School of GeographyFACULTY OF ENVIRONMENT
School of GeographyFACULTY OF ENVIRONMENT
GEOG5060 GIS & Environment
Dr Steve Carver
Email: [email protected]
School of GeographyFACULTY OF ENVIRONMENT
Outline:
• terminology, types and sources of error
• why is it important?
Lecture 1: Error and uncertainty
School of GeographyFACULTY OF ENVIRONMENT
Introduction
• GIS, great tool but what about error?• data quality, error and uncertainty?
• error propagation?
• confidence in GIS outputs?
• NCGIA Initiative I-1• major research initiative?
• dropped because too hard?
• Be careful, be aware, be upfront...
School of GeographyFACULTY OF ENVIRONMENT
Terminology
• Various (often confused terms) in use:
• error
• uncertainty
• accuracy
• precision
• data quality
School of GeographyFACULTY OF ENVIRONMENT
Error and uncertainty
• Error
• wrong or mistaken
• degree of inaccuracy in a calculation
• e.g. 2% error
• Uncertainty
• lack of knowledge about level of error
• unreliable
School of GeographyFACULTY OF ENVIRONMENT
Accuracy vs. Precision
Imprecise
Precise
Inaccurate
Accurate1
43
2
YO!
4
School of GeographyFACULTY OF ENVIRONMENT
Question…
What does accuracy and precision mean for GIS co-ordinate systems?
School of GeographyFACULTY OF ENVIRONMENT
Quality
• Data quality
• degree of excellence
• general term for how good the data is
• takes all other definitions into account
• error
• uncertainty
• precision
• accuracy
School of GeographyFACULTY OF ENVIRONMENT
Types and sources of error• Group 1 - obvious sources:
• age of data and areal coverage
• map scale and density of observations
• Group 2 - variation and measurement:
• positional error
• attribute uncertainty
• generalisation
• Group 3 - processing errors:
• numerical computing errors
• faulty topological analyses
• interpolation errors
School of GeographyFACULTY OF ENVIRONMENT
Northallerton circa 1867
Northallerton circa 1999
Age of data
School of GeographyFACULTY OF ENVIRONMENT
Scale of data
Global DEM
European DEM
National DEM
Local DEM
School of GeographyFACULTY OF ENVIRONMENT
Digitiser error
• Manual digitising
• significant source of positional error
• Source map error
• scale related generalisation
• line thickness
• Operator error
• under/overshoot
• time related boredom factor
School of GeographyFACULTY OF ENVIRONMENT
Vector to raster conversion error
• coding errors
• cell size
• majority class
• central point
• grid orientation
• topological mismatch errors
• cell size
• grid orientation
School of GeographyFACULTY OF ENVIRONMENT
Original Original raster
Tilted Shifted
Effects of grid orientation
School of GeographyFACULTY OF ENVIRONMENT
Attribute uncertainty
• Uncertainty regarding characteristics (descriptors, attributes, etc.) of geographical entities
• Types:
• imprecise (numeric) or vague (descriptive)
• mixed up
• plain wrong!
• Sources:
• source document
• misinterpretation (human error)
• database error
School of GeographyFACULTY OF ENVIRONMENT
Generalisation
• Scale-related cartographic generalisation
• simplification of reality by cartographer to meet restrictions of:
• map scale and physical size
• effective communication and message
• can result in:
• reduction, alteration, omission and simplification of map elements
• passed on to GIS through digitising
School of GeographyFACULTY OF ENVIRONMENT
Cartographic generalisation
1:3M
1:500,000
1:25,000
1:10,000
City of Sapporo, Japan
School of GeographyFACULTY OF ENVIRONMENT
Question…
An appreciation of error and uncertainty is important because…
School of GeographyFACULTY OF ENVIRONMENT
Handling error and uncertainty
• Must learn to cope with error and uncertainty in GIS applications
• minimise risk of erroneous results
• minimise risk to life/property/environment
• More research needed:• mathematical models
• procedures for handling data error and propagation
• empirical investigation of data error and effects
• procedures for using output data uncertainty estimates
• incorporation as standard GIS tools
School of GeographyFACULTY OF ENVIRONMENT
Question…
What error handling facilities are their in proprietary GIS packages like Arc/Info?
School of GeographyFACULTY OF ENVIRONMENT
Basic error handling
• Awareness
• knowledge of types, sources and effects
• Minimisation
• use of best available data
• correct choices of data model/method
• Communication
• to end user!
School of GeographyFACULTY OF ENVIRONMENT
Quantifying error
• Sensitivity analyses
• Jacknifing
• leave-one-out analysis
• repeat analysis leaving out one data layer
• test for the significance of each data layer
• Bootstrapping
• Monte Carlo simulation
• adds random noise to data layers
• Simulates the effect error/uncertainty
School of GeographyFACULTY OF ENVIRONMENT
Monte Carlo simulation
1. inputs characterised by error model
2. add random ‘noise’ to input
3. run GIS operations on randomised data
4. store results
5. re-run steps 2 thru 4 100 times
6. create composite results map to:
• assess sensitivity of result to random noise
• derive confidence limits
School of GeographyFACULTY OF ENVIRONMENT
Coping with uncertainty
Epsilon model:
1. inputs characterised by error model
2. use required confidence limit to define buffer distance and buffer inputs
3. run GIS operations on buffered data
4. store results
School of GeographyFACULTY OF ENVIRONMENT
Boolean AND
Inclusive AND
Exclusive AND
Exclusive/Inclusive AND
Inclusive/Exclusive AND
Epsilon modelling
School of GeographyFACULTY OF ENVIRONMENT
Conclusions
• Many types and sources of error that we need to be aware of
• Environmental data is particularly prone because of high spatio-temporal variability
• Few GIS tools for handling error and uncertainty… and fewer still in proprietary packages
• Need to communicate potential error and uncertainty to end users
School of GeographyFACULTY OF ENVIRONMENT
Workshop
• Handling error and uncertainty in GIS
• demonstration of jacknifing and bootstrapping methods
• issues of legality and liability?
School of GeographyFACULTY OF ENVIRONMENT
Practical
• Monte Carlo simulation
• Sea level rise and coastal re-alignment/inundation
• Use different resolution terrain models (OS Landform Profile and Panorama) to assess risk of coastal flooding via reclassification
• Error modelling based on Monte Carlo simulation
• Produce maps of 100, 95 and 80% credibility regions