Spatial Data and GIS and Spatial Data Analysis

1
Spatial Data and GIS andSpatial Data Analysis

• Yaji Sripada

2
In this lecture you learn

• What are spatial data and their special
  characteristics?
• GIS
• Spatial data analysis tasks and techniques
• Applying region growing approaches to
  segmentation of area data

3
Introduction

• In many domains we process information in
  relation to its spatial location
• E.g., epidemiological studies are dominated by
  geographical distribution of infected cases
• Dr Snows study of London Cholera epidemic
• engineering designs have a strong spatial basis
• CAD/CAM systems deal with locations of components
  in a design
• Image processing involves segmenting pixel data
  in relation to their location to identify objects
  of interest
• Position aware devices such as mobile phones
  allow us to track individual movement

4
Geo-referenced Data

• Data those are related to geographic locations
  are said to be geo-referenced
• Dr Snows data is geo-referenced
• Census data is geo-referenced
• Most of our decisions are based on geo-referenced
  data
• Weather at a location drives our decision to plan
  a picnic at that location
• Supermarkets decide the size and type of a new
  store after thoroughly analysing the
  characteristics of the neighbourhood
• Building the informational and computational
  infrastructure to support storing, retrieving,
  analysing and visualising geo-referenced data is
  the job of computer scientists
• Support for geo-referenced data in MySQl (version
  4.1 onwards)

5
GIS

• GIS refers to
• Geographic Information System
• Or Geospatial Information System
• GIS offers
• generic (application independent) functionality
  required for supporting decision making with
  geo-referenced data
• Data storage and retrieval
• Data analysis
• Visualization
• GIS combines
• Data analysis and Visualization for helping users
  understand geo-referenced data
• Therefore is an ideal example for our course
• The focus is on offering generic functionality to
  help users understand data rather than make
  decisions for them like expert systems

6
GIS (2)

• Advancement of Geographic Information Systems
  (GIS) and Global Positioning System (GPS) have
  allowed us to study most data in relation to its
  spatial location
• We are now in a position to formulate well formed
  spatial queries or hypotheses
• Technology is available to answer such queries or
  test those hypotheses
• All of us will use more and more geo-referenced
  data in the future

7
GIS Modules
Main Modules of a GIS
Spatial Visualization (Maps)
Spatial Data Analysis
Spatial Database
8
Characteristics of Spatial Data

• We use spatial data in this course in its
  restrictive sense of geo-referenced data
• Spatial Data has two kinds of attributes
• Spatial attributes location information
• E.g. longitude and latitude for points and
  boundary information for areas
• Non-spatial attributes
• E.g. rainfall or house prices
• We are mainly interested in the non-spatial
  attributes
• But want to study them taking their location
  (spatial attributes) into consideration
• Relationships among non-spatial attributes are
  explicit
• But relationships among spatial attributes are
  implicit

9
Characteristics of Spatial Data (2)

• Objects with similar attributes usually are
  located nearby spatially
• Everything is related to everything else but
  nearby things are more related than distant
  things first law of Geography
• In spatial statistics this property is called
  spatial auto-correlation
• Recall auto-correlation from time series data
• Data values are not independent
• Most geographic locations are unique (spatial
  heterogeneity)
• Therefore global parameters do not always
  accurately describe local values

10
Characteristics of Spatial Data (3)

• Special properties of spatial data
• Auto-correlation
• Spatial heterogeneity
• Implicit spatial relationships
• Modelling spatial data needs to be different from
  modelling ordinary data
• Data modelling influences data manipulation
• Querying
• Analysis
• Visualization

11
Concept of Modelling

• Common sense view
• Representation of something at a level of
  detail suitable for its purpose
• For example, an architects model of a bridge
• Architects model brings the bridge to life even
  before its construction
• Formal View
• Modelling function translates some source domain
  into its corresponding target domain
• Target domain is used (because it is simple in
  some sense than the source domain) for analysis
• An inverse modelling function should be available
  for translating results of analysis from target
  domain to the source domain

12
Modelling Spatial (geographic) Data

• Two fundamentally distinct views
• Absolute space
• Space exists in itself and objects are located in
  this absolute space
• You first create space and put objects in that
  space
• Relative space
• Space is one of the attributes of objects related
  to other objects
• You first define objects and they create space as
  a result of their relative locations and
  interactions
• Both these views are used in GIS for modelling
  spatial data

