Discrepancies between Different Northern Hemisphere Summer Atmospheric Data Products

Abstract

Northern Hemisphere summer (July?August) data from the NCEP?NCAR and ECMWF 40-yr Re-Analysis (ERA-40) reanalyses are compared with each other and with Trenberth's sea level pressure (SLP) dataset. Discrepancies in SLP and 500 hPa are mostly confined to a band connecting North Africa and Asia. In the NCEP?NCAR reanalysis, there is a negative offset in SLP over North Africa and Asia prior to the late 1960s, together with a similar problem in 500-hPa height, and in Trenberth's data there is a negative offset in SLP over Asia prior to the early 1990s. Both these offsets magnify the linear trend from 1958 to 2002 over North Africa and Asia in the NCEP?NCAR and Trenberth datasets. On the other hand, the interannual variability in the three datasets is highly correlated during the periods between these offsets. Compared to SLP and 500-hPa height, there is a more extensive area of discrepancy in 2-m temperature that extends eastward from North Africa across the subtropics into the Pacific, with an additional area of discrepancy over the Arctic and parts of the American continent. At 500 and 100 hPa, the biggest differences in the temperature time series are found in the Tropics, with a marked jump being evident in the late 1970s in the NCEP?NCAR, but not in the ERA-40, reanalysis that is almost certainly associated with the introduction of satellite data. On the other hand, all three datasets agree well over Europe. The summer North Atlantic Oscillation (NAO), defined here as the first EOF of summer mean SLP over the Euro-Atlantic sector, agrees well between the different datasets. The results indicate that the upward trend in the summer index in the 1960s is part of a longer-period interdecadal cycle, with relatively high index values also being found during the 1930s. The running cross correlation between the central England temperature record and the summer NAO shows a strong correlation throughout the last half of the twentieth century, but much reduced correlation in the early part of the twentieth century. It is not clear whether the change in correlation is real, or a data artifact, a topic that requires further research.