How AI and Machine Learning Are Transforming Pangolin Conservation

Published 11 June 2026 | 8 min read

Pangolins are the most trafficked mammals on Earth. All eight species face severe population declines driven by poaching, habitat destruction and the illegal wildlife trade. Traditional conservation methods struggle to keep pace with the scale of these threats. Increasingly, researchers are turning to artificial intelligence and machine learning to bridge the gap, deploying tools that process vast quantities of data, detect hidden patterns and respond to threats in near real time.

From the bushveld of South Africa to international customs halls, AI is reshaping how we protect pangolins. This article examines the key technologies driving this transformation, with a focus on the southern African context.

AI-Powered Camera Trap Image Recognition

Camera traps are indispensable for monitoring elusive, nocturnal species like pangolins. A single camera trap deployment can generate tens of thousands of images over a few months, and national programmes accumulate millions. The bottleneck has always been the same: someone has to look at every photograph and identify what triggered the camera.

Convolutional neural networks (CNNs) now automate this process. Trained on large datasets of labelled wildlife images, these models recognise species based on body shape, texture and movement patterns. Pangolins' distinctive overlapping keratin scales and bipedal gait create strong visual signatures that CNNs detect reliably.

Platforms such as Wildlife Insights allow researchers to upload images for automated classification. The system assigns a species label and confidence score to each image, routing pangolin detections to human reviewers while filtering out empty frames automatically. This reduces manual review workload by up to 90 per cent -- essential for monitoring across Limpopo, KwaZulu-Natal and Mpumalanga where field teams are stretched thin.

Machine Learning for Wildlife Trafficking Detection

Pangolin scales are smuggled in shipping containers, concealed inside legal cargo and moved through multiple transit countries. Machine learning is being deployed at several points along this supply chain to improve detection rates.

Customs scanning and cargo screening

At ports and border crossings, X-ray scanning systems enhanced with machine learning can identify the distinctive density signatures of pangolin scales within sealed containers. Trained on X-ray images of known contraband, these models learn to distinguish biological materials from legitimate cargo. South African customs operations at Durban harbour and OR Tambo International Airport have explored AI-assisted screening to augment human inspectors.

Dark web and encrypted platform monitoring

Criminal networks use encrypted messaging, dark web marketplaces and social media to arrange trade. Machine learning tools crawl these platforms, identifying listings through image recognition and natural language processing that detects code words and euphemisms. Organisations including TRAFFIC and the International Fund for Animal Welfare (IFAW) use these tools to build intelligence on active networks and share it with law enforcement agencies including INTERPOL.

Acoustic Monitoring with AI

Microphone arrays deployed in pangolin habitats record the soundscape continuously. AI models trained on labelled audio data scan these recordings for specific sounds of interest.

Identifying pangolin vocalisations

Although generally quiet, pangolins produce vocalisations including hissing, puffing and distress calls. Machine learning classifiers detect these sounds within hours of continuous audio, providing evidence of pangolin presence where camera traps may not be deployed. This is particularly useful for arboreal species in dense forest canopy.

Detecting poaching activity

Acoustic AI also detects indicators of poaching: chainsaw noise, off-road vehicles at night, hunting dogs and gunshots. When detected in protected areas, alerts reach ranger teams in near real time. Piloted in several South African reserves, these systems function as acoustic tripwires complementing traditional anti-poaching patrols.

Predictive Modelling for Habitat and Poaching Hotspots

Conservationists use machine learning to build predictive models addressing two critical questions: where are pangolins most likely to occur, and where are they most at risk?

Species distribution modelling

By combining pangolin occurrence records with environmental variables -- vegetation type, soil composition, rainfall, temperature and proximity to termite mounds -- machine learning generates detailed habitat suitability maps. For Temminck's ground pangolin in South Africa, these models have identified previously unsurveyed areas with high habitat suitability, guiding new camera trap deployments and field surveys.

Poaching risk mapping

Separate models integrate socioeconomic data, road density, historical poaching incidents and enforcement patrol coverage to predict where poaching is most likely. These risk maps help conservation managers allocate limited ranger resources more effectively, mirroring techniques from the Protection Assistant for Wildlife Security (PAWS) system used in several African countries.

