AFNI datasets can be made of data stored as integers or floating point numbers among a couple other less used types. If the datasets are in integer format, each sub-brick volume may be scaled by a floating point factor. Although this format saves disk space it also suffers from precision issues, average non-zero error in precision is shown in the warning. For a good detailed description of the warning, see this previous thread:
[
afni.nimh.nih.gov]