[GRASS-user] How much training areas is enough?

[maning sambale]

Hi,

In doing remote sensing classification, we are advised that more
training areas is best, but how much really?  There must be a threshold
where it is pointless to add more since it will not contribute
significantly to the classifcation process.

I would like to ask what are the standards/best practices in defining
and limiting training areas for classification in GRASS.  Some
literature would advise > 100 pixels per infomation class.  Are there standards?

For example, I have an image composed of {x} number bands of
approximately {y} hectares that I need to classify to {z} classes.
How much training area should be sufficient?

Any ideas?

cheers,
maning

-- 
|---------|----------------------------------------------------------|
| __.-._ |"Ohhh. Great warrior. Wars not make one great." -Yoda |
| '-._"7' |"Freedom is still the most radical idea of all" -N.Branden|
| /'.-c |Linux registered user #402901, http://counter.li.org/ |
| | /T |http://esambale.wikispaces.com|
| _)_/LI
|---------|----------------------------------------------------------|