Accuracy Flashcards
What is accuracy?
Accuracy expressed how closely spatial information matches reality. It does this by comparing the information I question to some reference information that is considered to represent reality more accurately. This can occur for individual values or through a test statistic.
Give some examples of how one can determine accuracy?
One may determine accuracy of the building footprint obtained from aerial photography by comparing them to ground surveys. This comparison can occur point wise, with a measure of each polygon or for the set of all footprints.
One could describe the accuracy of land cover classification of a satellite image by comparing it with some ground truth pixel by pixel or a particular land use category.
When do inaccuracies occur?
Because spatial information results from conceptualising reality in terms of fields, objects and networks and events, inaccuracies can get introduced in all these phases.
Conceptualisation can produce many errors, by treating a phenomena as a set of unconnected objects rather than a network.
Observation errors can result from bad equipment and transcription errors, or from assigning the wrong measurement units to values.
Analysis can introduce inaccuracies, by applying inadequate computations or combining data with incompatible granularities.
What is the notion of a true value?
The truth of some spatial information is normally not known and can only be estimated from observations the notion of a true value is called bias. Results from repeated measurements distribute regularly around the true value and measuring can be considered a radon process affected by ransoms errors with no systematic error.
How do you minimise systematic errors?
Minimising systematic errors is done through calibrationC which ideally reduces them to a level below granularity.
How is accuracy represented
Metadata in spatial, temporal and thematic accuracy
RMSE example of standard errors permitted with USGS topo maps
How is accuracy data used?
Mean, median and standard deviation from multiple measurement
Error propagation
What are the key issues to accuracy?
Best practice of reporting information required starting only as many digits after the decimal place as the combined effect of granularity and accuracy.
Exactitude is not truth
Systematically recording metadata is not done
