Summary Full Waveform Inversion (FWI) is an appealing technique for time-lapse imaging, especially when the prior model information is included into the workflow. After baseline reconstruction, several strategies such as: differential, parallel difference, and sequential difference can be used to assess the physical parameter changes. Using the synthetic Marmousi data-sets, we study which strategy could be more robust and give more accurate time-lapse velocity changes in the presence of noise. We illustrate that the sequential difference method, starting from a reconstructed baseline model and inverting the monitor data-set, can give a better result in the case of random ambient noise. However, the differential approach could also be interesting if the time-lapse response can be preserved from the noise level.