Hello ESRI Community,
Question: What is the process to automate the ground truth verification process for thousands of stratified random sampling points on historic imagery? In my mind, an ideal solution would be to intersect with a creditable source dataset (ie. calculate the attributes of a field to equal the value of the authoritative LULC map at intersecting points).
In a recent analysis, I had to reduce my ground truth points to 10% of what I should have sampled to make it more manageable (~240 points versus ~2400 x 12 images = ~28,800 stratified random sampling points that need to be ground verified).
Background: I have recently completed an extensive Land use/ Land cover (LULC) analysis using historic Landsat data (12 images sourced from Landsat 3, 4-5, 7, 8, and 9, from all three sensor types: MS, TM, and OLI TIRS). Each of the 12 images were classified using supervised classification methods while leveraging the Random Trees algorithm on training samples.
As you may (hopefully) know, the most important part of performing LULC analyses is to report your users and producer's accuracy of your classified images. Do to this massive realization of accuracy reporting, I had to reduce my analysis down to 2 images with approximately 480 ground truth points. I would love to have been able to calculate this on all 12 images.
Part of verifying "ground truth" points, is to manually click through many stratified random sampling points and verify with a source OTHER than your source data. In a perfect world, this would be done via mobilizing to the site via GPS and verifying with boots on the ground. Obviously, that is unrealistic, especially with historic imagery.
What sorts of tips, tricks, ideas do you have to make this process more efficient? Any ideas for verifying ground truth LULC types for historic imagery (1970s-2023)?
Thank you!
