Discriminating between climate observations in terms of their ability to improve an ensemble of climate predictions.

In view of the cost and complexity of climate-observing systems, it is a matter of concern to know which measurements, by satellite or in situ, can best improve the accuracy and precision of long-term ensembles of climate projections. We follow a statistical procedure to evaluate the relative capabilities of a wide variety of observable data types for improving the accuracy and precision of an ensemble of Intergovernmental Panel on Climate Change (IPCC) models. Thirty-two data types are evaluated for their potential for improving a 50-y surface air temperature trend prediction with data from earlier periods, with an emphasis on 20 y. Data types can be ordered in terms of their ability to increase the precision of a forecast. Results show that important conclusions can follow from this ordering. The small size of the IPCC model ensemble (20 members) creates uncertainties in these conclusions, which need to be substantiated with the larger ensembles expected in the future. But the larger issue of whether the methodology can provide useful answers is demonstrated.