TDWG2026
Home / Talks / SAL A
talk · Tuesday 22 September · 11:15–⁠11:27 · SAL A

The WildLIVE Portal: Towards Scalable Wildlife Monitoring through the Integration of Citizen Science and Human-in-the-Loop AI Workflows

Rajapreethi Rajendran · AI for Biodiversity Data

Recording time 2:45:14–2:57:01Open on Vimeo ↗

Summaries and key moments are machine-produced. Check the recording for the speaker’s exact words.

The short versionWildLIVE combines a model registry, provenance-tracked human verification and FAIR publication to move camera trap data from raw images to publishable data.

Overview

What this was about

Rajapreethi Rajendran presented the WildLIVE portal, which supports the lifecycle of camera trap wildlife monitoring: data management (Camtrap DP import/export, a Wildlife Monitoring Ontology built on SOSA and Web Annotation), a model registry supporting internally and externally hosted models ('bring your own model'), taxonomic name resolution through Catalogue of Life's ChecklistBank API, and human-in-the-loop verification by citizen scientists and experts. The ontology distinguishes classification from verification with full provenance, and project-level confidence thresholds can auto-approve high-confidence annotations. A GeoEngine-based virtual research environment connects results to Earth observation data, and data are published as FAIR digital objects with RO-Crate and FAIR Signposting.

Why it matters. It addresses trust in AI annotations by preserving the machine annotation and appending verification history, while making outputs machine-actionable.

Key ideas

In the room

  • Model registry versions models; currently SpeciesNet and a custom jaguar model (Bolivia, South Africa).
  • Catalogue of Life ChecklistBank provides stable species identifiers across models with different class names.
  • Each annotation has one provenance object to which verification steps are appended.
  • Consensus/threshold setting at project or capture-event level auto-approves annotations above chosen confidence.
  • Virtual research environment by GeoEngine gives access to Sentinel-2, Copernicus land cover, burn data, etc.
  • Planned: active learning, separating MegaDetector for blanks/humans, IUCN Red List-driven location obscuring, direct GBIF mobilisation.
Jump in

Notable moments

In their words

Transcript

Automatically generated captions can contain mistakes, especially in names and technical terms. Times are relative to the room recording.

Read the transcript ↓
Loading transcript…
Keep exploring

Related talks.

See the map ↗
talk · 24 Sept · 14:15–⁠14:30 · SAL C

Camtrap DP: extending data standard designed for camera trapping research to support AI workflows and new data sources

Camtrap DP fits data from any stationary sensor, and converging community feedback shows how to extend it for AI provenance and multimodal monitoring.

Karolina Kuczkowska
talk · 22 Sept · 09:47–⁠09:59 · ODIN

Resilient Attribution Infrastructures for Biodiversity Science: Archiving and Risk Mitigation in Bionomia

Small niche infrastructures can be resilient by planning for their own end: keep things simple and push versioned data and reusable code outward.

David Shorthouse
talk · 22 Sept · 08:57–⁠09:09 · SAL A

Delivering AI for biodiversity data in the public sector: data handling, security and humans-in-the-loop when small teams build at pace

Small public-sector teams can use AI tools productively if they set clear boundaries between prototype, dev, test and production and keep expert humans in the loop.

Rachel Wiles
Programme abstract

The WildLIVE Portal: Towards Scalable Wildlife Monitoring through the Integration of Citizen Science and Human-in-the-Loop AI Workflows

Rajapreethi Rajendran

Global biodiversity monitoring delivers an ever-increasing volume of recorded data, with even medium-sized projects gathering data in the order of hundreds of thousands of images annually. These vast amounts of data frequently overwhelm the manual review capacity of domain experts and small research teams. As a result, the scale of these observations creates significant bottlenecks that hinder timely data annotation and analysis for conservation efforts. AI-generated baseline annotations can reduce this workload, but they still require expert review before they can be treated as reliable scientific evidence. The primary challenge lies in establishing trust and quantifiable confidence in these annotations, regardless of their source. To address this challenge, Senckenberg’s WildLIVE portal combines AI workflows with Human-in-the-Loop (HITL) refinement. The platform supports multi-model object detection, including open-vocabulary approaches in which detections can be guided by textual category descriptions rather than being limited to a fixed label set. Through an active-learning infrastructure, the platform prioritises observations most in need of review, directing citizen scientists and domain experts toward images where labelling effort can have the greatest impact on model confidence. These evaluations are preserved as layered annotations linking back to AI4WildLIVE’s FAIR Digital Object-based data model. By integration of an AI-assisted annotation pipeline, active learning, AI-readiness of data and citizen science engagement, the WildLIVE portal establishes a shared, human and machine operable collaborative framework through which biodiversity monitoring data becomes a trusted, community-wide scientific resource.

Submitted to the conference programme before the talk, so the talk itself may differ.