In the doc strings it says that all noise types with the suffix _sigma are mixed together to contribute to the total of the temporal variance of noise. However, this is not true for drift_sigma in the latest version of fmrisim. Docstrings need to be updated to accommodate this nuance.