Trying to use Multiband imagery to identify permeable surfaces

Question

Hello, I am using this tutorial https://learn.arcgis.com/en/projects/calculate-impervious-surfaces-from-spectral-imagery/ to find permeable surfaces for possible tree planting in our city. At the Classify the imagery Step 3 Click Run the process goes on for a few hours, about 20 or so, which is to be expected considering the size of the area. It throws errors like " ERROR 003436: No training samples found for these classes: Roof(Black Grey), ImperviousShadow, ArtificialTurf, Crop.
Failed to execute (TrainSupportVectorMachineClassifier).
Failed at February 24, 2022 1:25:05 PM (Elapsed Time: 48 minutes 34 seconds)", but I did show it crop land in the sample and saved it.

Second run threw " ERROR 003436: No training samples found for these classes: Roof(Black Grey), ImperviousShadow, ArtificialTurf, Crop.
Failed to execute (TrainSupportVectorMachineClassifier).
Failed at February 25, 2022 6:01:58 PM (Elapsed Time: 47 minutes 33 seconds)", again I showed it shadows.

When I finally got it to run and produce results they were wildly inaccurate; identifying open water as crop land for example. I followed the directions so I am not sure what the issue is; is it that I am trying to process a large series of images?

IvanPotic · Answer

I had the same problem with the "ERROR 003436".

I want to confirm that @Taren_Esri first suggestion solved my problem.

I clipped the satellite imagery twice. First clip was wider area of Interest and second clip was the "area of interest", significantly smaller part. I was trying to classify smaller clipped part ignoring the fact that I noticed that when zooming to layer I have some invisible buffer (no data or some other problem that arose during clipping). Reverting the classification process to the wide clipped footage saved the day and the new classification process went smoothly. I used previously collected training samples from correctly formatted shape file.

Taren_Esri · Answer

For the sake of future reference, I recently worked on a case that resolved this same issue. If you encounter this error message, check two things. The first is that you are running your analysis on the same and original raster you trained your classes on. This means it can't have been clipped, masked, etc.  The second is ensuring your input training sample feature class is actually split and individual features that are appropriately labeled for their class. This means no multipart features. You can run Multipart to Singlepart to split them up if needed. Hope this helps someone!

dvandommelen · Answer

I've received this same error. Anyone know an answer?