13
Relational Data Model

• Relational databases model data into a connected
  set of relations
• Each relation is a collection of tuples
• Tuple1 -gt (location1,temperature1,rainfall1)
• Tuple2 -gt (location2,temperature2,rainfall2)
• For certain applications, relational models are
  often criticised for impedance mismatch between
• the relational database storing the data
• the object oriented code manipulating that data
• For spatial data this mismatch is a problem
• The inherent structure of spatial data is not
  captured by the relational model

14
Field-Based Models

• Information space is viewed as a collection of
  fields
• Temperature field, rain fall field and wind speed
  field form a weather information space
• Data attribute values are computed by functions
  of locations
• Temperature1 Temperaturefield(location1)
• Tempearture2 Temperaturefield(location2)
• RainFall1 RainFallfield(location1)
• The field is the function, not the set of values
• Field is the first-class entity in this kind of
  modelling

15
Field-Based Models (2)

• Field-based model is a function on location
• So we need location data as independent variable
• Given a region of space (geography) we need a
  framework to partition that space into locations
• Tessellation of space
• For example using grids
• A field based model then a function that maps
  each location to its attribute value
• Useful for modelling data from continuous spatial
  processes
• Temperature fields, elevation data

16
Object-based Models

• One or more tuples from the relational model can
  be lumped together as data values corresponding
  to an object
• All the tuples that have temperatures below zero,
  rainfall above 10mm describe an object
• The object then has spatial reference
• The above weather conditions could be true for a
  region of geography
• Object is the first-class entity in this kind of
  modelling
• Useful for modelling data from discrete spatial
  processes
• Administrative units, rivers

17
Object-based Models(2)

• Object-based model maps directly to the
  object-oriented model we are familiar in
  computing science
• Objects have attributes some of which happen to
  be spatial and therefore have values related to
  space (or geography)
• Field-based models also can be mapped to
  object-oriented models but not directly
• Field-based and object-based models are
  complementary not competing
• Both are useful for different contexts

18
Spatial Databases

• Connected set of Themes (corresponding to
  relations/tables in relational model)
• Each of these is a collection of geographic
  objects
• Geographic objects have two components
• Description non-spatial attributes
• Spatial component spatial attributes
• Geometric attributes such as location and shape
• Topological attributes such as adjacency
• Two example themes
• Countries (name, population, georegion)
• Languages (language,georegion)

19
Countries
20
Languages
21
Queries on Spatial databases

• Familiar operations from relational algebra can
  be defined on themes
• Theme projection
• ?population,geo(Countries)
• Theme selection similar to relational selection
• s populationgt50(Countries)
• Theme union similar to relational union
• You can work these out yourself

22
Spatial Join

• In a relational database, join queries help users
  to connect or link or join tables
• Spatial databases allow users to join themes
• These are called theme overlays
• An object of one theme is joined with an object
  of the other theme if their geometries interset
• In our example, the resulting theme will show all
  the rows and columns of both the tables
• You can work it out yourself

23
Special Queries

• Some queries to spatial databases are more
  complicated than the relational queries
• Window query select the objects that overlap a
  given window or area
• Point query select the objects that contain the
  given point
• Clipping select the objects with the exact
  intersection of the geometry of the object and
  the given window
• To process such queries GIS possesses geometric
  and topological sense
• We will not go into the details here

24
Visualization of Spatial Data

• Results of theme operations are not very useful
  if shown as tables
• They are normally shown as maps in GIS
• Theme overlay is the main operation for creating
  maps in GIS
• Data belonging to the required themes is
  retrieved from the database and plotted as
  overlays in a GIS (you will learn to use overlays
  in the practical)
• As discussed with other visualizations
  geo-visualization (or map drawing) too has two
  aspects
• Designing the map
• Rendering the map

25
Visualization of Spatial Data (2)

• Maps can be rendered using
• Vector graphics
• Raster graphics
• This distinction can be traced back to the
  distinction between
• Object-based data models (Vector models)
• Field-based data models (Raster models)
• Many modern GIS systems allow mixing and matching
  these two modes to render maps
• Google maps overlay vector based spatial
  information on top of raster satellite image
• This is the approach we use in our practicals
  where we write java code for visualizing spatial
  data

26
Spatial Data Analysis

• Techniques to analyse data taking into
  consideration their location information.
• Results of spatial data analysis change if
  spatial distribution of data changes
• How data varies in space?
• There are many stages of spatial data analysis
• Pre-processing or Smoothing
• Exploratory Spatial Data Analysis
• Model building
• For event prediction and hypotheses testing
• For communication
• Very similar to the stages involved in processing
  time series

