This is an example of aspects to consider when choosing observing condition constraints. The science drivers for each program are distinct, of course, and thus the appropriate set of conditions for your own observations may be quite different. The example is taken from the NIRI System Verification plan.

Consider a program to take an H-band (1.7µm) spectrum across the star-forming core of the galaxy M82 to measure the relative H and He recombination line strengths. Good spectral resolution is required, both to separate the Brackett-series H lines and to achieve good line-to-continuum contrast against the underlying stellar emission. NIRI in its f/14 configuration was selected as spectral resolution is more important than slit throughput for this program.

• Image quality - the target is spatially extended and thus the flux through the slit is very insensitive to observing conditions. To maximise the likelihood of the observation being executed, I chose "any" (100%-ile) conditions. 


• Sky transparency (cloud cover) - although the program is designed to measure relative line fluxes, photometric conditions are required because I also wish to derive the dust extinction using data at other wavelengths. I chose 50%-ile conditions ("photometric"). Had I not been requiring photometry I would likely have selected 70%-ile ("thin cirrus"), even though it would increase the integration time, to improve the likelihood of observation execution.


• Sky transparency (water vapour content) - from the transmission spectra of this wavelength region, there is very little sensitivity to water vapour content. I chose "any" conditions.


• Sky background - Aside from the period soon after sunset and before sunrise, the variation in OH background is adequately small. I chose 80%-ile conditions.

Note that the statistical likelihood of execution of this observation, if all of the observing conditions are truly uncorrelated, is 100% * 50% * 100% * 80% = 40% of the time when the target is accessible in the sky. (In fact we expect some mild correlation and so this is a slight underestimate). Note that 40% is not the probability of my observations being completed successfully in the queue since a project will be allocated time when conditions are right for that project.