The bhangmeter is a device for remotely measuring the yield of an atomic explosion within the earth's atmosphere. It is a relatively simple electronic circuit consisting of a fast-response photodiode or phototransistor, a very fast timer circuit, and (usually) some additional circuitry such as a high-pass filter. Early (manual) bhangmeters simply reported a plot of photonic energy over time, whereas more complex (automatic) bhangmeters are able to determine when such a plot has reached a local minimum and begun to rise for a second time - and possibly also contain circuitry to compare the resulting time readout against a predetermined (calibrated) curve. Bhangmeters were initially designed for U.S. atomic tests. Later, as they became more reliable and smaller, they were 'packaged' into NUDETS for deployment on aircraft and satellite platforms. Bhang is, of course, a pun: it is the Hindi name for the leaf of the cannabis sativa plant (i.e. weed, smack, dope, maryjane) and, of course, onomatopoeia for the event the instrument is designed to monitor.

The bhangmeter operates due to a particular characteristic of high-energy (i.e. nuclear) explosions. Since the majority of the energy release from such a detonation is in the form of electromagnetic radiation, the very first indication of such an event at any distance will be a bright flash. This is the outpouring of energy from the supercritical nuclear reaction (fission and/or fusion, although at the outset always fission). In addition to visible light, however, the explosion will produce a storm of very high energy photons in the X-ray range. Since the atmosphere is not entirely transparent to X-rays, the air immediately surrounding the detonation point will absorb them and as a result it will heat up very quickly and very intensely. As the atmosphere reaches the state of plasma, two things happen. First, it begins to expand, but that happens fairly slowly in the timescale of the atomic explosion. Second, and more important, it becomes opaque to photons because it has become 'saturated' with energy and its transmissivity drops. At the point of saturation, however, when it can't absorb any more energy, it begins to re-radiate the energy as a blackbody.

The atmosphere is still expanding, however - this is the birth of the characteristic fireball. As it expands, it cools down once more to the point where it is once again transparent to photons and, more immediately, it continues to radiate the energy it has absorbed. This results in a second flash - one much longer if less intense than the original.

So, if you were to look at a (notional) graph of the brightness of an atomic explosion over time, you would see something like the following:

  ^
  |         **
  |        *  *
B |       *    *  
r |             *                         *****
i |      *       *                    ****
g |                *               ***
h |                  **          **    
t |     *               **     **
n |                        *** 
e |     
s |
s |    *
  |     
  |    
  |   *
  |  
  |***A<-----B-------------->C 
  ------------------------------------------------>
 0   Time in Milliseconds (ms)

   Notional Bhangmeter Readings During Detonation

In the above graph, the detonation begins at time point A. The brightness of the event shoots up from zero to its maximum brilliance. Then, as the surrounding atmosphere reaches saturation at around time point B, the brightness begins to drop, quickly but not as fast. At time C or thereabouts, the fireball has begun to reradiate the energy it is containing (blackbody radiation) and to expand, allowing energy to pass through it. Thus, the energy detected at the bhangmeter begins to rise again.

The important bit here is that given a standard atmospheric environment (which we mostly have, here on Earth in the outdoors) the resulting curve will be roughly similar for detonations of the same size. If the explosion is larger, then it will take longer to begin radiating a second time - because even as the fireball begins to expand, there is still enough energy being pumped out from its center to keep it opaque! Hence, the smaller the explosion, the shorter time interval B-C; the larger, the longer. This breaks down if the explosion is, say, underwater where the water's absorptive and radiative properties are dramatically different, or in outer space where there is no surrounding atmosphere to absorb the initial X-ray storm. It turns out that for atmospheric detonations from the surface up to several thousand meters of altitude and above 5 kilotons, the bhangmeter readings tend to be accurate to within +/- 20% when estimating yield - and as far as open sources are aware, no false bhangmeter alerts have ever been recorded with the possible exception of the Vela incident. At low yields (below 5-10kt), some tower shots diverged from predicted values; analysis indicated that this was likely due to the presence of large amounts of mass within the initial fireball (the tower and support structures) in relation to the energy yield.

The United States was careful to keep multiple bhangmeters of a standard design aimed at many of its early nuclear tests, specifically the Tumbler-Snapper series and the Ivy King shot. This gave U.S. atomic weapon engineers a 'baseline' curve for detonations of various size. Later bhangmeters were programmed with this information, and could output an estimated yield immediately as well as recording and transmitting the base data. Eventually, these sensors were placed aboard satellites in orbit, where they could keep watch on the planet's surface and look for the characteristic 'double flash' of a nuclear explosion - as well as provide estimated information about its yield. The Vela satellites were the first to have bhangmeters installed, and their successors the Defense Support Program birds had more sensitive and complex versions. The GPS satellite constellation also has bhangmeter-based NUDETS aboard. Those are (or were) calibrated, according to the FAS, by hitting the satellite with a ruby laser pulse from Sandia National Laboratory and using GPS atomic clock time to measure pulse and event timing.

I'm not going to try to give exact values, but the U.S. testing series showed bhangmeter intervals (time B-C) of between 4 and 100 milliseconds. We can infer that the 4 ms time were the smallest explosions, and that 100 ms were the largest ones measured. This also indicates that if you can distinguish the double flash with your naked eye, you are a) quite a distance away from it if you're not blind and b) likely looking at a very large yield detonation.

Sources:

Document WT-562, Operation TUMBLER-SNAPPER: Bhangmeter Mod II (Nevada Proving Grounds) Project 12.1. 1952.
Report on the 1979 Vela incident. Carey Sublette; Nuclear Weapon Archive (http://nuclearweaponarchive.org/Safrica/Vela.html)
Alert 747: Sandia Laboratories report on the Sept. 22, 1979 Vela satellite event.
The Federation of Atomic Scientists newsletter (https://fas.org/sgp/othergov/doe/acnt/1993cd.pdf)
Discussion with personnel from Los Alamos National Laboratories (unclassified).