The "Rand Month Year" sites are 50 sites randomly drawn from the sites the bison actually occupied according to the survey from that month-year. Those sites are *not* located at random within some polygon (like a home range polygon).
Usage: One would normally use *all* the use data, but that's overwhelming for this visual, educational example - so we just took a random sub-sample of size 50 from the full "use" dataset for that Month/Year.
"Available" locations are randomly located within some pre-defined area. In our simple example, we've assumed that the available area is the whole rectangle shown (i.e. the park, and surrounding area). The "50 random sites" layer are randomly located within that rectangle.
Sometimes, researchers will define a more narrow "available area", such as 99-percentile home range. This narrows the scale at which the habitat selection question is asked.
Note that under this "Use/Availability" design, the available locations could be either be used or not-used, we don't know. This is as opposed to "Case/Control" sampling where we have "used" and "non-used" locations. Case-Control is rare in wildlife studies, because its very difficult to determine that a site is "not used".