Using reflectance spectra and Pl@ntNet to identify herbarium specimens: a case study with Lithocarpus.

Area of Science:

• Botany
• Taxonomy
• Digital Herbariology

Background:

• Digitization of plant collections is increasing data accessibility.
• Traditional taxonomic identification in collections has declined, leading to more specimens identified only to family or genus.
• This risks widening the data gap for understudied species.

Purpose of the Study:

• To compare the effectiveness of hyperspectral reflectance and computer vision for herbarium-based plant species identification.
• To assess the performance of spectral data for identifying Lithocarpus species, considering data volume, discrimination ability, and accuracy with close relatives.
• To evaluate Pl@ntNet (a computer vision approach) against spectral data for specimen identification.

Main Methods:

• Used Lithocarpus species as a case study.
• Compared species classification accuracy using leaf reflectance spectra versus computer vision (Pl@ntNet).
• Assessed spectral data requirements for optimal classification, species discrimination, and identification of closely related taxa.

Main Results:

• Lithocarpus herbarium specimens were accurately identified to species using limited spectral datasets.
• Spectral identification accuracy was only 14% lower than Pl@ntNet, despite not using reproductive structures.
• Close relatives were more frequently confounded in spectral identification.

Conclusions:

• Rapid, non-destructive leaf reflectance measurements show promise for plant identification in herbaria.
• Hyperspectral reflectance, combined with computer vision, can help fill identification gaps, especially for specimens lacking reproductive features.
• This approach can complement existing methods to improve the completeness of digitized plant collection data.