Twice daily (00Z and 12Z) data are time-filtered using a centered, weighted, "running mean" type of filter. Specifically,

FF_i = sum_j=0,n of [w_j * UF_(i-(n/2)+j-1) ] (function TIMFILc for the CCM processor)

is applied where FF_i is the filtered data at some point i in space, w_j is the jth weight, UF is the unfiltered data, n represents the number of weights, and (n/2) is the nearest integer value of the number of weights divided by 2. The weights used are:

w = < -0.003, -0.033, -0.037, -0.013, -0.132, -0.188, 0.172, 0.468, 0.172, -0.188, -0.132, -0.013, -0.037, -0.033, -0.003 >

Application of the filter results in a loss of the total number of time series points (at each end), so a January time filter would be applied to an original, unfiltered time series extending from 12Z on December 28 of the prior year to 00Z on February 4 of the current year. That gives exactly 7 data points on either side of the time series, allowing for the resulting time-filtered series to be (n-1) time samples shorter than the original, extended, unfiltered series. That is, the filtered series would have exactly 31*2 time samples for January.

This filter is a band-pass filter which serves to retain fluctuations with periods between 2 and 8 days. A time average is applied to the already time-filtered data to obtain a monthly mean of the band-passed series.

For an example of a specific application of these data, please refer to :

Trenberth, K.E., 1991, Storm Tracks in the Southern Hemisphere, Journal of the Atmospheric Sciences, V.28, pp. 2159-2178.