Abstract Detail

Biodiversity Informatics & Herbarium Digitization

Cardenas, Santiago D [1], Clement, Wendy [2].

Comparing herbivory in invasive and native honeysuckles using manual and automated methods.

Invasive species threaten native ecosystems, agriculture, and existing biodiversity at large. Understanding how introduced species become successful invaders is important in combating their invasion1. The enemy release hypothesis is one explanation, suggesting that an introduced species loses its native predators upon colonizing a new area, allowing it to become a successful invader2. Lonicera tatarica is one of 11 invasive honeysuckle species in the United States and has become widespread across the country3. Of the 17 species of honeysuckle native to the US, L. canadensis has a range partially overlapping with L. tatarica. We took two approaches to studying invasive species biology in the invasive L. tatarica by focusing on herbivory. First we compared the levels of herbivory present in native and invasive ranges of L. tatarica in Russia and the eastern United States, respectively. Second, we compared levels of herbivory between L. tatarica and L. canadensis occurring in the northeastern US. Manual annotation of 111 herbarium specimens over a 70 year time span was done using GIMP, where a 4x4 cm grid was digitally added to each specimen4. Grid cells with at least 25% leaf coverage were manually scored for the presence of margin feeding, interior feeding, piercing, and mining damage. Comparing probabilities of herbivory per specimen demonstrated that L. tatarica experienced more total herbivory, margin feeding, interior feeding, and piercing damage in its native range as compared to its invasive range. No difference was recovered when comparing herbivory of L. tatarica and L. canadensis, though we note a relatively small sample size for L. canadensis (n = 25). These results suggest that L. tatarica experiences less herbivory in the invasive part of its range, perhaps due to a loss of natural predators from its native distribution. We also explored the performance of an automated method of herbivory scoring. The machine learning pipeline GinJinn was trained on 113 L. tatarica herbarium specimens to detect margin feeding, interior feeding, and mining damage. After using the model to detect herbivory on 44 L. tatarica specimens, we show that this automated approach can be used to quantify and compare herbivory in herbarium specimens, providing the opportunity to expand sampling beyond what is possible using our manual approach. This study also demonstrates the power of digitized herbarium specimens in supporting ecological studies, as we were able to access specimens and their metadata from across the world.
1Holmes et al. 2009, 2Colautti et al. 2004, 3Woods, KD. 2021, 4Meineke et al. 2018

1 - The College of New Jersey, Department of Biology, 2000 Pennington Road, Ewing, New Jersey, 08628, United States
2 - The College of New Jersey, Department of Biology, 2000 Pennington Road, Ewing, NJ, 08628, USA

herbarium specimen
invasive species
Machine Learning.

Presentation Type: Oral Paper
Number: BI&HD I001
Abstract ID:702
Candidate for Awards:None

Copyright © 2000-2022, Botanical Society of America. All rights reserved