GUS98

1
Reasoning about the Environment
Chris Skelsey Keith Matthews
Macaulay Land Use Research Institute
2
Land Cover Data

• Describes nature of land cover over some
  geographical area

• Varied uses

3
Land Cover Data

• Derived from an interpretation of varied data and
  information sources
• Aerial photographs
• Soil maps
• Knowledge of local management
• Knowledge of seasonal cycles
• Satellite imagery
• Can automation play a more significant role?

4
Established Software

• GIS and remote sensing packages
• Arc/Info
• Smallworld
• ER-Mapper
• PCI
• Erdas Imagine

5
Problem Complexity

• Procedural, quantitative functionality

1988
Areas of forest-felling
1995
6
Artificial Intelligence (AI)

• Production rules
• Frame systems
• Semantic networks
• Neural networks
• Fuzzy logic
• Dempster-Shafer theories

7
Limitations of AI Approaches

• Data-specific and method-specific
• Single software environment
• Real-world domain complexities prevent these
  applications scaling-up
• Most remain within the research community
• Still need greater software flexibility

8
A Prototype AI Toolkit

• ETORA
• Developed within G2
• Blackboard reasoning
• Re-use of established software
• ARC/INFO and PV-WAVE servers
• Implementation of endorsement theory

9
A Prototype AI Toolkit

• Disparate multi-source data
• Quantitative and qualitative knowledge
• Dynamic solution strategies

• Use of 3rd-party, established software
• Full reasoning explanations, associated with
  end-products

10
A Map Revision Problem

• Land Cover of Scotland (1988) dataset
• Requires revision

Arable
Heather moor
Coniferous- plantation
11
A Map Revision Problem

• SYMOLAC solves a simple problem, but
  demonstrates the flexibility of ETORA
• Produces a useful product despite real-world
  complexities

12
In Summary

• Automation is becoming increasingly important
• Recognised need for AI technology
• AI approaches are often problem-specific, or
  adopt unsuitable software platforms

13
Some Conclusions

• Prototype ETORA toolkit offers flexibility to
  solution designer
• Exists potential to automate a greater number of
  processes involved in land cover data production

14
LADSS

• Land Allocation Decision Support System
• Evaluates economic impacts of land use strategies
• Use of genetic algorithms
• Bridge to the Smallworld GIS

15
Why use G2?

• Flexible knowledge-representation
• Object-orientated concepts
• Ability to visualise the reasoning processes
• G2-Gateway