In the NCEP/NCAR reanalysis data, the amplitude of the Northern Hemisphere storm track exhibits pronounced inter-decadal variations, with the decadal mean storm track intensity during the 1990's being nearly 40% stronger than that during the late 1960's and early 1970's. Since "both large and subtle changes in the observing system can create noise modes that can be as large as the sought after (climate) signal" (Kistler et al., 2001: The NCEP/NCAR 50-year reanalysis), it is not clear how real such inter-decadal storm track variations are.

In order to see whether the inter-decadal variations are real or are introduced into the reanalysis data by changes in the observational network, radiosonde observations along the storm track maxima are examined. Observations from all radiosonde stations in the Northern Hemisphere mid-latitudes which reported regularly since the 1960's are analyzed and compared to the NCEP/NCAR reanalysis data. Our results suggest that radiosonde observations are, by and large, consistent with the suggestions that substantial interdecadal variations occurred over a mid-latitude band running from the Eastern Pacific, across Central Canada, over the Atlantic, extending into Northern Europe and Russia. Unfortunately there are no radiosonde stations over the Western and Central Pacific, and very limited amount of observations over the Atlantic, thus variability over the oceanic regions can not be fully confirmed.

While the radiosonde observations are, to leading order, consistent with the results based on the NCEP/NCAR reanalysis data, direct analyses of radiosonde observations also indicate that there are substantial biases in the storm track amplitudes computed from the reanalysis data which are dependent on geographical locations as well as changing with time. Apparently, changes in the observing system had introduced distortions in the spatial pattern as well as temporal evolution of the storm track variations. It can be easily shown that storm track amplitudes, which are based on variances of observed fields, are not optimally estimated in the current data assimilation system. A question is thus raised: How can the data assimilation system be modified to give better estimates of variances and covariances? Efforts are now underway to generate a model-independent analysis of storm track variations based on radiosonde and aircraft observations as an alternative estimate of storm track variability.