Understanding Intraspecific Trait Variability Using Digital Herbarium Specimen Images

Abstract

Plant traits are vital to quantify, understand and predict plant and vegetation ecology, including responses to environmental and climate change. Leaf traits are among the best sampled, with more than 200,000 records for individual traits. Nevertheless, their coverage is still strongly limited, especially with respect to characterizing variation within species and across longer time scales. However, to date, more than 3000 herbaria worldwide have collected 390 million plant specimens, dating from the 16th century. At present, the herbarium specimens are rapidly digitized and the images are made openly available to facilitate research and biodiversity conservation. In this study, we determined the potential of the digitized herbarium specimens images to: overcome limitations of data availability for quantitative leaf traits such as the area, length, width along with petiole length and use the trait values to understand the intraspecific variability across spatio-temporal scales. overcome limitations of data availability for quantitative leaf traits such as the area, length, width along with petiole length and use the trait values to understand the intraspecific variability across spatio-temporal scales. For the study, initially, specimen metadata was analysed from various online resources such as the Global Plants Database, Natural History Museum Paris, iDigBio and Global Biodiversity Information Facility (GBIF). Based on the completeness of the metadata, image availability, and the ease of measuring the leaf traits, we selected Salix bebbiana, Alnus incana, Viola canina, Salix glauca, Impatiens capensis, Chenopodium album, and Solanum dulcamara for the study. The semi-automated tool TraitEx (Gaikwad et al. 2019) was used to measure quantitative leaf traits such as the leaf area, perimeter, width, length and petiole length. Finally, excluding duplicates, we downloaded 17383 digital herbarium specimen images from iDigBio and GBIF, which included specimens from the 17th century to the present. However, about 5000 had insufficient information or quality issues, including not-yet-identified duplicates, or no intact leaves. For each selected image we measured four leaf traits - area, length, width and perimeter of the leaf blade - on up to 5 leaves. In sum, we collected about 120,000 trait records from 32009 leaves. Comparison of measured leaf traits to data from the TRY Plant Trait database (Kattge et al. 2019) revealed that we could improve the database for studying intraspecific trait variability by several orders of magnitude (from less than 10-100 records per species to >1000). The variation of trait records within the seven species shows reasonable patterns, which improves trust in the data quality. The extracted trait measurements were used to analyse the intraspecific variability for the species across different spatio-temporal resolutions. Machine learning method (random forest) was used to perform the analysis and the results revealed the imprint of spatial and temporal climate variation, including long term trends and climate change as well as seasonality effects, on leaf area. Through this study, we demonstrate the high benefits of digitizing herbarium specimens and reusing it for research studies to improve ecological knowledge and predictability of size-related leaf traits.

Reversing extinction trends: new uses of (old) herbarium specimens to accelerate conservation action on threatened species

“…Corney et al ., 2012; Ott et al ., 2020; Weaver et al ., 2020), describe and annotate them (Reeb et al ., 2018; McAllister et al ., 2019; Younis et al ., 2020), count plants’ elements (e.g. the number of flowers per inflorescence) (Davis et al ., 2020b), and evaluate intraspecific trait variability (Kommineni et al ., 2020), as well as to relate plant traits to environmental factors to analyze complex relationships and trends (Václavík et al ., 2017; Schneider et al ., 2018a; Park et al ., 2020; Yost et al ., 2020). Moreover, machine learning is used for species identification and classification, because of its contribution to the automatic collection of morphological characteristics and plant recognition at different taxonomic levels (e.g.…”