WIND FLOW MODELING - Wind Resource Assessment: A Practical Guide to Developing a Wind Project - page 178

correlation is found. From this relationship, a linear equation y = mx + b could be

( )

derived and then applied to predict the speed at any other point in the area.
Statistical models are appealing because they are well grounded in measurement
and are fairly simple and transparent, unlike numerical wind flow models, which
often seem like “black boxes.” And they can work surprisingly well, particularly for
wind climates driven by synoptic-scale winds (i.e., where thermally driven mesoscale
circulations are largely absent), which tend to exhibit the clearest relationships
between wind speed and certain topographic indicators such as elevation and
exposure. Figure 13-2 illustrates the relationship observed at 74 towers in seven wind
resource areas between variations in wind speed and downwind exposure, defined as
the difference between the elevation of a given point and the average elevation out
to 3000 m in the downwind direction.
One of the potential limitations of statistical methods is that they can produce
larger-than-expected errors when making predictions outside the range of conditions
used to train the model. Suppose, for example, that one has data from three towers at
varying elevations along a ridgetop. Will the relationship between mean speed and,
say, elevation implied by these three towers hold when predicting the speed off the
ridgetop? Not necessarily, because the topographic influence on the wind flow may
be very different at the top compared to that at the slopes. In this respect, statistical
models can be less reliable than numerical wind flow models, which are designed to
produce plausible (if not accurate) results in a wide range of conditions.
Determining the accuracy of a statistical model is a particular challenge of this
approach. To derive an objective estimate of the uncertainty, it is necessary to divide
the data set into two groups: one to train the model and the other to validate the model.

4.0

3.0

2.0

1.0

0.0

-300 -200 -100 0 100 200 300

-1.0

-2.0

-3.0

-4.0

Difference in downwind exposure (m)

Figure 13-2. Data from pairs of 74 towers in seven wind resource areas indicate a significant
relationship between the differences in mean speed and downwind exposure. Such a statistical
relationship can be used to predict variations in wind speed across a project area. Source: AWS
Truepower.