A team of experts in artificial intelligence and animal ecology has developed a new cross-disciplinary technique to better research wildlife species and make better use of the huge amounts of data that are already being collected due to new technology. Their findings were published today in Nature Communications.
Animal ecology has entered the age of big data and the Internet of Things. Due to advanced technology such as satellites, drones, and terrestrial equipment such as autonomous cameras and sensors placed on animals or their surroundings, unprecedented amounts of data on wildlife populations are now being acquired. These data have become so simple to collect and transmit that they have reduced the distances and time necessary for studies while also reducing the distracting presence of humans in natural environments. Today, various AI methods are available to analyze enormous datasets, but their characteristics are often broad and unsuitable for investigating the specific behavior and appearance of wild animals. A group of scientists from EPFL and other institutions has described a novel approach to overcoming this challenge and producing more accurate models by combining advances in computer vision with ecologists’ experience. Their findings, published today in Nature Communications, provide new perspectives on the use of artificial intelligence to aid in the conservation of endangered species.
increasing interdisciplinary knowledge
Wildlife research has progressed from a local to a global scale. Today’s technology allows for more precise estimates of wildlife populations, a better understanding of animal behavior, a reduction in poaching, and a reversal of biodiversity loss. Ecologists may use AI, particularly computer vision, to extract critical elements from images, videos, and other visual forms of data to efficiently identify wildlife species, count individual animals, and glean specific information from massive datasets. Many of the generic systems currently used to analyze this type of data operate like black boxes, failing to take advantage of all available animal behavior and biology information. Furthermore, they are difficult to personalize, can suffer from poor quality control, and maybe exposed to ethical issues associated with the handling of sensitive data. They also have several biases, particularly regional biases; for example, if all of the data used to train a given program was collected in Europe, the program may not be suitable for other parts of the world.
“We wanted to get more scholars interested in this issue and combine their efforts to move this burgeoning discipline forward.” Prof. Devis Tuia, the study’s lead author and head of EPFL’s Environmental Computational Science and Earth Observation Laboratory, believes AI can be a game-changer in wildlife research and environmental protection. To reduce the margin of error of an AI program that has been trained to recognize a specific species, for example, computer scientists must be able to draw on the knowledge of animal ecologists.
These experts can specify which characteristics should be considered in the program, such as whether a species can survive at a given latitude, whether it is important for the survival of another species (via a predator-prey relationship), or whether the species’ physiology changes over time. For example, new machine learning algorithms can automatically identify an animal, such as a zebra’s unique stripe pattern, or their movement dynamics in the video can be a sign of identity. Prof. MackenzieMathis, the Bertarelli Foundation Chair of Integrative Neuroscience at EPFL and a co-author of the study, says, “This is where the intersection of ecology and machine learning comes into play: the field biologist has vast domain knowledge about the animal being studied, and our job as machine learning researchers is to collaborate with them to build tools to find a solution.”
Disseminating information about existing initiatives
Tuia, Mathis, and others addressed their research issues at numerous conferences over the last two years, and the concept of strengthening linkages between computer vision and ecology arose. They recognized that such collaboration could be extremely beneficial in preventing the extinction of certain wildlife species. A few projects have already been undertaken in this approach, some discussed in a recent paper published in Nature Communications. For example, Tuia and his colleagues at EPFL have created software that can detect animal species based on drone photographs. It was recently put to the test on a bunch of seals.
Meanwhile, Mathis and her colleagues have released DeepLabCut, an open-source software program that enables scientists to estimate and monitor animal positions with surprising precision. Over 300,000 individuals have already downloaded it. DeepLabCut was created for lab animals. However, it may also be used on other species. Other institutions’ academics have built programs as well, but they cannot share their findings because there is no actual community in this field. Others in the scientific community may be unaware of these programs or whether one is best suited to their specific study.
Having stated that, the first steps toward such a community have been done via numerous internet forums. The Nature Communications paper, on the other hand, is aimed at a larger readership of researchers from around the world. “A community is gradually taking shape,” Tuia explains. “Up until now, word of mouth has been the primary method of establishing an initial network.” We began two years ago with the people who are now the article’s other lead authors: Benjamin Kellenberger at EPFL, Sara Beery at Caltech in the United States, and Blair Costelloe at the Max Planck Institute. In Germany.”