Automated leaf physiognomic character identification from digital images

Research into the relationship between leaf form and climate over the last century has revealed that, in many species, the sizes and shapes of leaf characters exhibit highly structured and predictable patterns of variation in response to the local climate. Several procedures have been developed that quantify covariation between the relative abundance of plant character states and the states of climate variables as a means of estimating paleoclimate parameters. One of the most widely used of these is the Climate Leaf Analysis Multivariate Program (CLAMP). The consistency, accuracy and reliability with which leaf characters can be identified and assigned to CLAMP character-state categories is critical to the accuracy of all CLAMP analyses. Here we report results of a series of performance tests for an image-based, fully automated at the point of use, leaf character scoring system that can be used to generate CLAMP leaf character state data for: leaf bases (acute, cordate and round), leaf apices (acute, attenuate), leaf shapes (ovate, elliptical and obovate), leaf lobing (unlobed, lobed), and leaf aspect ratios (length/width). This image-based system returned jackknifed identification accuracy ratios of between 87% and 100%. These results demonstrate that automated image-based identification systems have the potential to improve paleoenvironmental inferences via the provision of accurate, consistent and rapid CLAMP leaf-character identifications. More generally, our results provide strong support for the feasibility of using fully automated, image-based morphometric procedures to address the general problem of morphological character-state identification.