Quantifying leaf herbivory: A guide to methodological trade-offs and best practices

Leaf herbivory is a ubiquitous ecological interaction that varies significantly in intensity across species, habitats, and biogeographic regions. Although quantification of leaf damage is crucial for understanding many ecological processes, the accuracy and precision of various damage estimation methods used by researchers, including visual estimation, digital image analysis, and artificial intelligence, have not been evaluated and compared. We use a phylogenetically diverse group of tropical plants to compare the accuracy and precision of damage estimation methods and use the results to provide a guide to herbivory estimation that balances the advantages and disadvantages of each method. We found that visual estimation tended to overestimate herbivory levels compared to digital methods but was 15 times faster and improved in accuracy and speed with training. Conversely, deep-learning algorithms underestimated herbivory relative to image analysis with ImageJ when it was on the margin, but showed similar accuracy for damage inside of leaf margins. Our results indicate that while visual methods allow for rapid assessment of large sample sizes and are suitable for detecting broad patterns of damage, image analysis is crucial for accurate and precise quantification. The disadvantages of each method, however, can be minimized through proper training and efficient use of each tool, and we therefore provide a guide of practical approaches to herbivory estimation.