drift group
Machinbird, as I see it, indeed there were several teams implementing different models with different physical basis and data assimilation capabilities, but at one point, the BEA had to estimate a search area (the output of the work for phase III) using all the different backdrifting estimates and their accuracy level. For this latter, the work to quantify the errors between the real trajectories of the buoys and the simulated ones is interesting (the mercator team produced a refined model through this process). The only thing I regret is the absence of montecarlo simulations of the backdrifted trajectories (to get a better idea of the sensitivity of the different models) instead of single trajectories with fixed parameters.