A multi-stage FWI workflow uses multiple-contaminated FWI models to predict surface-related multiples. A method embodying the present technological advancement, can include: using data with free surface multiples as input into FWI; generating a subsurface model by performing FWI with the free-surface boundary condition imposed on top of the subsurface model; using inverted model from FWI to predict multiples; removing predicted multiples from the measured data; using the multiple-free data as input into FWI with absorbing boundary conditions imposed on top of the subsurface model; and preparing a multiple free data set for use in conventional seismic data processing.