the
st_intersects() will check which portion of raster is intesected by vector. So the st_intersection()
intersects vector with raster by vectorizing only the necessary part of the raster [8]:
CREATE TABLE vector_srtm_inter AS
SELECT id, (gv).geom AS the_geom, (gv).val
FROM ( SELECT id, ST_Intersection(rast, the_geom) AS gv
FROM srtm_tiled, vector_data
WHERE ST_Intersects(rast, the_geom)
) inter;
2 http://docs.oracle.com/cd/B28359_01/appdev.111/b28398/geor_ref.htm#CHEDCICI.
25

In summary, the intersection is much slower than one in GeoRaster for two reasons:
• To intersect vector and raster data, firstly raster has to be polygonized, then the result is the
intersection between these polygons with vector data. In this process, polygonizing is a complex
operation.
• In Oracle, the raster data is sampled using rectangular windows. So the intersection is only
performed between these MBRs with vector data while the vector data intersected by the raster
data in PostGIS Raster are in some forms [15].
26

Chapter 4
PostGIS Raster
This section deals with representing raster structure and explaining from how raster is stored to
the mechanisms used to index, retrieve raster in PostgreSQL. It is also important to address the way
that raster and vector layers overlap - a relevant feature that makes PostGIS Raster more robust than
ORACLE GEORaster.
4.1 Implementation
PostGIS Raster chooses to implement a minimal raster data structure. That means that a raster
type is as simple as a single type and is stored in a single table. This data structure is very similar
to the one of PostGIS vector data and very different from the Oracle Spatial SDO_GEORASTER and
SDO_RASTER raster structures. In PostGIS Raster:
• One table with a column of type raster is referred to as a raster coverage.
• One table row with a column of type raster corresponds to one tile.
The choice of such simple data structure will facilitate the achievement of four objectives described
in Chapter 3. With a single raster type, it implies that each raster attribute is a complete and self
sufficient georeferenced raster. So raster attributes composing a table are not necessarily related to
each other to form a significant coverage. This also means that:
• Rasters from the same table may have different size.
• The upper left corner and the pixel size of a raster are not necessarily the same values as those
of another rasters. They may therefore snap to different grids.
• Finally, different rasters may overlap like polygons in a vector layer may overlap. This is a
fundamental feature to implement meaningful vector to raster conversions in which all attributes
of a vector are conserved in the resulting raster table. In this case, if the vector features and the
data values part of the resulting raster features do not necessarily overlap, the nodata values of
the raster features will overlap.
In addition, PostGIS Raster provides also a raster_columns table to help application get a quick
overview of raster tables and information (such as metadata) related to the rasters stored in these
tables.
4.2 Structure
PostGIS Raster implements only one type of raster instead of two like SDO_GEORASTER and
SDO_RASTER in Oracle Spatial. It supports no metadata when a raster contains all data about itself
such as attribute information, georeference information, band information and band data. Figure
?? describes these components in details [6]. As shown in this figure, a raster has no mask and
27

Attribute information
Table 4.1: PostGIS Raster structure.
Description Storage
version Raster format version unsigned 16 bit
integer
band number Number of raster bands stored in
the raster
unsigned 16 bit
integer
width Width of the raster unsigned 16 bit
integer
heigh Height of the raster unsigned 16 bit
integer
Georeference information
Description Storage
pixelsizex Pixel size in the x-direction in the
same map units as the coordinate
system
pixelsizey Pixel size in the y-direction in the
same map units as the coordinate
system
upperleftx X-coordinate of the center of the upper
left pixel
upperlefty Y-coordinate of the center of the upper
left pixel
64 bit double
64 bit double
64 bit double
64 bit double
rotationx Rotation about x-axis 64 bit double
rotationy Rotation about y-axis 64 bit double
srid Y-coordinate of the center of the upper
left pixel
32 bit integer
Band information
(one set per band)
Description Storage
isoffline Flag specifying if raster data is
stored in the database or as a file
in the file system
1 bit
hasnodatavalue Flag specifying if nodatavalue exists
or not
1 bit
pixeltype Pixel type of the band 4 bits
nodatavalue Nodata value of the band Depend on band
pixel type
Band data (one
set per band) for
in-db raster
Description Storage
values[] An array of pixel values Depends on band
pixel type and size
Band data (one
set per band) for
out-db raster
Description Storage
bandnumber Number of the out-db band unsigned 8 bit integer
path Path to the out-db raster file string
28

Figure 4.1: PostGIS Raster implementation.
no pyramid propeties when a mask can be created as an independent band and reduced resolution
coverages can be stored as a separate layer.
So a single raster has everything essential to do basic GIS raster operations. It is thus not necessarily
related to other rasters in the same table to be georeferenced. This choice makes PostGIS
Raster very simple and flexible.
4.3 Spatial Reference Identifier
A spatial reference system (SRS) or coordinate reference system (CRS) is a coordinate-based local,
regional or global system used to locate geographical entities. A spatial reference system defines a
specific map projection, as well as transformations between different spatial reference systems.
A spatial reference system identifier (SRID) is a unique value used to identify projection and local
spatial coordinate system. These coordinate systems form the heart of all GIS applications. In other
words, SRID values associated with spatial objects can be used to constrain spatial operations. For
instance, spatial operations (such as intersection, transformation and etc.) can not be performed
between spatial objects with different SRIDs. Or, the result of any spatial method derived from two
spatial objects is valid only if they have the same SRID that is based on the same unit of measurement
and projection.
4.4 Raster Data
Raster data is a matrix of cells (or pixels) organized into rows and columns (or a grid) where each
cell contains a value representing information. Unlike vector format, data stored in raster format can
be used to represent various real world phenomena like [16]:
• Land use or soils data which is classified as discrete data.
• Temperature, elevation or spectral data that comes from satellite images and aerial photographswhich
are classified as continuous data.
• Pictures such as scanned maps or photographs of drawings and buildings.
4.4.1 Usage
With a simple data structure, raster is exceptionally useful for a wide range of applications. In
GIS, the uses of raster data falls under four main categories:
• Base maps A common way to use raster is as a background for displaying other vector layers.
Three main sources of raster based maps are orthophotos, satellite imagery and scanned maps.
29

Figure 4.2: Information representation using raster data [16].
An thophoto, thophotograph or thoimage is an aerial photograph geometrically corrected such
that the scale is uniform: the photo has the same lack of distortion as a map. Unlike an uncorrected
aerial photograph, an orthophotograph can be used to measure true distances because
it is an accurate representation of the Earth’s surface, being adjusted for topographic relief 1 ,
lens distortion 2 and camera tilt 3 . When orthophotographs are displayed underneath other
layers, they provide user with confidence that map layers are spatially aligned and represent real
objects.
Figure 4.3: Basemap raster for road data [16].
• Continuous maps Due to the matrix of cells structure, rasters are well suited for representing
and storing data that changes continuously across a surface. Elevation values measured from
the Earth’s surface are the most common application of surface maps. For example, raster in
Figure 4.4 displays elevation, where green cells show lower elevation and red, pink and white cells
show higher elevation. Other values, such as rainfall, temperature, concentration and population
density, can also be used in spatial analysis as cells representing surfaces are regularly spaced.
• Thematic maps Rasters representing thematic data can be derived from other data through
analyzing operations. A common analysis application is classifying a satellite image by landcover
categories. Basically, this activity groups the values of data into classes (such as vegetation
type) and assigns a categorical value. Figure 4.5 is an example of a classified raster data that
expresses the human use of land. Thematic maps can also result from geoprocessing operations
that combine data from various sources such as vector, raster and terrain data. For example,
user can process data through a geoprocessing model to create raster data that maps suitability
for a specific activity.
• User-defined attributes Digital photography and scanner are output devices that capture real
world phenomena into digital world for long-term storage. Raster data is means by which digital
1 http://en.wikipedia.org/wiki/Topography.
2 http://en.wikipedia.org/wiki/Barrel_distortion.
3 http://en.wikipedia.org/wiki/Camera_tilt.
30

Figure 4.4: Surface map raster [16].
Figure 4.5: Thematic map raster [16].
photographs, scanned documents or scanned drawings are conserved. Via raster representation,
these real phenomena can also be used as user-defined attributes of a geographic object. For
example, a parcel layer may have scanned legal documents identifying the latest transaction for
that parcel. Or, a layer representing cave openings may have pictures of the actual cave openings
associated with the point features. Concretely, a real tree represented by the digital picture in
Figure 4.6 can be used as an attribute to a landscape layer that a city may maintain.
4.4.2 Cell Representation
Figure 4.6: Tree attribute [16].
Raster data is made up of a matrix of cells (or pixels), where each cell contains a value. In other
words, the cell values are used to represent phenomenon described by the raster data such as a category,
magnitude, height or spectral value. The category can be a land-use class such as grassland, forest
or road. A magnitude might represent gravity, noise pollution or percent rainfall. Height (distance)
can represent surface elevation above mean sea level, which can be used to derive slope, aspect and
watershed properties. Spectral values are used in satellite imagery and aerial photography to represent
light reflectance and color.
Cell values can be either positive or negative, integer or floating point. Integer values are best used
31

for discrete data while floating-point values are suited for continuous data (such as surfaces). Cells
can also have a noData value to indicate the absence of data.
Depend on represented data, there are two ways to apply a value to a cell. Generally, when the cell
value represents a measure as in the case of elevation, it is applied at the center of the cell. However,
in most cases, the cell value is resulted from a sampling of a phenomenon, so it is applied for the whole
cell square.
Figure 4.7: On the left side: cell values applied at the center point. On the right side: cell values
applied for the whole square.
Raster data is stored as an ordered list of cell values. For example, such a raster in Figure 4.8 will
be stored as a list of 80, 74, 62, 45, 45, 34 and so on.
Figure 4.8: Raster cell values [16].
The area represented by each cell consists of the same width and height and equals a portion of
the entire surface represented by the raster. For example, a raster representing elevation may cover
an area of 100 square kilometers. If there were 100 cells in this raster, each cell would represent one
square kilometer of 1 km in width and 1 km in height.
The dimension of the cell can be as large or as small as needed to represent the surface conveyed
by the raster and the objects within this surface. In other words, the cell size determines how coarse
or fine the objects in the raster will appear. The smaller the cell size is, the smoother the raster will
be. However, when decreasing the cell size to get better image, the number of cell will increase as the
surface size is maintained. A large number of cells will take long time to process and demands more
storage space. In contrast, if a cell size is too large, information may be lost or subtle patterns may
be obscured. For example, if the cell size is larger than the width of a road, the road may not exist
within the raster dataset. In Figure 4.10, it shows how does the polygon look like as it is represented
at various cell sizes.
The location of each cell is defined by the row or column where it is located within the raster
matrix. Essentially, the matrix is represented by a Cartesian coordinate system, in which the rows of
32

Figure 4.9: Raster cell [16].
Figure 4.10: Raster resolution [16].
33

the matrix are parallel to the x-axis and the columns to the y-axis of the Cartesian plane. Row and
column values begin with 0.
Figure 4.11: Pixel or cell location.
The extent of a raster then is defined by the top, bottom, left and right coordinates of the rectangular
covered by the raster, as shown in Figure 4.12 .
4.4.3 Zones and Regions
Zones
Figure 4.12: Raster extent [16].
Any two or more cells with the same value belong to the same zone. A zone consists of cells that are
adjacent, disconnected or both. Zones whose cells are adjacent usually represent a single feature such
as a building, lake, road or water body. Assemblages of entities, such as forest stands in a state, soil
types in a county or single-family houses in a town, are features that will most likely be represented
by zones made up of many disconnected groups of connected cells.
Every cell in a raster belongs to a zone. Some raster data contains only a few zones while others
contain many.
Regions
Each group of connected cells in a zone is considered a region. A zone that consists of a single
group of connected cells has only one region. Zones can be composed of as many regions as necessary
to represent a feature. In the example of Figure 4.13, zone 2 consists of two regions, zone 4 of three
regions and zone 5 of one region.
34

4.4.4 Representing Features
Figure 4.13: Raster zones and regions [32].
In raster data, the cell typically represents the predominant phenomenon of the area covered by
a cell, whereas vector data can accurately identify individual features. This is due to the fact that
when representing geographic features using raster data, features are nothing more than collections
of cells with the same attribute values and thus lose their unique identities. Raster data is best used
when the primary problem concerns with the locational relationships of the phenomena represented
by geographic features and not the features themselves.
Discrete Data
Discrete data, which is sometimes called thematic, categorical or discontinuous data, is used to
describe discrete objects and can be expressed in both vector and raster formats. A discrete object has
always precise boundaries. So it is easy to define where the object begins and where it ends as well as
how to compose them from discrete features like points, lines and polygons. A lake is a discrete object
surrounded by landscape. Other examples of discrete objects include buildings, roads and parcels.
Points
Figure 4.14: Discrete data represented in raster format [33].
A point is represented by an explicit x,y coordinate in vector format. However, in raster format,
it is a single cell, the smallest unit of a raster. By definition, a point has no area but will represent
an area when it is converted to a cell. For example, a telephone pole or the location of an endangered
plant will occupy the entire area covered by a cell. Therefore, the smaller the cell size, the smaller the
area and thus the closer the representation of the point feature.
As a raster point has the same size as the size of the cell. So the cell size has to be chosen small
enough to capture sufficient input points for the desired analysis. In the case if two or more points
fall within the same cell, the value of one of these points will be randomly selected when assigning a
value to the cell.
35

Lines
Figure 4.15: Raster point representation [34].
In vector format, a line is an ordered list of x,y coordinates whereas in raster format it is represented
as a chain of connected cells with the same value. When there is a break between the chain of same
valued cells, it corresponds a break in the line. A break can also be used to define two roads or two
rivers that do not intersect.
Figure 4.16: Raster line representation [34].
Converting a vector line to raster data is similar to converting a vector point to raster data. For
any line that passes over a cell, this cell will receive the attribute value of this line. If multiple lines
pass through a single cell, one of the lines will be randomly selected to represent that cell.
The same as vector point, the width of vector line will become the width of the cell. For example,
if the line that is being converted represents a road and the cell size is one meter, then the road will
be one meter wide in the output raster data.
Polygons
A vector polygon is an enclosed area defined by an ordered list of x,y coordinates in which the first
and last coordinates are the same. In contrast, a raster polygon is a group of contiguous cells with
the same value that most accurately describe the shape of the area. Examples of polygonal features
include buildings, ponds, soils, forests, swamps and fields.
The accuracy of the raster representation depends on the scale of the data and the size of the cell.
The finer the cell resolution and the greater the number of cells, the more accurate the representation.
Continuous Data
Continuous data is used to describe continuous surfaces. A continuous surface represents phenomena
in which each location on the surface has a measure of the concentration level or a value describing its
relationship with a fixed point in space or with an emitting source. Continuous data is also referred
to as field, nondiscrete or surface data.
One type of continuous surface, that is derived from those characteristics that define a surface, is
a surface where each point in the surface is its measured from a fixed reference point. These include
elevation (the fixed reference point is a sea level and the measure at each point is its height above the
36

Figure 4.17: Raster polygon representation [34].
sea level). Another example is temperature where the measure at each point is temperature at a place
on the Earth.
Figure 4.18: Raster elevation data for a part of the province of Quebec [8].
4.5 Spatial Resolution
The detail level of phenomena represented by a raster is often dependent on the spatial resolution
or cell size of the raster. The cell must be small enough to capture the required detaild but it has to
be also large enough so that computer storage and analysis can be performed efficiently. Smaller cell
sizes result in larger raster data set and greater storage space to represent an entire surface, which
often require longer processing time.
Figure 4.19: Spatial resolution.
37

