Evaluation of Air Temperature Uncertainty Estimates of the NOAA/CIRES/DOE 20th Century Reanalysis Over Greenland

Abstract Due to the high sparsity of in situ observations, reanalysis data sets are often the only source of climate information for the 20th century, particularly in the Arctic. The NOAA/CIRES/DOE 20th Century Reanalysis v3 provides uncertainty estimates, whose reliability we evaluate by comparing with non‐assimilated daily and monthly historical air temperature observations in Greenland. We find that at least for one weather station, the reanalysis correctly estimates the relative difference in uncertainty in different seasons, with higher uncertainties during winter. However, neither are the absolute values of the uncertainty of air temperature generally reliable nor are the day‐to‐day variability and the long‐term trends in uncertainty adequately captured. The error of the reanalysis—both with and without removing constant biases—is remarkably constant over time. Based on this, we show that using as little as a single year of observations—regardless of when those observations were made—provides more reliable information on the reanalysis uncertainty over the whole 20th century than the uncertainty estimates provided by the reanalysis product itself. We hypothesize that the lack of reliability of the reanalysis spread is at least partly influenced by the prescribed uncertainty of sea surface temperature boundaries. We conclude that uncertainty estimates of historical reanalysis for Greenland need careful evaluation.