Coronagraphic imaging of exoplanets and circumstellar environments using ground-based instruments on large telescopes is intrinsically limited by speckles induced by uncorrected aberrations. These aberrations originate from the imperfect correction of the atmosphere by an extreme adaptive optics (ExAO) system; from static optical defects; or from small opto-mechanical variations due to changes in temperature, pressure, or gravity vector. More than the speckles themselves, the performance of

more »

... -contrast imagers is ultimately limited by their temporal stability, since most post-processing techniques rely on difference of images acquired at different points in time. Identifying the origin of the aberrations and the timescales involved is therefore crucial to understanding the fundamental limits of dedicated high-contrast instruments. In previous works we demonstrated the use of a Zernike wavefront sensor called ZELDA for sensing non-common path aberrations (NCPA) in the VLT/SPHERE instrument. We now use ZELDA to investigate the stability of the instrumental aberrations using five long sequences of measurements obtained at high cadence on the internal calibration source. Our study reveals two regimes of decorrelation of the NCPA. The first, with a characteristic timescale of a few seconds and an amplitude of a few nanometers, is induced by a fast internal turbulence within the enclosure. The second is a slow quasi-linear decorrelation on the order of a few 10 −3 nm rms/s that acts on timescales from minutes to hours. We use coronagraphic image reconstruction to demonstrate that these two NCPA contributions have a measurable impact on differences of images, and that the fast internal turbulence is a dominating term over to the slow linear decorrelation. We also use dedicated sequences where the derotator and atmospheric dispersion compensators emulate a real observation to demonstrate the importance of performing observations symmetric around the meridian, which minimizes speckle decorrelation, and therefore maximizes the sensitivity to point sources in difference of images.