Detecting change between 2 aerial images (leaf off)

[SarahKendell]

Hello - this past winter, the forest I work experienced some heavy snow loading in December which caused significant damage to areas of our forest. We are now in the middle of assessing these areas for damage. My question relates to finding the best way to detect change over a larger area in an efficient manner rather than digitizing 30,000+ acres of damage. We have imagery available from Spring of 2022 (leaf off) and had another flight flown this spring over the area with heaviest damage, Spring 2023 (leaf off). I've worked through a few of the tools available in Arc to assess change, but am not having any success. Our imagery only has RGB bands. I've also tried to classify the image, but I believe since the imagery is leaf off, it not able to pick up difference in a down tree versus a standing (similar color value). 

Any ideas or suggestions would be greatly appreciated! I've attached some screenshots of the imagery for reference.  

[JeffreySwain]

Great question and thanks for the description. I would think you are on to something about the imagery if you are studying forests. Most times when opting to collect 'leaf off' imagery, the purpose is to see what is under the trees. Since you are trying to assess the damage to the trees, perhaps 'leaf on' would be preferred. If the 'downed trees' are recognizable in the imagery, they you can try to classify them as a class rather than more traditional land cover classes. You may even have some luck using deep learning - object detection to find them as well. If it were me, I would try and see if I could use a segmented image to locate the damaaged trees, create a classified thematic image in both, and then perform the change detection. If you can't recognize the 'damaged trees' well enough, then I would consider trying to acquire 'leaf on' imagery and then the damaged areas will be readily visible since the foliage would not be there where it is damaged. 

[PavanYadav]

@SarahKendell if you like to use Deep Learning to detect changes, you might find the following useful. 

https://developers.arcgis.com/python/guide/how-change-detection-works/
https://doc.arcgis.com/en/allsource/latest/analysis/geoprocessing-tools/image-analyst/detect-change-...

attaching workflow in pdf format. we plan to include these steps in the upcoming release. 

Cheers 

Pavan

[SarahKendell]

Thanks for the information Jeffrey and Pavan. Progress on this is temporarily stalled while I figure out if someone has an image analyst extension in my organization.