AI discovers flora's hidden characteristics to aid save species

Posted on 4 July, 2023 by benyamin chahkandi

Summary:To research and mitigate the effects of climate change on flora, scientists from UNSW and the Sydney Botanic Gardens have trained AI to access data from millions of plant specimens maintained in herbaria around the world.

Associate Professor Will Cornwell, the study's principal author, adds that herbarium collections are incredible "time capsules of plant specimens." It's no longer feasible to go through items manually because the National Herbarium of New South Wales alone receives over 8000 new specimens every year.  

Contrary to regularly observed interspecies tendencies, the team's analysis of over 3000 leaf samples using a new machine learning technique revealed that within a single species, leaf size did not increase in warmer climes.

This study, which was published in the American Journal of Botany, shows that factors other than climate have a significant impact on the size of leaves within a plant species. It also shows how AI can be used to transform static specimen collections and quickly and efficiently document the effects of climate change.

Herbarium collections are now available online.

Herbaria, or botanical libraries, are collections of plant specimens that date back to at least the 16th century.

"In the past, collecting plants to store in a herbarium was a worthwhile scientific endeavour. Every record has a date, location, collector, and potential species identification, according to A/Prof. Cornwell, a researcher from the School of BEES and a part of the UNSW Data Science Hub.

A few years ago, there was a push to migrate these collections online in order to improve scientific collaboration.

"The herbarium collections were kept secure in little boxes at certain locations, but today's world is extremely digital. There was an effort to scan the specimens to create high resolution digital reproductions of them in order to share information about all of the great specimens with the scientists who are now dispersed throughout the world.

Over 1 million plant specimens from the National Herbarium of New South Wales were converted into high-resolution digital photographs as part of the largest herbarium imaging project ever carried out at the Sydney Botanic Gardens.

"The digitization effort took more than two years, and soon after it was finished, Dr. Jason Bragg, one of the researchers, called me from the Sydney Botanic Gardens. He was interested in finding out how we could use some of these high-resolution digital photographs of the Herbarium specimens to apply machine learning.

"I was excited to work with A/Prof. Cornwell in developing models to detect leaves in the plant images, and then to use those big datasets to study relationships between leaf size and climate," said Dr. Bragg.

Leaf sizes are measured using "computer vision".

Dr. Bragg at the Sydney Botanic Gardens and Brendan Wilde, a UNSW honours student, worked with A/Prof. Cornwell to develop an algorithm that could automatically identify and gauge the size of leaves in scanned herbarium samples of two plant genera: Syzygium (commonly known as lillipillies, brush cherries, or satinas), and Ficus (a genus with about 850 species of woody trees, shrubs, and vines).

Convolutional neural networks, sometimes referred to as computer vision, are a subset of artificial intelligence, according to A/Prof. Cornwell. In essence, the procedure teaches the AI to perceive and recognise a plant's parts the same way a human would.

To educate the computer that this is a leaf, this is a stalk, and this is a flower, "we had to build a training data set," says A/Prof. Cornwell. Therefore, we simply trained the computer to find the leaves and measure their size.

"Numerous people have measured the size of leaves, therefore the practise is not new. However, a recent improvement is the speed at which these specimens may be analysed and their unique properties recorded.

A deviation from regularly occurring patterns

In the realm of botany, it is a generalisation to say that plants have larger leaves in wetter regions, such as tropical rainforests, as opposed to drier climates, like deserts.

And we observe that pattern in leaves amongst species all throughout the world, according to A/Prof. Cornwell. "The first test we ran was to see if we could recreate that relationship using the machine-learned data, and we were successful. Do we observe the same phenomena inside species today that we have access to so much more data than we did in the past?

For Syzygium and Ficus plants, the machine learning method was created, tested, and used to investigate the link between leaf size and climate within and among species.

The team's findings from this experiment were unexpected: while this pattern may be seen between various plant species, it is not consistently shown within a single species around the world. This is most likely due of a distinct mechanism known as gene flow that takes place within species. Localised plant adaptation is weakened by this process, which might also prevent species-specific relationships between leaf size and climate from forming.

Using AI to forecast future responses to climate change

Although not pixel perfect, the machine learning method utilised here to identify and quantify leaves provided levels of accuracy acceptable for investigating relationships between leaf attributes and climate.

But because the world is changing so quickly and there is so much data, A/Prof. Cornwell argues that these machine learning techniques can be used to effectively document the effects of climate change.

Additionally, the machine learning algorithms can be trained to spot tendencies that human researchers might not see right away. This could result in new understandings of plant evolution and adaptations as well as forecasts of how plants would react to upcoming climate change consequences.
 

source: www.sciencedaily.com/releases/2023/06/230620113755.htm