Meteorites, those elusive remnants of our early solar system, have long intrigued scientists with the secrets they hold about the universe’s formative years. However, hunting for these celestial treasures has traditionally been a labor-intensive and time-consuming endeavor. Researchers often resorted to manually scouring vast landscapes, which could take considerable effort and yield limited results. Fortunately, the convergence of technology and innovation is now transforming meteorite hunting. Scientists are increasingly turning to the power of drones and machine learning to revolutionize the way we locate freshly fallen meteorites.
Seamus Anderson, a planetary scientist at Curtin University in Perth, Australia, notes that a typical meteorite-hunting team of six individuals can search an area of approximately 200,000 square meters in a single day. However, the challenge lies in the imprecise location of meteorite clusters, often covering millions of square meters. As a result, the traditional search method proves to be a painstakingly slow process.
Around 2016, Anderson began exploring the concept of using drones as a tool to expedite meteorite discovery. What started as an idea eventually evolved into a Ph.D. project. In 2022, Anderson and his colleagues celebrated their first successful recovery of a meteorite spotted with a drone. Subsequently, they discovered four more meteorites at a different site, presenting their findings at a meeting of the Meteoritical Society in Los Angeles on August 17.
The integration of drones into meteorite searches has significantly accelerated the process. Anderson explains that the shift from human-driven efforts to drone-based searches reduces the time required from approximately 300 days to a mere dozen or so. This approach not only enhances efficiency but also adds an element of excitement and adventure to the work.
Challenges in drone-assisted meteorite hunts
Nonetheless, there are challenges associated with this innovative approach. Anderson’s team conducts their drone-assisted meteorite searches in remote regions of Western Australia and South Australia. The process commences with information from ground-based cameras that monitor meteoroids entering Earth’s atmosphere, providing vital data about fall sites. Once alerted, the researchers embark on a challenging journey, often traversing rough or non-existent roads that can take more than a day to navigate.
Upon arrival at the fall site, the team deploys their primary drone, flying it at an altitude of roughly 20 meters. The drone’s camera captures an image of the ground every second, with data downloaded approximately every 40 minutes when the drone lands for battery replacement. A typical day of flying can generate over 10,000 images, which are subsequently digitally divided into millions of smaller sections, each measuring 2 meters on a side.
The critical element in this process is a machine learning algorithm specially trained to recognize meteorites based on images of genuine meteorites or terrestrial rocks artificially painted black to mimic meteorites. While the algorithm proves effective, it is not infallible. It automatically discards most sections (typically more than 99 percent) that do not exhibit meteorite-like characteristics. Nevertheless, this still leaves around 50,000 sections for human manual review.
Many of these sections contain items that are distinctly non-meteoritic in nature, such as animal excrement, discarded tin cans, snakes, or even sleeping kangaroos. Anderson clarifies that these objects trigger false positives because the algorithm lacks familiarity with them. Consequently, it falls upon the research team to painstakingly sift through these potential meteorite candidates.
For those sections that still appear promising to the human eye, a smaller drone is deployed to investigate at a much lower altitude of about one meter above the ground. Finally, the research team physically visits the location to examine and verify potential meteorite findings.
Expanding possibilities: The future of meteorite discovery
The researchers are committed to refining their algorithm to reduce the likelihood of erroneously flagging items like animal waste or kangaroos as meteorites. Additionally, they aim to make their computer code open-source, facilitating its use by other researchers interested in this groundbreaking approach.
Anderson envisions the potential for drone-assisted meteorite searches in Antarctica, a region known for its rich meteorite deposits. However, he acknowledges that Antarctica presents an entirely new set of challenges, including ensuring the durability of sensitive electronic equipment in the harsh, frigid conditions and overcoming logistical hurdles associated with working in such a remote and challenging environment. As he aptly puts it, “Antarctica is a whole different beast.”