Evaluating multi-model ensembles

Guidance for product user
esd diagnostics for common EOFs (compare sd vs mean of PC loadings) for modes of variability.
– compare observations with 90% confidence intervals assumin binomial distribution in number outside(GCMs/RCMs – maps)
– compare observed trend(s) (eg reanalyses) against distribution of observed trends.
– rank statistics of reanalysis/GCMs (Wilcoxon type)
– a summary synthesis based on the above for temperature, precipitation, slp.
– automated with subsetting of ensemble to check if bad models deteriorate the results – guidance to the selection of ensemble.

graphic: zoomable maps with scores for each grid box + a global statistic.

[help]