Choosing an appropriate cell size is not always simple. For example, the spatial resolution needed
for applications againsts with the requirements for quick displaying, processing time and storages. In
GIS, the choice is made for the least accurate raster data. For example, for a raster data derived from
30-meter resolution Landsat imagery, a digital elevation model (DEM) at a higher resolution, such as
10 meters, may be unnecessary.
Figure 4.20: Raster in different spatial resolutions [11].
However in most of applications, determining an adequate cell size is not a static process and
depends on what data expected to obtain. In the other hand, a raster can always be resampled to
have a larger cell size but any greater detail can not be obtained by resampling the same raster at a
smaller cell size. So, it is useful to store a copy of the raster at its smallest cell size, then it will be
resampled to match the use in application to increase analysis processing speed.
4.6 Spatial Resolution Versus Scale
Spatial resolution refers to the dimension of the cell size that represents an area on the ground. If
the area covered by a cell is 5 x 5 meters, the resolution is 5 meters. The higher the resolution of a
raster, the smaller the cell size and thus the greater the detail. This is opposite to scale.
A scale model represents a copy of an object that is larger or smaller than its actual size. To do
this, a relative proportion (a scale factor) of the physical size of the original object is maintained for
the restitution. The smaller the scale, the less detail shown. So, an ortho photograph displayed at a
scale of 1:2,000 (that means 1 unit on model is equal to 2,000 units on real object) shows more details
than one displayed at a scale of 1:24,000. However, if the same orthophoto has a cell size of 5 meters,
the resolution will remain the same no matter what scale it is displayed at, as the physical cell size
does not change.
Image shown in Figure 4.21 illustrates well these effect. On the left, the scale of the image (1:50,000)
is smaller than the scale of the one on the right (1:2,500) but their spatial resolutions (cell size) are
the same. Another way to display images is to change the spatial resolution and maintain the scale
like the ones in Figure 4.22.
4.7 Bands
PostGIS Raster supports multiple bands. So besides the ordinates row and column, each raster can
have the third ordinate band and is identified as a cell in the space of three dimensions (see Figure
4.23). Band is numbered from 1 to n, where n is the highest band number. Each band has information
about itself such as band data, a pixel type and a nodata value (as described in Figure ??). The pixel
type indicates the number of bits used to express the color of a single pixel as well as the nodata value.
Band data is a two-dimensional matrix of cells. The storage used for band data is band sequential
(BSQ).
38

Figure 4.21: Rasters at different scales [11].
Figure 4.22: Same scaled rasters at different spatial resolution [11].
Figure 4.23: Raster bands.
39

The multi-band conception is resulted from the fact that, in practice, a raster image can contain
multiple bands. For example, electromagnetic wave data from remote sensing devices is grouped into
a certain number of channels, where the number of possible channels depends on the capacities of the
sensing device. Or, multispectral images contain multiple channels and hyperspectral images contain
environ 50 or more channels. Then by using raster type to model such data in a geographic information
system, each channel is expressed by a band.
Some rasters have a single band of data, while others have multiple bands. Basically, a band
is represented by a single matrix of cell values and a raster with multiple bands contains multiple
coincident matrices of cell values representing the same spatial area. An example of a single-band raster
dataset is a digital elevation model (DEM). Each cell in a DEM contains only one value representing
surface elevation. Most satellite imagery has multiple bands, typically containing values within a range
or band of the electromagnetic spectrum.
There are three main ways to display single-band raster data:
• Using two colors: In a binary image, each cell has a value of 0 or 1 and is often displayed using
black and white. This type of display is often used for displaying scanned maps with simple line
work such as parcel maps.
• Grayscale: In a grayscale image, each cell has a value from 0 to 255 or 65535. These are often
used for black and white aerial photographs.
• Color map: One way to represent colors on an image is using a color map. A set of values is
coded to match a defined set of red, green and blue (RGB) values.
Figure 4.24: Three main ways to display single-band raster [35].
With multiple bands raster, every cell location has more than one value associated with it. Each
band usually represents a segment of the electromagnetic spectrum collected by a sensor. Bands can
represent any portion of the electromagnetic spectrum including ranges not visible to the eye such as
the infrared or ultraviolet sections. The term band originated from the reference to the color band on
the electromagnetic spectrum.
Figure 4.25: Multi-band raster [35].
When a map layer is created from a multi-band raster image, user can choose to display the image
from a single band of data or from multiple bands. A combination of any three available bands
40

in a multiband raster can be used to create RGB composites. When displaying image using RGB
composites, obviously more information is conveyed from the dataset than in the case of one band.
Figure 4.26: RGB composite from a three-band raster [35].
A satellite image, for example, commonly has multiple bands representing different wavelengths
from the ultraviolet through the visible and infrared portions of the electromagnetic spectrum. Landsat
imagery, for example, is data collected from seven different bands of the electromagnetic spectrum.
Bands 1 to 7, including 6, represent data from the visible, near infrared and mid infrared regions.
Band 6 collects data from the thermal infrared 4 region.
4.8 Pixel Types
Figure 4.27: Multi-band raster image [35].
Besides band data, each band contains also information about nodata value and pixel type, which
is a string value. Below are different pixel types supported in PostGIS Raster [6]:
"1BB" = 1-bit boolean
"2BUI" = 2-bit unsigned integer
"4BUI" = 4-bit unsigned integer
"8BSI" = 8-bit signed integer
"8BUI" = 8-bit unsigned integer
"16BSI" = 16-bit signed integer
"16BUI" = 16-bit unsigned integer
"32BSI" = 32-bit signed integer
4 http://en.wikipedia.org/wiki/Infrared.
41

"32BUI" = 32-bit unsigned integer
"32BF" = 32-bit float
"64BF" = 64-bit float
4.9 Nodata Values
Cell values can be either positive or negative, integer or floating point. A noData value is used
to represent cell values that are either unknown or meaningless. In the other words, a noData value
represents the absence of data. Sometimes there are areas in a raster that user does not want to
display. These can include borders, backgrounds or other data considered to not have valid values. In
some cases, these areas are expressed as noData values, although in the other cases, they may have
real values.
When performing operations on raster data, there are typically two ways that noData are treated
for each cell: noData value is returned for the position of corresponding cell, noData is ignored and
can not be computed. So when calculating the statistics for a raster, user can decide to ignore or not
any cells with noData value.
Any analysis functions in PostGIS Raster always takes the nodata value into account. This means
that the areas with nodata value will be omitted from the operation. For example in intersection
operation, if a vector completely overlaps a nodata value area of one tile, the operation will set false
for this tile and vector intersection and returns an EMPTY GEOMETRY result [8]. If the vector partially
overlaps a nodata value area and intersect other parts of the tile, the operation will set true for this
intersection and returns only parts of the tile with significant values, omitting parts with nodata
values.
Each raster band has its own noData value whose range depends on the associated band pixel
type. The noData value is specified in the band information (see Figure ??).
4.10 Blocks or Tiles
When importing a raster image into PostGIS database, users can choose to split the raster into
small regular blocks by adding the -k option to the raster2pgsql.py Python loader, otherwise the
entire raster is loaded into a single block. The block size is a parameter that follows just after the
option -k (such as 128x128). All blocks are registered in a regular blocked table.
Each block is stored in a row of the table and contains all data about itself (such as attribute
information, georeference information, band information and band data) as the mother raster, except
its size is equal the size of mother raster divided by block size. So each block by definition can be
identified as a tile. This also means that in PostGIS Raster, there are no difference between rasters,
tiles and blocks.
Sometimes, the dimension size (row and column) of the mother raster may not be evenly divided
by the block size. PostGIST Raster adds padding to the boundary blocks that do not have enough
original cells to be completely filled. The padding cells have the same cell type as other cells and have
values equal to zero. Padding makes each block have the same block size.
Figure 4.28: Raster tiles [23].
42

Therefore, when the -k option is applied for a raster image, it implies that:
• The multidimensional matrix of cells representing raster image is partitioned into small blocks
(or tiles) for large-scale storage and optimal retrieval and processing.
• All imported blocks have the same width and height.
• All blocks do not overlap and their upper left corners follow a regular block grid.
• The global extent of the raster layer is rectangular and not rotated.
There is, however, no mechanism to enforce these criteria and adding, modifying or deleting a row
from the table might break this regular blocking. So some blocks can be missing in a regular blocked
table. Missing blocks are assumed to be filled with the proper nodata value of each band when they
are displayed in a map.
4.11 Pyramids or Overviews
4.11.1 Concept
The Pyramids concept is used to reduce resolution representations of a raster image to improve
performance in a way that Pyramids can speed up the display of raster data by retrieving only the
data at a specified resolution that is required for the display. For instance, the level resolution of image
is related to the amount of time that an application needs to retrieve and display it, particularly over
the Web. The lower the image resolution, the faster it can be displayed. For example, in the case
where the image details are not important as user has "zoomed out" considerably, loading a lower
resolution image is more adequate than loading entire image with a very high resolution.
Figure 4.29: Pyramids (or Overviews) in PostGIS Raster [24].
When users zoom in, levels with lower resolutions are required to draw successively smaller areas.
However, by using pyramids, performance is maintained because the database server can choose the
most appropriate pyramid level (or the most appropriate resolution) of the raster data to quickly
display it on the screen. In PostGIS Raster, at each level resolution, a copy of image is created and
maintained as a separate raster. So without pyramids, the entire raster must be read from disk and
resampled to a smaller size each time. This is called display resampling process.
Pyramid levels represent reduced resolution images that require less memory space. A pyramid
level of 1 indicates the original raster data, there is no reduction in the image resolution and no
change in the storage space required. Values greater than 1 indicates increasingly reduced levels of
image resolution and reduced storage space requirements. Each successive level of the pyramid is
downsampled at a scale of 4:1 as in Figure 4.30. Pyramids only need to be built once per raster. After
that, they are accessed each time the raster data is viewed. The larger the raster data, the longer it
takes to create the set of pyramids. This also means that more time is saved in the long run.
4.11.2 Creating Pyramids
In PostGIS Raster, a pyramid can be created at the same time as loading a raster image:
python raster2pgsql.py -r *.jpg -t raster_table -s 26986 -l 4 -k 100x100 -I
-o raster_overview_4.sql
43

Figure 4.30: Pyramids at different levels and scales [36].
The above command generates an sql file that will load all jpegs in current folder into a new
table called raster_table, where each raster record is a 100x100 pixels width/height. The -l will
create an overview table for raster_table with pyramid level 4. The overview table will be called
o_4_raster_table and contains less raster records than the table raster_table has. Each raster
record in o_4_raster_table is a 100x100 pixels width/height but at a lower resolution.
The following script will execute the sql file to load the data into postgisdb:
psql -d postgisdb -f aerials_overview4.sql
4.12 Masks
A mask is a special single band raster where each pixel having either the value of 0 or 1. It is
used to define an irregularly shaped region inside another image. The 1-bits define the interior of the
region and the 0-bits define the exterior of the region.
A mask can be formed from a non blank raster. Each band of a non blank raster can also have a
separate bitmap mask associated with it. Thus, there can be at most n+1 bitmap masks associated
with a non blank raster, where n is the total number of bands of the raster. A bitmap mask can also
be edited or updated independently.
A mask attached to a raster must have the same number of rows and columns as any raster bands
and must precisely cover the same area. It uses the same SRS (the spatial reference system) as that
of the raster itself. Logically, a bitmap mask is not an integral part of the raster image but rather an
additional piece of information. Like pyramid, It is stored as a separate raster and independent the
original raster.
Masks are generally used by applications in the following ways [4]:
• A mask can be used to determine which part of the image is displayed because sometimes users
might want to mask out a part of a raster. An example might be one where a raster layer
describes elevations ranging from below sea level to mountain top. If only certain elevations
above sea level are interested, a mask can be applied to the raster layer to visually simulate
a rise in sea level. A such mask will be computed from the source raster. For example, for
every cell greater than or equal to 100, its value is set to 1, otherwise its value is set to 0. The
output is a new raster containing cells with either a 0 (black) or 1 (while) value. The black areas
(respectively the white areas) represent everything below (above) the elevation of 100 meters.
Combining this mask with a suitable background, the effect of raising sea level is effectively
illustrated as show in Figure 4.31.
• When a mask is used as a noData mask in GIS application, the pixel value corresponding to the
exterior (0-bits) of the mask will be treated as noData value. For this purpose, this value can
be registered as a noData value in the raster band information.
44

Figure 4.31: In the right side: The mask. In the left side: the mask with the transparency set to 70%
overlays the digital elevation model raster [10].
4.13 Arrangements of Raster Layers
PostGIS Raster uses only one table for the raster data, the georeference and the metadata as
everything is stored in a single raster attribute. Thus raster attributes composing a raster table
are not related to each other to form a significant coverage. This choice makes the PostGIS Raster
structure very similar to the existing PostGIS vector structure. This also means that raster attributes
from the same table can be of different sizes, can snap to different grids and can overlap like polygons
in a vector layer can overlap.
Hence, how does a raster coverage (or a raster table) look like depends on the structure and the
relationships between the rasters that it contains. The following list represents different possible raster
table arrangements supported in PostGIS Raster:
1. A raster table may be referred to as an image warehouse of untiled and eventually unrelated
images. In this case, these images may or may not overlap since every image has its own
georeference. They may also have no georeference at all.
Figure 4.32: Image warehouse of untiled and unrelated images (4 images) [6].
This type of arrangement is intended for non-geographical users that employ raster data for web
sites or any other usage. That is because in this case, georeference information is not necessary
to be used. Furthermore, this soupless opens the door to other raster processing functions that
could be build from non-geographical users.
Figure 4.33: Image warehouse of cars [22].
2. A raster table may be an irregular tiled raster coverage. Its extent might not necessarily be
45

ectangular because there might be some missing tiles and they might be different sizes. Tiles
of the table are not the same size and do not overlap.
Figure 4.34: Irregular tiled raster coverage (36 tiles))[6].
3. A raster table may be a regular tiled raster coverage. However, its extent might not necessarily
be rectangular when there are some missing tiles. Tiles should be the same size and do not
overlap.
Figure 4.35: Regular tiled raster coverage (36 tiles) [6].
4. A raster table may be a rectangular regular tiled raster coverage. Its extent is necessarily
rectangular and there are no missing tiles. These tiles are all the same size and do not overlap.
This is the traditional way of displaying a coverage.
Figure 4.36: Rectangular regular tiled raster coverage (54 tiles) [6].
5. A raster table is a tiled image (using the option -k when loading the image). Its extent is
necessarily rectangular and there are no missing tiles. These tiles are all the same size and do
not overlap. This type is different from type 4. in that it does not represent a complete coverage.
Other images forming the rest of the coverage are stored as tiled images in other tables.
PostGIS Raster provides the padding mechanism to support regular tiled raster images (or
regular blocking raster images) because raster applications are often optimized to deal with
arrangements 3., 4. and 5..
46

Figure 4.37: Tiled images (2 tables of 54 tiles) [6].
6. A raster table may be a raster coverage resulting from any analysis operations (such as ST_AsPolygon(),
ST_Intersection()) that imply rasterization of a vector coverage. From the rasterization, each
vector feature becomes a small raster with the same extent as the original vector feature. This
type of coverage is not necessarily complete nor rectangular. Tiles should be of different sizes
and might overlap. It all depends on the characteristics of the vector layer being rasterized. For
example, a continuous layer of mutual exclusive geometries (without gaps or holes like a forest
cover) would result in a raster coverage in which significant pixels (with data values) would not
overlap, but non-significant pixels (nodata values) would. On the other end, a discontinuous
layer of mutual exclusive geometries (like a lake or building layer) would result in a coverage of
mostly disjoint raster objects. In practice, this arrangement is identical to a vector layer in the
sense that all pixels of each raster have the same value.
Figure 4.38: Raster object coverage (9 raster objects corresponding 9 rows in a raster table) [6].
All these arrangements are possible and PostGIS Raster does not impose one over the other even
though types 3. and 4. and 6. are the most practical for a GIS analyst. This is due to the fact
that users might add or delete rows (or tiles) of a raster table and PostGIS Raster must support
variable-sized tiles for vector to raster conversion. So it is very difficult to enforce a certain type
of configuration.
The raster arrangements play a fundamental role in implement meaningful vector to raster
conversion (see Section 4.18), where all attributes of a vector are conserved in the resulting
raster table. Example, 10 lake vector features with attributes like name, type, area and etc. are
converted to 10 lake raster features with the same attributes. In this case, if the vector features
and the data values, part of the resulting raster features do not necessarily overlap, the nodata
values of the raster features will overlap (see Figure 4.39). These characteristics enable an easy
integration of raster with vector.
In PostGIS, the conversion from raster to vector can be done using ST_DumpAsPolygons() function
as shown in Figure 4.40. So it is not matter whether the layers are in raster or vector format
47

