DIGITAL LEAF PHYSIOGNOMY: CALIBRATION OF A NEW METHOD FOR RECONSTRUCTING CLIMATE FROM FOSSIL PLANTS

For almost 90 years, paleotemperature proxies based on the physiognomy (size and shape) of plant leaves have relied almost exclusively on one character, the presence vs. absence of teeth; moreover, the biological basis underlying these proxies have remained obscure. If leaf physiognomy could be more fully described, using mechanistically-informed characters, significant improvements to current leaf-climate methods should be possible. The leading hypothesis to explain the adaptive significance of teeth is that they serve as ‘hot spots’ for photosynthesis and transpiration early in the growing season, which would be advantageous for plants growing in colder environments to maximize their shorter growing seasons. We tested this hypothesis quantitatively by measuring gas exchange along leaf margins for two greenhouse-grown floras. Results support the ‘hot spot’ hypothesis and the use of tooth-based leaf-climate methods: leaf-margin gas exchange is most vigorous early in the growing season, and toothed margins are more active than untoothed margins.

We developed a new method for exploring relationships between physiognomy and climate, called digital leaf physiognomy, which readily computes physiognomic variables such as tooth area and perimeter: area ratio. 1460 leaves from a transect of present-day forests spanning a mean annual temperature (MAT) of 5 to 26 °C from the U.S. east coast and Panama have been processed using the digital leaf physiognomy approach (1-6 images per species; 17-135 woody dicot species per site; 17 sites). Many variables correlate strongly with MAT, including number of teeth (r²=0.79; P < 0.001) and shape factor (4p ´ area / perimeter²; r²=0.71; P < 0.001). The standard error of the best multiple linear regression is ± 1.9 °C, which is more accurate than the univariate regression involving only presence vs. absence of teeth (± 3.1 °C). Even when restricted to only those variables applicable to fragmentary fossils, the standard error of the best model is still ± 2.1 °C. Automated resampling routines indicate that one image per species for at least 20-30 species per site is sufficient for precise MAT estimates. Digital leaf physiognomy shows great promise for facilitating breakthroughs in new, mechanistically-grounded leaf-climate methods.