Q. J. R. Meteorol. Soc. http://doi.org/qbj (2013)

Incomplete and uneven global data coverage may cause a bias in temperature reconstructions. The polar regions and Africa are poorly sampled in the commonly used HadCRUT4 data set, which for the past few decades has covered around 84% of the Earth surface. The Arctic is a key source of uncertainty owing to a lack of observations and its recent accelerated warming relative to other regions.

Kevin Cowtan, of the University of York, UK, and Robert Way, of the University of Ottawa, Canada, investigate two alternative methods for reconstructing global temperature through extrapolation of unsampled areas. These methods are judged on their ability to reconstruct omitted observation values, and are both found to provide good temperature reconstructions. Uncertainties in temperature estimates are reduced using either method, compared with omission of the unobserved regions.

This work suggests that there has been an underestimation of recent temperature trends arising from incomplete coverage, which has produced a cold bias in the HadCRUT4 data set.