27
Data quality - Smoothing

• Data quality is a serious issue in spatial
  databases
• Inaccuracies in measurement of location
  information
• E.g.Inaccuracies due to approximations in GPS
• Inaccuracies due to integrating data
  (particularly in a GIS) from different sources
  each of which using a different approximation of
  location information
• Simple smoothing techniques such as mean and
  median filters (refer to lecture 4) are still
  useful

28
Exploratory Spatial Data Analysis (ESDA)

• ESDA involves identification of data properties
  and formulating hypotheses from data
• Visualization of data using GIS is particularly
  suited for ESDA
• Results from ESDA often form input to subsequent
  stages of analysis
• ESDA is an important step in the development life
  cycle
• Developers gain lot of understanding of the
  underlying phenomena by performing ESDA
• As a result developers have better understanding
  of user requirements
• Therefore helps them in making better system
  design to fulfil user requirements

29
Spatial Data Types

• Three Types
• Data referenced to a point
• E.g. Location information of a restaurant
• Data referenced to a path
• E.g. Path information from my home to University
• Data referenced to an area
• E.g. information about a region bounded by a
  polygon
• We can transform point data into area data by
  aggregating values over all the points in an area
• Different data analysis tasks and techniques are
  employed for each of these data types

30
Points Data

• Event prediction
• E.g. given the spatial distribution of crimes in
  an area, predict the likely location of a future
  crime
• Given some actual observations predict unknown
  values at intermediate locations by interpolation
• Spatial regression

31
Paths Data

• Finding least cost path over a route map.
• Navigation systems on modern cars find paths and
  communicate the path information graphically and
  by speech
• A navigation system is a good example of the kind
  of systems we are interested in this course
• They analyse spatial data to extract important
  information plus
• They also communicate the extracted information
  in different forms to suit the user

32
Area/Lattice data

• Public domain is flooded with this type of data
• E.g. census data is available for public as
  aggregated values over a census tract
• Scrol Scotlands Census Results Online
• Weather parameters such as temperature and
  rainfall are reported as aggregated values over a
  region such as Grampian and Lothian
• Disease count data where counts of a disease are
  recorded for regions or counties
• Technology to analyse and communicate this type
  of data has large impact on public life

33
Segmentation

• Analysis of area data to find regions that have
  similar values of one or more non-spatial
  attributes
• E.g. segmentation finds areas in a country with
  high family income
• Visualizations of segments is done using maps
  with different segments shown in different
  colours
• Many computational approaches to segment area
  data
• Partitioning
• Hierarchical
• Density-based
• Grid-based and
• Model-based

34
Typical area analysis problem

• Input
• a table of area names and their corresponding
  attributes such as population density, number of
  adult illiterates etc.
• Information about the neighbourhood relationships
  among the areas
• A list of categories/classes of the attributes
• Output
• Grouped (segmented) areas where each group has
  areas with similar attribute values
• Visualizations using maps do not need
  segmentation process
• Census Website has plenty of examples
• http//www.statistics.gov.uk/census2001/censusmaps
  /index.html
• Textual presentation of segmented data requires
  segmentation
• Textual presentations useful for visually
  impaired users

35
Similarity with image segmentation

• Spatial segmentation is performed in image
  processing as well
• Identify regions (areas) of an image that have
  similar colour (or other image attributes).
• Many image segmentation techniques are available
• E.g. region-growing technique

36
Region Growing Technique

• There are many flavours of this technique
• One of them is described below
• Assign seed areas to each of the segments
  (classes of the attribute)
• Add neighbouring areas to these segments if the
  incoming areas have similar values of attributes
• Repeat the above step until all the regions are
  allocated to one of the segments
• You will work with a version of this technique in
  the practical 6

37
Spatio-temporal data analysis

• Many spatial data sets have a temporal dimension
  as well
• Census data from several census activities (UK
  collects census every 10 years) is
  spatio-temporal
• Weather data for a region collected over a period
  of time is spatio-temporal
• Spatio-temporal data analysis is concerned with
  data variation in space and time
• Graphical animations of spatial displays can help
  visualize spatio-temporal data

38
Summary

• GIS combines data analysis and visualization
  seamlessly
• Spatial data analysis is concerned with data
  variation in space
• How data changes with location
• Spatial data analysis is different because of
  auto-correlation and heterogeneity in spatial
  data
• Area data is ubiquitous and segmentation of area
  data can be achieved by region growing approaches