A method of marine time-lapse seismic surveying of a subsurface formation, comprises providing a baseline survey, providing a monitor survey that includes information about changes in the subsurface relative to the baseline survey, recording a repeat survey so closely in time to one of either the baseline survey or the monitor survey that changes in the subsurface can be ignored but under different near-surface conditions from said one survey, computing a short-time survey difference between the repeat signals and signals comprising said one of either the baseline survey or the monitor survey, computing a monitor survey difference, matching the short-time survey difference and the monitor survey difference to derive a matched noise survey difference, subtracting the matched noise survey difference from the monitor survey difference, and outputting a noise suppressed survey difference based on the result of the subtraction.