Reanalyses-based tropospheric temperature estimates: Uncertainties in the context of global climate change detection

Uncertainties in estimates of tropospheric mean temperature were investigat ed in the context of climate change detection through comparisons of the Na tional Centers for Environmental Prediction-National Center for Atmospheric Research (NCEP-NCAR) 40-yr reanalysis (1958-97), the National Aeronautics and Space Administration Data Assimilation Office (NASA/DAO) 14-yr reanalys is (1980-93), the European Centre for Medium-Range Weather Forecasts Reanal ysis Project (ERA) 15-yr reanalysis (1979-94), and the satellite microwave sounding unit channel 2 (MSU Ch2) (1979-97) temperature data. The maximum o verlap period for comparison among these datasets is the 14 full years Janu ary 1980 to December 1993. This study documents similar shifts in the relat ive bias between the MSU Ch2 and the ERA and the NCEP-NCAR reanalyses in th e 1991-97 period suggesting changes in the satellite analysis. However, the intercomparisons were not able to rule out the changes in the reanalysis s ystems and/or the input data on which the reanalyses are based as prime fac tors for the changes in the relative bias between the MSU and ERA and NCEP- NCAR reanalyses. These temporal changes in the relative bias among the reanalyses suggest th eir limitations for global change studies. Nonetheless, the analysis also s hows that the pattern correlations (r) between the MSU Ch2 monthly mean fie lds and each of the reanalyses are very high, r > 0.96, and remain relative ly high for the anomaly fields, r > 0.8, generally >0.9. This result sugges ts that reanalysis may be used for comparisons to numerical model-generated forecast fields (from GCM simulation runs) and in conjunction with "finger print'" techniques to identify climate change. In comparisons of the simple linear trends present in each dataset for the 1980-90 period, each of the reanalyses had spatial patterns similar to MSU Ch2 except that the NCEP-NCAR reanalysis showed smaller "positive" (warming ) trends in comparison with the MSU while the ERA reanalysis showed larger positive trends. The NASA/DAO reanalysis showed a mixed pattern. Many regio ns of the globe are identified that showed consistent warming/cooling patte rns among the major reanalyses and MSU, even though there were disagreement s in the exact magnitude among the analyses. The spatial patterns of linear trends changed, however, with the addition of three years of data to exten d the trend analysis to the 1980-93 period. This result suggests that such simple linear trend analyses are very sensitive to the temporal span in the se relatively shea datasets and thus are of limited usefulness in the conte xt of climate change detection except, however, when the signal is large an d shows consistency among all datasets. The long record (1958-96) of seasonal mean 2-m temperature anomalies from N CEP-NCAR reanalysis is well correlated with gridded analyses of station-bas ed observed surface temperature, with correlations between 0.65 and 0.85. I t is argued that these correlations might suggest an upper limit to the mag nitudes of the pattern correlations that might be obtained by correlating o bserved surface temperature analyses with those from multiyear GCM simulati on runs made in the context of fingerprint climate change detection.