From protectai.org

Amherst Island Wind Info
Point Source
From whywind.org

Back in my Basic Acoustics: Dispersion section, I discussed how the dimensions of both the source and the medium affect how sound dissipates. When a noise modeling program is used, it must make some assumptions about how the noise will disperse, and these assumptions have a direct impact on how the noise topo map is drawn. Modeling programs differ in their sophistication. Some allow point and line sources, some allow area sources. Some allow hemispherical, cylindrical and in-between noise dissipation. As far as I know, the program used for many projects, including Tug Hill and Wolfe Island - Cadna/A, from Scantek, Inc. - is not sophisticated enough to allow for all the complications of the real world, especially the complications of wind turbines.

One issue that seems important to me is the assumption of a point source. In the case of wind turbines typical values are 100 to 110 dBA, and are always assumed to originate at a point. This is in spite of studies that show that most of the noise is generated by the blades, radiating away from them at an angle both in a forward and backward direction. The picture below shows a picture of the noise, and Oerlemans and Bowdler provide more details.

If the blades were small or if you were far enough away this simplifying assumption might be reasonable. But when you are only 350 meters away and the blades have a diameter of almost 100 meters it is not reasonable. A series of pictures might make the problem clearer. Below I show the top view of a turbine with a small circular hub in the middle. The normal simplifying assumption is that all the noise radiates from this small hub. In reality, as shown in the picture above, most of the noise radiates from the blades. Further, the noise departs at an angle to the blades' disk, as shown in Bowdler and the pictures below. If the noise was perfectly contained within the dashed lines it would act like a laser beam and wouldn't dissipate very quickly at all. Obviously there is some spreading over 3 dimensions as shown, but it no longer spreads evenly as though from an idealized point.

Just as an exercise, I've drawn the spread in a 25% ratio below. I have no basis to choose 25%, and obviously the spreading goes beyond that, but I'd bet this isn't totally divorced from reality. This assumed spread would lead to a prediction that the noise directly under the turbine would be less than out in front of it, which does agree with observations.

If we then extend the lines back from the disk they would establish a new virtual source point 200 meters behind it. Since the sound spreads in both directions there would actually be two virtual points, on each side of the disk, but I'm only showing one.

If we then run all the models using the new virtual point as the source, we would get a different set of noise levels, one that arguably is closer to reality than the current methodology, as shown in the chart below.

Now, obviously, this is all a mental exercise, and the reality of my specific assumptions is debatable. What is not debatable is that the simplified models that are used to place these large and permanent structures are not accurate, and experience has shown that people are suffering as a result.

As a further complication, the large area where the noise originates may well lead to additional turbulance and shadowing, which in turn would almost certainly lead to more noise and amplitude modulation (the thumping and swishing that neighbors complain about).