An Evaluation of Monthly Mean MSU and ECMWF Global Atmospheric Temperatures for Monitoring Climate

Abstract

Monthly mean brightness temperature anomalies derived from channel 2 of the microwave sounding units (M5Us) on board NOAA satellites over the past decade are examined and compared with both weighted and pressure-level ECMWF monthly mean temperatures for the 96 months of 1982?89. Very good agreement between the MSU and channel 2 weighted ECMWF anomalies is found over most of the globe with correlation coefficients over 0.9, but the agreement falls off over the tropics, the South Atlantic, and high latitudes of the Southern Hemisphere. The ECMWF analyses agree best with the MSU data in regions of good radiosonde coverage, while lower correlations are found in regions where the analyses depend more heavily on satellite data. Systematic errors introduced into the analyses by the retrieval techniques applied to the radiance data largely explain this apparent contradiction. Additionally, changes to the analysis-forecast system at ECMWF over the decade appear as apparent changes in climate, and these discontinuities most strongly affect the tropics and are evident in regions of fewer observations. To the extent that the weighted ECMWF data data with the MSU brightness temperatures the vertical dependence of the MSU data can be examined. Correlations of the MSU data with ECMWF temperature anomalies at individual pressure levels are highest at 300 mb over the globe, a level apparently less affected by the frequent changes and improvements at ECMWF. Over regions of good data coverage, such as the Northern Hemisphere landmasses and Australia. the MSU anomalies correlate very highly with all levels of the troposphere up to 200 mb. Thus, the MSUs appear to be an extremely useful tool for measuring global tropospheric temperature fluctuations on a monthly and longer time scale. Problems in the ECMWF temperature record since 1982 am examined in detail for the tropics. In September 1982 the introduction of diabatic nonlinear normal-mode initialization resulted in significant temperature increases in the tropical middle troposphere, especially at 500 mb. In May 1985 tropical temperatures at 700 mb (950 inb) increased (decreased) after the implementation of the T106 spectral model with major accompanying changes to physical parameterizations. Tropical temperatures near the tropopause decreased substantially after the May 1986 enhancement of the vertical revolution of the model from 16 to 19 levels with 3 new stratosphere levels. Problems at 1000 mb are present throughout the 96-month study period and are directly related to the manner in which analyzed temperatures are obtained at ECMWF. Comparisons with temperatures obtained from radiosonde stations in the tropics show that the ECMWF analyses have clearly improved with time, especially after May 1985 after which the 1000?200-mb temperatures show much greater coherence. These results show the importance of realizing the inherent problems with operationally based gridded datasets, and they strongly support the need for reanalysis of all data using a state-of-the-art four-dimensional data assimilation system.