Robust comparison of climate models with observations using blended land air and ocean sea surface temperatures

Climate models are used to estimate future global warming, and can be checked against the global warming we have seen so far. Most comparisons suggest that the world is warming a little more slowly than the model projections. We show that this arises in part from the way the comparison is performed, because the methods used in constructing the historical temperature record are different from those used for the models. When we do a like-with-like comparison the discrepancy is reduced by a third, or completely eliminated if the warming fluctuation of the last decade is omitted. The data and computer code for robust blending of climate model outputs are provided. A unix-like system is required (Linux or Mac), although a Windows system with MinGW may also do. The Climate Data Operators package is required. Python 2.6/7 is required along with the numpy/scipy and either the Scientific.IO or netcdf4-python libraries. Shell scripts require csh/tcsh. A rudimentary knowledge of the Unix command line is assumed. CMIP-5 model outputs are required. The full ensemble data are too large to distribute from this website, and so should be obtained from the data providers. A single model output is provided for testing purposes.