Model-Based Imputation of SST Images

G. Zheng, F. Murtagh and J.G. Campbell


  1. Daily SST data available, prepared for 82 x 70 0.1 degree pixels of the west African coast.
  2. Output from Ispramix OGCM, with assimilation, also available.
  3. Model incorporates meteorological and other external information.
  4. Over 70% of observed SST values are missing (over a year's data).
  5. Can we impute such missing values from the model?
    What is the best way to do this?
    How effectively can we do this?
  6. This can be taken as a nonlinear regression problem.
  7. Obviously neighborhood information is important: we must take that into account.
  8. Resolution scale is very important also in geo-referenced data: how do we take that into account? How do we choose appropriate relative resolution scales?