In the framework of the BRAVOSEIS project, eight broadband ocean-bottom seismometers (OBS) were deployed in the Bransfield Strait, Antarctica. The internal clocks of the data loggers were synchronised with a GPS signal before deployment. After the recovery 13 months later, the time difference (skew) between the OBS clocks and GPS was measured. Due to a malfunction of some data loggers, only four stations revealed a skew value. Furthermore, the clock drift turned out to be non-linear, so not only the total clock drift, but also the shape of the drift curves has to be determined for all stations. The evaluation of OBS clock drift by noise cross-correlations was described by Hannemann et al. (2013). We expanded this method for non-linear drift. Two onshore stations with GPS-controlled data loggers could be used, one temporary station of the BRAVOSEIS network and one permanent station of the Antarctic Argentinean Italian Network. Cross-correlations of long-period noise (9-100 s) were calculated, stacked and filtered in pairs between OBS and land stations. This resulted in one trace per day. The first of these traces was correlated with all other traces to determine the daily offset. A polynomial was fitted to the offset data to obtain a smooth function for the time correction. The quality of the cross-correlations between OBS and onshore stations varied widely. The four OBS in the northern part of the Bransfield Basin yielded excellent results, but the quality of the cross-correlations was not sufficient to determine the clock drift of the four southern stations precisely. Therefore, we corrected the data of the northern stations for clock drift and used them as a base for cross-correlations with the southern stations. One station suffered from a technical fault of the seismometer after five months; in this case, we correlated the hydrophone data with the vertical component of the neighbouring OBS station. Another station failed to synchronize the internal clock before deployment, here we used nearby airgun shots from a seismic survey to estimate the initial time. Noise cross-correlations have so far been used to calculate linear clock drifts within OBS networks. A combination of careful filtering, usage of data from nearby onshore stations and calculation of polynomial fits can expand this method for non-linear drifts.