Several different kinds of accurate potential vorticity (PV) inversion oper
ators, and the associated balanced models, are tested for the shallow water
equations on a hemisphere in an attempt to approach the ultimate limitatio
ns of the balance, inversion, and slow-manifold concepts. The accuracies ac
hieved are far higher than for standard balanced models accurate to one or
two orders in Rossby number R or Froude number F (where F = /u//c; /u/ = ho
w speed; and c = gravity wave speed). Numerical inversions, and correspondi
ng balanced-model integrations testing cumulative accuracy, are carried out
for cases that include substantial PV anomalies in the Tropics. The balanc
ed models in question are constructed so as to be exactly PV conserving and
to have unique velocity fields (implying, incidentally, that they cannot b
e Hamiltonian). Mean layer depths of 1 and 2 km are tested.
The results show that, in the cases studied, the dynamical information cont
ained in PV distributions is remarkably close to being complete even though
R = z at the equator and even though local maximum Froude numbers, F-max a
pproach unity in some cases. For example, in a 10-day integration of the ba
lanced model corresponding to one of the most accurate inversion operators,
"third-order normal mode inversion," the mean depth was 1 km, the minimum
depth less than 0.5 km, and F-max similar or equal to 0.7, hardly small in
comparison with unity. At the end of lo days of integration, the cumulative
rms error in the layer depth was less than 15 m, that is, less than 5% of
the typical rms spatial variation of 310 m. At the end of the first day of
integration the rms error was 5 m, that is, less than 2%. Here "error" refe
rs to a comparison between the results of a balanced integration and those
of a corresponding primitive equation integration initialized to have low g
ravity wave activity on day 0. Contour maps of the PV distributions remaine
d almost indistinguishable by eye over the 10-day period. This remarkable c
umulative accuracy, far beyond anything that could have been expected from
standard scale analysis, is probably related to the weakness of the spontan
eous-adjustment emission or "Lighthill radiation" studied in the companion
paper by Ford et al.