An intercomparison of the reanalysis datasets of NCEP and ECMWF is performed with respect to the response of the axial angular momentum M to the torques. While both sets satisfy the budget equations of M reasonably well (except for the time mean), this is not the case with respect to the budget equations for the difference of both sets, where the analysis data explain only a small fraction of the difference of the angular momenta in terms of the difference of the torques. It is hypothesized

more »

... the larger fraction of the difference is a manifestation of analysis error. The autocorrelation functions of the differences of the mountain and friction torques between both sets exhibit a long memory, which reflects errors in the low-frequency components of the datasets. Probability distributions of M are considered as well. It is shown that the mean torque for a given M has to be positive (negative) for negative (positive) deviations of M. It is found that the NCEP torques, as analyzed, satisfy this basic requirement. The distribution of the difference of the angular momenta cannot be explained on the basis of the corresponding difference of the torques.