Comparison of Tropospheric Temperatures Derived from the NCEP/NCAR Reanalysis, NCEP Operational Analysis, and the Microwave Sounding Unit

Abstract

The Climate Prediction Center has used atmospheric temperatures for data analysis from the National Centers for Environmental Prediction (NCEP) model since 1979. Unfortunately, model changes have adversely affected the stability of the climatologic fields, introducing time-varying biases in the anomaly patterns of the Climate Diagnostic Data Base (CDDB). Fortunately, NCEP has addressed this issue by rerunning a state-of-the-art model using fixed assimilation, parameterization, and physics in order to derive a true climatology and anomalies. The authors compare the previous CDDB temperatures with those derived from the stable reanalysis. Results show major improvements for climate diagnostics and monitoring. Also compared are the reanalysis temperatures with brightness temperature Tb observed by the Microwave Sounding Units (MSU), flown aboard the National Oceanic and Atmospheric Administration (NOAA) series of polar-orbiting satellites (TIROS-N to NOAA-14). This MSU dataset has a precision of about 0.02°C globally, and it is available from December 1978. Therefore, the 17 levels of the reanalysis level temperature were weighted to simulate the MSUTb in order to measure its precision over the 17-yr record. Global time series of the spatial correlations between full fields approach 1.0 throughout the entire record, whereas correlations for the anomaly fields can drop below 0.8 during the high sun season in the Northern Hemisphere. In 1994 the correlations drop below 0.65, which is the largest difference between the two datasets. An EOF on the global Tb differences from both datasets identified a relative drift beginning in 1991. The maximum loading was in the tropical Pacific, although it also extended over the tropical Indian Ocean and the Asian landmass. Results indicate that the reanalysis anomalies are getting progressively colder, relative to the MSU, during the early 1990s. The authors associate this drift with the changes in satellite retrievals and a reduction of Soviet Union data during its breakup. Additional sources of bias may be associated with aerosol contamination after the Mt. Pinotuba eruption and/or drift in the NOAA-11 sensor. Although there is a relative offset in the anomalies, the reanalysis temperatures have a better correspondence with the radiosonde network after 1990. Therefore it appears that the bias is associated with an improvement in the reanalysis input data during the last several years. Since changes in the datasets assimilated into the model can introduce a slight bias, new procedures should be developed to minimizes these effects in any future reanalysis. Finally, although the reanalysis has a slight drift in the later years, the comparison with the MSU spatial anomalies generally showed excellent results. The reanalysis represents a substantial improvement over the CDDB for monitoring climate variability.