Susceptibility distortion correction using synthetic fieldmap

Dear fMRIprep experts,
I am using fMRIprep 1.5.1rc2 with synthetic fieldmap correction. In some subjects, it yielded very large/unrelistic FD, DVARS and GS etc. We checked the raw data and it does not contain such large fluctuations. I wonder if this has occured to anyone before and what the cause might be.

Motion parameters (hence FD) are calculated before SDC, so that shouldn’t affect this. It looks like a severe motion correction failure in the last several volumes. Are there any artifacts in the data? You might try looking at it with MRIQC or some other quality control tool.