As for the bad results your seeing, it looks like something is failing during the decay of the threats. There’s a known bug that we’ve recently patched that is not released yet with doing this decay and convolution. A workaround might be altering the MAX_DIST in the threat table. Since your LULC is about 1200 x 1200 meters, do you think you could scale back your MAX_DIST which is in km in the threat table? The exponential decay for threat road especially. I might try 1,1,1 down the board. That would basically decay across the entire LULC have still, but may provide stability to the bugged algorithm.
You can look at intermediate/filtered_road_c.tif to see the bad output currently and after you re-run to see if it looks more appropriate.
Doug