THE CLIMATE ADJUSTMENT PROCESS - Wind Resource Assessment: A Practical Guide to Developing a Wind Project - page 170

However, the method only works well if there is at least 1 year of on-site observa-
tions. With less than a year of measurements, the seasonal dependence of the speed
distribution becomes a concern (as does the accuracy of the predicted mean speed).
The variance ratio method is another way of preserving the target site's variance.
The idea is that the slope and intercept of the linear equation y = mx + b are chosen

to reproduce the observed variance and mean

y σ y

σ x

x

σ y
σ x

y = x + − (12.7)

The mean values (overbars) are from the concurrent target and reference data sets.
The predicted long-term mean speed can be derived from this equation or from the
linear regression method. The latter is preferred since only linear regression considers
the correlation in determining the size of the MCP adjustment; otherwise, the variance
ratio method effectively assumes perfect correlation.
Although the variance ratio method will match the observed speed variance, there
is no guarantee that it will produce the correct speed frequency distribution in detail
since the relationship between the target and reference speeds may (and usually does)
vary with speed, direction, time of day, and other factors. Matrix ratio methods can
overcome this particular difficulty with appropriate binning; but even these methods
break down when the correlation within each speed and direction bin is not very strong,
that is, when there is not a one-to-one correspondence between a particular reference
speed and direction and the corresponding target speed and direction. This last short-
coming can be addressed by introducing random noise terms when reconstructing the
target data set, but this can only provide an approximate solution.

12.4.4 Direction and Other Parameters
It is usually not necessary to use MCP to predict the target directional distribution,
so long as there is at least a year of directional data from the target site. Where the
on-site observations are inadequate, the simplest solution is to find the mean offset
between the concurrent reference and target directions for each reference direction
sector and apply that offset to the full reference data record. This works well, as one
might expect, when the directions are highly correlated, but it can break down in less
ideal situations. A more general solution is to sample the directional distribution at
the target for each reference direction. This method can readily be combined with the
matrix sampling method described at the end of the previous section.
Other parameters, such as the observed temperature, can be adjusted to the historical
norm using a linear regression between the reference and target site in the same manner
as wind speed. If available, air pressure measurements can also be adjusted using this
method. The results can be used to adjust the estimated air density at the site.

12.4.5 Summary
While every method has its strong and weak points, it is generally best for the inex-
perienced analyst to stick with relatively simple, tried-and-true approaches. By this