top of page
Wildlife Conservation

Machine Intelligence for
Wildlife Conservation

We are living at the time of the sixth mass extinction, the one caused by human activity [1]. Since we humans are part of the earth’s complex ecosystem, the extinction of animal and plant life will drastically affect us. Therefore, we must look after nature and protect its wildlife. Machine Intelligence (MI) can support us in this task.

According to recent research [1], the sixth mass extinction of wildlife on Earth is on the rise. More than 500 species of land animals are likely to vanish from the earth’s surface within the next twenty years. That is the same number of species that disappeared over the entire 20th century. If not for us, an extinction on this scale would have taken thousands of years! But these 500 land animal species are only the tip of the iceberg: there are up to one million species of plants and animals which are threatened with extinction [2]. Scientists warn that if we do not rapidly intervene this mass extinction could become unstoppable and become the tipping point for the collapse of human civilization [1].

We humans depend on biodiversity for our health and wellbeing. It is a well-known fact, for example, that agriculture strongly relies on a healthy bee population for cross-pollination; major sources of medication are plants and certain animals (e.g., the cobra, for its venom); and a lot of countries have economies which are strongly dependent on tourism, which is often driven by stunning nature and diverse wildlife (even though, the tourism itself often causes environmental destruction). The rise in human population, destruction of habitats, trade in wildlife, pollution and the climate crisis endanger the fragile ecological stability of the earth. Because of the complexity of the ecosystem, the extinction of just a few species could result in a domino effect, in which other species would also disappear and which could ultimately lead to the extinction of humanity itself. The current COVID-19 pandemic is only one extreme example of how ravaging nature can be when it strikes back [3]. Thus, it is in our own interest to protect wildlife and ensure the stability of the earth’s ecosystem. MI can considerably support us in tackling this issue.

The fight against poaching

 

Poaching is one of the major risks to the survival of endangered species. One kilogram of ivory can fetch $2,000 on the black market and the same amount of rhino horn as much as $50,000 – 60,000 [4, 5]. Thus, it is not a surprise that the trade in wildlife has a value of, according to some estimates, between seven and twenty-three billion dollars each year, making it the fourth most lucrative crime in the world, after the trafficking of drugs, humans and arms [5]. Fighting such a large criminal business is not an easy task and overstretched and under-resourced park rangers often struggle to keep pace with the criminals. However, there is a light at the end of the tunnel: some rangers have already started using MI to make their efforts more efficient and effective.

Statistics on poaching of rhinos and elephants

Some statistics on poaching of rhinos and elephants. Source: [5]

At Carnegie Mellon University, for example, scientists have developed a Machine Learning (ML) system which helps rangers to plan their patrol routes more effectively [6]. The system, called PAWS (Protection Assistant for Wildlife Security), takes as input information about a protected area together with past patrolling routes and poaching activities in that area to generate new, randomized patrol routes. In this way, the system learns the poachers' behavior and can suggest routes that are likely to be the most effective for finding poachers. The system can also utilize game-theoretical reasoning to predict how poachers will adapt their behavior to the more sophisticated rangers’ patrol routes (as they evolve) and can incorporate topographical information about complex terrains so that the suggested routes are indeed accessible for rangers. This approach has already proven to be successful. As an example of benefit of PAWS, in Cambodia, it has helped rangers find and confiscate one thousand snares: twice the number they usually confiscate in the same period of time.

Another measure to make the rangers’ work more efficient is to identify in real time the locations in which poachers are active. The non-profit organization Wildlife Protection Solutions (WPS) [7] uses a network of cameras installed around protected areas. Photos shot by these cameras are analyzed by an Artificial Intelligence (AI) algorithm, capable of distinguishing animals from humans. If humans are detected in an area where they are not supposed to be, local rangers get an alert via SMS, so they can quickly react to unlawful activities. Another poacher detection system, which uses a similar approach, is TrailGuard AI, has been developed by the non-profit organizations RESOLVE and Intel [8]. Their Computer Vision AI system was tested in the Grumeti Reserve in Tanzania. In the test phase alone, it helped in the location of thirty poachers, leading to their arrest and the seizure of nearly 600 kg of illegal bushmeat. The next step will be to expand the application of such systems to further parks and countries.

