Automated acoustic classifiers provide insights into calling patterns of cicadas in tropical Australia
This study demonstrates that the deep-learning model BirdNET can effectively automate the detection and monitoring of five cicada species in tropical Australia, achieving high accuracy and providing ecologically consistent insights into their calling patterns over a 15-month period.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer
In the vast, sun-drenched landscapes of the tropics, the air often hums with the sound of insects. Among the loudest and most persistent of these are cicadas, creatures that spend years underground before emerging to sing for a few short months. For scientists, listening to these insects is more than just an auditory experience; it is a vital way to understand the health of an ecosystem. Insects play a crucial role in nature, acting as pollinators, food sources, and recyclers, yet their populations are declining worldwide. To protect them, researchers must know where they are, when they are active, and how many there are. Traditionally, counting these creatures has been a slow, laborious task. Scientists would have to walk through the bush, listen carefully, and manually identify each species by its song, a process that is difficult to scale and easy to miss. However, a new tool is changing the game: automated listening devices that record the soundscape day and night, paired with computer programs that can learn to recognize specific insect calls, offering a way to monitor biodiversity on a massive scale.
In the tropical savannas of northern Queensland, Australia, a team of researchers put this technology to the test. They wanted to see if a computer program, originally designed to identify bird songs, could be trained to recognize the calls of five different species of cicadas. These insects, including the lesser bladder, the eastern rattler, the corroboree cicada, the northern double-drummer, and the fairy dust squawker, live in a complex acoustic environment filled with wind, rain, and the songs of other animals. The researchers set up eighteen recording stations across the region, capturing sound continuously for fifteen months. This resulted in a massive library of audio, totaling about twenty-four thousand hours of recording. Instead of listening to every minute of this tape, the team used a deep-learning system to scan the recordings. They taught the computer what each of the five cicada species sounded like using a few clear example calls, and then let the system search through the thousands of hours of field recordings to find matches.
The results showed that the computer was remarkably good at the job. The system successfully identified the calls of all five cicada species with high accuracy, distinguishing them from the background noise and the songs of other insects. The researchers found that the program could reliably tell the difference between a true cicada call and a false alarm, even when the sounds were mixed together. This success meant they could now look at the data to see exactly when and where these insects were singing. They discovered that each species had its own unique schedule. The eastern rattler and the corroboree cicada were active during the day, with the eastern rattler building up to a loud chorus in the mid-morning and the corroboree cicada singing continuously from early morning until evening. The northern double-drummer had a more complex pattern, singing in the morning, taking a break, and then singing again in the afternoon and early evening. In contrast, the lesser bladder cicada was mostly an evening singer, calling as the sun went down, while the fairy dust squawker was active during the day but stopped before dusk.
Perhaps the most surprising finding was how these insects behaved when the weather turned bad. While it is often assumed that insects stop singing when it rains, the automated detectors showed that these cicadas kept calling even during light, moderate, and heavy rain. In fact, the researchers noted that the choruses sometimes seemed to get louder as the rain intensified. This suggests that avoiding rain to listen for these insects might cause scientists to miss a significant amount of activity. The study also revealed that the cicadas did not strictly avoid singing at the same time as each other, even when their songs overlapped. Instead of taking turns to avoid confusion, they relied on differences in the pitch and structure of their calls to be heard. This finding challenges the idea that insects always separate their singing times to avoid interference, showing instead that they can share the acoustic space effectively.
By proving that automated systems can accurately track these insects in the wild, the study offers a powerful new way to monitor biodiversity. The method used was efficient, requiring far less human time than traditional surveys while providing a detailed picture of insect life over long periods. The researchers found that the technology worked well even in the complex sounds of the tropical savanna, where many different species sing at once. This approach allows scientists to gather data on a scale that was previously impossible, helping to build a clearer understanding of how insect populations are changing. As the world faces a loss of insect diversity, having a tool that can listen to the forest and count its singers automatically provides a vital way to keep track of these essential creatures and the ecosystems they support.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.