SAR-derived wind speeds and wind farm wakes

Rémi,

After having looked at using SAR-derived wind speeds (10 mASL neutral wind) for estimating the freestream wind resource in a previous post, it is now time to look into using the same SAR data but for wind farm wakes analysis. A good site for this is the Belgian wind farm cluster, and in particular the Northwind wind farm. This plant has been operating since 2014, thereby all of the Sentinel 1A and 1B can be used, and on top of this there were two floating LiDARs deployed for 6 months to one year nearby in Dutch Waters. See the two maps below.

Overview of the offshore wind farm areas in the Southern North Sea.
The Belgian offshore wind farm cluster, as per October 2020 (it is almost complete, only Seastar is still under construction).

After having compared the SAR-derived wind speeds with in-situ measurements, it became clear that SAR-derived wind speeds are closer to real wind speeds (yet, slighty larger) during unstable conditions, than they are to real wind speeds in stable conditions (where the clearly underestimate the 10 m wind). For wake analysis, it becomes a little trickier, due to the fact that:

  • the wake deficit, for a given wind speed at hub height, is itself be dependent on the atmospheric stability;
  • the atmospheric stability may vary with height.

Given all of the above, it is necessary to not rely on SAR data only, but to complement them with additional measurement- and model data. In the following, I will be using:

  • floating LiDAR measurements from BWFZ01 qand BFWZ02 (just a bit, as secondary measurements);
  • measurements from the Europlatform KNMI weather station, located 37 km to the North-East of BWFZ01;
  • ERA5 model data (single levels, and pressure levels);
  • Dutch Offshore Wind Atlas (DOWA) data;
  • wind farm area GIS data from EMODnet;
  • and, of course, SAR data from satwinds.windenergy.dtu.dk.

Diagnostic plots

For the purpose of the analysis, I have select all of the Sentinel 1A/1B data (I do not feel confident with the other SAR datasets, yet) which cover the area surrouding the Northwind wind farm. I have then only selected wind directions in a small wind directional bin (225°N +/- 15°).

Then, for each SAR image, I have produced plots like the one below. If you can’t see the image correctly, I have saved all of the figures for this post in this online repository.

Example of diagnostic plot, to better understand the SAR image. See online repository for better quality image.

In this example, we have a well-behaved wind farm wake in near-neutral conditions. Both surface model data and altitude model data agree: the air is colder than the water, the ABL top is high (h = ~800m according to ERA5), wind speeds are over rated power (i.e. the thrust coefficient is small), and therefore the wake recovery is relatively quick.

Another example if shown below, this time the wind speed is smaller, and thereby the wake recovery takes a little longer.

Another example of wind farm wake in unstable/near-neutral conditions. See here for better quality image.

In stable conditions at the surface, well… , ahem…, happens what happens to all SAR-derived sea roughness data in stable conditions, that is: the radar picks up small scale turbulence (here the wake turbulence), and not the mean flow (too small friction). It leads to the kind of plots displayed below, with some clearly physical behavior (larger wind speeds in the wake than upstream)

An example of SAR artefact in stable conditions: larger wind speeds in the wake than upstream. See here for better quality image.

Atmospheric stability, at the surface, and at heights

As highlighted by the findings of the recent WIPAFF project (where a well-instrumented research airplane was flown into wind farm wakes in the German Bight), it is important not only to consider the stability in the surface layer (typically in inferred by deriving the Bulk Richardson number using a COARE-type of method), but also to know what the stability looks like at higher heights, close or near- the top of the wind farm.

See the example below: while at the surface there is almost no difference between the air and the sea temperature, at relatively moderate heights (100 to 300 m) is the atmosphere stable.

An example of stable conditions that would be hard to guess using only surface information. See here for better quality image.
Another example of stable conditions that would be hard to guess using only surface information. See here for better quality image.

Conclusion(s)

At this stage, it appears important to flag that while, when looking at large amount of data (like, a year of floating LiDAR), it is reasonable to use surface-layer based algorithms to retrieve the atmospheric stability; when analysing smaller sets of the data there is a risk to misclasify the atmospheric stability, and thereby the results of the analysis.

Further work involve the comparison of stability parameters derived from the different methods mentioned above, and the consequent validation of wind farm wake model recovery.

Comments/suggestions are welcome,

Rémi

Published by Rémi