The next problem is: how to construct the 
-set, that is, the set of
climatic parameters (climates) 
prescribing a climate scenario?
Let us assume that the local climate is not changing (at least during the last
two decades covered our observation interval from 1970 up to 1984 during 15
years). Thus, despite annual variation of crop yield as a consequence of
weather variation, the local climate is fixed, and there is only one point
(corresponding to this climate) in -space. Suppose we have the
model of this climate, the ``statistical weather generator'', which is a
stochastic parametric process. Its parameters are, in turn,
characteristics of the mesoclimate at the given site. One realisation of
this stochastic process (a trajectory) is considered as the weather at a given
site and in a given year.
The simplest hypothesis is the following:
Let us take 15 (for each vegetation period) time-series 
,
for the temperature, where 
is any point of time
within the vegetation period.
is the mean temperature in the course of a vegetation period for this site,
that is, a characteristic trait of the local climate. The average current
variance of the temperature is also calculated as
Using these values we calculate (by formula (5)) a statistical
consequence describing some standard daily temperature dynamics, which is
typical for the given site. As generalised characteristics for the local
climate we use the temporal averaging of the functions 
and
over the interval 
which is the vegetation
period, so that
are the ``vegetation'' temperature and variance. If the first value is the
mean temperature of the vegetation period, then the latter is a variance of a
seasonal temperature during a vegetation period.
In the same way, the mean precipitation for the vegetation period and its
variance are calculated. The ``weather generator'' is constructed in such a way
that the generated stochastic series do not differ statistically from the real
climatic time-series, that is, the statistical characteristics (means and
variances) of the generated series are equal to those of the local climate.
Thus, the next hypothesis may be formulated:
At the first stage we assume that as a result of climate change the mean
temperature 
and its variance 
are changed. For instance,
and 
for the Kursk region. The values of and
make up the set of climatic parameters (climates), that is, the
-set.
In accordance with different climate models (GCMs, paleoclimatic and
extrapolation models), the increase of mean summer temperature in the Kursk
region would be 
- 
C for the doubling of CO
scenario. The methods of ``optimal filtration'' (in case if some marginal
predictions are rejected) give us the interval equal to
- 
C. Concerning the change of variance, there are
only some qualitative estimations available. For instance, the estimations of
variance by GCMs show that, in general, the variance of summer temperatures
decreases [10]. It seems that this statement could be valid for the
polar and tropic regions, but it is doubtful for such a temperate region as the
Central Russia. On the other hand, statistical extrapolation of the observed
data shows the increase of variance. We prefer the latter.
In our calculation we shall use an old empirical rule of statistics
[11], which suggests that
.
Thereby, we obtain the following biased estimation:
C. Assuming that the variance either does not
change or rises, we obtain that the interval of possible values for
is equal to 
- 
C.