Evaluating LMA and CLAMP: Using information criteria to choose a model for estimating elevation

The morphology of leaves and composition of the flora respond strongly to the moisture and temperature of their environment. Elevation and latitude correlate, at first order, to these atmospheric parameters. An obvious modern example of this relationship between leaf morphology and environment is the tree line, where boreal forests give way to artic (high latitude) or alpine (high elevation) tundra. Several quantitative methods, all of which rely on uniformitarianism, have been developed to estimate paleoelevation using fossil leaf morphology. These include 1) the univariate leaf-margin analysis (LMA), which estimates mean annual temperature (MAT) by the positive linear correlation between MAT and P, the proportion of entire or smooth to non-entire or toothed margined woody dicot angiosperm leaves within a flora and 2) the Climate Leaf Analysis Multivariate Program (CLAMP) which uses Canonical Correspondence Analysis (CCA) to estimate MAT, moist enthalpy, and other atmospheric parameters using 31 explanatory leaf characters from woody dicot angiosperms. Given a difference in leaf-estimated MAT or moist enthalpy between contemporaneous, synlatitudinal fossil floras-one at sea-level, the other at an unknown paleoelevation-paleoelevation may be estimated. These methods have been widely applied to orogenic settings and concentrate particularly in the Western US. We introduce the use of information criteria to compare different models for estimating elevation and show how the additional complexity of the CLAMP analytical methodology does not necessarily improve on the elevation estimates produced by simpler regression models. In addition, we discuss the signal-to-noise ratio in the data, give confidence intervals for detecting elevations, and address the problem of spatial autocorrelation and irregular sampling in the data.