Sorry to be so slow in responding.
From Section 4.1:
"We model memory as a finite map from locations to a set of values called the storable values. For historical reasons, we call this the store."
We represent such a map as some data structure, just as we represented environments as a data structure. We use apply-store to look things up in this map, just like we used apply-env.
This way the interpreter does not make a commitment to the representation of the store, just like using apply-env allows the interpreter to avoid committing itself to any particular representation of the environment.
Hope that helps.