Natural History Collections at the Crossroads: Shifting Priorities and Data-Driven Opportunities

Abstract

Natural history collections face a critical juncture as environmental change and biodiversity crises accelerate. While collections data are key components of eco-evolutionary and environmental research in both fundamental and applied contexts, collecting strategies remain primarily taxonomically motivated. We argue that sampling strategies must evolve to better address broader ecological challenges, through emerging applications enabled by advances in data science and digital technology. Here, we propose a flexible framework using modern statistical approaches to optimise sampling design and research prioritisation. By considering biodiversity, environmental, spatial and temporal dimensions, we demonstrate how information theory and decision science tools can support strategic collecting, databasing and digitisation efforts. Our framework provides an evidence-based pathway for collections to enhance their role as critical research infrastructure for addressing 21st century environmental challenges. To illustrate this data-driven approach to research prioritisation, we present an example based on sampling design for wombats (Vombatus ursinus) in Australia.

Keywords: Bayesian decision theory; FAIR data principles; adaptive sampling design; conservation planning; data science; digital extended specimen; museum genomics; natural history collections; research prioritisation; value of information.

FIGURE 1

Information theoretic framework for research prioritisation in collections science. (A) Conceptual illustration of information gain, showing how uncertainty decreases as new data are collected. When data with higher information value is collected, the posterior distribution becomes more concentrated with additional observations, improving precision for outcomes of interest (e.g., species distributions, phenological patterns, or morphological variation). (B) Decision support framework combining Value of Information (VOI) and Need for Information (NFI) metrics to guide sampling design. This consensus plot illustrates a quadrant contrasting high and low VOI against high and low NFI, providing a structured approach for balancing information gain potential against ecological priorities when allocating sampling effort.

FIGURE 2

Data‐driven research prioritisation example—sampling design for the bare‐nosed wombat ( Vombatus ursinus ) in New South Wales, Australia, based on observation and specimen data from GBIF. (A) Value of Information (VOI) map showing expected information gain from additional sampling (darker blue = higher gain), calculated using a binomial model of annual observations in each 50 km grid, with expected information gain calculated using Kullback–Leibler divergence between current and updated probability distributions after simulated sampling. (B) Need for Information (NFI) map based on habitat loss percentiles (darker green = greater habitat loss), derived from the Habitat Condition Assessment System. (C) Cost of Information (COI) map using the Australian Bureau of Statistics' remoteness classification system (darker purple = metropolitan areas with higher accessibility and lower expected sampling costs). (D) Consensus diagram—VOI‐NFI quadrant analysis revealing priority sampling locations (upper right quadrant) where both information value and ecological need are high. In Panel A, grid cells (50 km) with insufficient data appear in grey (fewer than 2 total observations per grid). See Data S1 for details on methodology, data sources, and implementation.