Estimation of temperature and precipitation from morphological characters of dicotyledonous leaves.

The utility of regression and correspondence models for deducing climate from leaf physiognomy was evaluated by the comparative application of different predictive models to the same three leaf assemblages. Mean annual temperature (MAT), mean annual precipitation (MAP), and growing season precipitation (GSP) were estimated from the morphological characteristics of samples of living leaves from two extant forests and an assemblage of fossil leaves. The extant forests are located near Gainesville, Florida, and in the Florida Keys; the fossils were collected from the Eocene Clarno Nut Beds, Oregon. Simple linear regression (SLR), multiple linear regression (MLR), and canonical correspondence analysis (CCA) were used to estimate temperature and precipitation. The SLR models used only the percentage of species having entire leaf margins as a predictor for MAT and leaf size as a predictor for MAP. The MLR models used from two to six leaf characters as predictors, and the CCA used 31 characters. In comparisons between actual and predicted values for the extant forests, errors in prediction of MAT were 0.6°-5.7°C, and errors in prediction of precipitation were 6-89 cm (=6-66%). At the Gainesville site, seven models underestimated MAT and only one overestimated it, whereas at the Keys site, all eight models overestimated MAT. Precipitation was overestimated by all four models at Gainesville, and by three of them at the Keys. The MAT estimates from the Clarno leaf assemblage ranged from 14.3° to 18.8°C, and the precipitation estimates from 227 to 363 cm for MAP and from 195 to 295 cm for GSP.