Caching for where and what: evidence for a mnemonic strategy in a scatter-hoarder

1. What they did

Scatter-hoarding animals bury many small caches of food in different places and must remember where they put them, ideally before other animals steal them. People remember lists better when they group similar items together, a memory trick called “chunking”. The authors asked whether fox squirrels do something similar: do they group their caches by the type of nut, so that all the walnuts end up in one area and all the almonds in another?

They worked with marked, free-ranging fox squirrels on the University of California at Berkeley campus. Each squirrel was handed 16 nuts, one at a time, of four kinds: almonds, hazelnuts, pecans and walnuts. Two things were varied:

• Order. Nuts came either in runs of four of the same kind (RUNS) or in a mixed order where no kind came twice in a row (PSEUDO).
• Source. Either the squirrel was given the next nut where it had just cached (multiple locations, MULTI), or it had to come back to one central spot for each nut (central location, CEN).

Three experimenters, keeping 5 to 10 m away, handed out nuts and recorded the start point and every cache location with handheld Global Positioning System (GPS) units. The authors then mapped each squirrel’s caches in a Geographic Information System (GIS) and measured how much the cache areas for different nut types overlapped, and, for comparison, how much the areas for each group of four caches in sequence overlapped.

2. What they found

• The order of nuts, the source, and the two together all affected how much squirrels overlapped caches of different nut types. Sex made no difference.
• Central source: when squirrels had to return to one spot for each nut, they kept caches of different nut types largely separate, whether the nuts came in runs or in mixed order. In the mixed order, the caches were grouped by nut type rather than by the order they were made in.
• Multiple sources: when squirrels were given each nut where they had just cached, they seemed instead to avoid places they had already cached in that session. With runs, they never overlapped the caches of more than two nut types. With mixed order, which the authors expected to be the hardest on memory, they overlapped caches of at least three nut types.
• In the mixed-order trials, overlap was higher with multiple sources than with a central source, and higher by nut type than by sequence. In the runs trials, overlap was higher with a central source.
• Big nuts and small nuts: to test a simpler explanation (carrying big nuts farther than small ones), they compared the large nuts (pecans, about 8.2 g; walnuts, about 11.8 g) with the small ones (hazelnuts, about 3.2 g; almonds, about 3.4 g). In the runs trials, the caches of the two large nut types overlapped each other much more than those of the two small ones (52.3% versus 11.6%). In the mixed-order trials the two were similar (23.1% versus 21.3%).

3. What they concluded

• This is the first evidence that a scatter hoarder may use spatial chunking, organizing caches by what is in them, which could lighten the memory load and make caches easier to find again.
• When foraging from many places, squirrels appear to switch to a different and perhaps simpler rule: avoid the areas already used. Alternatively, chunking may break down there because keeping similar nuts together costs more energy when they are found far apart.
• In the wild, a squirrel at a heavily fruiting tree may behave like a central place forager, while a squirrel moving between food sources faces a constant choice of where to forage. Squirrels may adjust their caching depending on how they get their food.
• Squirrels may be grouping caches by the value of each nut, not only its species, since under some conditions they seemed to treat nuts simply as large or small.

6. Other technologies, skills or innovations

• Handheld GPS units (Garmin eTrex H or 10), with two experimenters alternating to take enough waypoints for each cache. GPS error was measured at 75 known locations (mean 1.3 m), and a 1.65 m buffer was drawn around each waypoint to hold the true cache.
• Overlapping cache areas (polygons) were calculated in ArcGIS 10.3.