Another promising application of Computer Vision is the use of drones to collect data in protected areas. Drone video can be processed by AI to identify humans or vehicles involved in suspicious activities. Indeed, there are already such systems in development. One example is a system which is currently being developed in a collaboration between UAVAid [9], a UK startup that develops drones for humanitarian assistance work, and Archangel Imaging, an AI surveillance company [10]. The use of AI-powered drones to detect poaching is in its early stages, but it is conceivable that soon solutions like these will replace the more expensive methods currently in use, such as surveillance helicopters.

An alternative to detecting poachers with images or videos is to do so from sound-monitoring. Such systems can be trained to detect specific sounds from acoustic recordings. Since the sounds typical of poaching activities, e.g., chainsaws, gunshots or motor vehicles, are very different from those of the natural environment, AI can easily identify them and alert rangers to poaching in real time. One example of such a sound-monitoring AI system was developed by Rainforest Connection. It can be used not only to prevent poaching but also in the fight against illegal logging [11].

Understanding wildlife

In order to be able to protect wildlife we need to better understand the behavior and population of animals. However, field observations can be very expensive and time consuming so that only a small fraction of animals can be observed and only a small amount of data can be collected. Gathering a critical mass of data for population analysis can take years. This is too slow for effective and efficient conservation actions to take place and it limits the scope, scale, repeatability, continuity, and return on investment (ROI) for these studies. MI-driven analysis platforms can improve the situation significantly. For example, the non-profit startup Wild Me provides a Computer Vision platform which can identify different species from photos or high-resolution satellite images [12]. Using this technology, it becomes easier to collect data about the population of a particular species and their distribution across the planet. It is even capable of identifying individual animals so that scientists can get better insights into their migratory routes.

Artificial Intelligence identifies individual animals for wildlife conservation

One of the Wild Me projects, flukebook, uses AI to identify and track individual whales and dolphins across hundreds of thousands of photos. Here, AI identifies an individual whale based on unique fin edges [13].

But what about animals that are difficult to catch on photos? For example, forest elephants are very shy creatures, which conceal themselves in dense forests most of the time.  Even though these elephants are quite large, they are masters of camouflage, so that you can pass by them, meters away, without noticing them. This behavior makes it very difficult to observe them and to count their population. The Elephant Listening Project tackles this problem with AI [14]. Researchers install recorders in forests in order to take audio of forest sounds and to identify voices of elephants. This system captures an enormous amount of audio data which, from quantity alone, would be difficult to process by humans. Here, an AI-driven sound monitoring system which has been trained to detect elephant voices comes into play. Using this approach researchers can process hundreds of thousands of hours of jungle sounds within a short period of time.

Conservation of plants

 

Animal population is a quite prominent and well-funded research topic. For example, you can find population information for a lot of animal species on the IUCN Red List of Threatened Species, which is the largest and most comprehensive listing of species’ conservation status [15]. However, if you are interested in the conservation status of plants, this information is much harder to find. Only a small proportion (ca. 5%) of all known plants is included in the Red List. This underrepresentation of plants can be largely attributed to the unequal distribution of resources: a charismatic panda tends to receive more attention and financial resources than the average wildflower. But this underrepresentation does not mean that plant species are of less importance. On the contrary they are vital for ecosystems, wildlife’s food chains, and agriculture. Thus, research on the conservation status of plants is at least as important as similar research for animals.

A team of researchers led by Dr. Anahí Espíndola from University of Maryland has tackled this problem using an MI approach [16]. First, they used open-source databases to collect geographical, environmental, and morphological information on plant species whose conservation status was already known. Then, they utilized these data to train an ML-model and make it capable of determining the conservation status of a given plant based on geographical, environmental, and morphological information of its species. Using this approach, researchers were able to make predictions on the previously unknown conservation status of more than 150,000 land plant species, worldwide. The results of this study indicate that a large number of unassessed species are likely to be at risk of extinction. In this way, researchers were able to identify geographical regions with a very high need of conservation attention. Many of these regions had not been identified as regions of concern before.

The ML approach developed here helps prioritize the current process of assessing the conservation status of species and provide conservationists with information as to where research and resources should be allocated to make their efforts more effective and efficient.

Conclusion

 

The examples mentioned above represent just a small fraction of the potential use cases conceivable for applying MI for the protection of wildlife. Indeed, at MI4People we believe that there are many more potential applications waiting to be discovered and applied. Thus, think creatively and use the power of Machine Intelligence in your wildlife conservation organization [17].

bottom of page