2024-UKCEH3-LU11 Mobilisation of plant images and data to address the taxonomic impediment of the Syzygium, the world’s largest tree genus

Overview

Aggregated vouchered plant occurrence data and associated images available in user facing portals enable easier access and reuse but many data and literature resources necessary for accurate identification remain scattered. Initiatives like the digital extended specimen (Fig. 1 adapted from Page 2013) that integrate heterogenous data (including distribution, phylogenetics and micromorphology) across institutional boundaries are starting to work out how to address this data disconnect but are still in the planning stages.  

Taxonomic output provides immensely valuable training data through e.g., the updating of specimen identifications, classifying specimens at different taxonomic ranks and enumerating their biotic and abiotic characteristics and variables (traits). This project produce such data through revision of a section of the world’s largest tree genus, Syzygium (Myrtaceae), through classical taxonomy, molecular systematics and study of range and functional traits. Syzygium (Low et al. 2022) is a highly appropriate model group for such a study as it is considerably species rich (~1,200 accepted names), but hugely understudied and in need of thorough revisionary work. The greatest gap in Syzygium systematics is in Syzygium subgenus Syzygium (c. 1000 species), this subgenus has recently been divided into informal ‘sections’ including the lineage including the type of the genus, Syzygium sect. Syzygium (est. 500 species). Recent study has proposed diagnostic characters for that section.  

This project aims to develop an extensible digital working environment, delivering image-based, robust, and taxonomically authoritative identification of plants at a global scale for biodiversity monitoring purposes (e.g., plot-based inventory and conservation assessment), to mobilise and interact with data at scale, and to recognise new species. Identifications enabled in the tool will be strengthened by explicit reference to currently scattered metadata associated with herbarium specimen images. Much of this associated information is currently linked only in the human brain; this tool will make these relationships explicitly available to other stakeholders. Integrating different kinds of data in this way makes it more explicit. Surfacing implicit data is a key aim of open science and will be an asset for training multi-modal machine learning models. Machine learning is a fast-moving field, the tool will enable integration with new approaches as they become available. This project will investigate effective ways to fuse the textual and visual information by combining large language models (LLMs) and Convolutional Neural Networks (CNNs) for clustering Syzygium species.

Figure 1: Existing (white lines) and future links (red) between different data classes for the tree genus Syzygium. Data classes linked in this way have been termed the ‘digital extended specimen’ .