드론과 AI가 야생동물 생존 및 관리 방식을 재작성하고 있다

인공지능(AI)의 힘이 점점 더 많이 멸종 위기 종을 보호하는 데 활용되고 있습니다.

The same technology that many fear could one day cause job displacement or even pose a threat to humanity is now being used to save animals. AI is coming to the defense of endangered species across the globe in a myriad of ways, including tracking movement patterns and water loss in wetlands and rivers, enhancing anti-poaching efforts, developing advanced warning systems, and counting species using classification and surveillance techniques.

Through all these efforts, AI has helped save the dwindling populations of 코끼리, 물고기, 천산갑, 코뿔소, 붉은 늑대, 플로리다 팬서, and many more.

AI는 방대한 데이터를 분석하고, 추세를 포착하며, 시간이 지남에 따라 생태계를 모니터링함으로써 취약한 종을 찾고, 식별하고, 보호할 수 있습니다. 기존 방법은 생태계를 교란하고 많은 시간, 인력, 자원을 필요로 하지만, AI는 이를 빠르고 효과적으로 수행합니다.

멸종 위기에 처한 종이 백만 종에 달하고 생물다양성이 급격히 감소하고 있는 상황에서, AI는 보존 노력을 지원하는 강력한 도구를 제공합니다. 효율성 향상, 빠른 데이터 처리, 자동화된 야생동물 모니터링, 향상된 위협 탐지, 실시간 알림, 더 나은 의사결정, 확장 가능한 데이터 공유 등은 멸종 위기 종을 보호하는 방식을 혁신할 수 있습니다.

그 결과, 연구자들은 생물다양성을 모니터링하고 멸종 위기 종을 돕기 위한 노력에 AI를 활용하고 있습니다. 

플로리다 대학교 연구진이 수행한 최신 연구가 바로 그 예입니다. 그들은 AI를 사용해 아마존에 숨겨진 최대 41,000마리의 거북이 둥지를 발견했습니다. 이 발견은 스마트 모델링과 드론을 통해 가능해진 세계 최대 규모의 거북이 둥지로 기록됩니다.

혁신적인 기술을 항공 이미지와 통계 보정과 결합함으로써 기존 카운팅 기법의 주요 한계를 해소하고 야생동물 모니터링을 보다 정확하게 만들었습니다.

“우리는 동물 개체수를 보다 효율적으로 모니터링하는 새로운 방법을 제시합니다.”라고 연구의 주요 저자이자 UF 식품·농업 과학 연구소(UF/IFAS) 산림·어업·지리과학 학교의 박사후 연구원인 이스마엘 브랙이 말했습니다. “비록 이 방법은 거북이 개체수를 세는 데 사용되었지만, 다른 종에도 적용할 수 있습니다.”

계절적 집합: 정확한 야생동물 개체수 파악의 핵심

개체군 동태를 연구할 때, 즉 종이 성장·감소·이동하는 방식, 포식자-피식자 관계와 종간 상호작용을 이해하고 서식지 전환 및 전 지구적 기후 변화의 영향을 분석할 때, 풍부함(abundance)은 생태학 및 보전 분야의 기본 변수입니다.

By monitoring it over time, we can also detect and predict trends in populations of invasive or threatened species.

While knowing how many species are out there helps in tracking changes, identifying threats, and measuring the success of protection or control efforts, estimating this abundance is very difficult, especially in large areas where species are rare, elusive, or widely dispersed. This makes it hard to find and count species accurately.

An effective way to enhance the efficiency and accuracy of these efforts to estimate and monitor abundance is by counting animals during the periods of spatial aggregation.

What this means is that several wildlife species show seasonal behaviours in which they concentrate in small areas to rest, mate, breed, nest, and interact socially, providing the perfect opportunity to count them. For instance, turtles come together to nest on beaches and sandbanks.

To sample these spatially aggregated wildlife populations, drones are being used as an efficient and less invasive method.

Drones, also known as unmanned aerial vehicles (UAVs) or remotely piloted aircraft (RPAs), have proven to be more precise and accurate in counting species gathered in one place. They also cause less disturbance to animals compared to ground-based surveys.

To use drones, flight paths are planned to cover the entire area where the species are gathered. Overlaps are maintained between successive photos and lateral strips, allowing all the collected images to be merged into a single orthorectified mosaic.

Combining many smaller images with distortions removed to create a large, highly detailed, high-resolution, map-quality image makes for an orthorectified mosaic.

Counting wildlife individuals in orthomosaics during aggregation events, however, is subject to unintentional errors, which can result in biased estimates.

While it is an expeditious, less invasive, and more precise way to count animals than doing so from the ground, this technique does not account for the fact that animals sometimes move during observation.

For instance, an animal may be hidden by vegetation or simply be somewhere else temporarily when the image is collected. Even if the animal is in the image, it may not be detected by the algorithm or a human observer. Another possibility is that moving animals appear multiple times in the photos.

An important factor here, according to the latest study, is that these species concentrations are commonly temporary, with individuals arriving and leaving over the course of days due to nesting, breeding, or migration, causing fluctuations in population size.

