From an engineering perspective, first we have to define what is 'audible'.
Sirens produce a single or dual continuous tone that may be detectable down to the threshold of the ambient noise. Because it's a steady tone, and has a narrow bandwidth, it may not need to be much louder than the background noise in order to be noticed.
In a small town environment, background noise may be between 50 and 65 dBA. In a city, it's closer to 75 dBA. (The A-weighting doesn't have a super significant impact here, even though sirens are rated using C weighting (e.g. dBC), because the weighting curve is mostly effective at lower frequencies than sirens use, although the C-weighting curve is the 'flatter' curve). So to be audible above broadband background noise, the sound level produced by the siren needs to be 65dBA or above, and in a downtown area it should be 70dBA or above.
For a point source, sound level decays with distance at a rate of -6dB per doubling of distance. However, obstacles and the environment obstruct sound propagation. The guideline used by FEMA in CPG 1-17 (the Outdoor Warning Systems Guide) is a rate of -10dB per distance doubled. This accounts for wind, terrain, foliage, and obstructions.
For any siren, you can determine the distance for audibility roughly based on the siren's rated 100' output, then at 200' the output is -10dB from that number, at 400' the output is -20dB down, at 800' it is -30dB down, etc., and this is what a FEMA CPG 1-17 application of a siren will be designed to satisfy. In the real world, the propagation loss will be somewhere between -6dB and -10dB per doubling of distance, so there is some extra margin built into the numerical calculation of siren range.
Where things get more murky is the difference between an ideal -6dB propagation loss and the FEMA -10dB guideline. Lower frequencies naturally penetrate foliage and wrap around obstacles more easily, because wavelengths at lower frequencies are larger. All else being equal, the siren with the lower tone (down to about 300-400 Hz) will produce a louder tone at the distance, but you also have to balance this against the sensitivity of the ear. This is where the A weighting curve comes in to play, and the difference between the A-weighting curve and the C-weighting curve can become significant. The A-weighting curve crosses the C-weighting curve at 1000 Hz; below this, at 200 Hz the A-weighting curve is about -12dB lower. So there is a need to balance the gained distance from longer wavelengths with the naturally decreasing sensitivity of the ear. At ~500Hz, the A-weighting curve is about -3dB from the C-weighted curve. So the ideal siren warning frequencies are in the neighborhood of 400-800 Hz, which is where many sirens operate, and if you could propagate a tone simultaneously at the lower and higher fundamental frequencies, you can produce a 'beat frequency' or 'difference tone' that is lower in frequency while being fringed by higher frequency harmonics (this is the source of the 'buzz' character of sirens like the Thunderbolt).
Download this:
https://www.dhs.gov/sites/default/files ... 15-508.pdf for some interesting reading.
Also, A-weighting curve vs C-weighting curve (vs "Z"-weighting curve which is the same as 'unweighted'):
https://www.nti-audio.com/en/support/kn ... asurements