For centuries, botanists have collected plants, pressed them, dried them, and stored them in herbaria. These specimens were originally created as physical records of species, but digitization is giving them an entirely new scientific role.

More than 145 million plant and fungi specimens from over 170 institutions in 40 countries have now been digitized. Once photographed and placed in searchable databases, these historical collections can be analyzed at a scale that would have been impossible for individual researchers.

Artificial intelligence is helping turn those archives into evidence of how nature has changed over time.

Eight Million Specimens Analyzed

Researchers trained a machine-learning system to recognize whether plants were flowering in digitized herbarium images. The model was then applied to around eight million preserved specimens representing roughly 200,000 species from different parts of the world. The scale of the experiment is striking.

According to machine-learning researcher David Williamson, the automated analysis took around one week. Completing the same work manually could have required roughly 40,000 hours — equivalent to about 20 years of full-time work. That does not mean artificial intelligence replaces botanists.

Instead, it allows researchers to process enormous collections quickly, leaving experts more time to investigate patterns, check unusual results, and ask questions that were previously too large to study.

The real advantage of AI is speed: it can reveal patterns hidden across millions of specimens while specialists remain responsible for interpretation.

Flowering Times Are Shifting

One of the clearest findings concerned flowering. Across the past century, global flowering times shifted by an average of about 2.5 days per decade.

The direction of the change was not identical everywhere. Some species flowered earlier, while others shifted later. What surprised researchers most was where the strongest changes appeared. Because some of the fastest temperature increases associated with climate change are occurring at high northern latitudes, scientists might have expected the most dramatic changes there.

Instead, some of the largest shifts were found in tropical regions. Plant ecologist James Speed explained that tropical flowering is often closely connected to rainfall rather than temperature alone. If changing climate conditions alter the timing or reliability of rainy seasons, plants may respond by flowering at different times.

Why Timing Matters

A few days may not sound dramatic, but flowering is part of a much larger ecological calendar. Plants, insects, birds, and other animals often depend on events happening at the right time. If a plant flowers earlier but its main pollinating insect has not adjusted its life cycle in the same way, the two species may become less synchronized.

That mismatch can reduce successful pollination. The effects can then spread further through ecosystems. Fewer pollinating insects can influence insect-eating birds, while changes in pollination may also affect crop production and wild plant reproduction.

Climate change does not have to eliminate a species immediately to disrupt an ecosystem. Simply changing the timing of biological events can create consequences across a food web.

Old Collections Gain New Value

Herbarium specimens are especially useful because many were collected decades or even centuries ago. Each specimen can preserve information about where a plant grew, when it was collected, and what stage of development it had reached. When millions of these records are combined, researchers can compare historical patterns with modern ones.

This turns museum collections into something resembling a global environmental time machine. Specimens that once served mainly as taxonomic references can now help answer questions about climate change, species distribution, flowering patterns, biodiversity loss, and even extinction risk.

AI Could Help Find Unknown Species

The same technology may also help scientists confront another enormous challenge: much of the planet’s biodiversity has still not been formally documented. More than 90% of fungal species, for example, are thought to remain unmapped. At the same time, species are disappearing rapidly enough that some may vanish before they are fully described.

Machine-learning tools can help by searching digitized collections for unusual specimens and flagging them for expert review. Statistical models can also estimate whether a species that has not been seen for a long time may be extinct or simply overlooked. That information could help scientists decide where field surveys are most urgently needed.

Why Humans Still Matter

AI can recognize patterns quickly, but it cannot independently decide whether those patterns are biologically meaningful.

Taxonomist Martin Cheek has argued that automated identification could save experts considerable time, especially when dealing with common species. Specialists could then concentrate more effort on poorly known organisms and threatened species. But expert verification would still be essential.

Williamson makes a similar point: machine-learning systems only perform as well as the information used to train them.

Researchers are still needed to design the questions, select suitable data, test the models, check errors, and interpret the results.

AI can accelerate scientific work, but it cannot replace the ecological and taxonomic knowledge required to understand what the results actually mean.

The Digitization Gap

There is another major limitation. Despite the rapid growth of digital collections, less than 16% of the world’s herbarium specimens are currently accessible in digital form. Some of the largest gaps are found in countries with exceptionally rich biodiversity. Collections there may be understaffed, poorly documented, or disconnected from international databases.

That creates a serious imbalance. Artificial intelligence can analyze only the material that has been digitized, meaning important regions and species may remain invisible.

A New Scientific Archive

The combination of historical specimens and modern machine learning is changing what museum collections can do.

A dried plant collected generations ago may now help scientists trace shifting rainy seasons, changing flowering dates, or disappearing species.

The technology is powerful not because it replaces centuries of botanical work, but because it unlocks it. Eight million specimens have already shown that natural-history collections contain far more information than researchers could once extract — and as digitization expands, those archives may become one of the most valuable tools for understanding how climate change is reshaping life on Earth.