The most accurate way would be to maintain the individual’s resource calendar to reflect their availability to work. Next, is by sub-groups.

However, here is a method that I found very effective many times - It is less accurate but very time efficient in maintaining such data:

1. Instead of defining varying availability, keep all of your resources available at 100% of the time, as a default.
2. Use a specific project (or any other Work Item) to log resources unavailability, as Work - you can choose the granularity that will best match your required accuracy and changing nature of the data. For example, a resource that works only 75% FTE on varying days, mark an on-going task with 25% load. or another example, allocate a resource 100% loaded on a 1 day repetitive task every Monday (if they work only 4 days a week, not on Mondays).
3. By doing so, you will be marking time as work load (vs marking it unavailable). 
4. This will require your and resource mangers’ focus to be on monitoring load on a regulars basis, and checking for exceptions on overloads (weekly is recommended).
5. When  an overload is apparent, the RM can consider whether this is a load that poses a risk to the schedule and advise that the duration of the task should be longer, to lower the load, or perhaps decide to offload other work (not in AW) to accommodate the plan.

The downside:

• your total capacity will show 100% for that resource - and the total FTE count will not show your expected number - this can be a no-go for some organizations; But remember - the availability of the resources will not be compromised (which is what you are really interested in typically).

The upside:

• you will save time in maintaining this - RMs can own the maintenance of the “availability load”, and you will gain a lot of flexibility in your scheduling! When defining a resource has only 80% availability by definition, the system will be fixed and rigid. If that resource actually does have flexibility to do overtime, work more hours in a given week, or drop some of the other work they do (that is not planned in AW, for example, business-as-usual work), you will not be able to show for that. The suggested method allows you to allocate more time than the usual availability when necessary.

In any case, regular monitoring needs to be applied and the resource manager will need to be aware of the real allocated work, along with “availability load” applied (which can also be hidden in the view for easy validation. Hope tis is helpful