I have a string of around 70 dNBR images designed to assess burn scarring as part of the June 2017 Knysna Forest fires. However, the images do not look correct, particularly when comparing them to unprocessed post-fire images containing only visible bands. I I have been asked to change the classification by eye, manually altering the breaks and to reclassifying the dNBR images into 4 features ranging from unburned to high-burn severity. Using various classification models found in articles has not been effective, even when changing the total number of features. I am now using ArcMap and not ArcGIS Pro and manually moving the natural break bars under classification in layer features for each image. The issue is I don't really know how a more appropriate dNBR image is supposed to look and therefore progress is slow. Any advice on where to start would be greatly apprenticed.
Working with Sentinel-2 and Landsat-8 images over months Jan-Sept 2017.