Reviewing the initial images, I found that the upper had some diamond shaped pattern, visible in the darker “sand” portion of the image. The two scans show fairly low contrast und different overall brightness as noted by @Harry.
I tested low contrast/low density/narrow histogram negatives with different lighting (Enlarger, Kaiser Plano, iPad…) and found that these react very strongly when converted with NLP’s adaptive algorithm. During the conversion, the width of the histogram is stretched (depending on darkest and brightest parts of the image) and this “amplification” picks up minute differences and makes them easily visible, even though NLP is tolerant to variations in exposure of the scans.
I propose/advise you take a series of captures, taken at the same aperture and iso settings and different times to produce scans of e.g. -1, 0, +1, +2 and convert them like shown here.
Also, it can help to deviate from standard procedures occasionally. Some of my negatives get me a better starting point when I don’t white balance them. NLP has soo many ways to tweak the conversion and converted image that it makes sense to (systematically) test different settings. This does not mean that you’ll find the silver bullet for all of your images though.