Likavung Returns to the Lijia Mountains: A 55-Township Workflow Linking Indigenous Forest Knowledge, Camera Monitoring and RAG for Human-Bear Coexistence

On September 10, 2026, the Taitung Branch of Taiwan's Forestry and Nature Conservation Agency announced the release of an adult female Taiwan black bear known as Likavung. The bear had been rescued on May 27 after being caught in a snare in the shallow mountain area of Meinong, Beinan Township. After treatment, health checks and behavioral assessment, she was released into the Lijia mountain area.

Two details in the official account are especially useful for Indigenous public-service design. People from the Xiabinlang community took part in the original rescue, while the release site was selected after discussion with and support from the Lijia community. The case therefore represents more than wildlife rehabilitation. It shows a governance chain in which community knowledge, public agencies and professional wildlife teams each have a role.

Treat community collaboration as structured data

A first human-bear coexistence service map can begin with a small number of fields: date and time, township, approximate location, source of the report, species, incident type, risk level, response status, responsible agency, community contact, last human verification date, public-access level and original source URL.

Community consultation should also be recorded as part of the workflow. A release location can affect everyday movement, forest use and cultural context. Useful fields therefore include whether community consultation has been completed, who verified the decision, which information may be public and which locations require reduced precision.

The official Likavung release is available from the Taitung Branch of the Forestry and Nature Conservation Agency.

Camera traps and Indigenous tracking knowledge can complement each other

Taiwan already has a strong example of this two-way monitoring model. In June 2026, the Chiayi Branch reported that the Tsou Hunters Association had recorded Taiwan black bears for a fourth consecutive year using infrared automatic cameras placed along an established patrol route near the southern end of the Tefuye Trail. The agency noted that the images can support future analysis of activity hotspots and behavior. See the Forestry and Nature Conservation Agency report.

The practical lesson is to preserve different forms of evidence rather than forcing them into one score. Seasonal food sources, tracks, claw marks, animal trails and long-term observations by hunters can be one evidence layer. Time-stamped camera images can be another. Rescue, release and official coexistence-program records can be a third. RAG can connect them through a shared incident ID while preserving provenance.

Location needs access tiers

Wildlife digitization can easily lead to over-publication of precise coordinates. Exact den sites, travel routes, camera locations and places connected to traditional hunting knowledge may be sensitive. A better design stores precision where operationally necessary but returns different levels of detail to different users.

Public interfaces can show a township or coarse grid. Authorized conservation or local-government users can obtain greater precision. Community access should follow community rules. The CARE Principles and Local Contexts Labels provide useful models for describing collective benefit, authority, responsibilities, provenance and conditions of use.

Cameras may also capture people

Camera traps placed on trails can record hunters, gatherers, visitors or residents. Governance should therefore be established before deployment: purpose, area, retention period, access to raw images, public-display rules and procedures for blurring or deleting human images. An ethical framework for camera-trap research recommends permission, purpose limitation, disclosure and community awareness. See the camera-trap code of conduct.

If AI is used to classify large image collections, it should provide candidate labels for human review. Low light, occlusion and similar-looking species can produce errors. Image classification should support, rather than replace, the official record.

A chatbot can focus on the next correct action

The first chatbot MVP does not need to answer every ecological question. It can help a resident who reports a bear near a farm building find official safety guidance, the correct reporting channel and the most recent verified status. When safety is at issue, the system should route the user directly to a responsible human service.

The IUCN SSC Guidelines on Human-Wildlife Conflict and Coexistence emphasize that wildlife coexistence involves ecological, social, cultural and livelihood dimensions. For the 55 Indigenous township areas, that means a useful digital service should connect official guidance, local contacts, recent verified events and community-governed information.

A four-week MVP is enough to learn

Week one can define fields, roles and access tiers. Week two can import 20 to 50 historical reports or monitoring records and attach source links and verification dates. Week three can test simple search or RAG using natural-language queries such as "bear seen near a farm building" or "tracks found near the village." Week four can review privacy, location sensitivity, retrieval errors and escalation to human responders.

Success can be measured by whether users find the correct contact faster, whether sources remain traceable, whether sensitive locations respect access rules, and whether errors can be corrected quickly. Only after those basics work is it worth adding more GIS functions, automated image classification or notifications.

Put Two-Eyed Seeing into a maintainable workflow

Likavung's return to the Lijia mountains and the Tsou hunters' long-term monitoring show a practical path for Two-Eyed Seeing in an operational setting. Indigenous knowledge and technical monitoring can remain distinct in origin and responsibility while supporting the same conservation workflow.

Yuan Media AI can assist the 55 Indigenous township areas by starting with indexing and demonstration: organizing official black-bear reporting and conservation entry points, providing sample incident cards, building source-grounded RAG, applying access tiers to sensitive locations, and keeping permission fields for community images and forest knowledge. Field decisions, release decisions, rescue, safety and cultural rules remain with communities, responsible agencies and professional teams. This keeps digital tools useful for local work while preserving the context and authority attached to forest knowledge.