We think it’s a combination of a few possible factors - naturally lower scattering from protein at these frequencies, a loss of coherence due to the membrane (basically I assume it’s a membrane protein based on that FSC alone), persistent misalignment, or overlapping boxes combined with symmetry. The latter can even make FSC never fall to zero (including soluble protein).
Almost all (all?) published membrane protein structures have such a bump. In better refinements (for the same dataset) it will be less pronounced. It’s also affected by masking. I’ve been actively investigating it, but no conclusions so far.
Map-to-model FSC, EM ringer score, and subjective map quality assessment should guide you as well as GS-FSC. Usually the 0.143 FSC reported is close to the subjective assessment in such cases, so I wouldn’t worry about it if that remains the case.