Вештачка интелигенција

Can Tree Planting Bring Back Animals?

Summary

In the world, efforts are being made to restore forests, whether through tree planting or the natural return of flora. However, as trees grow, it becomes difficult to determine whether bird communities, insects, frogs, and other forms of life are […]

Can Tree Planting Bring Back Animals?

In the world, efforts are being made to restore forests, whether through tree planting or the natural return of flora. However, as trees grow, it becomes difficult to determine whether bird communities, insects, frogs, and other forms of life are also recovering.

In Chocó, northwest Ecuador, one of the most biodiverse rainforests in the world, a group of German and Ecuadorian scientists measured the recovery of biodiversity in different types of land, including active pastures and farms, abandoned farms growing into young secondary forests, and old-growth forests.

Using a combination of acoustic monitoring and DNA-based research, the team discovered that species are returning to restored forests after several decades.

The researchers also found that analyzing the sounds of animals and insects provides information about the overall biodiversity of the forest, including silent creatures that do not sing, chirp, or croak.

Sound-based monitoring using artificial intelligence tools is a powerful and cost-effective technique for tracking biodiversity recovery in tropical forests. The authors say that these tools are necessary “to support conservation mechanisms that rely on forest restoration, such as ecosystem services payments, biodiversity offsetting, and credit markets.”

The research team established 43 plots along a “forest recovery gradient” in Chocó. On each plot, they used three methods of acoustic monitoring and one DNA-based method.

On each plot, the team set up sound recording boxes that captured all ambient sounds (such as rain and wind), as well as insects and vocalizing vertebrates like frogs, monkeys, and birds.

The first acoustic method involved experts listening to 28-minute recordings and identifying mammals, birds, and amphibians based on their calls. This method is time-consuming and limited to the amount of data that humans can process, making it costly.

The second method used recordings to create an acoustic index analysis, which describes characteristics like complexity and diversity of sounds. For example, the sound picture of a mature forest is more diverse and dense than that of a pasture.

Finally, deep learning, a computer model known as a convolutional neural network (CNN), trained on recordings of 75 known bird species. Once trained, the CNN effectively performed the work of bird experts in identifying calls from weeks of recordings instead of short segments.

“Instead of having to do this manually with experts, this (CNN) neural network really enabled us to scale up the model,” said Martin Schaefer, one of the study’s co-authors and director of the Ecuadorian NGO Jocotoco Foundation.

For the final investigation, the team used DNA to set up light traps to capture nocturnal insects. They then used metabarcoding techniques to analyze the DNA of the insects to get an idea of the diversity present.

It was found that the acoustic AI data correlated well with overall levels of biodiversity, even for species that were not directly detected through sound. Schaefer said this was expected, but they were pleased with how well the AI models functioned.

“I think the most important discovery is that AI models allow us to measure levels of biodiversity relatively well, even in simplified versions,” Schaefer said. “These AI models are also a good indicator for the recovery of species that are not heard in the forest soundscape.”

“The integration of sound analysis, neural network models, and non-acoustic biodiversity levels… notably insect data, is I think the most important part of this work,” Shaefer added.

There are several caveats to these methods. For example, acoustic monitoring cannot determine whether species (especially birds) are just passing through these plots or actively living and utilizing them. It also cannot ascertain the abundance of species on the plots.

To apply this method to a larger number of forests, AI models need to be trained on a greater number and diversity of animal sounds. Sound libraries like Xeno-Canto and Cornell’s Macaulay Library can be useful for this purpose.

FAQ