Page 152 - Fluid Power Engineering

W ind Resource Assessment 127

may be 10 or more turbine locations for each met-tower location. This
step is, therefore, used to estimate wind speed at locations where mea-
surements were not performed.
The models described in “Resource Estimation Models” are used
to extrapolate wind resources from measured locations to planned
wind turbine locations. WAsP is the most popular model that is used
for this purpose. Micrositing CFD models may be used when assump-
tions pertaining to WAsP do not apply.

Hindcasting/MCP of Measured Data
Wind measurement for few years provides a window into the wind
conditions at a site. However, this is only a short snapshot of the wind
conditions. It is not the wind condition over the life of a wind project
(typically, 20 years). Wind speed like any other weather parameter
has cycles. Few years of measurement must, therefore, be placed in
context of longer-term wind pattern. For instance, measurement may
be in a low-wind speed period and, if this measurement were used to
compute wind energy production over a period of 20 years, then the
prediction would underestimate the energy production.
Hindcasting is a process of generating wind conditions that pre-
dict 20 or more years of wind data. It uses measurement data and
long-term reference wind data. The process of hindcasting involves
a technique called measure-correlate-predict (MCP), which has the
following steps:

1. Inputs: Onsite wind speed measurement for a year or more,
and long-term reference data for 20 or more years. The long-
term reference data from neighboring airport or NCAR re-
analysis data. Use as many long-term datasets as available.
2. Correlate measurement data with long-term reference data-
sets for concurrent time period. If correlation is acceptable,
then choose these reference datasets for the next step.
3. Estimate wind speed for the historical period, which covers
the duration of reference time series. This is the hindcasting
step.
4. Convert the hindcast into a forecast, if necessary, or use the
hindcast for energy computations.

In most practical applications, the correlate and predict steps are per-
formed iteratively with several sources of long-term reference data.
Experience has shown that long-term reference data is a key determi-
nant to reduced uncertainty in prediction. Siddabathini and Sorensen 7
show that uncertainty can be reduced by proper choice of consistent
data within the long-term reference data; in the example presented