Date of Completion

7-28-2017

Embargo Period

7-28-2018

Advisors

Cynthia Jones, John Silander, Robert Bagchi

Field of Study

Ecology and Evolutionary Biology

Degree

Master of Science

Open Access

Campus Access

Abstract

The first chapter explores relationships between a suite of six functional leaf traits and leaf-level reflectance spectra in a unique biogeographical region, the Greater Cape Floristic Region (GCFR). Leaf level reflectances were measured for 1,862 leaves (1,120 species) sampled across communities in the GCFR. Functional traits were measured from populations of the same species from each sampling location. We show that the leaf reflectance spectra have vary and can be classified based on differences in leaf trait magnitudes, taxonomic group and sampling region. Second, we compare two statistical methods, Penalized Functional Regression (PFR) and Partial Least Squares Regression (PLSR) in their ability to predict leaf traits from spectra. PFR outperformed PLSR in prediction with significantly lower mean RMSE values when models were validated and compared in ten fold cross validation. Based on RMSE values from the model validation, PFR lowered model bias up to 39% in certain regressions suggesting significant improvements in prediction. We also introduce a new cross-validation method that structures folds of data based on spatial autocorrelation. Our results suggest that physical proximity of observations did not affect predictability for PFRs.

The second chapter focuses on leaf margins within a single genus. The function and purpose of leaf margins are still under debate. Most botanical work on the subject has primarily focused on temperate woody species in the northern hemisphere. In the second chapter, the leaf margins of the genus Pelargonium, a recent radiation centered in South Africa, were evaluated for their linear relationships with a wide suite of variables related to climate, growth form, and evolutionary history (represented by clade). We found that Pelargonium margins are not a significant predictor of mean annual temperature, which is the linear correlation often cited as a method of predicting paleo-climates from fossil leaf assemblages. In a multivariate regression, where variables were selected to best predict the mean leaf tooth density, we found that a combination of temperature range, seasonality, growth form categories and clade yielded the best model explaining 28% of the variation in tooth density. This generates interesting hypotheses of the true purpose of leaf margins within Pelargonium since they seem to be related to broad climatic features but filtered by growth form and evolutionary relationship.

Major Advisor

Cynthia Jones, John Silander

Available for download on Saturday, July 28, 2018

Share

COinS