Develop method to more accurately detect when mining began in a given pixel

[WassonMF]

We know that there is a lag (and a large window of uncertainty) around when the algorithm detects mining in a given pixel compared to when that mining actually occurred. It would be helpful to quantify how large the size and dispersion of that error is.

Preliminary thoughts are to look at the trajectory of NDVI within a given year for certain pixels using NDVI from individual scenes (rather than composites), however, this might be challenging to automate, given the cloudiness of most summer Landsat scenes in Appalachia.

Another approach might be to just set a number of control points for this specific purpose and go through them manually to assess actual timing of active mining compared to the output of the algorithm.

OR maybe there are far better ideas...