Considering diffraction effects does give a ballpark estimate in my case, even though the lens I'm using is not of good quality. The general argument is as follows. The angular width of the Airy disk from the center to the first minimum is:
2.44 lambda/D = 2.44 lambda F/f
where lambda is the wavelength of the light, D is a aperture, f the focal length and F the F-number.
The angular width of a photosite viewed from the lens is r/f, so it seems that we should take:
F < r/(2.44 lambda)
for lambda = 750 nm. But this is too severe a constraint, because most sensors have an anti-alias filter which to prevent Moiré patterns. It is good enough to prevent two green pixels from getting well into the same Airy disk width, so we can take the constraint to be:
F < sqrt(2)r/(2.44 lambda)
for lambda = 500 nm. My image sensor has r = 4.2 micrometers, so I should shoot at an F number below 5. But in practice I find that an F number of 7.1 yields better results, this is because the lens I use is quite far from top quality, the lens is not sharp at the pixel level, so the optimum is reached when diffraction effects are already occurring at that scale.
But this is, of course, only relevant if you can focus almost perfectly. In my experience, manual focus where you use magnification to check for optimal focus yields better results than auto-focus. This is something you need to check for your camera.
Then, to get to pictures with the most detail, you have to eliminate the noise as best as you can. Sharp pictures are no good if the smallest details are invisible due to noise anyway. This can be achieved by shooting at the base ISO (usually 100), exposing to the right and taking multiple pictures that you need to align and then average over.
Even when using a tripod this alignment step is necessary, because a one pixel shift amounts to just a camera rotation of r/f = 5*10^(-3) degrees. The shutter opening and closing is enough to cause such tiny shifts (say 0.05 pixels). This is not a big deal when averaging over a few pictures, but I usually take 25 pictures or more and then the very slow drift of the camera orientation must then be corrected for.
I always take my landscape pictures using a remote control (you can also use a timer), even when shooting short exposure pictures at daylight, as this avoids any motion of the camera-tripod set up when pressing the shutter. The larger the shifts are, the more unsharpness will be induced in the post-processing phase when doing the remapping to align the pictures.
Suppose that by exposing to the right allows you to expose for twice as long and you take 25 pictures, then that will reduce the noise by a factor of 2*sqrt(25) = 10, so you'll get a huge noise reduction. Note that you should process your pictures with the noise reduction set to off. Noise reduction will eliminate the noise in individual pictures at the expense of any hidden details that are buried deep below the noise floor.
Whether atmospheric seeing effects will be important is hard to tell. I've only noticed such effects when taking long distance pictures at night in the arctic region. The way I take my pictures using image stacking to get to almost zero noise pictures, it would be a tour de force to eliminate such effects by taking shorter exposure pictures (using larger aperture or higher ISO) as the alignment between different pictures would be a far more complex task.
