Page 156 - Fluid Power Engineering

W ind Resource Assessment 131

Correlation
Correlation of Daily
Time Series 1 Time Series 2 of Raw Data Average Data
10-min met-tower data 10-min met-tower >0.65 >0.75
(site A) data (site B)
10-min met-tower data Hourly airport data >0.6 >0.75
10-min met-tower data 6-h reanalysis data >0.55 >0.75
TABLE 7-5 Guideline for Determining if Two Wind Speed Time Series Share the
Same Wind Climate

measurement intervals, 10 min versus 6 h. Instead, correlation of daily
averages of the two time series may be an appropriate measure of pre-
dictability.
The correlation results for the Valentine example using Wind-
PRO are in Table 7-6. Correlations are computed for 12 sectors. Sec-
ond column contains the number of data points in each sector. Third
and fourth column contain statistics of measured wind speed. Fifth
and sixth column contain statistics of wind speed from NCAR data.
Seventh and eighth column contain statistics of wind speed ratio
(onsite/long-term reference). Ninth and tenth column contain statis-
tics of difference between wind direction between measured and ref-
erence data. Final column contains correlations of the raw data. Since
onsite measurement data is hourly and NCAR is every 6 h, raw data
points with the shortest time difference are chosen. That is, every
sixth point of Valentine data is matched with NCAR data; for in-
stance, on 3/21/1995 1:00 AM Valentine data is paired with NCAR
data for the same hour, Valentine data from 2:00 AM to 6:00 AM is ig-
nored, 7:00 AM Valentine data is paired with NCAR data for the same
hour, and so on. Correlations for different aggregates are presented in
Table 7-7. As expected, the aggregate correlations are much higher
than the raw correlation. Monthly average wind speed data is plotted
in Fig. 7-8.
Other than correlation, other simple tests can determine if the
long-term reference data is suitable for prediction:
Comparison of diurnal and monthly pattern. If the measured data

and reference data do not have similar diurnal and monthly
patterns, then the reference dataset may not be a good choice
Comparison of wind rose. Sectors with the highest wind speed

andhighestfrequencymaynotbethesamebetweenmeasured
and reference data, but should be within +/− 1 sector. If there
are larger differences, and the differences cannot be explained
by examining the contour and roughness, then the reference
data may not be a good choice. (See Fig. 7-9a to c.)