I am very new to remote sensing analysis.

I am conducting a supervised classification in Google Earth Engine. I want to verify that as I develop my training samples that it is ok for me to toggle between spectral band combinations (as the different combinations help me see more clearly what to identify as which landcover type). I think that doing so is standard practice, but I want to verify!

i.e. I want to confirm that when I am viewing the different layers, represented by different spectral combinations, and drawing polygons that I identify as different landcover types, that even though the visualization of the pixels are changing when I toggle between spectrum combinations, when I actually run the algorithm that trains the classificatoin based on my samples that the algorithm isn't running it based off of the premise that I used just one specific band combination.

1 Answer 1


Yes, it is perfectly acceptable and encouraged to use ancillary data to increase the accuracy of your training samples. However, be cautious about using data that was acquired at a different time period as your features of interest may have changed due to seasonal variations or alterations across the landscape.

In the following publication, vegetation indices were used to help with the training sample acquisition (p.630). Also, it is common to use derived LiDAR metrics to aid in generating training samples of certain features.

Poznanovic, A. J., Falkowski, M. J., Maclean, A. L., Smith, A., & Evans, J. S. (2014). An accuracy assessment of tree detection algorithms in juniper woodlands. Photogrammetric Engineering & Remote Sensing, 80(7), 627-637.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.