Data assimilation impact of in situ and remote sensing meteorological observations on wind power forecasts during the first Wind Forecast Improvement Project (WFIP)

Abstract

During the first Wind Forecast Improvement Project (WFIP), new meteorological observations were collected from a large suite of instruments, including wind velocities measured on networks of tall towers provided by wind industry partners, wind speeds measured by cup anemometers mounted on the nacelles of wind turbines, and wind profiles by networks of Doppler sodars and radar wind profilers. Previous data denial studies found a significant improvement of up to 6% root mean squared error (RMSE) reduction for short-term wind power forecasts due to the assimilation of all of these observations into the National Oceanic and Atmospheric Administration (NOAA) Rapid Refresh (RAP) forecast model using a 3D variational data assimilation scheme. As a follow-on study, we now investigate the impacts of assimilating into the RAP model either the additional remote sensing observations (sodars and radar wind profilers) alone or assimilating the industry-provided in situ observations (tall towers and nacelle anemometers) alone, in addition to routinely available standard meteorological data sets. The more numerous tall tower/nacelle observations provide a relatively large improvement through the first 3 to 4 hours of the forecasts, which diminishes to a negligible impact by forecast hour 6. In comparison, the sparser vertical profiling sodars/radars provide an initially smaller impact that decays at a much slower rate, with a positive impact present through the first 12 hours of the forecast. Large positive assimilation impacts for both sets of instruments are found during daytime hours, while small or even negative impacts are found during nighttime hours.

Impact of model improvements on 80 m wind speeds during the second Wind Forecast Improvement Project (WFIP2)

“…The improvements are analyzed in terms of the new physics (EXP vs. CNT runs) as well as horizontal grid spacing of the models (HRRRNEST vs. HRRR runs), first separately and then combining the impact of the two (HRRRNEST EXP vs. HRRR CNT). Finally, we evaluate the dependence of the improvements on the dominant meteorological phenomena of the area (Shaw et al, 2019), including cold pools (Whiteman et al, 2001;Zhong et al, 2001;McCaffrey et al, 2019), gap flows Mass, 2002, 2004), easterly flows (Neiman et al, 2018), mountain L. Bianco et al: Model improvements on 80 m wind speeds waves (Durran, 1990(Durran, , 2003, topographic wakes, and convective outflows (Mueller and Carbone, 1987).…”

Measuring the impact of additional instrumentation on the skill of numerical weather prediction models at forecasting wind ramp events during the first Wind Forecast Improvement Project (WFIP)

“…The percent improvement of the skill of RAP simulations at forecasting ramp events is shown when assimilating the combined WFIP remote sensors and in situ instruments (black bars), when assimilating the WFIP remote sensor observations (green bars), and when assimilating the WFIP in situ observations (magenta bars). Similar to what was found by Wilczak et al, the impact due to the assimilation of additional in situ observations has a significant initial impact that diminishes rapidly over several hours. In comparison, assimilation of the less numerous remote sensing observations has a smaller initial impact but remains positive for a longer forecast lead times.…”

“…A previous analysis of these model simulations found that the relative percent improvement in model forecast skill of the Experimental Run's RMSE and coefficient of determination of turbine hub‐height winds (at forecast hour 1) was 6%, for both the NSA and SSA, decreasing at longer forecast horizons . A later, second analysis investigated the impact of different components of the WFIP observing system, by separating the instrumentation into two groups, remote sensors (WPRs and sodars) and in situ (tall tower vector winds and nacelle anemometer wind speeds), and assimilating them each independently for a subset of two of the six data denial periods (13‐20 October 2011 and 7‐15 January 2012) . That analysis demonstrated that the large numbers of in situ observations had a significant initial impact that diminished rapidly after only several hours, while the less numerous remote sensing instruments had a smaller initial impact that improved the forecasts for a longer time, due to their observing a deeper layer of the atmosphere.…”

“…Both time series are then converted into power using a standard International Electrotechnical Commission (IEC) class 2 turbine power curve. 9,10 Model data that correspond to periods of missing data in observations are disregarded in the analysis as well as observational data corresponding to periods of missing data in the model.…”

Impact of model improvements on 80-m wind speeds during the second Wind Forecast Improvement Project (WFIP2)

“…The biggest improvements were from the boundary-layer, finite differencing, and the SGS drag, which improved the representation of turbulent mixing in stable boundary layers. The reader is referred to Olson et al (2019a;2019b) for more details on the differences between the CTL and EXP model configurations. For our analysis, in order to compare to the observations, the 80m wind field model output is horizontally bi-linearly interpolated to the 22 site locations using the 4 closest grid points.…”