It looks like its a combination of smell and memory.
Abstract. It has previously been assumed that grey squirrels, Sciurus cqrolinensis, cannot remember the locations of nuts they have buried, and hence must relocate nuts by their odour. This assumption was tested by measuring the accuracy of cache retrieval of captive squirrels. Each squirrel was released alone into an outdoor arena,where it cached 10 hazel nuts. After a delay of 2,4 or 12 days, each squirrel was returned to the arena and tested for its ability to retrieve nuts from its own cache sites and from 10 cache sites used by other squirrels. Although each squirrel's own caches were close to the caches of other squirrels, the squirrels retrieved significantly more nuts from their own sites than from sites used by other squirrels, after all delays. The retrieval accuracy of the squirrels under these conditions indicates that while grey squirrels can locate buried nuts by their odour, they can also remember the
individual locations of nuts they have buried.
A greater .number of a squirrel's own caches ('own') retrieval than caches of other squirrels ('other') relative to the numbers of each cache type available, was taken as positive evidence for memory of cache location.
Memory of cache locations, even in the absence
of competitors or predators, may have additional
adaptive value if it increases a squirrel's retrieval
efficiency. The squirrels in this study appeared to
minimize their retrieval searcli paths by running
directly from one patch oftrvo or three caches to the
next, harvesting the caches r"ith little ;e-tracing of
their path; this can be seen in the sequence of
retrievals illustrated in Fig. l. If so, this would indicate
that grey squirrels, like chimpanzees Pan
troglodytes (Menzel, 1973), can remember a series
of locations in relation to each other and use this
information to form a cognitive map, where information
about cache sites may be encoded
interesting implication of these results is that the
squirrels employed two methods to find caches:
they returned to sites where they had buried nuts
and they searched for the odour of buried nuts.
Under the conditions of this experiment, either the
first method was used more often than the second,
or it succeeded more often. Perhaps these methods
are used simultaneously: squirrels might sniff the
ground for odour cues while they are orienting to
the locations of remembered cache.
Grey squirrels remember the locations of buried nuts
There was also a study that found that the squirrels were organizing their nuts to make them easier to remember and find when gathered from a single source.
Squirrels either collected each nut from a different location or collected all nuts from a single location; we then mapped their subsequent cache distributions using GPS. The chunking hypothesis predicted that squirrels would spatially organize caches by nut species, regardless of presentation order. Our results instead demonstrated that squirrels spatially chunked their caches by nut species but only when caching food that was foraged from a single location. This first demonstration of spatial chunking in a scatter hoarder underscores the cognitive demand of scatter hoarding.
Caching for where and what: evidence for a mnemonic strategy in a scatter-hoarder