Is the instrumental drift of superconducting gravimeters a linear or exponential function of time?

The instrumental drift of the superconducting gravimeter in Membach, Belgium, is estimated using 9 years of co-located and episodic absolute gravity measurements. We show that the best model of the long-term drift of the SG-C021 is an exponential. The thermal levelers used to compensate tilts are unlikely to induce the observed drift. Rather, the capacitance bridge, magnetic variations, gas adsorption on the levitating sphere, or helium gas pressure variations around it are most likely the possible combined causes of the observed instrumental drift. In practice, either linear or exponential drift models are equivalent as long as the record duration does not exceed 10 years. For longer records, this study demonstrates that an exponential models the drift better than a simple linear trend.