Satellite Imagery Analysis with Deep Learning

Monitoring habitat loss across the vast landscapes where pangolins occur requires tools that operate at scale. Neural networks trained on multispectral satellite data classify land cover types, detect deforestation, identify agricultural encroachment and track vegetation health changes over time.

In South Africa, this is particularly relevant to monitoring land-use change in the savanna habitats of Limpopo, North West and Mpumalanga where Temminck's ground pangolin occurs. Satellite-derived change detection flags areas where mining, agricultural expansion or settlement growth encroach on known pangolin habitat, enabling timely intervention by conservation authorities.

Natural Language Processing for Online Trade Monitoring

Pangolin products are advertised on e-commerce platforms, social media and messaging applications, often using coded language to evade keyword filters. Natural language processing (NLP) offers a more sophisticated approach.

NLP models understand contextual meaning, identifying trade-related posts even when species names are replaced with slang or euphemisms. Multilingual models are critical, as the trade spans Mandarin, Vietnamese, English and French language communities. In southern Africa, NLP tools monitor social media activity related to traditional medicine markets, providing intelligence that informs enforcement priorities.

South African Conservation Technology Initiatives

South Africa hosts a growing technology sector alongside globally significant pangolin populations and well-established conservation institutions, placing it in a strong position to lead AI-driven conservation.

The African Pangolin Working Group (APWG), based in South Africa, coordinates research for all four African pangolin species. The APWG has partnered with university computer science departments to explore AI-assisted monitoring, including automated camera trap classification and habitat suitability modelling for provincial conservation planning.

SANParks (South African National Parks) has invested in technology-enhanced anti-poaching operations. While much of this infrastructure was built for rhinoceros protection, sensor networks, drone capabilities and data analytics platforms are being adapted for broader wildlife monitoring including pangolins.

South African universities -- including the University of Pretoria, Stellenbosch University and the University of the Witwatersrand -- are producing research at the intersection of conservation biology and machine learning, training practitioners who bridge ecological fieldwork and computational analysis.

Challenges and Limitations

Despite its promise, AI in conservation faces significant obstacles that must be acknowledged.

Data scarcity and bias

Pangolins are rare and secretive, meaning labelled training datasets are far smaller than those available for common species. This limits model accuracy. Furthermore, models trained on data from one region may perform poorly elsewhere -- a classifier built on South African bushveld images may not work in West African forests. Diverse, representative training data is expensive to collect.

Cost and connectivity

Advanced AI requires computing power, technical expertise and ongoing maintenance. Many African conservation organisations operate on constrained budgets with limited cloud access. Real-time applications depend on internet connectivity, yet many pangolin habitats have poor or no cellular coverage. Edge computing offers a partial solution but with reduced processing power.

Ethical considerations

Surveillance technologies can capture data about local communities alongside poaching detection. Ensuring conservation technology respects the rights and privacy of people living near pangolin habitats is critical, particularly in South Africa where historical land-use injustices remain sensitive.

Future Directions and Emerging Technologies

Several emerging technologies are likely to become increasingly important for pangolin protection.

Federated learning allows models to be trained across multiple institutions without sharing raw data, addressing data scarcity and privacy simultaneously. African conservation organisations could collaborate on building better classifiers without centralising sensitive location data.

Autonomous drones with thermal imaging and onboard AI can survey large habitat areas for both pangolins and poachers during night-time hours. Trials in South African reserves have demonstrated drone-based thermal detection of ground pangolins in open savanna.

Environmental DNA (eDNA) analysis is increasingly paired with machine learning for species detection from soil and water samples. This non-invasive method could complement camera trapping and acoustic monitoring.

These technologies are not silver bullets. They augment, rather than replace, the fieldwork, community engagement and policy advocacy that form the foundation of effective conservation. What AI offers is the ability to operate at scales and speeds previously impossible -- processing vast data streams and extracting actionable intelligence from noise. For the world's most trafficked mammals, that capability may prove decisive.