Herbivores usually consume a mere fraction of available plant biomass. Spatial patterns in feeding damage may be attributable to induced defences by the host plant; a damaged plant reacts by lowering its nutritional value, thereby forcing herbivores to move on before food gets worse. In this study, we test this general hypothesis on a specific model system: caterpillars of the alpine butterfly Parnassius smintheus feeding on lance-leaved stonecrop Sedum lanceolatum. We first describe spatial patterns in host distribution and feeding damage within alpine meadows. We then use laboratory experiments to test a key assumption behind the proposed mechanisms: that the host plant exhibits an induced response with a negative impact on larval performance, and that this response is activated with a delay. Finally, we relate the patterns observed to the actual behaviour of Parnassius larvae. Overall, we found the level of feeding damage to be low (on damaged plants, only 5% of all leaves were fed upon). Within meadows, both host plants and feeding damage were clumped at a small spatial scale. This pattern seemed directly explicable by the timing of the host's induced defence. Laboratory experiments revealed a delay of 1Á2 d before the defence reached a level affecting larval performance, and wild larvae switch plants more quickly than this. A simulation model demonstrated that the spatial distribution of host plant damage can be generated by a simple random walk, based on the empirically observed step frequency, length and turning angles. Hence, as the most parsimonious explanation for the observed level and pattern of host plant damage, we offer a scenario where induced changes in host-plant quality limits the time spent per plant, but the herbivore moves throughout the landscape without any particular directionality.

Herbivores usually consume a mere fraction of available plant biomass. Spatial patterns in feeding damage may be attributable to induced defences by the host plant; a damaged plant reacts by lowering its nutritional value, thereby forcing herbivores to move on before food gets worse. In this study, we test this general hypothesis on a specific model system: caterpillars of the alpine butterfly Parnassius smintheus feeding on lance-leaved stonecrop Sedum lanceolatum. We first describe spatial patterns in host distribution and feeding damage within alpine meadows. We then use laboratory experiments to test a key assumption behind the proposed mechanisms: that the host plant exhibits an induced response with a negative impact on larval performance, and that this response is activated with a delay. Finally, we relate the patterns observed to the actual behaviour of Parnassius larvae.Overall, we found the level of feeding damage to be low (on damaged plants, only 5% of all leaves were fed upon). Within meadows, both host plants and feeding damage were clumped at a small spatial scale. This pattern seemed directly explicable by the timing of the host's induced defence. Laboratory experiments revealed a delay of 1Á2 d before the defence reached a level affecting larval performance, and wild larvae switch plants more quickly than this. A simulation model demonstrated that the spatial distribution of host plant damage can be generated by a simple random walk, based on the empirically observed step frequency, length and turning angles. Hence, as the most parsimonious explanation for the observed level and pattern of host plant damage, we offer a scenario where induced changes in host-plant quality limits the time spent per plant, but the herbivore moves throughout the landscape without any particular directionality.