That’s a classic in problematic conversions. The majority of the image is either blue or white.
Try adding a little bit of the border prior to converting with NLP and you’ll get a much more balanced image to begin with. I have no idea on the interals of NLP and why blue is so problematic, but adding border help.
I would add couple of things to the excellent suggestion by @cvandebroek :
you may add 1 or 11/2 stop while capturing the film image. You want mask/rebate to be as light as possible without clipping . That makes for better colors.
On the side let me note that film holder apparently is not aligned with the camera in the best way. You can see that sky is sharper then the foreground in terms of capturing film grain and defects - you may want to address that. Also the back light illumination is not exactly uniform - as evidenced by that picture where curve is applied to the extreme. That may also lead to certain color disbalance in the final image. Hope this helps.
I had similar cases in which a very tight crop would help. Imagene a thirds grid and only use the center of it, cropping away the outer 8 parts, or crop in from left and right to only leave a small strip. Try to get as big vaiety of colours and tones into the visible part.
Thanks @VladS. The camera is indeed not 100% aligned, because the copy stand I’m using is not able to hold the camera perfectly straight. My light source is also on the cheaper end, so I guess this is why it’s not perfectly uniform.
However, as I’m mostly shooting with a P&S camera I’m quite happy with the results.