AquaSentinel: One Health Water Intelligence

AquaSentinel turns citizen observations and USGS hydrology into explainable early warnings, digital water twins, and human-verified resilience actions for urban waterways.

Comment

Inspiration

🎯 Track Selection: Track 6 — Resilience Informatics
(Focus: Early Warning & Resilience Planning through Predictive Environmental Tools)

Urban freshwater ecosystems are the lifeblood of biodiversity and human communities, yet monitoring them remains fundamentally reactive. By the time municipal authorities or public health teams detect contamination through manual grab sampling, toxic runoff and chemical plumes have already traveled downstream, polluted drinking water intakes, and damaged delicate aquatic habitats.

Aligned with the IEEE OneAquaHealth Global Hackathon, we built AquaSentinel specifically to address the core problem of Track 6: Resilience Informatics—the critical lack of predictive environmental tools for freshwater ecosystems. Our inspiration was to bridge the dangerous gap between citizen observations, hydrological telemetry, and municipal decision-making. We moved beyond passive historical charts to an active, predictive early-warning platform that gives communities and resilience planners hours of advance warning before water contamination harms ecosystems and human health.

What it does

AquaSentinel is an end-to-end Environmental Intelligence and Early Warning System that turns environmental telemetry and citizen reports into verified, human-governed resilience actions:

  • Multi-Source Ingestion & Telemetry Fusion: Continuously streams real-time physical telemetry from the USGS National Water Information System (NWIS)—monitoring turbidity, dissolved oxygen, specific conductance, temperature, pH, and discharge rates—fused with Open-Meteo precipitation data and geo-located citizen field reports.
  • Rolling Adaptive $z$-Score Anomaly Engine: Replaces crude static limits with dynamic moving statistical windows. The system calculates real-time deviations: $$\pm 2.0\sigma \implies \text{Warning Threshold}$$ $$\pm 2.5\sigma \implies \text{Critical Anomaly}$$ This reliably detects rapid flash pollution without generating seasonal false alarms.
  • 6-Stage One Health Transmission Chain: Tracks environmental risks across the complete ecological continuum: $$\text{Water Physico-Chemistry} \longrightarrow \text{Aquatic Ecology} \longrightarrow \text{Animal Exposure} \longrightarrow \text{Human Health} \longrightarrow \text{Risk Propagation} \longrightarrow \text{Resilience Response}$$ Each stage provides real-time biomarkers including cyanotoxin risk, benthic macroinvertebrate stress, enteric infection probabilities, and intake vulnerability indices.
  • Explainable AI (XAI) Additive Evidence Ledger: Instead of an opaque "black-box" score, AquaSentinel calculates an auditable, additive evidence ledger:
    • +31 Turbidity Anomaly (sustained $\ge 2.5\sigma$ deviation)
    • +21 Rainfall-Runoff Correlation (preceding 42mm precipitation)
    • +17 Citizen Field Observation Corroboration (unusual odor, foam, and discoloration)
    • +13 Historical Baseline Deviation (7-day rolling excursion)
    • Composite Risk Score: 82/100 (Critical) Integrated Google Gemini 1.5 Flash provides natural language hydrological interpretation with deterministic algorithmic fallbacks.
  • Digital Water Twin with 4.5-Hour Advance Lead Time: Simulates downstream hydrodynamic plume dispersion and travel velocity, giving municipal authorities a proven 4.5-hour advance early-warning window before peak contaminants reach municipal water intake infrastructure.
  • Human-in-the-Loop Governance & Multi-Channel Alerts: Keeps human operators in command. Municipal officers review triage evidence, verify risks, and trigger automated emergency dispatches (Telegram Bot API, SMS, and Email). Every action is sealed with SHA-256 cryptographic audit logs for non-repudiation.
  • Clinical HL7® FHIR® R4 Interoperability: Implements the international digital health standard, automatically exporting stream anomalies into canonical Observation and RiskAssessment JSON bundles with standardized LOINC codes (e.g. 82810-3, 37798-6) and UCUM units, validated against public HAPI FHIR R4 servers.

How we built it

  • Geospatial & Interface Architecture: Engineered with Next.js 14, React 18, TypeScript, and Tailwind CSS. Built an interactive watershed command console utilizing Leaflet GIS with live plume dispersion contours, monitoring stations, and municipal intake proximity rings.
  • Hydrological Anomaly Pipeline: High-performance TypeScript analytical engine computing rolling moving averages, dynamic variance thresholds, and cross-sensor correlation algorithms.
  • Hydrological Intelligence: Powered by Google Gemini 1.5 Flash running strictly server-side for ecological risk analysis, backed by instantaneous mathematical deterministic engines.
  • Interoperability Standard: Native HL7 FHIR R4 schema serializer mapping environmental observations to clinical electronic health record (EHR) ontologies, complete with a live top-level JSON hierarchy inspector.
  • Alerting & Auditing Stack: Direct webhook integrations with Telegram Bot API, Twilio SMS, and SMTP email services, with cryptographic SHA-256 state hashing.
  • Production Security: Hardened backend deployed on Render and Vercel with zero client-side credential exposure.

Challenges we ran into

  • Corroborating Noisy Citizen Science Data: Volunteer field reports can be geographically clustered or qualitative. We solved this by developing a spatial-temporal credibility weighting algorithm that validates citizen reports against upstream and downstream USGS gauge sensors.
  • Distinguishing Natural Runoff from Severe Contamination: Normal heavy rain causes natural spikes in turbidity. We tackled this by correlating the rate of precipitation with the rate-of-rise in turbidity and dissolved oxygen depletion, differentiating benign silt runoff from hazardous industrial and agricultural flushes.
  • Bridging Environmental Science to Clinical Health Standards: Translating hydrological sensor feeds into medical-grade HL7 FHIR R4 schemas was challenging. We meticulously cross-referenced USGS parameters with LOINC ontologies and UCUM unit definitions to achieve valid HAPI FHIR compliance.

Accomplishments that we're proud of

  • 4.5-Hour Predictive Lead Time: Validated an early-warning window of 4.5 hours on real-world USGS storm-runoff benchmarks (Potomac River basin event).
  • 100% Explainable AI: Eliminated the "black box." Every single municipal alert displays the exact additive mathematical evidence ledger (+31, +21, +17, +13) behind its score.
  • True One Health Interoperability: Created one of the first working pipelines that directly converts river sensor anomalies into clinical HL7 FHIR R4 resources ready for public health epidemiological systems.
  • Production-Grade Implementation: Delivered a fully responsive, deployed application with real USGS telemetry ingestion, interactive GIS mapping, and active multi-channel emergency alerting.

What we learned

  • Citizen science becomes exponentially more impactful when algorithmically fused with physical sensor telemetry rather than analyzed as an isolated dataset.
  • For municipal resilience planning, algorithmic transparency is everything: public health authorities will only act on early warnings if the AI clearly justifies its reasoning.
  • Standardized data formats (HL7 FHIR) provide the vital bridge needed to connect environmental ecology with public health infrastructure, truly realizing the One Health vision.

What's next for AquaSentinel: One Health Water Intelligence

  • European Hydrological Network Integration: Expand API connectors to ingest European water monitoring portals and directly interface with the OneAquaHealth Open Information Hub.
  • Low-Cost LoRaWAN Stream Nodes: Deploy affordable, solar-powered community edge sensor nodes for unmonitored local urban streams.
  • Predictive Pathogen & Vector Modeling: Incorporate temperature-stratified hydrological models to forecast vector-borne water pathogen incubation periods ahead of seasonal heatwaves.

Built With

Share this project:

Updates

Submission history