The resulting errors from this “open population” can give us wrong numbers, with the concerning part being that “these errors are widely overlooked in abundance estimations derived from orthomosaic counts of drone-based surveys.”

So, the University of Florida researchers wanted to create an approach that accounts for multiple sources of error. For that, they are using two types of datasets: resightings of marked animals and overall population counts.

Aerial Surveillance & Smart Modeling Revolutionizing Population Estimates

In collaboration with non-governmental New York-based Wildlife Conservation Society (WCS) researchers in Colombia, Brazil, and Bolivia, the project began with a focus on Giant South American River Turtles (Podocnemis expansa), also called the giant Amazon River turtle, river turtle, or simply the Arrau.

게시일 in the Journal of Applied Ecology, the research¹ was driven by the need to estimate the abundance of river turtles and have a monitoring protocol for them during the world’s largest known aggregation of freshwater turtles. 

River turtles have experienced historical declines, either disappearing from many tributaries of the Amazon and Orinoco Rivers or being present in much lower densities.

Their population has declined substantially, primarily because of their overexploitation by poachers for meat and egg consumption. As a result, their large aggregations have now become rare.

Still, there are some large populations of this species across its range, and some of them seem to be recovering, with their seasonal behavior providing an invaluable opportunity to monitor their populations. 

Thousands of these social creatures gather every year during the dry season (July or August) to nest in sandbanks of the Guaporé River, along the Brazil-Bolivia border. 

In order to estimate their numbers, previously experts relied on counting hatchlings once they emerged, based on which the number of females is extrapolated, using the average number of eggs per nest. This is an invasive and time-consuming method due to fencing the perimeter and manipulating hatchlings. 

Also, individual nests can’t be distinguished from each other, making it not only challenging but even impossible to estimate the numbers in areas with considerable mass nesting.

There’s another way, visual counts of adult turtles from the ground, but this one also presents the difficulties of constant movement and getting obstructed by each other.

Here, drones, which are being tested to survey river turtle populations, have been showing great promise as an efficient and precise method to estimate their population sizes during the nesting events, which is important to assess the population trends and the effectiveness of conservation actions.

So, the researchers applied the modelling approach they have developed to determine the population of river turtles when they come together for nesting. 

By accounting for multiple sources of errors, it offers a new method for ecologists to monitor at-risk animals with more accuracy.

The novel approach, according to researchers, offers several advantages, including the aerial image to count the river turtles without any obstructions. The use of a less invasive technique also reduces animal disturbance. 

Moreover, the approach provides a uniform approach that can be applied and compared across different sites and different years. Given these benefits, the researchers expect to see a protocol similar to theirs being used by government and non-government institutions to monitor the species.

드론 기술이 빈번한 오용에도 불구하고 하늘 높은 잠재력을 보유한 이유를 알아보려면 여기를 클릭하세요.

A Smart, Scalable, Error-corrected Model to Monitor Global Wildlife

To count the turtles, researchers marked the shells of 1,187 river turtles with white paint, and over a period of twelve days, they flew a drone overhead, following an exact path, back-and-forth, four times a day.

The drone took 1,500 pictures each time, which were stitched together using software. The researchers then reviewed the composite images. Each turtle was recorded by them as well as whether its shell was marked, and whether the animal was walking or nesting when photographed. 

Using this data, they developed probability models that account for multiple sources of error. It used mark-resight data and overall population counts to account for individuals unavailable for detection during flight, open population (the constant joining and leaving) during the nesting event, marked individuals detected in the mosaic with unidentifiable marks, and double counts due to the orthomosaic building process.

Thus, the team estimates that the daily nesting probability is 0.37 and that 35% of river turtles that used the sandbank at night are also present during the drone’s morning flight. 

Additionally, they found that 20% of the turtles walking in the orthomosaic are double counts, and the probability of identifying the mark was 0.78. This way, the novel approach provides a more accurate way to count wildlife using drones.

When counting the turtles, on the ground observers reported about 16,000 turtles, while researchers who reviewed the orthomosaics without accounting for errors counted about 79,000 turtles.

But using the technique, the researchers estimate the total abundance for the aggregation site to be 41,377 turtles. According to Brack:

“These numbers vary greatly, and that’s a problem for conservationists. If scientists are unable to establish an accurate count of individuals of a species, how will they know if the population is in decline or whether efforts to protect it are successful?”

While the estimates represent a large number of river turtles, the researchers note that it is likely to be a fraction of their historical populations in the Amazon region, based on historical records of exported eggs. Not to mention, the nesting event also continued for some days after the last drone flight.

As such, the study recommends extending the usage of the monitoring tool throughout the entire nesting period. Also, other sandbanks in the region should be included for a comprehensive estimate of the nesting population.

In regard to this, the research team plans to have more drone flights at the Guaporé River nesting site as well as in other South American countries where the river turtles gather, such as Colombia, and possibly Venezuela and Peru. This will help the team improve its monitoring methods.

