Robust Tropospheric Warming Revealed by Iteratively Homogenized Radiosonde Data

Results are presented from a new homogenization of data since 1959 from 527 radiosonde stations. This effort differs from previous ones by employing an approach specifically designed to minimize systematic errors in adjustment, by including wind shear as well as temperature, by seasonally resolving adjustments, and by using neither satellite information nor station metadata. Relatively few artifacts were detected in wind shear, and associated adjustments were indistinguishable from random adjustments. Temperature artifacts were detected most often in the late 1980s-early 1990s. Uncertainty was characterized from variations within an ensemble of homogenizations and used to test goodness of fit with satellite data using reduced chi squared. The meridional variations of zonally aggregated temperature trend since 1979 moved significantly closer to those of the Microwave Sounding Unit (MSU) after data adjustment. Adjusted data from 5 degrees S to 20 degrees N continue to show relatively weak warming, but the error is quite large, and the trends are inconsistent with those at other latitudes. Overall, the adjusted trends are close to those of MSU for the temperature of the lower troposphere (TLT). For channel 2, they are consistent with two analyses (Remote Sensing Systems, p = 0.54, and the University of Maryland, p = 0.32) showing the strongest warming but not with the University of Alabama dataset (p = 0.0001). The troposphere warms at least as strongly as the surface, with local warming maxima at 300 hPa in the tropics and in the boundary layer of the extratropical Northern Hemisphere (ENH). Tropospheric warming since 1959 is almost hemispherically symmetric, but since 1979 it is significantly stronger in ENH and weaker in the extratropical Southern Hemisphere (ESH). ESH trends are relatively uncertain because of poor sampling. Stratospheric cooling also remains stronger than indicated by MSU and likely excessive. While this effort appears not to have detected all artifacts, trends appear to be systematically improved. Stronger warming is shown in the Northern Hemisphere where sampling is best. Several suggestions are made for future attempts. These results support the hypothesis that trends in wind data are relatively uncorrupted by artifacts compared to temperature, and should be exploited in future homogenization efforts.