Waterfall stacking: Can you leave your tripod and ND filter at home?

Detail Man wrote:

A 1024x768 version of the (present) 512x384 section of the lower-right quadrant would be good.

Here is the basis of my thinking - and reasons that I think that the samples resemble random noise.

You sampled sixteen 0.004 Second "windows" at a repetition-rate of 9 samples per second. Only 3.6% of what occured (within the duration of each 0.004 Second sampling "window") was averaged. The rest of what occured was entirely unkown to the camera system.

(Assuming absolute camera stability), each photo-site on your image-sensor was illuminated (during the 0.004 Second "windows") by a limited sampling of a complex reflectance waveform containing various frequency components with random phase relationships (relative to the phase of some fundamental frequency of analysis), the time-domain sum of which represented some average value of luminance - but (only) within each of the 0.004 Second sampling "windows".

Due to the varying and unpredicatble phase relationships (relative to the phase of some fundamental frequency of analysis) that exist between the various frequency components of the reflectance waveform, the instantaneous time-domain sum of the frequency components that (at any given time) exist in the frequency-domain (even at less than 1/2 the sampling frequency of 4.5 CPS) does not represent the simple case of an amplitude-stable complex periodic time-domain waveform that can be sampled to yield consistent inter-sample results - because the various frequency components are being phase-modulated in essentially random ways.

Frequency components existing at greater than 1/2 the sampling frequency - which are quite possible - are aliasing, "wrapping around", and showing up below (as opposed to above) 4.5 CPS.

Assuming some amount of angular motion in the X and Y axes of the camera orientation shifting, the situation described above is complicated further by any (non IS-cancelled) hand tremors.

If the sum of all of the above-described effects do (for the reasons described) resemble randomly sampled results - which in turn result in random average illumination levels of individual photo-sites during the sampling "windows" - then the recorded signals themselves (and not just the resulting Photon Shot Noise) are (at least partially, if not quite significantly) random in nature.

In the case of a single continuous exposure, all that occurs is included within the averaging process - so the sampling uncertainty introduced by the random phase relationships described does not exist.