Summary and Conclusions
- Our problem is a special case of information fusion - how do we combine
information from an OGCM model and observed SST data?
- We also have here a spectial type of image registration problem.
- Imprecision in the data is catered for by nonlinear MLP regression.
The mapping of model to observed data is carried out by a nonlinear MLP.
- Validation follows learning, and is followed in turn by generalization.
- Our function mapping problem has the following characteristic:
Model data is relatively very smooth, Observed data is very choppy.
- Multiscale decomposition provides a framework for assessing an
appropriate, or best, resolution level to carry out the mapping.
We can "tune" the mapping.
- Once the strategy has been carefully prototyped, application to
many datasets is straightforward. Mean square error, MSE,
provides quality control criterion.
Text of paper available at: