Optimal settings for Crop Type Map


Are there any optimal/recommended parameters for crop type mask processor?

53 PM

I’m just wandering since according to this forum more training trees are better but with diminishing returns.



The performance will indeed vary depending the quality of your in-situ data and acquisitions. We had good results with the defaults, although at times we’ve seen indications of overfitting. Increasing the number of trees should be fine, but the max depth value is more important.

You should experiment with various combinations of parameters. Besides the quality metrics reported in the resulting product, an easy way to guess the quality of the classification is to look at the model size on disk. If it’s too large (say 1 GB or more), it’s a pretty good indication that overfitting occurred, or that your input data is noisy. But this is just a rule of thumb, of course.