Figure 4.39: Vector to raster conversion.
when working with analysis functions like ST_Intersection(). For instance, the intersection
operation can perform on vector and vector and results in a vector layer, or on vector and raster
with result as a vector/raster layer or on raster and raster with result as a raster layer.
Figure 4.40: Raster to vector conversion.
However, as each spatial object has a spatial reference identifier (SRID) and only spatial objects
with the same SRID can be used when performing spatial operations. So users have to ensure
that vectors or rasters should have the same SRID before executing some spatial operations on
them.
4.14 Data Formats
PostGIS Raster allows different forms to express raster data:
• Well Known TexT Well Known TexT (WKT) refers to the human readable text format. In the
context of PostGIS Raster, it is used when inserting and retrieving a raster from text format.
Example, the following request creates a raster from a line string with three bands (see Figure
4.41):
SELECT ST_AsRaster( ST_Buffer( ST_GeomFromText(’LINESTRING(50 50,150 150,150 50)’),
10), 200,200,ARRAY[’8BUI’, ’8BUI’, ’8BUI’], ARRAY[118,154,118], ARRAY[0,0,0])
Where ST_Buffer(geometry, radius) returns a new geometry that represents all points whose
distance from the geometry is less than or equal to radius.
Figure 4.41: Image formed by the raster [17].
48

• Well Known Binary Well Known Binary (WKB) refers to the binary form of the Well Known
Text described above. It is used when inserting a raster using ST_RasterFromWKB() (not yet
implemented) and retrieving a raster using ST_AsBinary().
Example:
SELECT ST_AsBinary(rast) As rastbin
FROM dummy_rast WHERE rid=1;
rastbin
-------------------------------------------------------------------------------
0010000000000000000000000000000000000000000000000000000100000000000000000003400
0000000000000000034000000000000000000000000000000000000000000000000001200000000
0012000024000
• Hexadecimal WKB Hexadecimal WKB (HEXWKB) is a hexadecimal representation of the WKB
form. This form is used by the Python loader raster2pgsql.py and when outputting the value
of a raster field like SELECT rast FROM raster_table.
Example:
SELECT rast from dummy_rast WHERE rid = 2;
---------------------------------------------------------------------------------
01000003009A9999999999A93F9A9999999999A9BF000000E02B274A4100000000771956410000000
0000000000000000000000000FFFFFFFF050005000400FDFEFDFEFEFDFEFEFDF9FAFEFEFCF9FBFDFE
FEFDFCFAFEFEFE04004E627AADD16076B4F9FE6370A9F5FE59637AB0E54F58617087040046566487A
1506CA2E3FA5A6CAFFBFE4D566DA4CB3E454C5665
4.15 Physical Storage
PostGIS Raster enables user to store raster images in the database (storing in-database raster)
or merely register them as external files (registering out-database raster). When registering a raster,
only the raster metadata (width, height, number of band, georeference and path to the actual raster
file) are stored in the database and not the pixel values.
4.15.1 Storing In-database Raster Versus Storing Out-database Raster
The choice made between out-database raster and in-database raster depends on user applications.
When user applications are read-only applications, that means that they only retrieve data and do
not prefer to modify data. Out-database raster storage facilitates these processes by providing:
• Fast application access: storing out-database raster stores the path of the raster image file
conserved in the filesystem. It can speed up the reading of rasters files (JPEG, GIF, PNG) for
web applications in the way that only the path is needed to read when retrieving raster data
and there is no need to convert it from the PostGIS Raster type to an image file.
• Fast loading database: in the first time of loading a raster coverage (a table that contains
a colon of type out-database raster or in-database raster), it is executed faster than normally
since only the raster metadata has to be loaded or copied from the database.
• Applicable SQL operators and functions: all operators and functions, that are available for
in-database raster data (except those involving write operations to the raster data like setting
pixel value), work also with out-database rasters.
49

Figure 4.42: Web application with out-database raster [23].
• Simplified backup: since raster images in the filesystem that is not expected to be edited
and only its metadata is registered in the database, in this way, out-database storage can avoid
useless database backup of large data.
In contrast, the traditional in-database raster is related to editing and analysis applications because
of the following advantages:
• Fast analysis: analysis operations performed on raster data (such as dum a raster as polygons
using ST_AsPolygon(), find intersection between raster and vector/raster with ST_Intersections()
and etc.) require working on pixel values. The structure of raster data is nothing else a grid
of two dimensions of cells, where value of each cell represents value of the corresponding pixel.
Obviously, execution time of these operations is faster on in-database rasters since there is no
need to extract the pixel values from an image file of JPEG or TIFF format.
• Single storage: no matter what if data belongs to raster or vector type, PostGIS Raster allows
a single dataset storage. There is no need to store the raster data separately from the vector
data, they can be placed alternatively in the same dataset.
• Multi-Users edition: multiple users can modify raster data at the same time as PostGIS
database handles concurrent editing on in-database rasters.
4.15.2 Registering Out-database Raster
Registering out-database raster is done via the -R option when loading raster image file into
database.
Example:
python raster2pgsql.py -r c:/imagesets/landsat/image.tif -t landsat -R
Where the key option -r specifies the input raster image, the -t option specifies the name of
rater table that contains the corresponding raster and the -R option indicates that the raster image
is registered as a filesystem.
4.15.3 Storing In-database Raster
Like geometry type, raster type is a complex PostgreSQL type composed of many attributes. A
raster attribute may be composed of many bands having a common size, pixel size and georeference.
Besides this, additional information is needed to interpret the image data, such as a pixel type and
a nodata value. The storage used for raster bands is band sequential (BSQ), which is one of three
primary methods for encoding image data for multiband raster images. BSQ is not itself an image
format but is a method for encoding pixel values of an image, where each line of the data is followed
immediately by the next line in the same band. The BSQ data organization is optimal for spatial
50

access of any part of a single band. It can also handle any number of bands and accommodates black
and white, grayscale, pseudocolor, true color and multi-spectral image data.
Figure 4.43: Storing in-database raster.
For example, the following SQL command will create a raster table containing two dummy rasters,
each one is stored in a single row using BSQ:
Create table dummy_rast(rid integer, rast raster);
Insert into dummy_rast(rid, rast)
Values (1,
-- Raster: 0 band, Nodata = 0
(’01’ -- little endian (uint8 ndr)
||
’0000’ -- version (uint16 0)
||
’0000’ -- nBands (uint16 0)
||
’0000000000000040’ -- scaleX (float64 2)
||
’0000000000000840’ -- scaleY (float64 3)
||
’000000000000E03F’ -- ipX (float64 0.5)
||
’000000000000E03F’ -- ipY (float64 0.5)
||
’0000000000000000’ -- skewX (float64 0)
||
’0000000000000000’ -- skewY (float64 0)
||
’00000000’ -- SRID (int32 0)
||
’0A00’ -- width (uint16 10)
||
’1400’ -- height (uint16 20)
)::raster),
-- Raster: 5 x 5 pixels, 3 bands, 8BUI pixel type, NoData = 0
(2, (’01000003009A9999999999A93F9A9999999999A9BF000000E02B274A’ || ’410000000077195
64100000000000000000000000000000000FFFFFFFF050005000400FDFEFDFEFEFDFEFEFDF9FAFEFEFC
F9FBFDFEFEFDFCFAFEF’ || ’EFE04004E627AADD16076B4F9FE6370A9F5FE59637AB0E54F586170870
40046566487A1506CA2E3FA5A6CAFFBFE4D566DA4CB3E454C5665’)::raster);
51

In this example, the dummy_rast table has two rows. Each of them corresponds to a raster tile.
When inserting raster data for the first row, the first byte 01 is used to define the order of bytes
organized in memory, called endianness. If it is a little endian, a computer records 0xA0B70708
in the order 08 07 B7 A0 with the byte lowest first. The rest specifies different components of raster
attribute like version, number band, scale, pixel size, skew, SRID, with and height. Each of them is
separated from each other by a double vertical bar ’||’. However, it is not always in this way, like the
one in the second row. The ’||’ can be placed anywhere in the whole hexadecimal string. As the first
raster tile has no band, there is no string for band data values.
In contrast, more invisibly, raster data in the second row is composed of three long strings but they
can be decomposed into small strings according to structure of a raster type. Such a decomposition
splits the first byte for ’endianness’, the next two bytes for ’version’ and etc. Or, data in the second
raster tile can be rewritten in the following way:
01 -- little endian
00 00 -- version (uint16 0)
03 00 -- nBands (uint16 3)
--Georeference information
9A 99 99 99 99 99 A9 3F -- scale X (float64 0.05)
9A 99 99 99 99 99 A9 BF -- scale Y (float64 -0.05)
00 00 00 E0 2B 27 4A 41 -- ipX (float64 3427927.75)
00 00 00 00 77 19 56 41 -- ipY (float64 5793244)
00 00 00 00 00 00 00 00 -- skewX (float64 0)
00 00 00 00 00 00 00 00 -- skew Y (float64 0)
FF FF FF FF -- SRID (int32 -1)
05 00 -- width (uint16 5)
05 00 -- height (uint16 5)
--Band information
--Band 1
0 -- isoffline, hasnodatavalue (no isoffline, no nodatavalue )
4 -- pixeltype (4 bit 8_BUI)
00 -- nodata value
--Band data values 5x5 pixels
FD FE FD FE FE
FD FE FE FD F9
FA FE FE FC F9
FB FD FE FE FD
FC FA FE FE FE
--Band 2
0400 --
--Band data values 5x5 pixels
4E 62 7A AD D1
60 76 B4 F9 FE
63 70 A9 F5 FE
59 63 7A B0 E5
4F 58 61 70 87
--Band 3
04 00
--Band data values 5x5 pixels
46 56 64 87 A1
50 6C A2 E3 FA
5A 6C AF FB FE
4D 56 6D A4 CB
3E 45 4C 56 65
As shown in Figure 4.44, the raster data stored in the second row of dummy_rast table can be
52

Figure 4.44: In-database raster.
directly seen from Quantum GIS (an open source desktop geographic information systems). For
instance, the first pixel value for the first, the second and the third band are respectively FD, 4E, 46
in hexadecimal. These values correspond to 253, 78 and 70 in decimal.
4.16 Indexing
PostGIS Raster creates indexes over raster data itself. In this way, user can do really complex
operations on raster data, like getting a set of geometry and geometry value from a given raster
band (ST_DumpAsPolygons() function). Unlike Oracle GeoRASTER that creates indexes over spatial
extent, it has more sense when PostGIS Raster creates indexes over raster data.
In PostGIS, loading a new raster table and creating indexes over the raster column can be done
by executing these two lines:
python raster2pgsql.py -r path/*.tif -t raster_table -s 4326 -k 50x50 -I
-o raster_sql.sql
psql -d database -f raster_sql.sql
The -s option georeferences the resulting rasters using SRID 4326.
The -k option splits image files into tiles of 50x50 pixels (one per row). All these tiles do not
overlap and their upper left corners follow a regular block grid. The global extent of the layer is
rectangular and not rotated.
The -I option informs the loader to create the spatial index on the raster table. This index
is very important as it restricts PostGIS Raster’s computing efforts only to the tiles involved in
spatial operations. For example, when finding the intersection of the raster_table and a set of
geometry points, this operation will be performed only on the tiles that actually intersect with the
geometry points. So it is much faster to search for those tiles if they are spatially rather than try
them one after the other sequentially from the raster table. If the -I option was not specified when
using raster2pgsql.py, the index can always be created afterward by executing the following SQL
command:
CREATE INDEX rast_gist_idx
ON raster_table
USING GIST (ST_ConvexHull(rast));
The ST_ConvexHull() function returns the convex hull geometry of the raster including pixel
values equal to BandNoDataValue(). Further, a spatial index can also be made on the geom_points
table to make sure that the intersection operation will be done as quickly as possible:
CREATE INDEX point_geom_idx
ON geom_points
USING GIST (the_geom);
53

One approach to make this intersection operation without PostGIS Raster is to convert those
raster images to shapefiles (a popular geospatial vector data format), import them in PostGIS and do
a traditional intersection query between geometries only. The problem is that converting only one of
those Shuttle Radar Topography Mission (SRTM) files is very difficult. The vector version of only one
of those files exceeds the limit of 2 GB imposed on shapefiles. Since it is mostly impossible to convert
those rasters to vectors, the PostGIS Raster approach is to load them into PostGIS and will vectorize
only ones that are needed to do the requested operation. This approach benefits greatly from the
indexing of raster data in the database.
4.17 Retrieving
As each storing type is specified for a class of application, it results in a strong difference in the way
they retrieve raster data.
4.17.1 In-database Raster Retrieving
There exist two modes when retrieving in-database raster data from PostGIS database:
1. ONE RASTER PER ROW
2. ONE RASTER PER TABLE
PostGIS uses the GDAL driver for exporting rasters via a connection string. The syntax of this
connection string is as follows:
PG":host=’’ port:’’ dbname=’’ user=’’ password=’’
[schema=’’] [table=’’] [where=’’]
[mode=’’]"
Certain fields can be omitted like password in some case. The table option specifies name of a
raster table to connect. The where option is used to filter results from the raster table. The mode
option determines which connection mode is evoked:
• Mode = 1 or ONE RASTER PER ROW mode (the default mode): in this mode, a raster table is
considered as a set of different raster images. Then where parameter is used to select which
raster images will be exported.
• Mode = 2 or ONE RASTER PER TABLE mode: in this case, a raster table is considered as a unique
raster image composed by a set of raster tiles (one raster tile per row). This mode is intended
for reading tiled raster.
When the connection is established, the GDAL driver uses gdal_translate command to create
an image file from the connected raster table:
gdal_translate -outsize 50% 50% "PG:host=’localhost’ dbname=’’
user=’’ password=’’ table=’table_name’ mode=’2’"
table_name.tif
The additional option -outsize will create an image file like the original image but at a different
size (a half size in the above example).
4.17.2 Out-database Raster Retrieving
PostGIS Raster enables user to simply register basic metadata of images stored in the filesystem
without having to actually load their data values into the database. This is out-database storage as
opposed to the more natural in-database storage. In this way, web or desktop applications:
• do not have to retrieve images from the database but instead can access them directly from the
file system (such as JPEG files).
• can use almost PostGIS Raster operators and functions on those rasters using GDAL driver
(out-database raster support is under development).
54

4.18 Conversion
In PostGIS, when using analysis functions, it is no matter whether the layers are raster or vector
type as PostGIS Raster provides a set of functions to support conversion between raster and vector
data.
4.18.1 Raster to vector
With ST_DumpAsPolygons() function, raster data is converted to vector form, along with a set of
geometric objects.
Example:
SELECT ST_DumpAsPolygons(rast)).geom, ST_DumpAsPolygons(rast)).val
FROM raster_table ;
The result of this function is a set of geomval formed by a geometry and a pixel value, where each
geometric object is the union of all pixels of the same band that have the same pixel value. So the
ST_DumpAsPolygons() function is useful for polygonizing rasters in which a single raster will be split
and transformed into multiple polygons. For example, as the circle in Figure 4.45 is composed by two
half-circles that have different values, when applying the ST_DumpAsPolygons() function, it results in
two geometric half-circles by definition.
4.18.2 Vector to raster
Figure 4.45: Conversion from raster to vector [22].
Inversely, a raster object can be build from a geometric object using ST_AsRaster() function. The
vector to raster conversion can be performed without loss of information as PostGIS Raster introduces
the concept of raster object. This concept postulates that geographic entities can be stored as raster
tiles at variable sizes instead of polygons. In the following demonstration, the request will output a
raster containing image of a black geometric circle taking up 150 x 150 pixels as the one in Figure
4.46.
Select ST_AsRaster(ST_Buffer(ST_Point(1,5),10),150, 150, ’2BUI’);
Figure 4.46: Conversion from vector to raster [22].
55

