T42 Tapered Truncation


We transform the data to T63 resolution by adding a zero coefficient 
for the extra mode. For fields which have significant variance at 
the truncation limit, such as divergences, ringing effects are 
introduced. When products are taken with such fields, experience shows 
that computational noise is present for all waves beyond about T42. For 
instance, for the energy divergence, wavenumber spectra typically fall 
off slightly in power beyond about total wave number 15 to wave 42, but 
then exhibit an increase which peaks at T63. Accordingly, all fields are 
truncated with a tapered filter at about T42 resolution before generating
comparisons and analysis of variance. The filter is applied in spectral 
space; at T31 has a response function of 0.93, the half power point 
(amplitude response 0.71) is at T36, and at T42 the response function is 
0.37 (1/e). The results are smoothed to T31 or T21 resoluion for plotting
purposes. 

The tapered function (Fn_t) is given by:

Fn_t = Fn*exp(-(n*(n+1))/(no*(no+1))^r)

where Fn is the untapered field, n is the total wave number, no = n-2, and r is a power to which Fn is tapered (it is like a diffusion in time). In the case of a T42 taper, we use n = 42, no = 40, and r = 4.