Thursday, February 19, 2015
More on real temperatures and models
but [perhaps a tease can be helpful. The scaling required adjusting the baseline and the trend slightly to fit the majority of the more accurate SST data. The divergence at roughly 1900 might be real or might be an artifact, but is is close to the uncertainty range of both products.
SWo why would I waste my time with all this?
The Marotzke and Forster 2015 Nature article compares models with HADCRUT4 "surface" temperature. The HADCRUT4 data is about 70% SST and about 30% land surface temperature. To get the best match of models to observation I have to create a combined 70% SST and 30% land model product since the Model tas appears to be a real attempt at a 2 meter or some real surface temperature. There isn't an equivalent observational product. Since most of the model errors appear to be in the SST portion of the programming, it makes sense to me to compare worst to worst before getting into fantasy to fantasy. If the Berkeley "scaling" is acceptable, that would provide some out of sample data for Marotzke and Forsester to expand this research "globally" or just land only if they don't like the scaling. Either way it would be nice to see apples with apples.