4.19 Intersection
The fact that PostGIS Raster achieves efficient conversion between raster and vector, is the base
to implement certain operations over data, whether it is in raster or vector format. These operations
are essential in PostGIS because they contribute to the accomplishment of its second goal described
in Chapter 3. This goal includes offering a single set of overlay SQL function operating seamlessly
on vector and raster data. Furthermore, in comparison to ORACLE GeoRaster (also in Chapter 3),
they play an important role that makes PostGIS Raster more performing than ORACLE GEORaster
and more adapted to various groups of users (experienced and inexperienced ones). Such operations
include ST_Intersection() that makes intersection between raster and vector data. Several possible
scenarios are available for this operation:
1. A vector and vector intersection with a vector layer as a result.
2. A vector and raster intersection with a raster layer as a result.
3. A vector and raster intersection with a vector layer as a result.
4. A raster and raster intersection with a raster layer as a result.
The result can be a raster or a set of geometry-pixel value pairs (as in the case of ST_DumpAsPolygons())
representing the shared portion of two geometries, two rasters or between a vectorization of a raster
and a geometry. These cases will be examined one by one in the following examples.
Example 1.1: A vector and vector intersection with a vector layer as a result is illustrated in Figure
4.47.
Figure 4.47: Vector and vector intersection with a resulting vector layer [22].
This example examines the first case of intersection between a geometric circle a with a geometric
vegetation cover (type 1 in blue and type 2 in green). The result is a half circle in vector format with
two attributes: a circle name and a cover type.
Example 1.2: A vector and raster intersection with a raster layer as a result.
In general, the intersection begins to select which raster tiles have to be loaded in order to construct
an intersected raster extent (using ST_intersects()) containing all intersected raster tiles. In the
first view, this paradigm makes the operation more efficient in the sense that the number of intersected
raster tiles is often a small number in compare to whole number of existing raster tiles (see Figure
4.48). In second view, the intersection operation works only with these intersected tiles. The raster
extent construction is just the first phase because true intersection will take pixel values (raster format)
or geometries (vector format) into account (using ST_intersection()).
If the output is a raster, then the operation will attach pixel values in the final result as one in
Figure 4.49. So the vegetation cover type represented by a raster is well known. In addition, the
vector attribute is also conserved and become an attribute of the resulting raster.
Example 1.3: A vector and raster intersection with a vector layer as a result.
If the final result is expressed in vector form, the intersecting raster tiles are first vectorized (using
ST_DumpAsPolygon()) into a set of geometry-pixel value and this set is then intersected with the
geometry using the ST_Intersection(vector, vector) function. At this stage, one more time, as
vectorization is a complex task (it is related to each pixel value) and time consume, the paradigm
56

Figure 4.48: Raster intersection paradigm [22].
Figure 4.49: Vector and raster intersection with a resulting raster layer [22].
sounds very useful because it helps PostGIS to vectorize only what is needed to do the requested
operation. As the intersection(vector,vector) → vector operation, the final result is a set of geometrypixel
value pairs representing the shared portion of the geometric and the vectorization of the raster
(see Figure 4.50).
Figure 4.50: Vector and raster intersection with a resulting vector layer [22].
Example 1.4: A raster and raster intersection with a raster layer as a result.
In the case of raster-raster intersection, the result is stored in a multiband raster. The extent of
the resulting raster corresponds to the geometrical intersection of the two raster extents. Nodata area
presenting in any band will result in nodata area in every band of the result. In other words, any pixel
intersecting with a nodata value pixel becomes a nodata value pixel in the result.
To obtain a result similar to the Intersection(vector/vector) → vector operation, the resulting
raster must be vectorized and moreover, this vectorization must take into account multiband.
Different possible scenarios will be one more time examined in the following part but in the case
of mutual exclusive polygons.
Example 2.1: A vector and vector intersection with a vector layer as a result. Here, in Figure 4.52,
the result is trivial.
Example 2.2: A vector and raster intersection with a raster layer as a result.
As in the previous case, the intersection operation results in different disjoint rasters, along with
additional attributes that come from input vectors (Figure 4.53). To obtain a result similar to the
intersection(vector,vector) → vector operation, the resulting rasters must be vectorized.
Example 2.3: A vector and raster intersection with a vector layer as a result.
57

Figure 4.51: Raster and raster intersection with a resulting vector layer [22].
Figure 4.52: Mutual exclusive polygons intersection with a resulting vector layer [22].
Figure 4.53: Mutual exclusive vector-raster intersection with a resulting raster layer [22].
Figure 4.54: Mutual exclusive vector-raster intersection with a resulting vector layer [22].
58

The final result is a set of geometry-pixel value pairs representing the shared portion of the geometric
and the vectorization of the raster (Figure 4.54). This result is the same as the one obtained
from the intersection(vector,vector) → vector operation.
Example 2.4: A raster and raster intersection with a raster layer as a result.
Figure 4.55: Mutual exclusive raster-raster intersection with a resulting raster layer [22].
The same, the results are stored in multiband rasters as different disjoint objects as shown in Figure
4.55. To obtain a result similar to the intersection(vector,vector) → vector operation, the resulting
rasters must be vectorized.
4.20 Temporal Raster
By definition, continuous fields are phenomena that are perceived as having a value at each point in
space and time. Temporal raster, an extension of raster, is a new type designed to take into account
the temporal aspect. In addition, a set of operations allowing interactions between this new type with
date and time [2] are also implemented.
PostgreSQL provides different languages to support new types defined by user, such as C, PL/pgSQL.
PL/pgSQL is a procedural language that allows user to create new types, complex functions and etc.
It is easy to learn and can do anything that can be done in C language. The rest of this section
introduces the new temporal raster type and different functions built using PL/pgSQL.
4.20.1 Structure
Temporal raster is defined as a new composite data type based on raster type and timestamp type:
st_rastertemp( rast raster, temp timestamp)
4.20.2 Operations
Operation ST_InTime is a constructor of temporal raster. It yields for each input raster, a
corresponding temporal raster whose value is a pair of a timestamp instant and a raster value. A
simple pseudocode for ST_InTime is shown below:
Example:
SELECT ST_InTime(raster, ’5/12/2011’)
FROM rasterTable;
Inversely, ST_Instant and ST_Value operations retrieve respectively the timestamp instant and
the raster value of the input temporal raster.
Example :
SELECT ST_Instant(tempRast), ST_Value(tempRast)
FROM tempRasterTable;
59

ST_Present and ST_Passes predicates check whether a temporal raster is created respectively at
an instant or a period of time.
Example :
SELECT ST_Present(tempRast, ’13/7/2001’::date)
FROM tempRasterTable;
SELECT ST_Passes(tempRast, ST_Range(’1/1/2008’, ’31/12/2011’))
FROM tempRasterTable;
Acting in the same way as ST_Present and ST_Passes but instead of returning a boolean,
ST_AtInstant and ST_AtPeriods operations return the temporal raster values.
Example :
SELECT ST_AtInstant(tempRast, ’13/7/2001’::date)
FROM tempRasterTable;
SELECT ST_AtPeriods(tempRast,ST_Range(’1/1/2008’,’31/12/2011’))
FROM tempRasterTable;
In addition, there are also some aggregate functions [13] that execute on multiple temporal raster
values. These values are grouped together as input. The output is a single value computed based
on certain criteria. An example of a simple aggregate function is ST_Sum that calculates the sum of
a sequence of temporal raster values. As each temporal raster is a matrix of values of each point in
space at a time instant, the computation of the sum of this sequence goes back to the computation of
the sum of values of each corresponding point. The output is a summing raster.
Pseudo code of ST_Sum:
WHILE (band

Finally, ST_Min and ST_Max, as indicate their names, return the temporal rasters that have respectively
a minimum or maximum timestamp instant. To find the temporal rasters that have the
minimal timestamp instant, firstly this minimum timestamp instant is found from a loop. Then a
second loop choose only those that match this timestamp as output.
Pseudo code of ST_Min:
WHILE r timestamp of temporal raster r
minTime = timestamp of temporal raster r
WHILE r

Chapter 5
User Guide
Sometimes storing data using raster is the only choice such as imagery is only available in raster
format. Moreover, users can get many other advantages resulting from the nature of raster data,
which is a matrix of cell values. For example, this simple data structure offers the ability to represent
continuous surfaces as a grid of cell values and so enables easy spatial and statistical surface analysis.
The ability to uniform storage as points, lines and polygons, which are all vector features, could also
be stored using raster data.
The following section introduces instructions about how to execute some basic operations such as
importing, exporting raster data, conversion and intersection between raster and vector data, along
with other abilities that PostGIS Raster could offer to users.
5.1 Store and Manage Rasters
5.1.1 Import Rasters
User can import rasters from existing raster images stored in file system using raster2pgsql
Python loader:
(1) raster2pgsql.py -r "c:/temp/mytiffolder/*.tif" -t mytable -s 4326 -o table_name.sql
(2) psql -d postgisdb -f table_name.sql -U -W
The first command (1) generates a sql file that loads file_name.tif with SRID 4326 into a
new table called mytable. The second command (2) runs the SQL script for loading the data into
postgisdb.
Or users can create their own rasters from SQL command:
CREATE TABLE dummy_rast(rid integer, rast raster);
INSERT INTO dummy_rast(rid, rast)
VALUES
-- Raster: 5 x 5 pixels, 3 bands, PT_8BUI pixel type, NODATA = 0
(1,(’01000003009A9999999999A93F9A9999999999A9BF000000E02B274A4
1000000007719564100000000000000000000000000000000FFFFF
FFF050005000400FDFEFDFEFEFDFEFEFDF9FAFEFEFCF9FBFDFEFEF
DFCFAFEFEFE04004E627AADD16076B4F9FE6370A9F5FE59637AB0E
54F58617087040046566487A1506CA2E3FA5A6CAFFBFE4D566DA4C
B3E454C5665’)::raster);
5.1.2 Get and Set The Raster Properties
• Upper left corner coordinates and transformation parameters
–Get the upper left corner coordinates of all rasters in the table:
62

SELECt rid, ST_UpperLeftX(rast)
FROM dummy_rast;
--Set the upper left corner coordinates of the raster whose rid number is 1
--from the table dummy_rast:
SELECT ST_SetUpperLeft(rast,-71.01,42.37)
FROM dummy_rast WHERE rid = 1;
• SRID and number of bands
–Get SRID of the raster whose rid is 1:
SELECT ST_SRID(rast) As srid
FROM dummy_rast WHERE rid=1;
–Set SRID of the raster whose rid is 1:
SELECT ST_SETSRID(rast) As newRast
FROM dummy_rast WHERE rid=1;
–Get number of bands of the raster whose rid is 1:
SELECT rid, ST_NumBands(rast)
FROM dummy_rast Where rid = 1;
• Get and set band properties –Pixel type and nodata value Get pixel type of band number 1 of
the raster whose rid is 1:
SELECT ST_BandPixelType(rast,1)
FROM dummy_rast
WHERE rid = 1;
Get no data value of band number 2 of the raster whose rid is 1:
SELECT ST_BandNoDataValue(rast,2)
FROM dummy_rast
WHERE rid = 1;
Set no data value of band number 2 of the raster, whose rid is 1, to 254:
UPDATE dummy_rast
SET rast = ST_SetBandNoDataValue(rast,2, 254)
WHERE rid = 1;
• Reproject rasters This operation will reproject the geographic locations of rasters into any appropriate
coordinate system for execution of some spatial operations. The reprojection is executed
using ST\_Transform(raster, srid, algorithm). This function uses specified resampling
algorithm such as NearestNeighbor, Bilinear, Cubic, CubicSpline or Lanczos. Default is NearestNeighbor.
The following query transforms all rasters of the dummy_rast table into WGS 84
(having the SRID number 4326) using Bilinear algorithm:
SELECT ST_Transform(rast,4326,’Bilinear’)
FROM dummy_rast ;
• Resample a raste Resample a raster with new dimensions, an arbitrary grid corner and a set of
raster georeferencing attributes. This operation is performed using ST_Resample(raster, width,
height, algorithm). New pixel values are computed using the NearestNeighbor, Bilinear, Cubic,
CubicSpline or Lanczos resampling algorithm. Default is NearestNeighbor.
63

SELECT rast As origin, ST_Resample(rast,100,100) As reduce_100
FROM dummy_rast;
With this query, all rasters will be resampled into new dimension 100x100 using NearestNeighbor
algorith by default.
• Rescale rasters Rescaling a raster corresponds to resample a raster by adjusting its scale (or pixel
size). New pixel values are computed using the NearestNeighbor, Bilinear, Cubic, CubicSpline
or Lanczos resampling algorithm. Default is NearestNeighbor. The following example rescales
rasters to a pixel size of 0.0015 using ST_Rescale(raster, scalexy, algorithm).
SELECT ST_Rescale(rast,0.0015)
FROM dummy_rast;
• Snap rasters to different grid This operation corresponds to resample a raster by snapping it
to a grid defined by an arbitrary pixel corner and optionally a pixel size. New pixel values
are computed using the NearestNeighbor, Bilinear, Cubic, CubicSpline or Lanczos resampling
algorithm. Default is NearestNeighbor. The simple example below uses ST_SnapToGrid(raster,
cornerx, cornery, algorithm) function to snap a raster to a slightly different grid defined by the
pixel corner (0.0002, 0.0002):
SELECT ST_SnapToGrid(rast,0.0002, 0.0002)
FROM dummy_rast;
5.1.3 Vector to Raster Conversion
The example below creates a raster containing a black circle using ST_AsRaster(geometry, width,
height, pixeltype) function. Its dimensions are 150x150 pixels and its pixel type is 2-bit unsigned
integer (2BUI):
SELECT ST_AsRaster(ST_Buffer(ST_Point(1,5),10),150, 150, ’2BUI’);
where ST_Buffer is used to return a geometry that represents all points whose distance from the
point (1,5) is less than or equal to 10 pixels.
5.1.4 Raster to Vector Conversion
Figure 5.1: Raster circle [17].
The conversion using ST_DumpAsPolygons(raster, band) returns a set of geomval (formed by a
geometry and a pixel band value) rows from a given raster band. Default band number is 1. Each
geometry is the union of all pixels that have the same pixel value.
SELECT (ST_DumpAsPolygons(rast)).*
FROM dummy_rast
WHERE rid = 1;
64

