2.397 GHz Link

My guess would be the downhill node has some local QRM that is impacting the received signal.  I was going to suggest trees impinging the downhill node, but I would expect that to also impact the down-to-up path as well.  

Good luck!  This is part of the interesting challenge of working on microwaves.  I had a similar situation over a challenging 3 GHz path that I attributed to a 5G site being turned on near one end of the path.  

73s,
Collier
NM7B

Is the downhill node near other things, such as homes or businesses? I'd suspect it's other wifi networks or 2.4GHz uses near that node. Other uses would include, but not be limited to, bluetooth, wireless security cameras and security systems, cordless phones... Keep in mind that 2.397 (ch -1) is only slightly outside the license-free ISM band, so, if there's a strong signal at the bottom end of that band, it could de-sense the node enough to raise the noise floor, and if it's bursty, it could easily cause a variable SNR like you see. I'm also guessing the UpHill node is physically more isolated from other RF sources. 

Given the effect isn't symmetrical, I'd generally say it's unlikely to be path-related variables, such as vegetation moving - you'd normally see that in both directions, though vegetation moving could also cause reflections of noise into the antenna. I notice the Downhill node is an LDF2. I presume you have that mounted to a dish, in which case it's going to be a lot narrower beamwidth antenna, which could magnify such effects on the noise floor, especially compared to the extremely wide pattern of the mANTBox. I suspect the dish isn't moving in the wind, or we'd see variance on the mANTBox side...