A reference level is defined as a level of known altitude at which temperature, pressure, and perhaps wind are specified as functions of time. This study is concerned with the optimum location of a reference level without wind information. Experiments were performed with the NCAR six-layer general circulation model to compare the usefulness of a surface reference level with an upper tropospheric reference level. We first performed a control integration using real atmospheric initial data. We then ran several comparison runs with initial conditions differing from those of the control run. The initial pressure distribution at the reference level was kept the same as the control run. The distribution of temperature pseudo-error employed in calculating the initial pressure distributions at the other levels was chosen to simulate possible error patterns in temperatures radiometrically derived from satellites. The initial conditions in all cases were in hydrostatic and geostrophic balance. Three data sets were used and the experiments were integrated to five or seven days. In addition, two horizontal distributions of initial temperature pseudo-error and two horizontal mesh lengths of the model were used for one of the three data sets. The results were examined using an rms difference of the distribution of pressure and meridional wind normalized (in the vertical) by the difference statistics derived from randomly chosen states of the model.
It appears that pseudo-error growth rates are nearly independent of the location of a reference level, but details of the pseudo-error patterns depend on the initial synoptic conditions. Pseudo-error growth rates differed depending on the manner in which the horizontal pseudo-error was initially distributed (but did not differ with the location of the reference level). The most significant change in the pseudo-error growth rates was observed when the mesh length was changed; halving the mesh length produced much faster growth rates, particularly in the lower layers.