5.2 Exporting Rasters
User can export existing rasters from PostGIS database as many small raster files named image1.tif,
image2.tif et etc. Each file corresponds to one row of a raster table.
gdal_translate "PG:host=’localhost’ dbname=’’ user=’’
password=’’ table=’table_name’ mode=’1’" c:/temp/image\#.tif
Or user can produce only one big raster file composed by available rasters of a raster table:
gdal_translate "PG:host=’localhost’ dbname=’’ user=’’
password=’’ table=’table_name’ mode=’2’" c:/temp/image.tif
5.3 Get Raster Statistics
• Get summary statistics The ST_SummaryStats(raster, band, exclude_nodata_value) function
returns summary statistics consisting of count, sum, mean, stddev (standard deviation), min
and max for a given raster band of a raster. Band 1 is assumed if no band is specified. By
default, exclude_nodata_value parameter is true.
SELECT rid, ST_SummaryStats(rast) As stats
FROM dummy_rast;
• Get pixel histogram The ST_Histogram(raster, band, categories, width[ ]) function returns a
set of (min, max, count, percent) records summarizing raster data distribution for a number of
categories(the number of resulting records) of a given band. Number of categories are autocomputed
if it is not specified. Band 1 is assumed if no band is specified. The width[] parameter
is an array that indicates the width of each category. If the number of categories is greater than
the number of widths, the widths are repeated.
SELECT rid, band, ST_Histogram(rast) As stats
FROM dummy_rast;
• Computes quantiles The ST_Quantile(raster, quantiles[]) function computes quantiles for a
raster in the context of the population. A pixel value could be examined to be at 25%, 50%,
75% percentile of the raster data that contains all possible pixel values.
SELECT ST_Quantile(rast, 0.75) As value
FROM dummy_rast WHERE rid=1;
• Get number of pixels The ST_ValueCount(raster, values[]) function returns a set of records
containing a pixel band value and the number of pixels in a given band that have a given set
of values. If no band is specified, band number 1 is used by default. By default nodata value
pixels are not counted and pixel band values are rounded to the nearest integer.
SELECT ST_ValueCount(rast) As value
FROM dummy_rast WHERE rid=1;
5.4 Display Rasters
Normally any softwares, that use GDAL to read raster and enable a database connection, are able
to display raster directly from PostGIS database. Examples of such software include:
• QGIS plugin
65

• gvSIG plugin
• MapServer through GDAL
Other softwares that do not support direct raster displaying but instead they allow displaying a
vectorization of the raster. This is the case of any softwares that are used to display vector PostGIS
queries such as OpenJump or ArcGIS 10.
– OpenJump
SELECT ST_AsBinary((ST_DumpAsPolygons(rast)).geom), (ST_DumpAsPolygons(rast)).val
FROM dummy_rast
WHERE rid=1;
– ArcGIS 10
Same as OpenJump but without ST_AsBinary():
SELECT (ST_DumpAsPolygons(rast)).geom, (ST_DumpAsPolygons(rast)).val
FROM dummy_rast WHERE rid=1;
5.5 Edit and Compute Rasters
• Set pixel value The ST_SetValue(raster, band, geometricpoint, newvalue) function returns a
new raster resulting from setting the value of a given band in a given column x and row y
pixel or at a pixel that intersects a particular geometric point. Band number starts at 1 and is
assumed to be 1 if not specified.
UPDATE dummy_rast
SET rast = ST_SetValue(rast,1, ST_Point(3427927.75, 5793243.95),100)
WHERE rid = 1 ;
• Reclass rasters The ST_Reclass(raster, band, reclassexpression, pixeltype, nodatavalue) function
creates a new raster by applying a valid PostgreSQL algebraic operation that defines a class
expression on the input raster band. The band number specifies the band to be changed. If it
is not specified, it is assumed to be 1. All other bands are unchanged. The new raster will have
the same georeference, width and height as the original raster. Use case: convert a 16 BUI band
to a 8 BUI and so on for simpler rendering.
ALTER TABLE dummy_rast ADD COLUMN reclass_rast raster;
UPDATE dummy_rast
SET reclass_rast = ST_Reclass(rast,2,’0-87:1-10, 88-100:11-15, 101-254:0-0’,
’4BUI’,0)
WHERE rid = 1;
• Clip rasters The ST_Clip(raster, band, geometry) function returns a new raster that is clipped
by an input geometry. If no band is specified all bands are returned.
SELECT ST_Clip(rast, ST_Buffer(ST_Centroid(ST_Envelope(rast)), 20), false )
FROM aerials.boston
WHERE rid = 4;
• Apply algebra operation The ST_MapAlgebraExpr(raster, band, pixeltype, expression) function
creates a new one band raster by applying a valid PostgreSQL algebraic operation on the input
raster band. Band 1 is assumed if no band is specified. The new raster will have the same
georeference, width and height as the original raster but will only have one band. If pixeltype
is given, then the new raster will have a band of that pixeltype. If pixeltype is passed NULL,
then the new raster band will have the same pixeltype as the input rast band.
66

Figure 5.2: Clipping raster effect [17].
ALTER TABLE dummy_rast ADD COLUMN map_rast raster;
UPDATE dummy_rast SET map_rast = ST_MapAlgebraExpr(rast,NULL,
’CASE WHEN rast < 0 THEN rast+10 ELSE NULL END’)
WHERE rid = 1;
Figure 5.3: Algebra operation on rasters [23].
5.6 Convert Rasters to GDAL Format
5.6.1 Raster to TIFF File
The ST_AsTIFF(raster, numberbands, compression) function returns the selected raster bands as a
single TIFF image (byte array). If no band is specified, it will try to use all bands. Where:
– numberbands: an array of bands used to export (max is 3 bands). The order of the bands is
RGB. For example, ARRAY[3,2,1] maps band 3 to Red, band 2 to green and band 1 to blue.
– compression: a compression type such as JPEG90 (or some other percent), LZW, JPEG or
DEFLATE9.
SELECT ST_AsTIFF(rast, ’JPEG90’) As rasttiff
FROM dummy_rast
WHERE rid=1;
5.6.2 Raster to JPEG File
The (ST_AsJPEG(raster, band/numberbands, quality) function returns the selected raster bands as
a single JPEG image (byte array). Where:
– band: a number that specifies a band for single band export.
– numberbands: an array of bands used to export (max is 3 bands). The order of the bands is
RGB. If more than 3 bands are specified, then only the first band is used. If only 3 bands are specified,
then all 3 bands are mapped to RGB. For example, ARRAY[3,2,1] maps band 3 to Red, band 2 to
green and band 1 to blue. If both band and numberbands are specified, band 1 is used by default.
– quality: a number ranged from 0 to 100. Default is 75.
67

SELECT ST_AsJPEG(rast) As rastjpg
FROM dummy_rast
WHERE rid=1;
5.6.3 Raster to PNG File
The ST_AsPNG(raster, band/numberbands, compression) function returns the selected raster bands
as a single PNG image (byte array). Where:
– band: a number that specifies a band for single band export.
– numberbands: an array of bands used to export. Bands are mapped to RGB or RGBA space.
For example, ARRAY[3,2,1] means map band 3 to Red, band 2 to green and band 1 to blue.
If both band and numberbands are not specified, then all bands are used if the raster has 1, 3, or
4 bands and. If it has 2 or more than 4 bands, then only band 1 is used.
– compression is a number from 1 to 9.
SELECT ST_AsPNG(rast) As rastpng
FROM dummy_rast WHERE rid=2;
5.7 Intersect Rasters with Vectors
One approach to make this operation is to convert all raster tiles to a set of geometries and do a
traditional intersection query between geometries only. The problem is that vectorizing all the tiles
will be too long and it is impossible to load a huge number of geometry in GIS software. Since it is
mostly impossible to convert those tiles to vectors, the PostGIS Raster approach will vectorize only
what is needed to do the requested operation.
In the intersection, if a geometry completely overlaps a nodata value area of one tile, st_intersects(rast,
geom) will return false for this tile/geometry couple and st_intersection(rast, geom)
will return an EMPTY GEOMETRY without vectoring this tile. If the geometry partially overlaps a nodata
value area and intersects other parts of the tile, st_intersects(rast, geom) will return true for this
tile/geometry couple and st_intersection(rast, geom) will return only values of parts of the tile
that intersects the geometry.
5.7.1 Intersect Rasters with Points
Practically, user can use intersection operation to extract ground elevation values for lidar (Light
Detection And Ranging) 1 points.
SELECT pointID, ST_Value(rast, geom) elevation
FROM lidar, srtm WHERE ST_Intersects(geom, rast)
Figure 5.4: Intersection of raster and geometric points [23].
In the above request, the intersection(raster,vector) operation is not necessarily needed. This is
due to the fact that each geometric point is represented by a pixel or a cell of a grid. So to extract
1 http://en.wikipedia.org/wiki/LIDAR.
68

elevation values at lidar points, only the ST_Value() function, that gives elevation values at particular
geometric points, is needed.
5.7.2 Intersect Rasters with Lines
In the case of geometric lines, this operation is used to extract elevation values for each road
segment in a road network.
SELECT roadID,
(ST_Intersection(geom, rast)).geom road,
(ST_Intersection(geom, rast)).val elevation
FROM roadNetwork, srtm
WHERE ST_Intersects(geom, rast)
User accesses the geometry and the value parts of the resulting set of geomval by surrounding
them with parentheses and adding .geom or .val at the end of the expression.
5.7.3 Intersect Rasters with Polygons
Figure 5.5: Intersection of raster and roads [23].
In the same way, user can extract temperature values for each polygon (circles in the following
example).
CREATE TABLE buff_temp AS
SELECT id, (gv).geom buffer, (gv).val temp
FROM (SELECT id, ST_Intersection(geom, rast) gv
FROM buffers, temperature
WHERE ST_Intersects(geom, rast))
Figure 5.6: Intersection of raster polygons [23].
Then, the result can also be used to compute the mean temperature for each polygon.
CREATE TABLE result AS
SELECT id,
sum(ST_Area(the_geom) * val) / sum(ST_Area(the_geom)) AS meantemp
69

FROM buff_temp
GROUP BY id
ORDER BY id;
5.8 Aggregate Functions
Figure 5.7: Different temperature values in a polygon [23].
The ST_Union(rast, band, expression) function returns a single band raster from the union
of a set of raster tiles. If no band is specified for the union, band number 1 is assumed. The
resulting raster’s extent is the extent of the whole set. The resulting raster pixel values are defined by
expression parameter. User can define their own expressions or use predefined functions like FIRST,
LAST, MIN, MAX, SUM, MEAN and COUNT. The default function is LAST when no expression
is specified. The LAST (respectively FIRST) function merges all rasters together by unioning their
extents and burning the last (first) raster without taking the other raster pixel values into account.
SELECT (ST_Union(rast, ’last’)
FROM dummy_rast;
5.9 Create a High Resolution Analysis Grid
Even though, to compute road and river length, mean temperature and etc, a vector format can
be used to store each cell of a grid as a polygon. Then necessaire operations are applied to the set
of polygons. It sounds like raisonnable but with a huge set of polygon (500 000 x 300 000 polygons)
resulting from a large area, these polygons are unmanageable. In contrast, by using raster, users can
choose to encapsulate them into small blocks as far as they want for easy analysis.
For example, all of USA can be stored as a grid of 500 000 x 300 000 pixels or cells (each cell
represents an area of 10 m). In raster format, these polygons are manageable in 15 000 000 tiles
of 100x100 pixels. The rest is to intersect vector layers with the raster tiles and apply appropriate
operations to the intersection result.
5.10 Create a Specialised Web or Desktop GIS Application
Complete SQL GIS
With the raster API, PostGIS is now a very complete SQL GIS as:
• It provides a single set of overlaid functions that work seamlessly on raster and vector files. So
an easy SQL API to manipulate and analyse graphical data.
• All geographic data (vector and raster) is spatially indexed. So there is no need to write complex
C,C++, Python or JAVA code to manipulate complex data.
70

Figure 5.8: Gird of cells of USA map [23].
• PostGIS provides several storage solutions for applications whether they are read-only applications
(storing out-database raster) or editing and analysis applications (storing in-database
raster).
With out-database raster, image files (JPEGs) can be stored as they are in any GDAL format.
So it speeds up geoprocessing (a GIS operation used to manipulate spatial data) by providing
direct access for web applications.
The traditional in-database raster, that stores raster data as a grid of pixel values, is intended
for analysis operations that require working on pixel values. Obviously, execution time of these
operations are faster than ones executed on out-database rasters since there is no need to extract
the pixel values from an image file (such as JPEG, TIFF and etc).
Lightweight Multi-users
As all geoprocessing is done in the database, these processes thus are kept close to the data where
the data should be (in a database). Hence, desktop and web GIS applications become simple SQL
query builders and data viewers.
Figure 5.9: Web GIS application [23].
71

Chapter 6
Application
Most of GIS applications often relate to the history of changes of phenomena like temperature,
precipitation and etc. This section intends to introduce a novice application that shows the evolution
in time of a set of temporal raster data, where each temporal raster is a matrix of values of points
in space at a time instant. Each temporal raster can also be referred as a map, hence the evolution
is considered as a sort of movie-map. The application is built on Quantum Geographic Information
System (QGIS). So, the first part of this section will introduce QGIS and its application. The second
part focuses on the time application.
6.1 Quantum GIS
Quantum GIS (QGIS) is a Geographic Information System application that provides geographical
data viewing, editing and analysis capabilities using graphical interface.
QGIS is cross platform and runs on Linux, Unix,Mac OS X and Windows. By using GDAL,
GRASS GIS, PostGIS and PostgreSQL as extensions, QGIS provides access to various vector, raster
formats and works on multiple databases. QGIS is written in C++ and allows integration of user
plugins developed using either C++ or Python.
6.1.1 Advantages
There are several reasons that make QGIS become one of the tools used in most GIS and preferred
than other GIS applications [20]:
• Quantum GIS is open-source.
• It offers a similar interface as market software like ArcGIS or SuperGIS.
• It can view, edit and create a wide variety of vector formats ( Shp, vectors Grass, PostgreSQL
/ PostGIS and etc.)
• Its software that evolves and improves continuously through custom plugins (Python or C++).
• It allows interoperability with external databases (such as PostGIS,MapServer and etc.).
• It is available in multi-language and has a large volunteer community.
• Its capability to run with additional tools (Openstreetmap, Google Maps and etc.).
• It connects with external sources of data (WFS and WMS servers).
• The graphical interface of QGIS is much cleaner and faster than one of GRASS GIS.
• Its possibility of using Quantum GIS as a graphical interface GRASS GIS through a plug-in.
• Quantum GIS requires a small memory space.
72

6.1.2 Utilisation
QGIS is designed with a graphical user interface to enhance user interaction. It consists of a layout
area, a layers panel, toolbars and a status bar (see Figure 6.1).
The layout area is where the map is displayed.
The layers panel contains a list of layers that are added to the project. In this area, user can make
a layer visible (not visible) by checking (unchecking) the box related to its layer name. Only the top
layer is displayed in layout area as this one masks bottom layers.
Figure 6.1: QGIS interface.
The status bar is located at the bottom right. It displays coordinates of the current position, the
scale of the area, and the projection property of the layer.
The toolbars describe the features found in the menu. They comprise four main parts (as shown in
Figure 6.2 from left to right): the file part for project management, the view part for repairing the area
being worked on, the layer part for layer management and the extra part for extension functionalities
(such as GRASS, WKTRaster and etc.).
Figure 6.2: QGIS toolbars.
In the layer management tool, user can import a vector or a raster layer from filesystem. Furthermore,
a remarkable feature, that makes QGIS become the first choice in comparison with GRASS, is
its ability to connect with PostGIS database. Then user can load a vector or a raster layer directly
from PostGIS into QGIS.
However, just only QGIS with version from 1.7 supports raster. This feature is integrated through
WKTRaster plugin.
73

6.2 Tools
Figure 6.3: PostGIS connection.
This part focuses on a new approach to represent geographic data such as temporal raster data.
This approach consists of exploiting the temporal features of raster to show the evolution in time of
real world phenomena, so called time travel. In addition, an extra tool that allows user to create a
specific color legend for a set of temporal raster data is also represented. The extra tool thus enhances
the time travel application.
6.2.1 Interface
At the beginning, the application starts with listing all of the tables that have a column of type
temporal raster in the Temporal raster table(s) window. These tables are retrieved directly from
PostGIS database as QGIS is capable of establishing a connection to PostGIS. Users will choose one
table for visualizing the evolution in time of temporal rasters existed in this table.
Figure 6.4: Temporal raster window.
74

