grity
I think, a line based on a polynomial regression, is a little bit to simple for a 3D problem
Grity, I think we realize that as a first order linear solution to a complex (chaotic) problem, there will be a lot of inaccuracy and uncertainty remaining, however given the choice of 6 days uncertainty and 30 hours uncertainty, reason says the one with the significantly shorter period of uncertainty will have much greater accuracy. The key question is, is the SAR detected slick relevant?
If the slick was not relevant, I would still expect another potentially relevant slick to show on the same SAR image taken 30 hours post accident. With an airframe that impacted with as much energy as AF447, I would expect tanks to rupture and the fuel to promptly be on the surface after impact, with a further trail of fuel coming up from the wreckage with a relatively short half life (perhaps 10 minutes)
FluidFlow made a number of simplifications from the drift group's methods, so a methodology that re-integrates those methods can't help but be yet more accurate.
The choice of 22 bodies with almost common recovery times as a data subset for analysis is a good idea. Uncertainties in this group come from initial depth of insertion in the water (3D flow vectors) , time until bodies reach the surface and begin moving with the surface flow vectors, and value of the resultant aerodynamic/hydrodynamic drag coefficients as bodies begin to float progressively higher in the water.
Hopefully consideration of these various factors will make only small adjustments in the X marks the spot prediction. If so, that location will be ready for the next search phase to check out.