I have this script in GEE that has worked successfully before for LULC mapping (with 8 classes). I used this script for mapping solar panels (2 classes: solar & non solar). The accuracy results from RF, SVM, CART algorithms are more that 90%. When I map the classification result, it is so wrong (instead of seeing solar panels where they actually exist, half of the province is classified as solar panels). How can the resulting map be so wrong if the accuracy results are so high? I share the scripts and all assets.


  • Your script is too long to review. Could you please share (also in the question body) a minimized version of your code? Commented Feb 23, 2023 at 12:18
  • Dear @KadirŞahbaz, I wanted to include the script here but it was also very long and might lost important parts. If you run the script, you will see the overall validation accuracy printed and the Final RF map. By looking at the map, it is obvious that it is wrong
    – Paris
    Commented Feb 23, 2023 at 12:29

1 Answer 1


I had similar situation with classifications in GEE. Now I create separately two sets of ROI for training and for control (not random from one set 80 /20) then accuracy is lower and closer to real situation

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.