Bias correction

The published forecasts correct each SEAS5 member's rainfall amount by mean–variance scaling (MVS) before the tercile probabilities are computed. This benchmark compares 13 other corrections with it, member by member, for every grid cell and target, in two steps: a screening of the amount correction alone, then a confirmation through the full probability pipeline that accounts for how a method is chosen. A suggestion engine combines that evidence with the independent seasons scored since; the published method changes only by an explicit decision.

• January initialization · FMAM and its months
  12 methods against Mean-Variance Scaling (operational); confirmation run; shadow and validation
  Candidates: 0 of 10 target–product combinations
• February initialization · FMAM and its months
  12 methods against Mean-Variance Scaling (operational); confirmation run; shadow and validation
  Candidates: 1 of 10 target–product combinations
• May initialization · JJAS and its months
  12 methods against Mean-Variance Scaling (operational); confirmation run; shadow and validation
  Candidates: 4 of 10 target–product combinations
• September initialization · ONDJ and its months
  12 methods against Mean-Variance Scaling (operational); confirmation run; shadow and validation
  Candidates: 1 of 10 target–product combinations

Full write-up: docs/47.