The Radar@Sea project is motivated by previous investigation works, along with shared experience of the operators of large offshore wind farms in Denmark (ie. DONG Energy and Vattenfall. The core idea of the project consists in installing and operating a Local Area Weather Radar (LAWR) offshore at a large wind farm (Horns Rev, see pic below), in order to research and demonstrate the resulting benefits for short-term forecasting, better control of power output fluctuations, and potentially better maintenance planning. Analysis of data acquired thanks to LAWR devices in the past and in other context, as well as investigation on its ability to track rain and local front events, with potential derivation of local wind fields, has revealed the potential interest of such devices as a new operational tool for wind power production management. The possibility of having images covering a radius of 60kms around the LAWR, with high spatial resolution, and updated every 1 to 5 minutes, makes it an ideal tool for having a real-time detailed knowledge of the meteorological conditions that may influence the power output at the wind farm in the following few minutes to few hours.
One of this LAWR devices was therefore installed at Horns Rev (on the DONG accommodation platform), with data collection starting early 2010. Since then the data collected (along with meteorological and production data at the level of the wind farm) is being used for demonstrating the improved predictability and controllability of the wind farm power output resulting from utilization of such device.
Two important meteorological aspects are to be considered. First of all, the role of rain and local fronts on the behavior of wind power fluctuations, as it has recently been observed and demonstrated that the rain intensity in the neighborhood of Horns Rev indeed had an influence on the potential magnitude of power fluctuations (at a few minutes to few hours temporal scale). In parallel, the possibility of having information about rain events and local wind fields (derived from radar images and Doppler radar data) may allow to develop and implement local forecasting models that would describe when and how fronts will hit the wind farm. It is a well-known issue that the current forecast methodologies based on time-series analysis and/or meteorological forecasts very often fail in predicting when a sudden rise or decrease in wind will affect offshore wind farms. This issue is often referred to as phase errors in the wind power forecasts. It is especially true for the case of the Horns Rev wind farm, being the first location to be hit in the typical westerly flows, without possibility (so far) to have upwind information on meteorological conditions. With the radius of visibility of the LAWR, predictability may be improved up to 2-3 hours. Improved predictability, at a high temporal and spatial resolution, would translate to a significant easing of power output control and in a more general manner of the power production management. A side benefit is that such forecasts would also ease planning of maintenance operation on site, thus contributing to the safety aspects for the human workforce involved.
A sample video explaining the project’s rationale
Here is a small sample video where one can compare the evolution of wind power generation at Horns Rev with the radar images collected over a 2-day period… The wind farm is represented by the small white square near the middle of the radar images, while the “white stain” on the right is the coast of Western Jutland in Denmark, where the Horns Rev wind farm is located.
The link to the video as posted on YouTube: http://www.youtube.com/watch?v=YShQDCdVykM
Project status at the end of 2011
The work on the evaluation of the different WRF model parameterisations was finalized, while more emphasis was then given to the assimilation of nacelle winds. The yaw angle, as a proxy of wind direction, was also assimilated. The work included the assimilation of these wind farm data into the 4-Dimensional Data Assimilation system of WRF. For that purpose, optimal pre-processing algorithms for the nacelle wind speeds were analysed as to which would yield best forecasting results. It could be shown, that due to the assimilation of upper air observations, the impact of the assimilatio lasted beyond 24 hours. The impact of the assimilation of wind farm data lasted up to 6 hours. This is also the time window where the pre-processing algorithms lead to different results.
As additional input to the project, a collection of new weather radar images was provided by the Danish Meteorological Institute (DMI). These images originate from the C-Band weather radar at Rømø (west coast of Jutland) and covers a circular area with a radius of 240km. They bring new opportunities as they enable the observation of rainfall systems at larger scales and also further ranges than the DHI radar. Based on all various datasets gathered through the project, the rationale of the research is developing into a hierarchical approach, where meteorological forecasts would serve as the first type of information to be looked at, then the DMI radar images (since having broader coverage), and then the DHI radar images (since having finer spatio-temporal resolution).
The data were processed and merged into one dataset covering 2010 and half of 2011. At the Horns Rev 1 (HR1) wind farm, the data provided by Vattenfall mainly include wind speed and power, since severe issues were observed for the case of wind direction. The clutter detection methods were improved and applied to both DHI and DMI images. The new methods allow to identify image pixels persistently affected by erroneous measurements. New values are then assigned to these pixels thanks to adaptive smoothing techniques. Based on these data, a statistical analysis was performed in order to illustrate the link between the presence of rain in the proximity of Horns Rev 1 and wind speed/power fluctuations of large amplitude. In particular, the analysis shows a higher wind speed variability (as observed at HR1) when small convective rain cells are approaching and going through HR1. This is particularly relevant since it confirms the expert observations of meteorologists. For producing this analysis, it was proposed to characterize the weather radar images in a statistical fashion. Different robust statistics were estimated in order to identify the scale of a given rainfall system, its spatial feature (2D correlogram), its reflectivity distribution (median, 99% quantile, etc.) and its motion (speed and direction). The estimated statistics will be used for the creation of a catalogue of events.