Time Travel
The time travel feature is available through a separate window. This window is divided into three
main areas: time , media and band areas. The time and media areas gather all of the necessary
options for controlling the evolution in time of temporal raster data. The band area relates to the
band properties.
Time Area
Figure 6.5: Time travel interface.
As shown in the time region, the first slider bar represents the evolution in time of a set of loaded
temporal rasters. The interval between two ticks represents the minimum time interval between two
any temporal rasters. It is expressed in day unit and its value can be changed from the spin box
Time interval. Just behind these ticks, on the left side is the minimum time instant and on the
right side is the maximum time instant that exists in the temporal raster set. The current time field
indicates the current time instant of the evolution as the slider moves.
The Begin time and End time fields express the evolution time interval in which all the temporal
rasters that have a time instant included in this interval will be counted in the evolution. For example,
user may choose to view the evolution beginning at some particular time instant (2001-07-13 08:17:34
in the figure) by entering this time instant in the Begin time field. Otherwise, it can be taken from
the calendar button, located just next to the Begin time field. The same for the End time field.
Media Area
The media area offers different ways to control the evolution : move forward to next temporal
raster (or map), move backward to previous map, move to begin map, move to final map, play and
stop temporal raster evolution (or movie). Furthermore, the second slider bar help user change the
speed of the movie. This speed varies from minimum 0.05s to maximum 2s.
Band Area
It depends on the number band that each raster possesses, the application proposes appropriate
options. In the one band case, the raster is a gray image because each pixel point is expressed by a
single value. Hence, the application provides user the possibility to load a color palette from a palette
file (Load a style field). The palette file is created from the color legend button (this feature is
described later). When rasters have multiple bands (color images), users have possibilities to view
overlapping multiple bands (Multi Band Color radio button) or each band individually ( Single Band
Color radio button). If the Single Band Color radio button is marked, a compose box with a list of
band numbers will appear to help user choose to visualize a desired band.
The Single Band radio button (respectively Multi Band radio button) is not an option for users.
It is used to inform users if rasters are single band (respectively multi band). Both of them are
75

Figure 6.6: Band area.
controlled by the application and choices made by users on these options have no effect either on
Multi Band or Single Band£ button.
6.2.2 Color Legend
The color legend is used to indicate the significance of color by numbers. It consists of a title, a
color bar and a common unit for numbers and tick notations. In the example below, the red color
area expresses high temperature and the blue color area expresses low temperature.
Figure 6.7: The temperature color legend.
As each temporal raster is a matrix of values of points in space, hence it is referred to as a map.
Each map possesses normally a minimum value and a maximum value. If Time travel is an application
that shows the evolution of a set of maps, then the color legend representing all these maps needs to
be computed from this set. It does not make sense when the color legend of a specific map is used
for a set of maps. This also means that the minimum value (respectively maximum) at the bottom
(respectively top) of the color legend is the minimum value of a set of maps.
The color legend is available through another window than Time travel window. The Output color -
palette file field let user enter the desired name for the created color palette file. In the Color palette
region, the left compose box contains all of the available color palettes that help users to freely choose
a satisfied one. Behind this box, the chosen color palette is shown.
The Apply button aims to apply the chosen palette to the current map for users to visualize, along
with creating a corresponding color palette file with the name given by the Output color palette file
field. Users can repeat this step as many times as the number of existing palettes. Once a color palette
file is created, its name is registered in the Created color palette(s) compose box. On the one
hand, these color palette files serve as style files for maps and are used to create corresponding color
legends on the other hand. Finally, the Create button creates the color legend of the current color
palette file. The result of the color legend is simply a PNG image file.
6.3 Implementation
The application is developed as a plugin [12] in QGIS using PyQt, the bridge between Python and
Qt4 graphics library.
76

Figure 6.8: The color legend interface.
QGIS retrieves the temporal raster table from PostgreSQL using a connection string and a SQL
command. The connection string is used to establish a connection to PostgreSQL. Its structure is
’PG: dbname=%s host=%s user=%s password=%s port=%s’. PostgreSQL then executes the SQL
command and outputs temporal raster table list as a result.
6.3.1 Time Travel
Loading Data
To show the revolution in time of a set of temporal rasters, first of all, data need to be loaded
from PostgreSQL into QGIS. There exist two approaches for loading temporal raster (or map). The
first one attempts to load each map, one after another from PostgreSQL database into QGIS. When
loading a map, there is a change from QGIS cursor to PostgreSQL cursor. That is because when QGIS
establishes a connection to PostgreSQL, PostgreSQL will execute a request to retrieve a temporal raster
row. When the map is loaded, necessary PyQt instructions like updating current time, progress slider
bar etc. are need to be executed to show the evolution. In turn, a cursor change from PostgreSQL to
QGIS occurs again before these instructions take place, as PyQt instructions are managed by QGIS
cursor. These steps are exactly what will happen to the next loaded map and so on until the final
loaded map. This approach sounds cumbersome and involves a lot of execution time due to cursor
changes.
Figure 6.9: The loading map sequential approach.
In the second approach, all maps are loaded into QGIS memory once and for all. Hence, only two
cursor changes are needed (one from QGIS to PostgreSQL and another from PostgreSQL to QGIS).
Then, the visibility of each map is simply handled by turning on or off the check box corresponding to
this map from the layers panel as described in the previous section. This approach reduces enormously
execution time. Another advantage results from this is that all maps can be organized in an order
(descending or ascending) by modifying the SQL command. The ordering can not happen in the first
approach as each map is individually loaded.
Pseudo code of function iniLayers that loads all layers into QGIS using the second approach:
def iniLayers(self):
Retrieve a list of temporal raster IDs to rids
Change QGIS cursor to PostgreSQL cursor
77

Figure 6.10: The loading map parallel approach.
For each rid in rids:
Retrieve a layer from PostgreSQL using a connection string
Try to add layer to qgis
Come back to QGIS cursor
Figure 6.11: Maps representing world climate [7] are loaded once time. Their visibility is handled by
turning on/off the check boxes list.
Media Control
Each loaded map is identified by an ID. All map IDs are saved in a list layerIds. Only maps that have
a time instant included in interval from Begin time to End time will be shown in the evolution. When
the play media button is triggered, the first map that satisfies the time condition will be displayed.
Subsequently, value in the Current time field is updated by time instant of the first map and the slider
in the Time region moves by a step. Each step is at least equals the tick interval (by definition of
tick interval in the previous section). Then a QTimer ctimer is necessary to increase the time interval
between two displayed maps instances. Otherwise, all maps that satisfy the time range condition will
be displayed nearly at the same time and human eyes can not recognize the succession of the maps.
Users can drag and drop the speed slider to change this time interval from 0.05s to 2s. After the
ctimer is time out, it triggers the displaying function displayLayer to display the second map and so
on until the final map is reach.
Pseudo code of function playMedia:
If a map satisfies the time condition
Display the map
Update current time field
Move slider
Trigger ctimer
Pseudo code of function displayLayer:
Retrieve a map from a map id
Turn on the visibility of this map
78

Band Properties
Figure 6.12: Media area.
Depend on the number of bands that a map possesses, the Single band radio button (respectively
the Multi band radio button) will be marked if the map has only one band (multi band).
Figure 6.13: Single band rasters.
In the case of one band, each map is coded in gray color (a pixel color is coded either in RGB or
CMYK color models that require at least three band). However, user can load a color palette from a
style file created from color legend option. A style file describes each pixel value by a RGB triple in
each row. Example of rows in a style file is as follows:
"17.00592,139,0,116,255,
17.56223,137,0,118,255,
..."
The first number is a pixel value. The next three numbers are a RGB triple. The final number
is alpha value that indicates the transparency level between an image and a background. In this
example, 255 will create a full transparency.
To set up a color palette for a map from a style file, firstly QGIS uses the file to create Color Ramp
Entries array object and then set this object for the maps.
Pseudo code to set up a color palette:
Create Color Ramp Entries array object
For each loaded map
Tell the map to use a QgsColorRampShader function
Get a pointer to the raster shader function
Set parameters for the QgsColorRampShader function using Color Ramp Entries array
Refresh the map
In multi band case, as maps are already coded in either RGB or CMYK color models, then there
is no option to create a color table. Instead, it offers the possibility for user to choose a specific single
band for visualization or switch to multi band mode. By default, color maps are displayed in multi
band color mode. When single band color mode is activated and a band number is chosen, the function
mulSinBandCol is used to turn on only the chosen band and turn off the others band for all maps.
Pseudocode for mulMulBandCol:
For each loaded map
Turn the map on MultiBandColor mode.
For single band color, pseudocode of mulSinBandCol is:
79

Figure 6.14: Maps representing world climate are loaded as single band rasters. They are turned into
color maps by loading a created color palette "worldclim.txt".
Figure 6.15: Map is loaded as multi band raster.
80

For each loaded map
Turn the map on MultiBandSingleBandPseudoColor mode
Set the chosen band as the band to be displayed
Set PseudoColorShader as color shading algorithm for the map
6.3.2 Color Legend
Figure 6.16: Visualisation of band number 2 of a multi band map.
The most important function in color legend construction is computeRasterStats. It computes the
minimum and the maximum pixel values from a set of temporal rasters (or maps). In fact, these values
are resulted from PostgreSQL through a connection. For instance, the minimum pixel value is computed
from the request SELECT min((ST_SummaryStats((rtt).rast)).min) FROM temporalRasterTable
(the temporal raster table is described in the interface Section). The (ST_SummaryStats((rtt).rast)).min)
part gives, for each temporal raster in the temporalRasterTable, a corresponding minimum pixel value.
Hence, the min((ST_SummaryStats((rtt).rast)).min) gives the minimum value of a set of minimum
values according to the definition of the minimum pixel value of a set of temporal rasters.
Each color palette is provided from an external tbl file. The tbl file describes a RGB triple of
color in each row. Hence the number of possible colors of a color palette is the number of lines used
to code colors. Example of a tbl file is something like:
"{ r g b }
102 47 0
153 96 53
..."
Each time, when users choose a color palette from Color legend window, the information related
to this palette, such as the number of colors, the RGB values array and the real data values, is saved
respectively into three variables nColo, arColo and arColV. The creation of a color legend file (a PNG
file) implies the use of these variables, along with the minimum and maximum pixel values of a set of
temporal rasters. For instance, the number of colors decides the color legend length, where each RGB
color is painted as a color ligne. The tick notations are given by variable arColV. The number of tick
notations depends on the number of colors that a palette has. If it has less than 10, the number of tick
notations is equal to the number of available colors. Otherwise, it will be a fixed number and equal
to 10. In this case, the number of colors between two tick notations is the division of the number of
colors by 10.
81

Figure 6.17: Maps with a color legend.
82

Chapter 7
Conclusion
Geography that evolves spatial organization and material character of the Earth’s surface is a
relevant source to decision makers in business and government. For example, underlying all of the
recommendations results that the demand for contributions from geography and the supply capacity of
current resources, are far out of line. Otherwise, faulty actions are taken and taken quickly, geography’s
contributions will be severely constrained or may decline in quality.
Even in the areas where geography is not contributing in a significant way, it could be seen as a
provider giving resources and information for decision makers. Geography provides a valuable way
of thinking about human and environmental issues. Geography integrates phenomena and processes
in particular places, in connection with the search, to contributes to science as an creative effort to
advance the frontiers of knowledge. For example, the relationship between population, society and
environmental resource has been a central issue for science. In this way, geography focuses on the
nature of that relationship and analysis processes to identify relations between changes in population,
environment and society.
In many ways, geography sees the world as a continuous surface of landscapes, movements and
interactions. Research on interdependencies between places can help to explain and predict spatial
interactions. For example, decisions to relocate are the important decisions made by households that
have, as a consequence, significant implications for the links between places. Recent work on the
nature of the migration and mobility process shows that decisions to move are related to age, family
composition and economic circumstances. Studies of interdependencies between places contribute also
to the understanding of several critical issues for society. For example, one of the best illustrations of
spatial interdependence addresses the spread of infectious diseases. The spread of such diseases can
be understood and predicted by using spatial modeling techniques related to location, synthesis and
scale [19].
So a GIS with only vector data type relates to geometric elements (such as roads, houses, trees,
etc.) is not enough. The raster data type made of a matrix of points, suitable to describe continuous
fields, phenomena that are perceived as having a value at each point in space and time, is not only
a necessary complement type of vector but also a powerful feature that makes GIS become a perfect
graphical tool. Current trends in GIS techniques suggest a future in which researchers, students,
business people and public policy makers will explore a world of shared spatial data. They will
request analyses from a rich menu of options, select preferred geographic area and spatial scale for
analysis and export results in multimedia formats. So, in developing GIS tools, developers need to
think more not only about context of problems, but also about knowledge and skill levels of users as
the users of tomorrow will be more numerous than at present.
Furthermore, more development is needed to turn GIS applications into robust, efficient and fully
functional tools. For example, advanced features like a query editor tool providing facilities for formulating
a query, a query viewer that displays result of geographic queries with various additional
features such as an intelligent zoom, different possible visualizations of the same data by navigating
in time and in different perceptions [1].
More importantly, with the demonstrated important role of geographers in decision making, institutions
and leaders in government, business, education and the communications media need to raise
83

their levels of awareness of geography’s value to science and society and find more effective ways to
publicize and utilize geography’s perspectives, skills and knowledge.
A multimodal interface is defined as an interface that uses two or more means of modalities to
interact with a computer. The usability of current GIS in the future will be improved using multimodal
interfaces as these interfaces can give users more expressiveness and flexibility as well as better tools
for controlling sophisticated visualization and multimedia output. For example, in planning and
management, current GIS require extensive time and training for proper operation. So crisis managers
may not have the time or training needed to use them effectively. In this case, multimodal interfaces
can be more intuitive than conventional interfaces.
By coupling the visually expressive nature of maps with multimodal interfaces, it will expand the
range of possible users to include non-GIS specialists as inexperienced users can mostly interact with a
map and express commands multimodally when describing spatial information about location, number,
size and orientation or shape of an object. What remains to be determined is which combination of
modes that will create more intuitive interfaces to maps and GIS.
In Human-Computer Interaction (HCI), speech and gesture are the most natural means of humanhuman
interaction. The Geo-Collaboration Crisis Management (GCCM) project is presently developing
a Dialogue Assisted Visual Environment for Geo information (DAVE_G) [21] that uses a combination
of speech and gesture as inputs. The key functionality of DAVE_G is its ability to coordinate
a collaborative dialogue between itself and users. For example, a user can say "DAVE", zoom to this
area ? and makes a circular gesture around a desired area. DAVE_G gives the answer back with
text, speech response and image display. In the future, it is being extended to become a tool for use
in emergency response centers.
84

