Does GridGIS require more intelligence than GIS? Claire Jarvis Department of Geography GEOGRAPHY.

21
Does GridGIS require more intelligence than GIS? Claire Jarvis Department of Geography GEOGRAPHY

Transcript of Does GridGIS require more intelligence than GIS? Claire Jarvis Department of Geography GEOGRAPHY.

Page 1: Does GridGIS require more intelligence than GIS? Claire Jarvis Department of Geography GEOGRAPHY.

Does GridGIS require more intelligence than GIS?

Claire JarvisDepartment of Geography

GEOGRAPHY

Page 2: Does GridGIS require more intelligence than GIS? Claire Jarvis Department of Geography GEOGRAPHY.

Solving environmental problems with the aid of GIS

e.g. attributes ofbiological reserve

required toevaluate

conservationstatus

Task-relatedknowledgeplus expertreasoning

Low-levelknowledge plus

simplereasoning

Command syntax– from user

manuals, on-linehelp systems

Page 3: Does GridGIS require more intelligence than GIS? Claire Jarvis Department of Geography GEOGRAPHY.

“With the vast power of a user friendly GIS increasingly in the hands of the non specialist, the danger that the wrong kind of spatial statistics will become the accepted practice is great”

(Anselin 1989)

Page 4: Does GridGIS require more intelligence than GIS? Claire Jarvis Department of Geography GEOGRAPHY.

“It is unwise to throws one’s data into the first available interpolation technique without carefully considering how the results will be affected by the assumptions inherent in the method”

(Burrough 1986)

Page 5: Does GridGIS require more intelligence than GIS? Claire Jarvis Department of Geography GEOGRAPHY.

“the time taken to explore, understand, and describe the data set should be amply regarded”

(Isaacs & Srivastava 1989)

Page 6: Does GridGIS require more intelligence than GIS? Claire Jarvis Department of Geography GEOGRAPHY.

(‘Intermediate knowledge’ concept after Bhavnani and John, 2000)

Body of knowledge required to solve/research environmental problems

using GIS

e.g. attributes ofbiological reserve

required toevaluate

conservationstatus

Task-relatedknowledgeplus expertreasoning

Low-levelknowledge plus

simplereasoning

Command syntax– from user

manuals, on-linehelp systems

Intermediateknowledge plus

(human) expertreasoning

e.g. statisticalassumptions of

individualGISciencetechniques

Page 7: Does GridGIS require more intelligence than GIS? Claire Jarvis Department of Geography GEOGRAPHY.

GIS technologies arguably need more intelligence to support their users, in a context where GIS is now much more

accessible to ‘naïve’ users

Does this position change when moving across to GridGIS?

Page 8: Does GridGIS require more intelligence than GIS? Claire Jarvis Department of Geography GEOGRAPHY.

GridGIS may open up regular or occasional GIS usage to a wider audience, both scientific and public

‘Doing’ GridGIS - Need for Intelligence (1)

• Lack of coherent body of documentation

Grid GIS will not come with a coherent ‘user manual’ of commands and terminologies

• Variation in audience

Page 9: Does GridGIS require more intelligence than GIS? Claire Jarvis Department of Geography GEOGRAPHY.

GridGIS may be encountered through an ‘expert’ user interacting with a well designed portal to develop a pre-specified workflow of known data and processing services

In the future GridGIS may equally be used, by an expert or otherwise, in a more explorative adaptive mode

• Variation in levels of interaction

‘Doing’ GridGIS - Need for Intelligence (2)

Page 10: Does GridGIS require more intelligence than GIS? Claire Jarvis Department of Geography GEOGRAPHY.

How?

Page 11: Does GridGIS require more intelligence than GIS? Claire Jarvis Department of Geography GEOGRAPHY.

Goal: Design of an ‘intelligent’ module that

sits between task and GIS

Focus task: Spatial interpolation

Domain: Creation of gridded meteorological surfaces for use in environmental models

Non-Grid Pilot Prototype

Page 12: Does GridGIS require more intelligence than GIS? Claire Jarvis Department of Geography GEOGRAPHY.

