Leaf area varies within and between species, and previous work has linked this variation to environment and evolutionary history. However, many previous studies fail to examine both these factors and often are data‐limited.
To address this, our study developed a new workflow using machine learning to automate the extraction of leaf area from herbarium collections of Australian eucalypts (Eucalyptus, Angophora and Corymbia). This dataset included 136,599 measurements, expanding existing data on this taxon's leaf area by roughly 50‐fold. Our methods were validated using field standard metrics of accuracy, and comparisons to manual measurements both from the present study and existing datasets.
With this dataset for the eucalypt clade, we observed positive relationships between leaf area and mean annual temperature and precipitation similar to those reported for the global flora. However, these relationships were not consistently observed within species, potentially due to gene flow suppressing local adaptation. When we examined these relationships at different phylogenetic levels, the slope of trait–climate associations within lineages converged towards the overall eucalypt slope at shallow phylogenetic scales (5–12 MY), suggesting that effects of gene flow relax just above the species level.
The strengthening of trait–climate correlations at evolutionary scales just beyond the intraspecific level may represent a widespread phenomenon across various traits and taxa. Future studies can unveil these relationships with the larger sample sizes of new trait datasets generated through machine learning.
Synthesis. Using machine learning, researchers are able to confirm current positive global relationships between leaf area and mean annual temperature and precipitation. Additionally, they were able to take this a step further and examine how it changes across time. Here they saw that at roughly 5–12 million years ago in the phylogenetic tree, the trait–climate slope begins to show significantly less variation. Overall, the study shows the potential of machine learning in ecology, with exciting new potential findings with its use.