Prioritizing data for improving the multidecadal predictive capability of atmospheric models

The prioritization accorded to observation types currently being considered for a space-based climate observing system is extended from a previous study. Hindcast averages and trends from 1970 through 2005 of longitude–latitude maps of 200-hPa geopotential height and of net downward shortwave and longwave radiation at the top of the atmosphere are investigated as relevant tests of climate models for predicting multidecadal surface air temperature change. To discover the strongest tests of climate models, Bayes’s theorem is applied to the output provided by phase 5 of the Coupled Model Intercomparison, and correlations of hindcasts and multidecadal climate prediction are used to rank the observation types and long-term averages versus long-term trends. Spatial patterns in data are shown to contain more information for improving climate prediction than do global averages of data, but no statistically significant test is found by considering select locations on the globe. Eigenmodes of intermodel differences in hindcasts may likely serve as tests of climate models that can improve interdecadal climate prediction, in particular the rate of Arctic tropospheric expansion, which is measurable by Earth radio occultation.