Using AI to support Wildlife Conservation

The West African country of Gabon is not only home to 2 million people, its spectacular tropical forest is the roaming land of wild animals, including 1,000 species of birds and 400 species of mammals, like mandrills, western lowland gorillas and forest elephants.

In fact, an estimated 70% of the world’s remaining forest elephants live there, and many conservationists believe the country offers the best hope for reversing the losses of Central Africa’s wildlife. 

Unfortunately, many animals that call the region home are currently threatened with extinction, with one of the biggest threats being uncontrolled logging and poaching. However,  the country has shown a strong political will to conserve its wildlife; In 2002, a network of 13 national parks was established covering 10 percent of the country. Later, in 2007, a National Parks Agency was created and is doing everything in its power to curb these pressing threats. 

How can AI help to protect precious wildlife?

Saving species on the brink of extinction in the world’s second-largest rainforest is a daunting  task. Conservationists have deployed automated cameras to monitor some of the iconic and most endangered species – African forest elephants, gorillas, chimpanzees, and pangolins. Though it took anywhere from months to years to analyze the data, which resulted in a system that was never up to date.

Camera traps and artificial intelligence can be used to tackle these problems, but until recently it was severely limited by technical obstacles, particularly when it came to transmitting camera trap data for real-time analysis. Due to these barriers, the impact of existing technology used by Gabon National Parks was limited. The management team sought  for technical support that would enable them to do more.

Valuable tech support provided by Appsilon

Appsilon, a company that provides R Shiny apps and ML solutions for enterprise business, stepped in to see what solutions could be deployed. As a part of  their Data4Good program, which supports scientists and organizations working on solutions to tackle biodiversity loss, Appsilon provided pro-bono services and developed an easy-to-use, open-source, scalable software tool called Mbaza AI. This  tool was built in collaboration with researchers at  the University of Stirling and The National Parks Agency of Gabon (ANPN). 

The images provided by camera traps are processed with Mbaza on the rangers’ or ecologists’ regular laptops. The tool has helped to speed up  several weeks of work, to completion in a single day. In this way, there is more time for further analysis of the data, leading to better insight into the ecosystem and thus better-informed decisions.

The camera traps and Mbaza software can be deployed in a variety of environments while requiring only a small training dataset of the regional wildlife. The result of the ML algorithm achieved 96% accuracy in predicting out-of-sample data for many of the species.

Mbaza AI is already having a real impact

When Mbaza AI was launched and tested in Lopé and Waka National Parks in Gabon in 2020, the algorithm was used to analyze more than 50,000 images captured by 200 camera traps spread across 7000 km2 of contiguous forest area. Currently there are hundreds, if not thousands, of camera traps deployed by different organizations in the forests of Gabon and other West and Central African countries, such as Salonga National Park in DRC (Democratic Republic of Congo) and Taï National Park on the Ivory Coast. The great news is that researchers around the world can freely access this tool because it is open-source. This means the possibilities for large-scale adoption are unlimited.

Dr Whytock of Sterling University predicts that Mbaza will be the core technology for these types of projects in the next 5 years. 

The Mbaza AI model can classify about 3,000 images per hour on desktop computers used by park rangers and ecologists in the field, without access to powerful cloud computing resources. 

The model’s performance has been carefully studied by ecologists and researchers at Appsilon and Sterling University, and published in a peer-reviewed journal.  
Ecological metrics such as:


  • Species richness – to quantify species count both temporally and spatially
  • Activity patterns – to determine activity and life behavior traits
  • Occupancy – a hierarchical model to account for imperfect detection


have been shown to be reliable, even when based on automated processing of the images with the model.

 

What’s next for Mbaza AI?

Increasing Mbaza AI adoption is the main priority for Appsilon this year. This aligns with the plans of Gabon’s government leaders, who want to legislate Mbaza AI as part of the country's main private sector biodiversity monitoring methods.

Moreover, Mbaza AI can be easily used as shown in Gabon by other countries in West and Central Africa without requiring any changes. However, by incorporating some adjustments to the Gabor model, this tool can be used in any country in the world. 

Mbaza AI left one important problem unresolved – transmitting images for real-time data analysis in remote areas so  as a result, a new model emerged from Appsilon’s ML Team and has been used by researchers from the Netherlands (Thijs Suijten and Tim van Deursen from Hack the Planet, Q42) in which theAI immediately tags an image and sends an ‘alert’ message over a satellite network with the tag and metadata. The pilot test has already proven to be a success, opening the door to further reducing the significant administrative burden.  

Mbaza AI is just the beginning of a long-line of AI solutions that can be used to protect wildlife, and we’re excited to see what the future holds in this area. 

 

Help NGOs make an impact at scale. Find out more about how to become a member of Tech To The Rescue.



Would you like to learn more
about Tech To The Rescue?

Learn how others are helping  organisations with Tech to the Rescue