Approach

• Construct a network of rules that assist the user to select an appropriate interpolation method according to:

– the task-related knowledge (or “purpose”) of the user;

– encoded intermediate knowledge gained from experts in interpolation.

• Trigger statistical diagnostics to run on the data sets when a rule requires them to be evaluated.

Page 13: Does GridGIS require more intelligence than GIS? Claire Jarvis Department of Geography GEOGRAPHY.

Elements of knowledge

Purpose

Domain

Function characteristic

sParameters

and assumptions

statistical

cognitive

knowledge

Page 14: Does GridGIS require more intelligence than GIS? Claire Jarvis Department of Geography GEOGRAPHY.

Proportion of case-based knowledge initially low, will

increase over time

Derived from the theoretical literature. These trigger

appropriate statistical diagnostic checks.

Extracted from the user, with supporting visualisation where

appropriate

These rules will be weighted lower than theoretical rules,

hence lower proportion overall.

Derived from the theoretical literature. These suggest

broadly suitable functions for certain types of data..

Contributions to the knowledge base

Case-based knowledge gathering

within the module

Use planned for the interpolated surface

Task related knowledge extracted

from the user

Rules regarding general characteristics

of interpolation methods

Rules regarding assumptions and

parameters for specific interpolation methods

Applications in the example domain by

literature

Page 15: Does GridGIS require more intelligence than GIS? Claire Jarvis Department of Geography GEOGRAPHY.

Implementation of a prototype intelligent module

• Stand-alone module;

• Software environment: Java & Jess;

• Knowledge acquisition: iterative approach;

• Knowledge structure: decision tree;

• Interface design: multi-modal.

Page 16: Does GridGIS require more intelligence than GIS? Claire Jarvis Department of Geography GEOGRAPHY.

visualise[y] ormodel [n]?

Yes No

Check fortrend

Smooth[y] orbumpy [n]

Corollary data[y] or [n]?

Data on grid [y]or at points[n]?

Honourdata

points[y][n]? No

Measurableauto-

correlation?

Smooth[y]or bumpy

[n]

Yes No

Yes No

Yes No

No

Yes

analyse

Collatoral data

Page 17: Does GridGIS require more intelligence than GIS? Claire Jarvis Department of Geography GEOGRAPHY.
Page 18: Does GridGIS require more intelligence than GIS? Claire Jarvis Department of Geography GEOGRAPHY.

Outputs

• Interpolation methods that might be and should not be considered for the data set;

• Any parameters required to interpolate the particular data set (e.g. distance decay parameter for Inverse Distance Weighting);

• The rationale of the decision process, so the 'intelligent interpolator' also acts as a learning tool.

Page 19: Does GridGIS require more intelligence than GIS? Claire Jarvis Department of Geography GEOGRAPHY.

Conclusions from the pilot• Previous work incorporating intelligence into GIS had been computer-intensive

or knowledge intensive -- prototype module offers a more balanced approach

• Successful verification and validation by users, but in a small trial only

• Needed wider testing to establish truly generic ability .‘The ultimate aim is to develop an intelligent partnership between user and machine, a relationship which

currently lacks balance.’ (Openshaw and Alvanides, 1999)

Page 20: Does GridGIS require more intelligence than GIS? Claire Jarvis Department of Geography GEOGRAPHY.

Incorporating ‘intelligence’ within (Grid)GIS – Questions (1)

• Should methods be selected mostly according to purpose and domain, or the characteristics of the data?

• How can purpose be encapsulated within an adaptive Grid processing system?

• Should intermediate knowledge be associated with GIS functions, or encoded as meta-data?

• How should we approach metadata regarding GIS services?

Page 21: Does GridGIS require more intelligence than GIS? Claire Jarvis Department of Geography GEOGRAPHY.

Incorporating ‘intelligence’ within (Grid)GIS – Questions (2)

• How far should a user be aware the decision making process, or should this be hidden?

• How do we build usable ‘case’ examples into a re-usable body of knowledge?

• How do we balance rules and case study information, to take the best from inductive and deductive approaches?

• How can we capture intelligence related to more complex processing tasks; the pilot applied to a small range of services that are likely in an applied context to form only part of a workflow?