Experiments

  1. Trial 1: we learned, tested, and applied the nonlinear mapping (MLP) to the model and observed SST datasets.
  2. Trial 2: we wavelet transformed the model and observed SST datasets into the sets
    Im = Wm(1) + Wm(2) + Rm
    and
    Io = Wo(1) + Wo(2) + Ro
    where W(1) is wavelet resolution scale 1, and R is the residual (or background).
  3. Trial 2 (cont.): we then learned, tested and applied the nonlinear mapping to the datasets
    Wm(1) and Wo(1)
    Wm(2) and Wo(2)
    Rm and Ro
    and combined the results.
  4. In Trial 2, we looked in particular at data filtering, through suppressing one or more resolution levels.
  5. Trial 3: mapping of Im onto Ro, and the mapping of Im onto Wo(2) + Ro. The best result is given by the first of these Trial 3 mappings.