If I understand your question correctly you are after defining what a sheltered bay is and what variables/parameters are important?

Well I am expecting both questions to have something to do with variables generated from the local topography, as a sheltered bay is commonly referred and/or defined as a protected location from wind or weather.

I imagine then it would have a series of spatial variables related to that. For example, solar radiation exposure (RE), wind direction/velocity (WV), tidal influence (TI) and so on.

For a predictive model or in GIS you could probably generate and/or identify locations along a coastline that has a low ratio of RE, WV and/or TI and generate sheltered locations based on that data.

Many thanks for your answers that is extremely helpful

I would add sea level changes to George's list.

I would reformulate the question. Is there a degree of sheltering that predicts the location and movement of prehistoric populations?

I would take a topographic map that I could correct for sea level, and I would digitize it in such a way that I could take each coastal pixel and develop a measure of exposure. Each point will be exposed to the sea from 0 to 360 degrees of arc. At 360 degrees you are on a small island. I would correct then adjust this. A coastal town with 0 degrees of exposure is at the back of a bay with an island completely obscuring the mouth of the bay. At 360 degrees of exposure, you are on a small island some distance from the mainland, and with no protection from any nearby land. You get to define what "nearby land" means. You will then correct the map to scale relevant to humans and scaled to the level of expected storms. A cove 30 meters wide may be sheltered until subject to 10 meter waves.

With this measure of exposure you can simplify, by defining "exposed" as anything with greater than X degree exposure, and "sheltered" as anything with Y degree exposure. Anything between X and Y is "uncertain." Now go through and for each value of X and Y such that X>Y, is there a value that does a reasonable job of predicting the probability of settlement? Also factor in distance to neighboring settlements, distance to a valued natural resource (the greater the reward, the more people will risk to gather it), and other factors you see fit.

Sounds like a fun project.