AI allows iNaturalist to map California plants with unprecedented precision

Using advanced artificial intelligence and citizen science data from the iNaturalist app, researchers at the University of California, Berkeley have developed some of the most detailed maps to date showing the distribution of California’s plant species .

iNaturalist is a widely used cell phone application, originally developed by UC Berkeley students, that allows people to upload photos and location data of plants, animals, or other life that they they meet, then collect their identity. The app currently has over 8 million users worldwide who have collectively uploaded over 200 million observations.

The researchers used a type of artificial intelligence called a convolutional neural network, which is a deep learning model, to correlate citizen science data on California plants with high-resolution remote sensing satellite or aerial images of the state . The network discovered correlations that were then used to predict the current range of 2,221 plant species across California, at scales of just a few square meters.

Botanists typically create high-quality species maps by carefully cataloging all the plant species in an area, but this is not feasible outside of a few small natural areas or national parks. Instead, the AI ​​model, called Deepbiosphere, leverages free data from iNaturalist and remote sensing planes or satellites that now cover the entire globe. With enough observations by citizen scientists, the model could be deployed in countries lacking detailed scientific data on plant distributions and their habitats to monitor vegetation changes, such as deforestation or regrowth after wildfires. .

The results were published September 5 in the journal Proceedings of the National Academy of Sciences by Moisés “Me” Expósito-Alonso, assistant professor of integrative biology at UC Berkeley, first author Lauren Gillespie, a doctoral student in computer science at Stanford University, and colleagues. Gillespie is currently receiving a grant from the US Fulbright Student Program to use similar techniques to detect patterns of plant biodiversity in Brazil.

“In my year here in Brazil, we experienced the worst drought on record and one of the worst fire seasons on record,” Gillespie said. “So far, remote sensing data has been able to tell us where these fires have occurred or where the drought is most severe, and with the help of deep learning approaches like Deepbiosphere, they will soon tell us what that happens to individual species in the field.”

“That’s a goal: to expand it to many places,” Expósito-Alonso said. “Almost everyone in the world has a smartphone now, so maybe people will start taking photos of natural habitats and it can be done on a global scale. At some point, this will allow us to have layers in Google Maps showing where all the species are, so we can protect them.

In addition to being free and covering most of the Earth, remote sensing data is also finer-grained and more frequently updated than other sources of information, such as regional climate maps, which often have a higher resolution of a few kilometers. Using citizen science data with remote sensing images – just basic infrared maps that provide only an image and temperature – could enable daily monitoring of hard-to-track landscape changes.

Such monitoring can help ecologists discover hotspots of change or identify species-rich areas in need of protection.

“Thanks to remote sensing, we obtain new images of the Earth with a resolution of one meter almost every few days,” Expósito-Alonso said. “These now allow us to potentially track changes in plant distributions and changes in ecosystem distributions in real time. If people deforest remote places in the Amazon, they can’t get away with it – this is reported by this prediction network.”

Expósito-Alonso, who moved from Stanford to UC Berkeley earlier this year, is an evolutionary biologist interested in how plants evolve genetically to adapt to climate change.

“I felt the need to have a scalable way to know where plants are and how they move,” he said. “We already know that they are trying to migrate to cooler areas, that they are trying to adapt to the environment that they are facing now. The main thing in our lab is to understand these changes and these impacts and to determine if plants will adapt.

In the study, researchers tested Deepbiosphere by excluding some iNaturalist data from the AI ​​training set and then asking the AI ​​model to predict plants in the excluded area. The AI ​​model had 89% accuracy in identifying the presence of species, compared to 27% for previous methods. They also compared it to other models developed to predict where plants grow in California and how they will migrate as temperatures rise and precipitation changes. One of these models is Maxentdeveloped at the American Museum of Natural History, which uses climate grids and georeferenced plant data. Deepbiosphere performed significantly better than Maxent.

They also tested Deepbiosphere against detailed plant maps created for select state parks. It predicted with 81.4% accuracy the location of redwoods in Redwood National Park in Northern California and accurately captured (with R²=0.53) burn severity from the 2013 Rim Fire in Yosemite National Park.

“What was amazing about this model that Lauren came up with is that you just train it with publicly available data that people keep downloading with their phones, but you can extract enough information to be able to create maps well defined in high resolution.” » said Expósito-Alonso. “The next question, once we understand the geographic impacts, is: ‘Will the plants adapt?’

Megan Ruffley, also of the Carnegie Institution for Science at Stanford, is a co-author of the paper.