AI and 8 million digitized plant specimens reveal how the climate is changing nature in large parts of the world

The Royal Botanic Gardens, Kew has released a report highlighting the impact of digitization and artificial intelligence on plant and fungal research. Four hundred researchers from 40 countries collaborated on the study, with NTNU contributing expertise from Trondheim. James Speed, a professor of plant ecology, and David Williamson, a postdoctoral researcher in machine learning for natural history, played significant roles.

Over the past century, global flowering times have shifted by an average of 2.5 days per decade. However, this change is not uniform, with the most significant shifts occurring in tropical regions. The rise in temperatures due to climate change is particularly pronounced in these areas, while flowering patterns in the far north follow a different trend.

Traditionally, naturalists collected plant and fungi specimens, preserving them in herbaria worldwide. In recent years, many of these specimens have been digitized and made available online, with more than 145 million plant and fungi preparations now accessible. This digital transformation has opened up new avenues for research, allowing scientists to analyze vast datasets that would previously have been impossible.

Using machine learning models, researchers can now determine whether plants are in flower from digitized herbarium specimens. Williamson and his team trained a model using 8 million preserved specimens from around the world, representing 200,000 species. The model processed data at a speed that would have taken a human around 40,000 hours, or 20 years of full-time work.

The analysis revealed that global flowering times have shifted, with both earlier and later occurrences. This shift is more pronounced in tropical areas, where the largest temperature increases due to climate change are also observed. In tropical regions, flowering is more closely tied to precipitation, and changes in rainfall patterns can lead to shifts in flowering times. This, in turn, can disrupt the relationship between flowering and pollinating insects, potentially affecting insect-eating birds and food production.

AI and machine learning cannot solve all problems related to species extinction and climate change, but they can provide scientists with powerful tools for mapping and understanding biodiversity. By making millions of herbarium specimens searchable and machine-readable, researchers can focus fieldwork on areas where it is most needed and calculate the probability of a species being extinct.

AI can also assist in identifying unknown species in digitized collections, expediting the naming process and aiding conservation efforts.

Despite the potential of AI in biodiversity research, it is essential to recognize that it is not a replacement for human expertise. Less than 16% of the world's herbarium specimens are currently digitally accessible, and the biggest gaps exist in countries with rich biodiversity where collections are understaffed and underdocumented. The ultimate goal is to leverage technological advancements while ensuring that human researchers remain integral to the process, guiding the models and validating the results.