Architecture and resource allocation in annual plants under environmental variability

Models of phenotypic plasticity frequently assume that organisms have perfect information regarding the selective environment; studies of adaptive bet-hedging assume that organisms have no such information. This dissertation examines the optimal response of the schedule of reproductive allocation in annual plants to intermediately unpredictable environments.; Chapter 1 develops an evolutionary algorithm that finds the optimal allocation schedule given any season-length distribution. Schedules optimized for constant environments were bang-bang: no graded allocation, or simultaneous allocation to growth and reproduction. Optimal gradedness increased with environmental unpredictability. Every schedule can be described as a linear combination of specialist schedules which are optimal under perfect predictability. The weighting vector describes the contribution of each specialist schedule to the optimal schedule. Weight vectors of the optimal schedules resemble the distributions in that broader distributions select for schedules whose weights are similarly broadly distributed.; Chapter 2 describes a common-garden experiment in which reproductive schedules were compared among 4 annual species of Deinandra: 2 coastal species and 2 inland species. The coastal sites have more predictable climates than the inland sites. Regular harvests yielded regressions estimating dry weights from linear measurements, and repeated nondestructive measurements of the same 10 individuals of each species yielded vegetative and reproductive growth schedules and reproductive allocation schedules. The reproductive growth schedule of D. fasciculata had insignificantly smaller entropy than did the other three. Examination of late-season vegetative growth suggests that gradedness in the inland species may occur either through early floral initiation or by late vegetative investment.; Chapter 3 examines the fitness benefit incurred by the plant when the allocation schedule is cued by a random variable predicting season length with arbitrary reliability. The benefit of plastic allocation, relative to the fitness of the optimal schedule for the uncued distribution, can be no greater than the mutual information—an information-theoretic measure of the entropic overlap between two random variables—of the cue and the season length. This result can be applied sequentially to components of a multidimensional cue, to estimate optimal complexity of perception.