Many applications in seismology rely on the accurate absolute timing of seismograms. However , both seismological land stations and ocean bottom seismometers (OBSs) can be affected by clock errors, which cause the absolute timing of seismograms to deviate from a highly accurate reference time signal, usually provided by GPS satellites. Timing problems can occur in land stations when synchronization with a GPS signal is temporarily or permanently lost. This can give rise to complicated, time-dependent clock drifts relative to GPS time, due to varying environmental conditions. Seismometers at the ocean bottom cannot receive GPS satellite signals, but operate in more stable ambient conditions than land stations. The standard protocol is to synchronize an OBS with a GPS signal immediately before deployment and after recovery. The measured timing deviation, called 'skew', is assumed to have accumulated linearly over the deployment interval, an assumption that is plausible but usually not verifiable. In recent years, cross-correlations of ambient microseismic noise have been put to use for correcting timing errors, but have been limited to interstation distances of at most a few tens of kilometres without reducing the temporal resolution. We apply noise cross-correlations to the evaluation of clock errors in four broad-band land stations and 53 wideband and broad-band OBSs, which were installed on and around the island of La Réunion in the western Indian Ocean during the RHUM-RUM (Réunion Hotspot and Upper Mantle-Réunions Unterer Man-tel) experiment. We correlate all three seismic components, plus a hydrophone channel in OBS stations. Daily cross-correlation functions are derived for intermediate distances (∼20 km) for land-to-land station pairs; stable, 10 d stacks are obtained for very large interstation distances up to >300 km for land-to-OBS and OBS-to-OBS configurations. Averaging over multiple station pairs, and up to 16 component pairs per station, improves the accuracy of the method by a factor of four compared to the single-channel approaches of prior studies. The timing accuracy of our method is estimated to be ∼20 ms standard deviation or one sample at a sampling rate of 50 Hz. In land stations, nonlinear clock drifts and clock jumps of up to 6 min are detected and successfully corrected. For 52 out of 53 OBSs, we successfully obtain drift functions over time, which validate the common assumption of linear clock drift. Skew values that were available for 29 of these OBSs are consistent with our independent estimates within their observational error bars. For 23 OBSs that lacked skew measurements, linear OBS clock drifts range between 0.2 and 8.8 ms d −1. In addition to linear drift, three OBSs are affected by clock jumps of ∼1 s, probably indicating a missing sample problem that would otherwise have gone undetected. Thus we demonstrate the routine feasibility of high-accuracy clock corrections in land and OBSs over a wide range of interstation distances.