Introduction
In this post, I will briefly show comparisons between wind speed values derived from SAR measurements, and, well, in-situ measurements. If you are not familiar with Satellite Aperture RaDAR (SAR) scatterometer measurements, see the nice introduction below: the radar measures the backscatter received from small waves at the surface, and from this signal it is possible to infer a wind speed value.
For doing detailed analysis of the offshore wind conditions at a site, wind analysts prefer (by far) using high-quality in-situ LiDAR- or cup measurements. Yet they are expensive, often the campaigns are short, and they only provide information at one point. Therefore, the wind energy science community has long sought for alternatives, and in the past decade much progress has been done with using SAR data, in particular using the ESA Sentinel 1A and 1B datasets.
There are several teams very active in SAR research for Wind Energy, see in particular:
- DTU Wind Energy (Merete Badger and her colleagues) in Denmark,
- HZG in Germany (Johannes Schulz-Stellenfleth).
- IFREMER and CLS in France (Alexis Mouche).
As explained in the video, the wind speed derived from the SAR backscatter, typically at 10 m above the surface, are produced using an algorithm called an Geophysical Model Function (GMF). As explained here (and elsewhere too), these functions assume that the atmosphere is neutral. This implies that the SAR derived wind speeds are larger than the real wind speed in unstable conditions in unstable conditions, and, conversely, smaller than the real wind speed in stable conditions. See the illustration below, where the equation to the right is only valid in the surface layer.
Reference in-situ measurements
I have chosen, in this post, to work with the Hollandse Kust Zuid A (HKZA) floating LiDAR measurements. See its location below. If you would like know how accurate these measurements are, please see this post, and the documentation in this folder.
The measurements show that the wind profile depends on the atmospheric stability, see the mean wind speed profiles below. Here, for assessing the atmospheric stability, I used the 2m air temperature, as well as sea surface temperature from ERA5, and used the methodology explicated in (Peña, 2008). Note: the 4 mMSL measurements are not LiDAR measurements, they come from a sonic anemometer mounted on the buoy.
Results
For the purpose of this analysis, I have downloaded SAR-derived wind speeds from the DTU website satwinds.windenergy.dtu.dk, and selected only the Sentinel 1 data. I could also have used the data from ESA (data available at scihub.copernicus.eu/dhus/#/home. These are the kinds of products I have used so far – I do not master the art of the GMF yet.
Then, I extracted the SAR wind speeds at the HKZA locations. Interpolating directly at the LiDAR location, or taking the mean over a wider area (some km squares) did not change the results of the analysis. I then interpolated the 10-minute measurement time series at the SAR time stamps, and created two 10 mMSL wind speeds from the measured time series:
- a “true” 10 mMSL wind speed, interpolated from the measurements;
- a “neutral” 10 mMSL wind speed, which I computed using the stability information (see the paper 2008 quoted above explains how to derive not only the Monin-Obukhov length, but also the friction velocity and the roughness length).
And here are the results of the comparison. As expected, the SAR wind speeds match better the 10 m wind in average, than they match the real wind. Yet, for stable and neutral conditions, this is at the price of a slightly larger scatter, likely due to the way I have interpolated the wind speed down to 10m (I would need to check this in more details).
It is therefore always needed to account for stability conditions when using SAR data for wind resource purposes; see for instance this nice work carried out during the NORSEWInD project: (Badger et al., 2016). The below maps of 100 m wind speeds testify of the magnitude of the correction:
Please note that above the surface layer, as explained in the article by Sven-Erik Gryning (2007) quoted earlier, that the wind shear decreases. This means that, for the purpose of energy yield assessment, one needs to account for the actual energy content of the wind profile and not only the hub height wind speed (which in this case is misleading, and leads to some overprediction of the wind turbine power; see (Dörenkämper et al, 2012).
Conclusion and suggestions
In this short post, we have seen that thanks to the multiple, publicly available floating LiDAR datasets in the North Sea, that it is possible to carry out comparisons between SAR-derived wind speeds and in-situ measurements. For the case of the HKZA dataset, the GMF function seems work well in average. As to the scatter in the comparison, it is not only caused by the GMF, but also by the short sampling time (almost instanteneous), and thereby the natural micro- and mesoscale turbulence.
In places where there are no in-situ measurements, and in unstable conditions, the SAR measurements can be used for doing wind resource assessments, along side mesoscale models. For example in the Taiwan Strait (where the atmosphere is almost never stable): some coastal areas appear to be sheltered due to the nearby orography – an effect which is difficult to model accurately.
In places with recurrent occurences of stable conditions, care need to be taken to account for the biases and caveats mentioned earlier. This is the case in the Baltic Sea, in New England, in California, … places where warm air flows over cold water. See the example of the SST and air temperature time series from a buoy nearby San Francisco:
Finally, SAR measurements have long been used for characterizing wind farm wakes. See the state-of-the art PhD thesis of Tobias Ahsbahs. The Figure below shows some comparisons between SAR-derived wind speeds and dual doppler radar data at the Westermost Rough wind farm. I shall come back to this topic in a future post – including a discussion of the results from the HZG team during the WIPAFF project (see in particular this article by Djath et al.).
That was all for now! Comments/suggestions are more than welcome.
Rémi