Bibliography
[1] Christine Parent, Stefano Spaccapietra, Esteban Zimányi. The MurMur project: Modeling and
querying multi-representation spatio-temporal databases. Science direct, 2006.
[2] Alejandro Vaisman, Esteban Zimányi. A multidimensional model representing continuous fields in
spatial data warehouses. Databases and Theoretical Computer Science. ACM Press, 2009.
[3] Qingyun (Jeffrey) Xie and Jayant Sharma, Jean Ihm. Oracle Database 11g GeoRaster: An Oracle
Technical White Paper, June 2007.
[4] Qingyun (Jeffrey) Chuck Murray, Qingyun (Jeffrey) Xie, Weisheng (Terry) Xu, Sophia Yuditskaya,
Zhihai Zhang, Hongwei Zhu. Oracle Spatial GeoRaster Developer?s Guide, 11g Release 1 (11.1):
GeoRaster Overview and Concepts. August 2008.
[5] Sandro Bimonte. Intégration de l’information géographique dans les entrepôts de données et
l’analyse en ligne : de la modélisation à la visualisation. University thesis. Informatique et Information
pour la Société - L?Institut National des Sciences Appliquées de Lyon. 207 pages, 18
Decembre 2007.
[6] Pierre Racine , Bborie Park, Sandro Santilli, Mateusz Loskot, Jorge Arevalo, Regina Obe,
David Zwarg. PostGIS Raster Beta Documentation. http://trac.osgeo.org/postgis/wiki/
WKTRaster/Documentation01. Last updated: 2012-03-10.
[7] Robert J. Hijmans, Susan Cameron, Juan Parra, Peter Jones, Andrew Jarvis, Karen Richardson.
WorldClim: Global Climate Data, 2005. http://www.worldclim.org/formats.
[8] Pierre Racine. PostGIS WKT Raster Tutorial 1: Intersecting vector polygons with large
raster coverage using PostGIS WKT Raster. 2005. http://trac.osgeo.org/postgis/wiki/
WKTRasterTutorial01. Last updated: 2011-06-20.
[9] Jay Busselle. Computer Technology: Raster Images versus Vector Images. http://www.
signindustry.com/computers/articles/2004-11-30-DASvector_v_raster.php3
[10] Gary Sherman. Spatial Galaxy: Exploring the depths of open source GIS and more. Using
the QGIS Raster Calculator, 25 January 2012. http://spatialgalaxy.net/2012/01/25/
using-the-qgis-raster-calculator/.
[11] Environmental Systems Research Institute. ArcGIS Desktop Help 9.3: Cell size of raster
data. http://webhelp.esri.com/arcgisdesktop/9.3/index.cfm?TopicName=Cell_size_of_
raster_data. Last undated: 2009-11-12.
[12] Martin Dobias. PyQGIS Developer Cookbook: Developing Python Plugins. http://qgis.org/
pyqgis-cookbook/plugins.html. Last undated: 2012-07-03.
[13] The PostgreSQL Global Development Group. PostgreSQL 8.1.11 Documentation: Create aggregate,
pages: 787-789.
[14] Jorge Arevalo. PostGIS Raster Tutorial Series. Comparing Oracle GeoRaster with PostGIS
WKT Raster (I). http://gis4free.wordpress.com/2010/07/19/oracle-georaster-part-i/.
Last undated: 2010-08-08.
85

[15] Jorge Arevalo. PostGIS Raster Tutorial Series. Comparing Oracle GeoRaster with PostGIS WKT
Raster (II). http://gis4free.wordpress.com/2010/09/01/oracle-georaster-part-ii/. Last
undated: 2010-09-01.
[16] Environmental Systems Research Institute. ArcGIS Desktop Help 9.3: What is raster data
?. http://webhelp.esri.com/arcgisdesktop/9.3/index.cfm?TopicName=What_is_raster_
data?. Last undated: 2009-11-12.
[17] Project Steering Committee, Refractions Research. PostGIS 2.1.0 SVN Manual: Raster Reference,
pages 359-467, 22 june 2012.
[18] Wikipedia Foundation. Crime mapping. http://en.wikipedia.org/wiki/Crime_mapping. Last
undated: 2012-06-27 21:58.
[19] Rediscovering Geography Committee Board on Earth Sciences and Resources Commission on
Geosciences, Environment and Resources National Research Council. Rediscovering Geography:
New Relevance for Science and Society. Geography’s Contributions to Decision Making. National
Academies Press Washington, D.C.,ISBN: 0-309-57762-4, 248 pages, 1997.
[20] CartoExpert.com. Cartographie Analyse Spatiale Géomatique SIG. Pourquoi utiliser QGIS ?.
http://www.cartoexpert.com/index.php/fr/formations/90-pourquoi-quantum-gis.html.
[21] Adrian Cox. Undergraduate Texts in Geography. Multimodal GIS Interfaces in Crisis Management
- Usability Study. The Pennsylvania State University, March 2005.
[22] Pierre Racine. PostGIS WKT Raster: Seamless operations between vector and raster layers. BAM
project, University Laval, July 2008.
[23] Pierre Racine, Steve Cumming. Store, manipulate and analyze raster data within the PostgreSQL/PostGIS
spatial database. FOSS4G 2011 Conference, Denver, Colorado, USA, 12-16
September 2011.
[24] Regina Obe, Leo Hsu. PostGIS 2.0 3D and Raster support enhancements. North Carolina GIS
Conference, 2011 (NCGIS 2011).
[25] Doxygen v1.7.1. GDAL Utilities. http://www.gdal.org/gdal_utilities.html. Last updated:
2012-01-07 12:57:28Z.
[26] Environmental Systems Research Institute. ArcGIS Desktop Help 9.3: Georeferencing: Assigning
map coordinates and spatial location. http://webhelp.esri.com/arcgisdesktop/9.3/index.
cfm?TopicName=Georeferencing_and_coordinate_systems. Last updated: 2009-01-20.
[27] Environmental Systems Research Institute. ArcGIS Desktop Help 9.3: Elements of geographic information.
http://webhelp.esri.com/arcgisdesktop/9.3/index.cfm?TopicName=Elements_
of_geographic_information. Last undated: 2009-01-20.
[28] Wikipedia Foundation. Raster graphics. http://en.wikipedia.org/wiki/Raster_graphics.
Last undated: 2012-07-15 04:26.
[29] Wikipedia Foundation. Vector graphics. http://en.wikipedia.org/wiki/Vector_image. Last
undated: 2012-07-18 04:07.
[30] Environmental Systems Research Institute. ArcGIS Desktop Help 9.3: How GIS represents
and organizes geographic information. http://webhelp.esri.com/arcgisdesktop/9.3/
index.cfm?TopicName=How_GIS_represents_and_organizes_geographic_information. Last
undated: 2009-01-20.
[31] Wikipedia Foundation. GIS and hydrology. http://en.wikipedia.org/wiki/GIS_and_
Hydrology. Last undated: 2012-06-20 01:27.
86

[32] Environmental Systems Research Institute. ArcGIS Desktop Help 9.3: Raster dataset zones
and regions. http://webhelp.esri.com/arcgisdesktop/9.3/index.cfm?TopicName=Raster_
dataset_zones_and_regions. Last undated: 2009-11-12.
[33] Environmental Systems Research Institute. ArcGIS Desktop Help 9.3: Discrete and continuous
data. http://webhelp.esri.com/arcgisdesktop/9.3/index.cfm?TopicName=Discrete_
and_continuous_data. Last undated: 2009-11-12.
[34] Environmental Systems Research Institute. ArcGIS Desktop Help 9.3: Representing
features in a raster dataset. http://webhelp.esri.com/arcgisdesktop/9.3/index.cfm?
TopicName=Representing_features_in_a_raster_dataset. Last undated: 2009-11-12.
[35] Environmental Systems Research Institute. ArcGIS Desktop Help 9.3: Raster bands. http://
webhelp.esri.com/arcgisdesktop/9.3/index.cfm?TopicName=Raster_bands. Last undated:
2009-11-12.
[36] Environmental Systems Research Institute. ArcGIS Desktop Help 9.3: Raster pyramids.
http://webhelp.esri.com/arcgisdesktop/9.3/index.cfm?TopicName=Raster_pyramids.
Last undated: 2009-11-12.
[37] Wikipedia Foundation. Remote sensing. http://en.wikipedia.org/wiki/Remote_sensing.
Last updated: 2012-07-20 11:16.
[38] Wikipedia Foundation. Cartography. http://en.wikipedia.org/wiki/Cartography. Last updated:
2012-07-12 19:45.
[39] Wikipedia Foundation. Geology. http://en.wikipedia.org/wiki/Geology. Last updated:
2012-07-26 06:42.
[40] Wikipedia Foundation. Geophysics. http://en.wikipedia.org/wiki/Geophysics. Last updated:
2012-06-15 14:38.
87

Appendix A
PostGIS Raster Utilisation
A.1 Dependencies Installation
– PostgeSQL version 7.2 or higher.
– PostGIS version 1.3.5 or higher.
Other necessary component is GDAL as PostGIS Raster uses GDAL to driver for images conversion,
transformation, reprojection and connecting to raster files and Python loader:
– Python version 2.5 or higher.
– GDAL for Python version 1.6 or higher.
– GDAL PostGIS Raster Driver.
A.2 Installing and Configuring
A.2.1 Linux
PostGIS Raster is an extension of PostGIS and need to be installed separately. The last version of
PostGIS Raster can be downloaded here.
The next step is to configure PostGIS Raster by the following commands:
./configure --with-raster
make
sudo make install
A.2.2 Windows
Instructions can be found on this Web site. The option ’–with-raster’ should be added when
invoking configure scripts.
The full instructions about how to install and configure PostGIS Raster on Windows is available
here.
A.3 Creating Database
The following code will create a new PostGIS Raster database in PostgreSQL:
createdb
createlang plpgsql
psql -U -f postgis/postgis.sql -d
psql -U -f spatial_ref\_sys.sql -d
psql -U -f raster/rt_pg/rtpostgis.sql -d
88

Appendix B
Python Plugin in QGIS
B.1 Necessary Files
The directory structure of a plugin is as follows:
PYTHON_PLUGINS_PATH/
testplug/
__init__.py
plugin.py
metadata.txt
resources.qrc
resources.py
form.ui
form.py
Where:
__init__.py: contains some basic information about the plugin such as its name, version and
main class.
plugin.py: plugin body that contains main code for all the actions describing the plugin.
metadata.txt: Required for QGIS >= 1.8.0. Contains general information, version, name and
some other metadata used by plugins website and plugin infrastructure. Metadata in metadata.txt is
preferred to the methods in __init__.py. From QGIS 2.0 the metadata from __init__.py will not
be accepted and the metadata.txt file will be required.
resources.qrc: a .xml document containing relative paths to resources of the forms.
resources.py: The translation of the resources.qrc file to Python.
form.ui: a form in QT-Designer with its resources.qrc.
form.py: The translation of the form.ui file to Python.
There are also two automated ways of creating the basic files (or skeleton) of a typical QGIS Python
plugin: the first one using a QGIS Plugin Builder and the second one come from a web application.
B.2 Python Code
B.2.1 __init__.py
def name():
return "My testing plugin"
def description():
return "This plugin has no real use."
def version():
89

eturn "Version 0.1"
def qgisMinimumVersion():
return "1.0"
def authorName():
return "Developer"
def classFactory(iface):
# Load TestPlugin class from file testplugin.py
from testplugin import TestPlugin
return TestPlugin(iface)
The classFactory() function is called when the plugin is loaded into QGIS. It receives reference
to instance of QgisInterface and return instance of TestPlugin plugin (the TestPlugin class will be
described later in plugin.py file).
def category():
return "Raster"
Normally, user-created plugins are placed into Plugins menu but from QGIS 1.9.90, they can also
be placed into Vector, Raster, Database or Web menus using category entry. For example, the above
plugin will be available from Raster menus.
B.2.2 Plugin.py
from PyQt4.QtCore import *
from PyQt4.QtGui import *
from qgis.core import *
# initialize Qt resources from file resouces.py
import resources
class TestPlugin:
def __init__(self, iface):
# Save reference to the QGIS interface
self.iface = iface
def initGui(self):
# Create action that triggers plugin
self.action = QAction(QIcon(":/plugins/testplug/icon.png"), "Test plugin", self.iface.ma
# Code in function run will be executed
QObject.connect(self.action, SIGNAL("triggered()"), self.run)
# Check if Raster menu available
if hasattr(self.iface, "addPluginToRasterMenu"):
# Raster menu and toolbar available
# Add action to toolbar and menu item
self.iface.addRasterToolBarIcon(self.action)
self.iface.addPluginToRasterMenu("\&Test plugins", self.action)
else:
# There is no Raster menu, place plugin under Plugins menu as usual
self.iface.addToolBarIcon(self.action)
self.iface.addPluginToMenu("\&Test plugins", self.action)
def unload(self):
# Check if Raster menu available
90

if hasattr(self.iface, "addPluginToRasterMenu"):
# Remove action from Raster toolbar and menu
self.iface.removePluginRasterMenu("\&Test plugins",self.action)
self.iface.removeRasterToolBarIcon(self.action)
else:
# Remove the Plugin menu and toolbar
self.iface.removePluginMenu("\&Test plugins",self.action)
self.iface.removeToolBarIcon(self.action)\\
def run(self):
# Main code for all the actions describing the plugin
print "TestPlugin: run called!"
The functions that must exist are initGui() and unload(). These functions are called when the
plugin is loaded and unloaded.
B.2.3 Resources.qrc
In initGui(), an icon icon.png from resource file resources.qrc is used. The structure of this
file is as follows:
icon.png
The pyrcc4 command will generate a Python file that will contain the resources:
pyrcc4 -o resources.py resources.qrc
91

Appendix C
Code Implementation
C.1 Temporal Raster Functions
PostgreSQL allows also abilities for users to define new types and their own functions. Then
CREATE AGGREGATE can be used to provide the desired features that are not provided by PostgreSQL.
An aggregate function is identified by its name and input data type. An aggregate function is made
from one or two ordinary functions: a state transition function sfunc and an optional final calculation
function ffunc.
An temporary variable of type stype is used to hold the current value of the state transition
function. At each input data item, the state transition function is invoked to calculate a new state
value. After all the data has been processed, the final function is invoked once to calculate the
aggregate result. If there is no final function then the aggregate returns the current state value as
final result. sum is an example of this kind of aggregate, where for each input, sum adds its value to
current state value and returns the sum of all input value as final result.
A final function is specified in the case where the final result is different from the input data type.
Average aggregate function is an example of this case and will be examined thereafter. If the final
function is declared "strict", then it will not be called when the current state value is null. Instead a
null result will be returned automatically.
Example of creating the aggregate function ST_Avg:
CREATE AGGREGATE ST_Avg(st_rastertemp) (
SFUNC=array_append,
STYPE=st_rastertemp[],
FINALFUNC=ST_AvgTempRaster
);
CREATE OR REPLACE FUNCTION ST_AvgTempRaster(arrayRastTemp st_rastertemp[])
RETURNS raster AS
$$
DECLARE
newraster raster;
oldRast1 raster;
oldRast2 raster;
width int;
height int;
cx int;
cy int;
size int;
s int;
value double precision = 0;
intType boolean = true;
92

BEGIN
newraster := arrayRastTemp[1].rast;
width := ST_Width(newraster);
height := ST_Height(newraster);
size := array_upper(arrayRastTemp,1);
-- Determine pixel type of temporal rasters
IF ST_BandPixelType(newraster) = ’32BF’ OR ST_BandPixelType(newraster) = ’64BF’
THEN intType = false;
END IF;
-- Check if all temporal rasters have same dimensions
IF ST_Width(oldRast1) != ST_Width(oldRast2) OR ST_Width(oldRast1) != ST_Width(oldRast2)
THEN
-- If not then the function raises error
RAISE EXCEPTION ’ST_SumRaster: Attempting to sum a band with different width or height
END IF;
FOR cx IN 1..width LOOP
FOR cy IN 1..height LOOP
FOR s IN 1.. size LOOP
value := value + ST_Value(arrayRastTemp[s].rast,cx,cy);
END LOOP;
value = value/size;
-- Cast the result to an appropriate type
IF intType = true
THEN newraster := ST_SetValue(newraster,cx,cy,value::int);
ELSE newraster := ST_SetValue(newraster,cx,cy,value);
END IF;
value := 0;
END LOOP;
END LOOP;
RETURN newraster;
END;
$$
LANGUAGE ’plpgsql’ IMMUTABLE STRICT;
C.2 Temporal Raster Application
Python code for loading layers into QGIS:
def iniLayers(self):
# Retrieve a connection string
connstring=str(conn.getConnString(self,self.connection))
# Retrieve a list of layer IDs
rids=conn.listRIDs(self, connstring, str(self.table))
# Change QGIS cursor to PostgreSQL cursor
QtGui.QApplication.setOverrideCursor(QtGui.QCursor(QtCore.Qt.WaitCursor))\\
for rid in rids:
# Split table name from schema
connstring = str(conn.getConnString(self,self.connection))
table=str(self.table).split(".")
93

name=""
# Setting table name and mode
if len(table)>1:
# Checking if there is schema in the table name
name+=str(table[0])
connstring+=" schema="+str(table[0])
connstring+=’ table=’+str(table[-1])
name+="."+str(table[-1])
# Query for row mode
connstring+=’ mode=1 where=\’rid=’ + str(rid) + ’\’’
# Object rlayer will contain loaded layer
rlayer=None
# Retrieve a layer from PostgreSQL using the connection string
rlayer = QgsRasterLayer(connstring, name)
# Try to add layer to qgis.
if rlayer:
if rlayer.isValid():
status = QgsMapLayerRegistry.instance().addMapLayer(rlayer)
self.legend.setLayerVisible(rlayer,False)
# Each layer is identified by an unique layer id
self.layerIds.append(rlayer.id())
else:
QtGui.QMessageBox.warning(self,"Error", "Could not load "+connstring)
# Come back to QGIS cursor
QtGui.QApplication.restoreOverrideCursor()
94

