Thorn scrub country in the far south of Texas, mesquite and cattle, turbines standing over all of it.
On top of some of the nacelles sits a gray box with a perforated face.
It is a speaker, and it is shouting into the dark at frequencies no person will ever hear.
Below, crews walk transects at dawn looking for what fell in the night.
Half as many bats lay under the noisy machines.
Not every species.
Why noise works on an animal that navigates by noise
A bat builds its picture of the world from echoes of its own calls, which means the picture is only as good as the returning sound.
Flood the same band with loud broadband noise and the returning echo is buried, the way a voice disappears beside a jackhammer.
The animal does not die of the noise and is not harmed by it. It simply loses its map and leaves.
That distinction matters for approval, because a device that drives an animal away is regulated very differently from one that injures it.
The emitters used here cover 20 to 50 kilohertz at about 120 decibels measured a yard from the source.
That band is chosen because it overlaps the calls of most North American bats, which is also why it cannot be tuned narrowly to one species.
The device does not repel anything. It blinds it.
How the trial was actually built
This is the detail most coverage flattens, and it changes what the result means.
Sixteen turbines took part at a facility of 255, in a county on the Lower Rio Grande.
Eight ran as treatment and eight as control on any given night, and the assignment swapped between nights.
That design is what makes the comparison hold, because weather, moonlight and insect activity land on both groups equally.
Searching ran from the last day of July to the end of October, in two consecutive years, which covers the autumn migration when losses peak.
Never 16 at once. Always 8 against 8.
The species breakdown that carries the finding
Across all bats, the reduction came to 50 percent, with a confidence range of roughly 38 to 61.
Split by species the picture changes. The Brazilian free tailed bat, which dominates the local sky in enormous numbers, fell by about 55 percent.
The hoary bat fell by 78 percent, and that is the number worth carrying away.
The hoary is a solitary long distance migrant and takes the heaviest bat losses at turbines across North America, so a method that works best on it is worth more than one that works best on a locally abundant species.
The gap between those two figures is larger than the confidence interval around either, so it is not noise in the sampling.
The widely quoted 54 percent is one species.
The relatives it did not reach
The northern yellow bat showed no statistically detectable change at all.
It is a close relative of the hoary, in the same group of tree roosting bats, and it flew through the same noise on the same nights.
Why one responds and its cousin does not is unresolved. Call frequency, flight height, foraging style and how each species weighs a degraded echo against an attractive insect swarm are all candidates.
Separate cage work with tethered and free flying bats found the same pattern of uneven response, with some species backing off consistently and others varying by frequency and by season.
Moving one blade is a routing problem, and a 240 foot blade that clipped a trestle showed how little margin these jobs carry.
The cage trials and the species responses are described in a research journal.
Deterrence is species specific and nobody can yet predict which.
Where this leaves an operator
Halving bat losses without stopping a turbine is a real result. The alternative is curtailment, holding the rotor still below a cut in wind speed, and that costs generation on exactly the low wind nights bats fly.
A deterrent runs without giving up megawatt hours, so the economics favor it wherever it works.
The catch is that a site cannot know in advance whether its own bats belong to the responsive group.
That argues for measuring the community first and for reporting results by species rather than as a single headline figure.
It also argues against treating one number as the property of the whole technology, because this one already belongs to two animals and not to a third.
Site conditions decide how any machine performs, which is why one Arctic turbine stands on steel piles kept deliberately frozen.
The species figures, the trial design and the emitter specification are set out in the published study.
The losses fell by half, and the honest sentence names which bats and which did not.
