I’ve done a little timelapse work (mine was with fairly basic equipment - results here - I do intend to revisit this fun pastime with better resources in the future).
There are two main considerations:
Fidelity of capture
Aesthetics of finished output
When you watch something like David Attenborough’s Private Life Of Plants, you’re looking at timelapse video that was captured in a studio setting (even though it looks wild) with very highly controlled lighting setups - most often, all natural light is excluded for the moment of frame capture, and the scene is illuminated by flash.
Obviously if you’re capturing the outside world, you won’t have the luxury of controlling the lighting - so the best you can do is to capture well-focused shots, free from obstruction, and other interference. AaronX is right - it’s better to capture more frequent frames than less, as you can always discard what you don’t need, but never recreate what was never photographed.
As to the aesthetics of the output - lighting of a natural scene will vary from one moment to the next (for example, when a cloud obscures the sun), as well more obviously varying as the day progresses, and also across many days as the weather changes, and the seasons progress.
If you want a smoothed output, you have to make some design choices - here are a few example scenarios (for the sake of simplicity, let’s assume the output video will be 25 frames per second):
A day in ten seconds: You want to compress a 24 hour period into ten seconds - that’s 250 frames in total - you could just capture one frame every 5 minutes and 46 seconds, then string them together, but as you observe, you’ll end up with something jiggly and frantic, so you could instead do something like capturing only 50 frames for the period (one every 29 minutes or so), then composing them into a video where each image notionally occupies 5 frames, but with smoothed transitions between them - instead of switching abruptly from one image to the next, the images flow or fade into one another
A year in a minute: You want to show a years’ worth of change in a screen minute - that’s 1500 frames of video - and each day will occupy about 1/6th of a second - clearly you probably can’t bother with day and night, or the screen will just be flashing at 12Hz, so you could:
Capture one frame per day and use the same transition method as above.
or
Capture a collection of frames each day (say, ten frames over the course of the same minute each day), then compose an averaged image for each day (which will help to eliminate things like people and animals flicking in and out of frames), then compose the 365 averaged images into a film (using transitions again to smooth between days)
or
Capture a series of images at intervals of 24 hours plus 4 minutes (compose video as above) - when you view these in sequence, you’ll be progressing through the year and the day at the same time (that is, the sun will appear to rise and set, across the course of a whole year’s change). Make sure that the capture process is ignorant of any daylight saving clock changes, or you’ll get an abrupt blip somewhere.