Donner et al components, and (iii) low-frequency variability indicating a true longrange dependent process. In the presence of such multi-scale variability, common estimators of long-range memory in time series are prone to fail if applied to the raw data without previous separation of timescales with qualitatively different dynamics.