A knowledge representation system is presented for studying hummingbird behaviour in a laboratory environment. It is shown that a set of procedural rules can be developed, based on numerical, symbolic and heuristic techniques to aid in discovering how a hummingbird learns a simple spatial grid pattern of food sources. The use of different types of tree-like data structures facilitates a systematic representation of the knowledge fragments and allows a thorough cross-examination of both the experimental designs and the hypotheses. The problem of analyzing a non-uniformly sampled time series of behaviour observations is discussed and a solution proposed that uses a mathematical matching algorithm called warping. A trajectory of individual feeder visitations is generated by a bird behaviour model. The technique of warping is used to test if this trajectory can be mapped to another generated by a bird foraging. The two-dimensional analogue of the warping technique is applied to the spatial grid in order to evaluate the degree of spatial specialization in the bird's foraging behaviour. A correlation measure is applied to groups of pairs of rule combinations to ascertain which of these account for most of the observed behaviour. It is shown that by using a collection of different types of similarity measures a procedural approach can be formulated to aid in the representation of the knowledge accumulated by a hummingbird during the course of a spatially distributed foraging experiment. These procedures are arranged in a hierarchy of choices and implemented in an interpreter which formed the basis for an expert system in hummingbird spatial foraging. Experimental applications of these numerical algorithms and data structures are presented. The system was then tested on a complete series of behaviours by testing five different individuals on the same design. The procedural algorithms were calibrated on the first individual and then applied to subsequent individuals to test the knowledge representation derived from the first case. The results from this experiment suggest that a knowledge representation system composed from these rule fragments can be developed into a grammar that would standardize the testing of all spatial foraging experiments. In addition it is indicated that representing knowledge as a hierarchy of procedural options is of use in testing the way in which experimental knowledge is gathered. The implications of this knowledge of spatial foraging can be tested interactively as an experiment progresses. / Graduate and Postdoctoral Studies / Graduate

I consider two types of foraging situations. In the first type, the forager knows the location and quality of no more than one food item or habitat at a time. I call this myopic foraging. In the second situation, the forager may know the location and quality of more than one food item or habitat at a time. I call this periscopic foraging. I develop theoretically both models and the predictions generated by each. Myopic models predict that foragers should have a so-called "bang-bang" control with respect to the choice of food types. Periscopic models predict that foragers will have a continuous control with respect to the choice of food types. I experimentally tested for the presence of each type of control in a field study that employed hummingbirds feeding on artificial resources. As predicted by the theory, the transition of behavior from picky to opportunistic was significantly sharper for the myopic than for the periscopic foragers. Furthermore, theory predicts that there should be some range of relative values of the rich and poor food types over which the myopic foragers are opportunistic, while the periscopic foragers exhibit a partial preference. This prediction was supported by the data. I predicted that the partial preferences of periscopic foragers would result from the hummingbirds exploiting those poor quality feeders which were located nearest to the best foraging path among rich feeders. The data supported this prediction. Periscopic foragers also performed as predicted by becoming more selective on rich feeders as the densities of both rich and poor feeders increased. I developed a model of optimal sampling behavior that hypothesized birds have evolved in an environment of exploitative competition. The model predicted a rule of departure from a resource patch that depended on the presence or absence of nectar in a sampled as well as the expected quality of the nectar. Hummingbirds performed according to the model's predictions.

I tested the twin-threshold model, a risk-sensitive foraging model incorporating both a starvation threshold and a higher reproductive threshold. The model predicts risk-adverse foraging when an animal's energy state is close to the starvation threshold and risk-prone foraging when the animal's energy state is close to the reproductive threshold. Wild rufous hummingbirds (Selasphorus rufus) were presented with a choice of three artificial flower types that had either no, moderate or high variability around a common mean. I manipulated energy state by changing either the mean nectar volume or altering the cost of foraging (long versus short corollas). When the energy state of hummingbirds was close to the reproductive threshold they preferred the variable options. When the energy state of hummingbirds was close to the starvation threshold they preferred the nil option. Hummingbirds responded in a manner consistent with the twin-threshold model. / x, 90 leaves ; 29 cm.

Central place foragers, such as territorial hummingbirds, feed from resources that tend to be constant in space and to replenish with time (e.g. nectar in flowers). The ability to remember both where and when resources are available would allow these animals to forage efficiently. Animals that feed at multiple locations would also benefit from forming routes between these multiple locations. Hummingbirds are thought to forage by repeating the order in which they visit several locations following a route called a “trapline”, although there are no quantitative data describing this behaviour. As a first step to determining how and if wild free living hummingbirds forage by traplining, I decomposed this behaviour into some of its key components. Through five field experiments, where I trained free-living hummingbirds to feed from artificial flowers, I confirmed that territorial hummingbirds will, in fact, trapline. Birds will use the shortest routes to visit several locations and will prioritize those locations that are closest to a usual feeding site. Additionally, even though hummingbirds can learn to use temporal information when visiting several patches of flowers, the spatial location of those patches has a larger influence in how these birds forage in the wild. Since male and female hummingbirds were thought to forage differently I also tested whether there were sex differences in the types of cues they use when foraging. Contrary to expectation, female hummingbirds will also use spatial cues to relocate a rewarded site. Using the foraging ecology of rufous hummingbirds to formulate predictions as to what information these birds should use has lead me to discover that these birds forage in a completely different way than previously thought.