Because of the relatively wide transmitter antenna beamwidths at medium fre
quency (MF), measurements are expected to represent an integration over spa
ce as well as time. For historical (and practical) reasons, the normal reco
rd lengths at Saskatoon have been similar to5 min (4.5-4.8 min). This is ap
proximately the Brunt-Vaisala (B-V) period, equivalent to 1 cycle of the sh
ortest expected gravity wave (GW) period, so some GW energy may be lost by
averaging. In addition the GW oscillations may add "noise" to the measured
wind or even prevent a successful analysis, which assumes stationary statis
tics. On the other hand, if the spatial averaging has removed those waves c
ontributing to high-frequency fluctuations in wind, then reducing the recor
d length increases the noise level rather than giving us more information r
egarding these GW. This paper describes two multiple record length experime
nts performed with the Saskatoon MF radar. Conclusions are that in the fall
, below say 80 km, there is an inverse relationship between wind variance a
nd record length, as would be expected from elementary statistics if the va
riance were strictly analysis noise. Presumably, this is because GW amplitu
des are negligible here. In the summer, short period GW contribute signific
antly to the variance from 60 to 100 km. If we assume a form for the (unkno
wn) GW spectrum, e.g. -5/3 log-log slope (VanZandt, 1982), then measurement
s with different record lengths can be used to separate the analysis noise
and GW parts of this variance. Finally, the small amount of 90 s record win
d data examined does not reveal obvious coherent features at short periods
which are not also seen in 5 min records. On the other hand, their spectra
do enhance some features which are less obvious in those of 5 min data, and
interestingly, no apparent change is seen in spectral slope at the B-V fre
quency. (C) 2001 Elsevier Science Ltd. All rights reserved.