“By combining information from multiple surveys, we can detect population trends, and the Wildlife Conservation Society will know where to invest in conservation actions.”

– Brack

While the framework developed was initially driven by the need to improve the monitoring of river turtles, the researchers noted that it is “very versatile and can be readily used or adapted to several different contexts.”

Besides river turtles, the developed methodology can also be applied and adapted to the conservation efforts involving other threatened species surveyed using drone-based orthomosaics.

For instance, previous drone monitoring studies clipped seals’ fur, marked mountain goats and bison with paintball pellets, and attached collars to elk to track their movement during counts.

Ultimately, the new model can be used for the efficient and timely monitoring of abundance in wildlife conservation and management programs.

Investing In Conservation Tech

The AI darling NVIDIA Corporation (NVDA ) is playing a big role in saving animals and our planet. 
Its GPUs power many of the deep learning models used in image recognition, object detection, and environmental monitoring software. The company even promotes the usage of AI for the global good, including biodiversity research.

NVIDIA Corporation (NVDA )

Now, among the companies utilizing Nvidia’s tech, the AI research institute Ai2 has developed EarthRanger to make more informed operational decisions for wildlife conservation in real time. The world’s largest elephant database is trained on NVIDIA Hopper GPUs. It also displays data on a large number of wildlife, aggregated from radios, satellites, camera traps, acoustic sensors, and more data sources. 

Ai2 recently also released an open-source AI model named Atlantes to analyze more than five billion GPS signals a day emitted from nearly 600,000 ocean-going vessels and predict what any of these vessels is doing with about 80% accuracy. If a vessel is engaged in illicit fishing, the model sends alerts to the coast guards. The 4.7M parameter transformer-based model, Atlantes, is trained on NVIDIA H100 Tensor Core GPUs and PyTorch.

Rouxcel Technology’s AI-based RhinoWatches are trained and optimized using NVIDIA accelerated computing. It is deployed across over 40 South African reserves and is being expanded in Kenya and Namibia. The company is currently developing AI models for more species, including the critically endangered pangolins.

The NVIDIA CUDA and Jetson modules, meanwhile, are being used for edge AI and data processing by OroraTech, which combines data from satellites, cameras, aerial observations, and local weather information to monitor animal poaching and wildfires and provide alerts in real time.

But that’s not all. Over the years, Nvidia tech has been used for many other interesting experiments, including de-extinction. For instance, Colossal Biosciences has been using gene editing technology, AI models, and the NVIDIA Parabricks software suite to bring back the dodo bird, the woolly mammoth, and the Tasmanian tiger.

Besides wildlife, Nvidia technology is helping scientists, researchers, and developers gain a better understanding of climate, oceans, and space.  

With a market cap of $4.39 trillion, the full-stack computing infrastructure company’s shares are currently trading at $180.95, up over 34% YTD. 

The company’s share price has surged more than 59% over the past three months. Just on the last day of July, the stock hit a 52-week high of $183.30, which shows continuing strong investor confidence in the company and its future prospects.

With that, it has an EPS (TTM) of 3.10 and a P/E (TTM) of 57.98, while the dividend yield offered is 0.02%.

For the first quarter ended 2025년 4월 27일, Nvidia reported revenue of $44.1 billion. The main driver of it is data centers, accounting for $39.1 billion of the revenue, which makes up a whopping 89% of total company sales. This was fueled by the explosive demand for AI.

This growth has been despite Nvidia facing geopolitical setbacks with export restrictions on its H20 chips in China. These chips are likely to return to China with the Trump administration assuring the company that it would be permitted to resume sales. Nvidia also announced a new “fully compliant” GPU for China.

However, Nvidia may still struggle to regain its former market share, with Bernstein forecasting Nvidia’s AI chip market share in China to drop from 66% last year to 54% this year.

Latest NVIDIA Corporation (NVDA) Stock News and Developments

Conclusion

건강하고 안정적인 지구를 유지하려면 멸종 위기 종을 보호하는 것이 필수적이며, 이들의 손실은 전체 생태계에 연쇄적인 영향을 미칠 수 있습니다. 멸종 위협이 가속화됨에 따라 효과적인 모니터링을 구현하는 것이 그 어느 때보다 중요합니다.

여기서 드론과 스마트 모델링 기술의 통합은 큰 변화를 의미합니다. 종 모니터링의 정확도와 효율성을 향상시킴으로써, 이러한 기술 혁신은 가장 취약한 야생동물을 보호하기 위해 더 빠르고, 더 똑똑하며, 더 전략적으로 행동할 수 있게 해줍니다.

투자할 최고의 드론 기업 목록을 보려면 여기를 클릭하세요.

References:

1. Brack, I.V., Valle, D., Ferrara, C., Torrico, O., Domic‑Rivadeneira, E., & Forero‑Medina, G. Estimating abundance of aggregated populations with drones while accounting for multiple sources of errors: A case study on the mass nesting of Giant South American River Turtles. Journal of Applied Ecology, first published 17 2025년 6월. https://doi.org/10.1111/1365-2664.70081