Researchers use deep transfer learning to study nest site fidelity in painted stork in Delhi zoo
A study published on March 1, 2026 in the journal Royal Society Open Science has demonstrated that deep transfer learning — a form of artificial intelligence — can accurately identify individual painted storks (Mycteria leucocephala) from photographs and confirm that specific birds return to the same nesting spots across breeding seasons, a behaviour known as nest site fidelity.
The research was conducted by scientists affiliated with the Salim Ali Centre for Ornithology and Natural History (SACON), a premier avian research institution under the Wildlife Institute of India (WII), located in Coimbatore, Tamil Nadu.
Traditional individual bird identification requires physical tagging, banding, or colour-ringing — invasive and labour-intensive methods. Deep transfer learning allows researchers to train AI models on annotated bird photographs and then identify individuals non-invasively at scale.
The study used pre-trained convolutional neural network (CNN) models, fine-tuned on painted stork images — a technique called transfer learning, where knowledge gained from training on one large dataset (such as ImageNet) is "transferred" to improve performance on a specialised smaller dataset.
Establishing nest site fidelity behaviour in colonial waterbirds is important for conservation planning: it helps determine the significance of specific nesting colonies and informs decisions on protected area boundaries and disturbance management.
This study is part of a growing global trend of applying AI and machine learning to wildlife monitoring and conservation — reducing the cost, risk, and ecological disturbance associated with traditional field methods.
Deep Transfer Learning — Concept and Conservation Applications
Transfer learning is a machine learning technique in which a model trained for one task is repurposed for a related but different task. Deep transfer learning specifically refers to this approach using deep neural networks — multi-layered architectures that can learn complex, abstract representations of visual, acoustic, or numerical data.
Key Details
- In wildlife research, the AI model is typically pre-trained on ImageNet (a dataset of over 14 million labelled images of general objects) and then fine-tuned on a domain-specific dataset of wildlife photographs — requiring far fewer labelled training images than training from scratch.
- Convolutional neural networks (CNNs) — such as ResNet, VGG, or EfficientNet — are particularly effective for image-based tasks; they learn hierarchical visual features (edges → textures → shapes → individual patterns) that enable species and individual identification.
- Beyond individual ID, AI is being applied to: acoustic monitoring of bird calls (BirdNET, Cornell Lab), radar-based detection of nocturnal migration, thermal camera analysis for nocturnal wildlife surveys, and drone-based population counts.
- India is developing AI applications for wildlife monitoring: WII uses satellite tracking and camera-trap analytics for tiger and elephant surveys; SACON focuses on avian research.
- The Technology Development Board (TDB) and Department of Science and Technology (DST) fund AI-for-biodiversity research under the National Science & Technology Entrepreneurship Development Board schemes.
Deep transfer learning enables researchers to track individual birds across seasons non-invasively — a capability that was practically impossible at scale before AI, opening new frontiers in behavioural ecology and conservation planning.
Painted Stork (Mycteria leucocephala) — Ecology and Conservation Status
The painted stork is a large waterbird belonging to the family Ciconiidae (storks). It is a characteristic resident of the Indian subcontinent, found across South and Southeast Asia, typically associated with wetlands, shallow lakes, and flooded agricultural fields.
Key Details
- IUCN Red List status: Near Threatened (population decreasing).
- Total estimated population: 25,000–35,000 individuals; predominantly in South Asia (India, Sri Lanka, Nepal, Bangladesh) and parts of Southeast Asia.
- Painted storks are colonial breeders: they nest in large mixed-species heronries (called rookeries) along with cormorants, herons, egrets, and other storks, typically in trees near water bodies.
- Nest site fidelity — the tendency to return to the same nesting colony or tree year after year — is a well-documented behaviour in many colonial birds; it reduces the cost of finding safe sites and leverages accumulated local knowledge.
- Key nesting sites in India: Koonthankulam Bird Sanctuary (Tamil Nadu), Sultanpur National Park (Haryana), Keoladeo Ghana National Park (Bharatpur, Rajasthan — a UNESCO World Heritage Site and Ramsar wetland).
- Threats include wetland drainage and conversion, disturbance from human activity at nesting colonies, pesticide accumulation in fish (their primary prey), and loss of nesting trees.
The AI-based study targeted painted storks specifically because their colonial nesting behaviour and declining population status make accurate individual tracking a priority for evidence-based conservation management in India's wetland ecosystems.
AI in Biodiversity Monitoring — Global and Indian Context
Biodiversity monitoring — systematically tracking the status and trends of species, ecosystems, and genetic diversity — is a core obligation under the Convention on Biological Diversity (CBD). AI-powered tools are transforming the scale and cost-efficiency of monitoring.
Key Details
- The CBD's Kunming-Montreal Global Biodiversity Framework (GBF, 2022) includes Target 21 — ensuring availability of quality data for biodiversity monitoring and decision-making, and explicitly calls for use of digital sequence information (DSI) and technology for biodiversity assessment.
- India's National Biodiversity Mission and the Biodiversity Heritage Sites programme (34 sites designated as of 2024) require ongoing monitoring that benefits from AI-assisted species identification.
- GBIF (Global Biodiversity Information Facility) — an open-data platform to which India contributes records — hosts millions of verified species occurrence data points, many now verified using AI tools like iNaturalist's computer vision.
- In India, the eBird platform (Cornell Lab) and iNaturalist have enabled citizen science-driven data collection; AI back-ends now classify bird species from uploaded photographs with high accuracy.
- CAMPA (Compensatory Afforestation Fund Management and Planning Authority) funds are increasingly directed toward technology-assisted monitoring of compensatory planted forests.
The painted stork study is an example of how Indian researchers are deploying state-of-the-art AI to address specific conservation questions — aligning with global biodiversity monitoring obligations under the CBD and India's National Biodiversity Action Plan.
- Study: "Individual identification and confirmation of nest site fidelity in painted stork (Mycteria leucocephala) using deep transfer learning," published March 1, 2026 in Royal Society Open Science.
- Research institution: Salim Ali Centre for Ornithology and Natural History (SACON), Coimbatore, Tamil Nadu (under Wildlife Institute of India, WII).
- Species studied: Painted stork (Mycteria leucocephala); IUCN status: Near Threatened; population 25,000–35,000.
- Method: Deep transfer learning using pre-trained convolutional neural networks (CNNs) fine-tuned on painted stork images.
- Key behaviour confirmed: Nest site fidelity — individual birds return to the same nesting spots across breeding seasons.
- Painted stork family: Ciconiidae; predominantly non-migratory; makes local and seasonal movements.
- Breeding season in India: August–November (north India), October–April (south India).
- Keoladeo Ghana National Park (Bharatpur): UNESCO World Heritage Site and Ramsar wetland; a major painted stork nesting colony.
- CBD Kunming-Montreal GBF (2022) Target 21: Calls for quality biodiversity data using digital and technological tools.
- GBIF: Global open-access biodiversity data platform; India is an active contributor.