Appendix D
Raster Sources
Raster data is collected and used by a variety of geographic information technologies, including
remote sensing, airborne photogrammetry, cartography, and global positioning systems. The collected
data is then analyzed by digital image processing systems, computer graphics applications, and computer
vision technologies. These technologies use several data formats and create a variety of products.
This section briefly describes some of the main raster data sources and their applications.
D.1 Remote Sensing
D.1.1 Introduction
Remote sensing obtains information about an area or object through a device that is not physically
connected to the area or object. For example, the sensor might be in a satellite, balloon, airplane,
boat, or ground station.
There are two main types of remote sensing: passive remote sensing and active remote sensing.
Passive sensors detect natural radiation that is emitted or reflected by the object or surrounding areas.
Reflected sunlight is the most common source of radiation measured by passive sensors. Examples
of passive remote sensors include film photography, infrared,charge-coupled devices, and radiometers.
Active collection, on the other hand, emits energy in order to scan objects and areas whereupon a
sensor then detects and measures the radiation that is reflected or backscattered from the target.
RADAR and LiDAR are examples of active remote sensing where the time delay between emission
and return is measured, establishing the location, height, speed and direction of an object.
D.1.2 Applications
Remote sensing applications include environmental assessment and monitoring, global change
detection and monitoring, and natural resource surveying.
Remote sensing makes it possible to collect data on dangerous or inaccessible areas. Remote sensing
applications include monitoring deforestation in areas such as the Amazon Basin, glacial features in
Arctic and Antarctic regions, and depth sounding of coastal and ocean depths. Military collection
during the Cold War made use of standoff collection of data about dangerous border areas. Remote
sensing also replaces costly and slow data collection on the ground, ensuring in the process that areas
or objects are not disturbed.
Orbital platforms collect and transmit data from different parts of the electromagnetic spectrum,
which in conjunction with larger scale aerial or ground-based sensing and analysis, provides researchers
with enough information to monitor trends such as El Niño and other natural long and short term
phenomena. Other uses include different areas of the earth sciences such as natural resource management,
agricultural fields such as land usage and conservation, and national security and overhead,
ground-based and stand-off collection on border areas.
By satellite, aircraft, spacecraft, buoy, ship, and helicopter images, data is created to analyze and
compare things like vegetation rates, erosion, pollution, forestry, weather, and land use. These things
95

Figure D.1: Synthetic aperture radar image of Death Valley [37].
can be mapped, imaged, tracked and observed. The process of remote sensing is also helpful for city
planning, archaeological investigations, military observation and geomorphological surveying.
The data collected by remote sensing is often called geo imagery. The wavelength, number of
bands, and other factors determine the radiometric characteristics of the geo images. The geo images
can be single-band, multiband, or hyperspectral, all of which can be managed by PostGIS Raster.
These geo images can cover any area of the Earth (especially for images sensed by satellite).
Generally, remote sensing works on the principle of the inverse problem. While the object or
phenomenon of interest (the state) may not be directly measured, there exists some other variable
that can be detected and measured (the observation), which may be related to the object of interest
through the use of a data-derived computer model. The common analogy given to describe this
is trying to determine the type of animal from its footprints. For example, while it is impossible to
directly measure temperatures in the upper atmosphere, it is possible to measure the spectral emissions
from a known chemical species (such as carbon dioxide) in that region. The frequency of the emission
may then be related to the temperature in that region via various thermodynamic relations.
The quality of remote sensing data consists of its spatial, spectral, radiometric and temporal
resolutions. The temporal resolution can be high, such as with meteorological satellites, making
it easier to detect changes. For remote sensing applications, various types of resolution (temporal,
spatial, spectral, and radiometric) are often important.
D.2 Photogrammetry
D.2.1 Introduction
Photogrammetry is used to determine the geometric properties of objects from measurements
made on photographic images. In the simplest example, the distance between two points that lie on
a plane parallel to the photographic image plane can be determined by measuring their distance on
the image, if the scale (s) of the image is known. This is done by multiplying the measured distance
by 1/s. Most photogrammetry applications use airborne photos or high-resolution images collected by
satellite remote sensing. In traditional photogrammetry, the main data includes images such as black
and white photographs, color photographs, and stereo photograph pairs.
Photogrammetry rigorously establishes the geometric relationship between the image and the object
as it existed at the time of the imaging event, and enables you to derive information about the
object from its imagery. The relationship between image and object can be established by several
96

means, which can be grouped in two categories: analog (using optical, mechanical, and electronic
components) or analytical (where the modeling is mathematical and the processing is digital). Analog
solutions are increasingly being replaced by analytical/digital solutions, which are also referred to as
softcopy photogrammetry.
The main product from a softcopy photogrammetry system may include digital elevation models
(DEMs) and orthoimagery. GeoRaster can manage all this raster data, together with its georeferencing
information.
D.2.2 Application
Photogrammetry is used in different fields, such as topographic mapping, architecture, engineering,
manufacturing, quality control, policeinvestigation, and geology, as well as by archaeologists to quickly
produce plans of large or complex sites and by meteorologists as a way to determine the actual wind
speed of a tornado where objective weather data cannot be obtained. It is also used to combine live
action with computer-generated imagery in movie post-production; The Matrix is a good example of
the use of photogrammetry in film (details are given in the DVD extras).
This method is commonly employed in collision engineering, especially with automobiles. When
litigation for accidents occurs and engineers need to determine the exact deformation present in the
vehicle, it is common for several years to have passed and the only evidence that remains is crime
scene photographs taken by the police. Photogrammetry is used to determine how much the car in
question was deformed, which relates to the amount of energy required to produce that deformation.
The energy can then be used to determine important information about the crash (such as the velocity
at time of impact)
D.3 Cartography
Cartography is the science of creating maps, which are two-dimensional representations of the
three-dimensional Earth (or of a non-Earth space using a local coordinate system).
Advances in electronic technology in the 20th century ushered in another revolution in cartography.
Ready availability of computers and peripherals such as monitors, plotters, printers, scanners (remote
and document) and analytic stereo plotters, along with computer programs for visualization, image
processing, spatial analysis, and database management, have democratized and greatly expanded the
making of maps. The ability to superimpose spatially located variables onto existing maps created
new uses for maps and new industries to explore and exploit these potentials. See also: digital raster
graphic.
These days most commercial-quality maps are made using software that falls into one of three
main types: CAD, GIS and specialized illustration software. Spatial information can be stored in
a database, from which it can be extracted on demand. These tools lead to increasingly dynamic,
interactive maps that can be manipulated digitally.
With the field rugged computers, GPS and laser rangefinders, it is possible to perform mapping
directly in the terrain. Construction of a map in real time, for example by using Field-Map technology,
improves productivity and quality of the result.
Today, maps are digitized or scanned into digital forms, and map production is largely automated.
Maps stored on a computer can be queried, analyzed, and updated quickly.
There are many types of maps, corresponding to a variety of uses or purposes. Examples of map
types include base (background), thematic, relief (three-dimensional), aspect, cadastral (land use),
and inset. Maps usually contain several annotation elements to help explain the map, such as scale
bars, legends, symbols (such as the north arrow), and labels (names of cities, rivers, and so on). Maps
can be stored in raster format (and thus can be managed by GIS), in vector format or in a hybrid
format.
97

D.4 Digital Image Processing
D.4.1 Introduction
Figure D.2: Relief map Sierra Nevada (Spain) [38].
Digital image processing is used to process raster data in standard image formats, such as TIFF,
GIF, JFIF (JPEG), and Sun Raster, as well as in many geo image formats, such as NITF, GeoTIFF,
ERDAS IMG, and PCI PIX. Image processing techniques are widely used in remote sensing and
photogrammetry applications. These techniques are used as needed to enhance, correct, and restore
images to facilitate interpretation; to correct for any blurring, distortion, or other degradation that may
have occurred; and to classify geo-objects automatically and identify targets. The source, intermediate,
and result imagery can be loaded and managed by GeoRaster.
Digital image processing is the use of computer algorithms to perform image processing on digital
images. As a subcategory or field of digital signal processing, digital image processing has many
advantages over analog image processing. It allows a much wider range of algorithms to be applied
to the input data and can avoid problems such as the buildup of noise and signal distortion during
processing. Since images are defined over two dimensions (perhaps more) digital image processing
may be modeled in the form of multidimensional systems. A digital image processing allows the use
of much more complex algorithms for image processing, and hence, can offer both more sophisticated
performance at simple tasks, and the implementation of methods which would be impossible by analog
means. application.
D.4.2 Digital Camera Images
Digital cameras generally include dedicated digital image processing chips to convert the raw data
from the image sensor into a colour-corrected image in a standard image file format. Images from
digital cameras often receive further processing to improve their quality, a distinct advantage that
digital cameras have over film cameras. The digital image processing typically is executed by special
software programs that can manipulate the images in many ways. Many digital cameras also enable
viewing of histograms of images, as an aid for the photographer to understand the rendered brightness
range of each shot more readily.
D.5 Geology, Geophysics and Geochemistry
Geology, geophysics, and geochemistry all use digital data and produce some digital raster maps
that can be managed by GeoRaster.
• In geology, the data includes regional geological maps, stratum maps, and rock slide pictures. In
98

geological exploration and petroleum geology, computerized geo stratum simulation, synthetic
mineral prediction, and 3-D oil field characterization, all of which involve raster data, are widely
used.
Geologists use a number of field, laboratory, and numerical modeling methods to decipher Earth
history and understand the processes that occur on and in the Earth. In typical geological
investigations, geologists use primary information related to petrology (the study of rocks),
stratigraphy (the study of sedimentary layers), and structural geology (the study of positions
of rock units and their deformation). In many cases, geologists also study modern soils, rivers,
landscapes, and glaciers; investigate past and current life and biogeochemical pathways, and use
geophysical methods to investigate the subsurface.
In modern times, geology is commercially important for mineral and hydrocarbon exploration
and for evaluating water resources; is publicly important for the prediction and understanding
of natural hazards, the remediation of environmental problems, and for providing insights into
past climate change; plays a role in geotechnical engineering; and is a major academic discipline.
Figure D.3: William Smith’s geologic map of England,Wales, and southern Scotland. Completed in
1815, it was the first national-scale geologic map and by far the most accurate of its time [39].
• In geophysics, data about gravity, heat flow, seismic wave transportation or vibrations, the
magnetic field, and other subjects is saved, along with georeferencing information.
Geophysics is applied to societal needs, such as mineral resources, mitigation of natural hazards
and environmental protection.Geophysical survey data are used to analyze potential petroleum
reservoirs and mineral deposits, locate groundwater, find archaeological relics, determine the
thickness of glaciers and soils, and assess sites for environmental remediation.
• In geochemistry, the contents of multiple chemical elements can be analyzed and measured. The
triangulated irregular network (TIN) technique is often used to produce raster maps for further
analysis, and image processing is widely used.
Geochemistry is the science that uses the tools and principles of chemistry to explain the mechanisms
behind major geological systems such as the Earth’s crust and its oceans. The realm
of geochemistry extends beyond the Earth, encompassing the entire solar system and has made
important contributions to the understanding of a number of processes including mantle convection,
the formation of planets and the origins of granite and basalt. is the slow creeping
motion of Earth’s rocky mantle caused by convectioncurrents carrying heat from the interior of
the Earth to the surface.
99

Figure D.4: Age of the sea floor. Much of the dating information comes from magnetic anomalies [40].
100

Appendix E
Raster Resolution
When working with raster data, there are four types of resolution often encountered: spatial
resolution, spectral resolution, temporal resolution and radiometric resolution.
Spatial resolution refers to the size of the smallest object that can be resolved on the ground. In a
digital image, the resolution is limited by the pixel size, i.e. the smallest resolvable object cannot be
smaller than the pixel size.
The spatial resolution of raster data refers to the cell size. A higher spatial resolution implies that
there are more pixels per unit area. Therefore, the graphic on the left represents a higher spatial
resolution than the graphic on the right.
Figure E.1: Raster data represented at different resolution or cell sizes [11].
Spectral resolution describes the ability of a sensor to distinguish different wavelength intervals in
the electromagnetic spectrum. The higher the spectral resolution, the narrower the wavelength range
for a particular band. For example, a grayscale image (a single band) records wavelength data over
the visible portion of the electromagnetic spectrum. However, it has a low spectral resolution. A color
image (with three bands) collects wavelength data from three smaller parts of the visible portion of
the electromagnetic spectrum: the red, green, and blue parts. Therefore, each band in the color image
has a higher spectral resolution than the single band in the grayscale image. Advanced multispectral
and hyperspectral sensors collect data from up to hundreds of spectral bands, resulting a very high
spectral resolution data.
Temporal resolution refers to the frequency with which images can be captured over the same
place on the earth’s surface. It is also known as the revisit period, which is a term most often used
for satellite sensors. Therefore, a sensor that captures data once every week has a higher temporal
resolution than one that captures data once a month.
Radiometric resolution describes the ability of a sensor to distinguish objects viewed in the same
part of the electromagnetic spectrum. This is synonymous with the number of possible data values
in each band. For example, a Landsat band is typically 8-bit data and an IKONOS band is typically
11-bit data. Therefore, the IKONOS data has a higher radiometric resolution.
101

Appendix F
Continuous Surfaces
One type of continuous surface, that is derived from those characteristics that define a surface, is
a surface where each point in the surface is its measured from a fixed reference point. These include
elevation (the fixed reference point is a sea level and the measure at each point is its height above the
sea level) and aspect (the fixed reference point is one of four directions: north, east, south and west
and the measure at each point is the direction of a normal at that point).
Figure F.1: Elevation represented in raster format [33].
Another type of continuous surface includes phenomena that progressively vary across a surface
from a source. Examples of such phenomena include fluid and air movement. These surfaces are
characterized by the way in which the phenomenon moves. The first type of movement is through
diffusion in which the phenomenon moves from areas with high concentration to those with less
concentration until the concentration level evens out. Surface characteristics of this movement type
include salt concentration moving through the ground or water, contamination moving away from a
hazardous spill or a nuclear reactor and heat from a forest fire. In this type of continuous surface,
there has to be a source, so-called source-concentration surface. The concentration is always greater
near the source and decreases following a function of distance and the intermediate substance.
In the source-concentration surface above, the concentration at any position is a function of the
capability of the phenomenon to move through the environment. Another type of concentration surface
is governed by the inherent characteristics of the moving phenomenon. For example, the movement of
the noise from a bomb blast is governed by the inherent characteristics of noise and the environment
that it moves through. Mode of movement can also limit and directly affect the surface concentration
of a feature, as is the case with seed dispersal from a plant. The means of movement, whether it be
bees, man, wind, or water, all affect the surface concentration of seed dispersal for the plant. Other
concentration surfaces include dispersal of animal populations, potential customers of a store (cars
being the means of locomotion and time being the limiting factor) and the spreading of a disease.
102

Figure F.2: Source-concentration surface represented in raster format [33].
103