AI and Digitization Unlock Centuries of Plant Flowering Data

Herbarium curators are transforming centuries-old collections of pressed plants into dynamic climate datasets by combining high-throughput digitization with artificial intelligence. By rapidly photographing millions of specimen sheets and deploying computer vision models to identify reproductive structures like buds, flowers, and fruits, researchers can extract precise morphological traits at an unprecedented scale. Combined with automated text recognition to pull historical collection dates and locations from specimen labels, this technology allows scientists to map century-long shifts in plant flowering phenology and understand how flora responds to global climate change.

High-throughput digitization is the foundation of modern herbarium research. Historically, extracting data from physical specimen sheets required manual inspection, limiting studies to a few hundred or thousand plants. Today, herbaria use industrial-grade conveyor systems coupled with high-resolution cameras and specialized lighting rigs. Curators place mounted plant sheets onto moving belts that automatically capture standardized digital images at rates of thousands of specimens per day. This process rapidly converts fragile, physical archives into accessible, machine-readable image repositories.

Once the images are captured, curators must extract the historical context from the specimen labels. Each sheet contains a label with the plant's species name, the exact date it was collected, and the geographic location. Many older labels are handwritten, faded, or typed in inconsistent formats. Curators use Optical Character Recognition (OCR) and Natural Language Processing (NLP) algorithms to transcribe this text automatically. The AI parses the unstructured text and converts it into standardized digital database fields, assigning exact GPS coordinates to historical location descriptions in a process known as automated georeferencing.

With the temporal and spatial data secured, morphological trait extraction AI is deployed to analyze the plant matter itself. Curators utilize deep learning algorithms, specifically Convolutional Neural Networks (CNNs), trained to recognize specific plant structures on the digitized sheets. These computer vision models scan the high-resolution images to detect, segment, and count reproductive organs. The AI distinguishes between unopened flower buds, fully open blossoms, and developing fruits or seeds, effectively scoring the exact phenological stage of the plant on the day it was picked.

Training these AI models requires collaboration between botanists and data scientists. Curators manually annotate thousands of digital specimen sheets, drawing bounding boxes around flowers and fruits to create a reliable ground-truth dataset. Once trained on this data, the AI can analyze millions of unexamined specimens in a matter of days. Advanced models can even account for the visual challenges of pressed plants, such as overlapping leaves, discoloration from drying, and physical damage accumulated over decades of storage.

By merging the AI-extracted trait data with the digitized label information, researchers construct massive timelines of plant development. A single dataset can now span 100 to 200 years and cover entire continents. Scientists analyze these datasets by comparing the collection dates of actively flowering specimens from the nineteenth and twentieth centuries against those collected in recent decades, while controlling for latitude, elevation, and local temperature anomalies.

This automated pipeline has revealed clear, measurable shifts in flowering phenology linked to industrial-era warming. Across millions of specimens, the data consistently shows that many plant species are flowering significantly earlier in the spring than they did a century ago. Curators and ecologists use these precise historical baselines to identify which species are adapting to shifting seasons, which are failing to respond, and where potential ecological mismatches may occur between flowering plants and their native insect pollinators.