Number of samples
The number of samples required depends on:
- number of map classes / mapping categories
- level of detail/rigour required in assessment.
Apply sampling theory to determine number of samples per image information class, e.g., at least 50 + observations per class for 80-95% expected accuracy. A commonly applied rule of thumb specifies a minimum of 50 samples per cover class.
The number of samples is a BALANCE between a number that is statistically sound and practically attainable.
Example formula to determine number of samples N required
N = required sample size, Z = # of standard deviations associated with a certain percentage or confidence level, E = allowable error, p = expected percent accuracy , q = (100- p)
For example,
We may require; p = 85% , q = (100 - p ) = 15%, E = 5% and Z = 2 standard deviations
