Number of polygons for a training dataset
With Google Earth Engine (or any other way to create a dataset), how many polygons per class would be required for adequate accuracy? Is there something like a rule of thumb?
Thanks and cheers,
Often the number of samples proposed for validation derives from non-statistical analysis or empirical values or just experience. For example, the approach proposed by Congalton (1) who defines a number of samples per map class or strata according to the number of classes and the size of the test site. This “rule of thumb” shall give an orientation to estimate the appropriate number of total required samples.
1. Congalton, R.G., Green, K., 2009. Assessing the accuracy of remotely sensed data: principles and practices, 2nd ed. ed. CRC Press/Taylor & Francis, Boca Raton.