R. Huth et al., Time structure of observed, GCM-simulated, downscaled, and stochastically generated daily temperature series, J CLIMATE, 14(20), 2001, pp. 4047-4061
The time structure of simulated daily maximum and minimum temperature serie
s, produced by several different methods, is compared with observations at
six stations in central Europe. The methods are statistical downscaling, st
ochastic weather generator, and general circulation models (GCMs). Outputs
from control runs of two GCMs are examined: ECHAM3 and CCCM2. Four time ser
ies are constructed by statistical downscaling using multiple linear regres
sion of 500-hPa heights and 1000-/500-hPa thickness: (i) from observations
with variance reproduced by the inflation technique, (ii) from observations
with variance reproduced by adding a white noise process, and (iii) from t
he two GCMs. Two runs of the weather generator were performed, one consider
ing and one neglecting the annual cycle of lag-0 and lag-1 correlations amo
ng daily weather characteristics. Standard deviation and skewness of day-to
-day temperature changes and lag-1 autocorrelations are examined. For heat
and cold waves, the occurrence frequency, mean duration, peak temperature,
and mean position within the year are studied.
Possible causes of discrepancies between the simulated and observed time se
ries are discussed and identified. They are shown to stem, among others, fr
om (i) the absence of physics in downscaled and stochastically generated se
ries, (ii) inadequacies of treatment of physical processes in GCMs, (iii) a
ssumptions of linearity in downscaling equations, and (iv) properties of th
e underlying statistical model of the weather generator. In downscaling, va
riance inflation is preferable to the white noise addition in most aspects
as the latter results in highly overestimated day-to-day variability. The i
nclusion of the annual cycle of correlations into the weather generator doe
s not lead to an overall improvement of the temperature series produced. No
ne of the methods appears to be able to reproduce all the characteristics o
f time structure correctly.