ESTIMATES OF MEAN ANNUAL TEMPERATURE USING LEAVES AND WOOD FROM MODERN SITES ALONG A TEMPERATURE GRADIENT IN EASTERN NORTH AMERICA

The relationship between leaf morphology or wood anatomy and climate has been recognized qualitatively for over 85 years. These relationships have been quantified recently using regression and multivariate statistical approaches calculated from modern leaf, wood, and climate data, primarily from North America and Asia. These equations are statistically significant, demonstrating that these equations are robust. Accordingly, these equations are used to estimate the paleotemperature of fossil sites. Over a dozen different equations have been calculated from modern leaf morphology, wood anatomy and temperature data, leading to confusion as to which equations are best suited to determine the paleotemperature of a given fossil site. One solution to the question of which paleotemperature equation to use is to rigorously test each equation for precision and validity on modern floral sites not used in the original statistical analyses.

A database of leaf morphology, wood anatomy, and climate variables from modern floral localities along a temperature cline in eastern North America has been created to test these equations. The floras are located from New Hampshire to southern Florida, and experience mean annual temperatures between 4.5°C and 22.3°C. The floras belong to a common geographic province and are floristically similar, thus limiting the chance that variation in leaf morphology between floras is due to major differences in topography, phylogeny or precipitation between sites. Fifteen localities within this temperature cline have been sampled.

Simple linear regression equations, multiple regression equations, and canonical correspondence analysis were tested on the fifteen sampled modern floras to determine the validity of these equations to predict paleotemperature of sites with analogous floras. Initial results show that many of the approaches give equivalent results. Average absolute error for all approaches ranges from 2.1°C to 3.3°C, with a median error ranging from 1.3°C to 2.5°C. In several cases, the temperature of a site is consistently over- or underestimated regardless of the equation used. These initial results suggest that one of several approaches could be used to determine the paleotemperature of fossil sites with floras analogous to modern eastern North American floras.