An old-school approach to taking care of windmill echoes

Confronted with the challenge of having to correct the pixels contaminated by direct and multipath echoes from wind farms, we elected to try an old-school approach: identify and replace the pixels by interpolation using “clean” neighbors. This reduced the problem to the identification of contaminated echoes.

Wind mills have one nice property: although they do move, their position and characteristics (height, size of the blades, etc.) remain fixed, and are generally well documented in publically-available documents. Given radar characateristics, one can then use that information to determine the location of wind mills on the radar data. And since multipath effects show up only on azimuths where wind mills are found, their potential location can also be determined in advance. Hence, the location of all potentially contaminated pixels and of definitely clean pixels is known and can serve as the basis for the correction approach.

Given a database of wind mill characteristics, our algorithm determines the list of pixels that may need to be corrected, and of those that can be used to interpolate echo information from. Echoes over the wind mills are always declared contaminated. Multipath echoes being generally weak, they only affect data in the absence of significant weather echoes in their neighborhood. Cells declared “potentially contaminated” are then replaced by interpolation from clean nearby cells.