Lecture Notes for Clarke, K. C. Analytical and Computer Cartography

Lecture 9: Attribute Data Structures

"...we query the cartographic objects representing the geographic phenomena by their attributes reflecting geographic properties" (Clarke, p. 159)

Sequence of Development

First generation GIS and computer mapping systems largely ignored attributes
Second generation systems were entity-by-entity based, but had topology
Allowed separation of attribute data from map data structure e.g. PAT files
Attributes could also be attached to points and chains etc.
Third generation systems used DBMS or spreadsheets, and allowed management and linkage
Most common data model for attribute data was the relational

Attribute Data Models

Data management most commonly uses the logical data model of the flat file, but physically, data can be placed in many different combinations of files and physical data structures.

Several models for attibute data have developed over time. The hierarchical and network data models were short-lived, and quickly replaced by the relational. More recent hybrids, and the now dominant object-oriented model are considered.

The Hierarchical Data Model (e.g. IBM's IMS)

Maps geographic files onto the file system hierarchy of computer systems, i.e. directory-> file.
Assumes that geographic relations are nested and hierarchical.
Some (e.g. administrative, elevation relations by contour line) are, many are not

Major problem is hierarchy inversion.
For example, if the hierarchy included CITY, it would invert with COUNTY in the case of New York.

The Network Data Model (e.g. DEC's DBMS-10)

Basis of model is now the relationship between entities.
Maps closely onto computer science concept of linked list
Fits neatly into the Point->Line-> Area model, e.g. DLG, TIGER
Pointers must be stored with the entity objects, e.g. to other objects
Data model can be thought of as a network map, like and airline or street network
Problems arise when relations are complex, e.g. polygon overlap, or pointers are self-references

The Relational Data Model

The relational model has dominated the DBMS field for the last 2 decades
Data are divided between flat files stored in separate physical files
The relational manager maintains relations (links) between records across files
This is do-able because each record has one attribute that identifies it uniquely
This is called the KEY or INDEX field (attribute)
For example, a stream could be a line entity by itself, a state boundary and a county boundary
Flat files could be structured by stream, state, and by county, and each would reference the same entity object, but in different contexts (relations)
Very powerful search, selection, and maintenance is then possible.
Duplicates must be "weeded" as files are restructured.
RDBMSs allow new flat files to be created by combining records from others, e.g by JOINING subsets of records
Examples are the INFO database, Microsoft ACCESS, INGRESS, Informix.
Important entension is SQL

Some Attribute Data Models and Structures

The Hypergraph Structure

Extended form of the relatioal model is the entity-relationship model.
Similar to Bouille's Hypergraph structure
Unites set theory and topology
Fundamental geographic concepts: objects, class, attribute, relations.
Hypergraph drew linkages between abstract data types based on concepts.
Hypergraph of edges & links came from topology
Applications to road networks

The Entity-Relationship Structure

Nyerges introduced Chen's entity-relationship model into cartography in 1980
Shapiro & Haralick implemented system in 1982
Combined merits of relational and network models
Non-fixed length records (e.g. strings) possible in tables.

Object-Oriented Models

OOP data bases and programming based on identical concepts
Objects are primitives, that have classes (templates) and instances
Complex objects can be built from simpler ones
OO systems support
- inheritance
- encapsulation (incorporation of rules)
- hidden-layer
OOPS are leading to a new generation of software
Basis for Avenue, C++, Smallworld

Hybrid Structures

Many systems combine methods or use hybrids to achive goals
QTM indexing
Vaster
Expert systems-based methods
Phenomenon-based methods
"Geographic" model remains elusive

Data Structures and Models for Analytical and Computer Cartography

Attribute data management most pertinent to GIS
What about analytical and computer cartography?
Automated mapping requires retrieval of attributes for specific purposes e.g. choropleth mapping
Also sometimes requires management and conversion (e.g. COMPUTE)
Many recent applications require "smart maps" e.g. vehicle navigation
Analytical cartography more concerned with analysis and management
Specific map purposes (e.g. analysis) require specific structures and models.

"The power of both map and attribute data structures is how well they support cartographic transformations."