Vertical Structure of Midlatitude Analysis and Forecast Errors

The dominant vertical structures for analysis and forecast errors are estimated in midlatitudes using a small ensemble of operational analyses. Errors for fixed locations in the central North Pacific and eastern North America are selected for comparing errors in regions with relatively low and high observation density, respectively. Results for these fixed locations are compared with results for zonal wavenumber 9, which provides a representative sample of baroclinic waves. This study focuses on deviations from the ensemble mean for meridional wind and temperature at 40°N; these quantities are chosen for simplicity and because they capture dynamical and thermodynamical aspects of midlatitude baroclinic waves. Results for the meridional wind show that analysis and forecast errors share the same dominant vertical structure as the analyses. This structure peaks near the tropopause and decays smoothly toward small values in the middle and lower troposphere. The dominant vertical structure for analysis errors exhibits upshear tilt and peaks just below the tropopause, suggesting an asymmetry in errors of the tropopause location, with a bias toward greater errors for downward tropopause displacements. The dominant vertical structure for temperature analysis errors is distinctly different from temperature analyses. Analysis errors have a sharp peak in the lower troposphere, with a secondary structure near the tropopause, whereas forecast errors and analyses show a dipole straddling the tropopause and smooth vertical structure, consistent with potential vorticity anomalies due to variance in tropopause position. Linear regression of forecast errors onto analysis errors for the western North Pacific is used to assess the nonseparable zonal-height structure of errors and their propagation. Analysis errors near the tropopause rapidly develop into a spreading wave packet, with a group speed that matches the mean zonal wind speed of 31 m s . A complementary calculation for the regression of 24-h forecast errors onto analysis errors shows that forecast errors originate from analysis errors in the middle and upper troposphere. These errors rapidly expand in the vertical to span the troposphere, with a peak at the tropopause.