Mean annual temperature estimation based on leaf morphology: a test from tropical South America

Are models that predict mean annual temperature (MAT) from leaf morphology applicable globally? Fifteen models that predict MAT from leaf morphology were tested on thirty floral samples from tropical South America to determine the degree to which models based on published data that are primarily from other regions are applicable to floras from tropical South America. The models included are based on regional data from North America, South America, and Asia. Of the fifteen models tested, five are simple linear regressions, six are multiple linear regressions, two are canonical correspondence analyses, and two are correspondence analyses followed by nearest neighbor analyses. For the seven modern floras with MAT ≤21°C, every model overestimates MAT. For the 23 modern floras with MAT >21°C, all models produce variable results without a systematic error. The range of average model errors is 2.7–7.3°C, while the absolute extremes of error are 0 and 15.1°C. Average 95% predictive confidence intervals range from 1.6 to 6.9°C. Predicted MAT falls within the published standard error of the model for 0–67% of the South American test floras. Evaluating the seven sites with MAT ≤21°C separately from the 23 sites with MAT >21°C shows that no equation accurately estimates MAT of the majority of low-temperature sites, but that four equations accurately estimate >50% of high-temperature sites. The results suggest that at least for sites of unknown or high elevation, mean annual temperature of fossil floras from tropical South America may be better predicted from models based on the leaf morphology of tropical South American floras.