How Guarayú Communities Restored Paúros: A Water-Resilience Map Taiwan's 55 Indigenous Areas Can Adapt

The Guarayú experience in Bolivia is useful to Taiwan's 55 Indigenous areas not because it asks communities to return to an old system, but because it shows how traditional water knowledge, youth participation, scientific tools and local governance can be connected into a resilience system that communities can maintain themselves.

Mongabay reported in August 2026 that Guarayú people in Ascensión de Guarayos in the Bolivian Amazon have long used natural spring-fed pools and groundwater outlets known as paúros. Before the severe 2022 drought, local youth, students and teachers began locating, cleaning and restoring water sources that had gradually fallen out of routine use. By June 2022, eight paúros had been restored. Months later, when the main municipal reservoir dried up, residents returned to the paúros for water. The case is documented by Mongabay.

The importance of the paúros lies in more than the physical pools. Guarayú knowledge connects water, forest vegetation, inherited experience, community maintenance responsibilities and cultural meaning. Different water sites can carry different care practices. Before restoration work begins, local knowledge holders can explain which vegetation may be cleared, which should remain and what forms of use are appropriate. A digital system therefore needs fields for interpretation authority, seasonality, disclosure limits and maintenance responsibility alongside coordinates and water observations.

From youth water-source restoration to local protection

The paúros revival did not end as a one-time environmental activity. Fundación Socioambiental Semilla describes how youth groups later worked with neighborhoods, schools, environmental specialists, legal professionals and local government to connect water sources through the Yande I Yar urban ecological corridor. Local protection rules were approved in 2025, and the work expanded toward interpretation trails and locally organized guiding. See Fundación Socioambiental Semilla.

This pathway links traditional knowledge, nature-based solutions, youth work and local institutions rather than treating them as separate projects. It also creates a practical maintenance structure: people know who checks a site, who explains its cultural context and how information can be updated after a drought or heavy-rain event.

International development programs are also combining decentralized source protection with modern infrastructure. The Inter-American Development Bank's WATERSHARED Accelerator in Bolivia aims to improve rural water access while combining infrastructure, water-source conservation agreements and community operation and maintenance capacity. The useful lesson is a hybrid approach in which natural sources, engineering and community stewardship support one another.

Start with a water-resilience service map for the 55 Indigenous areas

A first version does not need a large new platform. One township or community can begin with 10 to 20 locally approved nodes. Nodes might include public reservoirs, rainwater systems, springs, stream intake points, irrigation channels, storage tanks, community care-facility backup water and agricultural water sources where disclosure has been approved.

Each node can record its type, main use, public location precision, seasonal status, maintainer, last human verification date, alternative source, official monitoring link, next action during an abnormal event, human contact and cultural-use conditions. The goal is not to expose every water point; it is to make maintenance and routing clearer.

Location precision should be tiered. Ordinary public facilities may display exact locations. Sources that are important for community life, ecology or culture can be shown only at village, watershed or coarse-grid level. Ceremonial, family-governed or otherwise restricted knowledge can remain as internal codes with no public coordinates.

For traditional-knowledge RAG, permissions should also be separated: public display, RAG retrieval, research, model training, generation, external sharing and later withdrawal are different uses. The CARE Principles and Local Contexts are useful references for authority to control, responsibility, ethics, provenance and community-specific conditions.

Use official water data as background signals and preserve local verification

Taiwan already has strong machine-readable water information. The Water Resources Agency Water Resources IoT portal provides rainfall, river level, urban flooding, groundwater, flow and related sensor data. The Central Weather Administration Open Data Platform provides weather, rainfall and warning information. These can form an official background layer for a local water-resilience map.

The AI layer does not need authority to declare a source safe or potable. A more reliable workflow is to use official rainfall, water-level or drought signals to create a pending-verification list. Local maintainers can then report whether a spring has weakened, a channel is blocked, an agricultural area lacks water or an access route remains usable.

A Chatbot response can display both the official data timestamp and the last human verification date. That distinction helps readers understand which information is a sensor reading and which is a community field observation. If the observation is stale, the service can say that it needs confirmation rather than filling the gap from model memory.

Let RAG find evidence, explain sources and request updates

A local water Chatbot can begin with practical questions: Which public water or backup facilities are available in this area? When was the last field inspection? Which official rainfall station is closest? Who is the responsible contact for an irrigation node? If a record is outdated, the system can flag it for review.

The RAG data model can separate nodes, observations, sources and permissions. High-frequency sensor values belong in observation records; stable names and uses belong in node records; government APIs, field interviews and documents belong in source records; location precision and cultural conditions belong in permission records. This prevents rapidly changing measurements from repeatedly duplicating cultural descriptions.

UNESCO has long encouraged participatory ways to connect Indigenous and local knowledge with scientific information for drought, flood and climate-risk management. Its Indigenous Knowledge for Climate Risk Management work is a useful reference for keeping local participation central while bringing scientific information into the same decision environment.

Extend the model with moisture tracking, satellite data and local observation

In May 2026, UN-Water described SIWI and PIK work on Indigenous Lands Moisture Tracking, which uses advanced atmospheric moisture-tracking models to estimate how Indigenous lands contribute to rainfall generation and cross-regional moisture transport. The project highlights the role such lands can play in stabilizing rainfall during dry periods. See UN-Water.

This creates a useful Two-Eyed Seeing research direction. A Two-Eyed Seeing Lab could place long-term community observations—such as when a spring weakens, which plant signals precede water stress or how many days after rain a source recovers—on the same timeline as official rainfall, water levels, satellite vegetation indicators and watershed data. AI can help organize correlations and event cards, while interpretation remains with knowledge holders and field specialists.

The value of this approach is not to turn cultural observations into anonymous model features. It is to make different forms of evidence visible together, preserve their provenance and allow communities to decide which relationships are appropriate to investigate or share.

A 90-day MVP that one community can actually use

During the first 30 days, one community or small watershed can identify 10 to 20 water-related nodes that may be publicly or conditionally indexed, along with maintainers and cultural permissions. During days 31 to 60, the pilot can connect one or two official APIs and add a simple map, status field and RAG query interface.

During days 61 to 90, the team can test four situations: drought, heavy rain, water interruption and irrigation demand. The test should check whether users can find the right data, see its date, reach a person when needed and confirm that sensitive locations remain protected.

Evaluation should not focus on the number of AI answers. More useful measures include the share of records with traceable sources, the proportion of stale records, the completion rate for human field checks, the time needed to find the correct contact, whether sensitive information is correctly classified and whether each event leaves a record that improves the next season's preparation.

The most transferable message from the Guarayú paúros experience is that traditional knowledge and new technology can work side by side when communities retain authority over meaning, access and final decisions. For Taiwan's 55 Indigenous areas, a small water-resilience service map can connect elder knowledge, youth field checks, agricultural needs, official water data, GIS, RAG and Chatbots into a practical public-service tool for both drought and heavy-rain conditions.