Estimation of monthly and annual local wind speed anomalies through the use of high resolution numerical simulations

IInformation about the sign and magnitude of local monthly and annual near-surface wind speed anomalies has significant value to the wind power industry. In order to know the potential viability of a wind power project, a developer needs to know the departure of the mean wind speed from the long-term climatological value during the period for which onsite observations are available. This facilitates a more accurate assessment of the long-term power production potential of a site. For the same reason, wind anomaly information is also important in assessing the performance of operational wind power plants. Unfortunately, reliable long-term measurements are often only sporadically available or completely unavailable at many of the sites of interest. This challenge invites the use of a high-resolution numerical weather prediction model to simulate both the longer-term climatology and the departure from climatology on a monthly or annual basis.

For this study, a 10-year series of simulations were generated with the Mesoscale Atmospheric Simulation System (MASS) model for the period 1997 through 2006. The simulations were made on a grid covering the conterminous United States, for the purpose of generating a baseline dataset from which wind speed anomalies can be calculated. MASS has been widely used to generate wind resource maps for the wind energy industry. The simulations were made on a 60 km coarse grid and a 20 km nested grid, using NCEP/NCAR Global Reanalysis (NNGR) data for initial and lateral boundary conditions. Rawinsonde data from the Integrated Global Rawinsonde Archive (IGRA) were the only observations used in the objective analyses of the initial fields. To produce the most accurate possible trends and anomalies, the rawinsonde station list was limited to stations with high data recovery throughout the period; in addition, only fixed pressure levels for which observations were available throughout the period were assimilated. This is one key factor that separates our approach from NNGR and other reanalysis methods, which assimilate a changing set of observational data. The different physical parameterization schemes and numerical techniques used in MASS, as well as differences in the grid resolution, may also contribute to differences between the results from the MASS-based dataset and those from other (e.g. NNGR) datasets. No attempt was made in this investigation to explicitly determine the relative contributions of different factors that could cause differences between the datasets. For each month, two series of simulations were made. One series began with a cold start simulation on the 1st of the month and the other was initiated on either the 15th or 16th. Each cold start simulation was followed by a series of 12 hr simulations that assimilated IGRA rawinsonde data at each new 0000 or 1200 UTC initialization.

The results of these simulations were compared to surface, rawinsonde and three tall tower wind speed measurements in regions of both simple and complex terrain to assess the quality and utility of the simulated data for the calculation of wind speed anomalies (and other potential uses). The results generated by this simulation procedure were also compared to results obtained by extracting wind speed data for the same locations from the NNGR and NARR datasets. Each of the methods (MASS, NNGR and NARR) had significant wind speed biases for some stations. However, on the whole, the MASS anomalies were more accurate than NNGR and NARR anomalies by a substantial margin.

When compared to 10 m surface observations, the monthly and yearly anomalies produced by the MASS simulations had better agreement with anomalies calculated from measured data than those from the NNGR or NARR datasets. MASS also produced better anomalies for all three 60 m tower locations, despite the fact that MASS wind speed errors were higher for one of them. Results from one of the towers illustrate the advantage of higher resolution at sites with complex terrain – MASS (20 km) performed better than NARR (32 km) and far better than NNGR (190 km). But there are some sites (e.g. Beckley, WV) for which even 20 km resolution was not sufficient to produce a useful simulation of the measured anomalies. Further investigation will be necessary to determine what is required to adequately simulate the wind speed anomalies at these sites.

With only a few exceptions, the NARR dataset performed poorly in these comparisons. There appears to be a serious and heretofore unrecognized discontinuity in low-level wind speeds in early 2002 in the NARR dataset. This analysis indicates that the pre-2002 NARR wind speeds are systematically too low, but that the later wind speeds appear to be more reasonable. The pre-2002 NARR wind speed bias clearly had an impact on the overall performance of the NARR dataset for this application, but even in the later years the NARR results were often worse than NNGR in spite of their higher resolution.

The presentation will include comparisons of the simulated monthly and annual anomalies from all three sources (i.e. MASS, NNGR, NARR) with those obtained from tall tower measurements, rawinsonsde observations and observed 10 m wind data from several METAR stations. It will also include an assessment of the impact of complex terrain and variable surface properties on a model's ability to accurately simulate anomalies.