Good catch Suhel! I've been waiting for somebody to start complaining about this :) Requesting everyone to please spot more issues. There are so many trojans planted everywhere which I've given up on flagging for my own sanity because if nobody else flags them then it means that nobody is looking at the data carefully enough and which would mean that nobody really cares about what the data is saying! Believe me, that's an existential reason for sleep loss for me!
In this case, the data seems probably okay but we need better metrics. We need to be tracking not just tree cover gain/loss but the type of loss. This was pointed out by Ravindra during the solvability workshop as well. Whether it's happening on forest fringes or spotty over crop lands, for example. In the first case, it's eroding the forest edges and in the second case its farmers getting rid of trees on their land. People have studied both these kinds of losses separately btw (papers we've posted in the past on the group), and we can infer them from the data too -- we don't have forest boundaries but a proxy can be large contiguous tree patches of 1ha or more, and we can count tree loss in these patches vs. spotty tree losses. We in fact have a github issue created on this too.
Some deep dive screenshots are attached: Ghatagaon with KYL filters turned on for areas seeing significant tree cover gain and tree cover loss. The tree cover gain is spotty in many places, but large patches of new tree cover gain are clearly orchard plantations -- you can see the gridded tree cover in the google maps. The tree cover loss from MWS reports about these areas show loss on fringes but also spotty losses. So bottomline, we need to segregate tree cover gain/loss metrics into spotty or patched. And spotty gain/loss can be further classified as whether it's on farm lands or shrub lands, where gain/loss on shrublands used mostly for grazing would indicate changing dynamics for livestock. This is in fact a pattern we'll put up on KYL soon that some students are working on.
regards
Aadi