Downscaled Climate Projections

Fitting Statistical Models - Cross Validation

- Use station observations and NCEP Reanalysis for 1950-2009
- Do separate fit for each calendar month
- Parameters are estimated using Maximum Likelihood
- All fits are Cross Validated:
- Set aside 3 years of the 60 years of data
- Fit a model to remaining 57 years of data
- Test model on the 3 years that were set aside
- Repeat so that all 3 year chunks have a chance to be set aside
- Number of skillful predictors while cross validating determines the number of predictors for final fit with all 60 years
- The empirical predictor weights are also cross validated at the same time as the statistical models

Home