The problem
Guyana's 459 km coastline sits below sea level behind a mangrove and seawall defence, while offshore petroleum traffic through the Demerara and Essequibo approaches grows every year. Hazard reports from fishers and coastal villages arrive by phone call or word of mouth, are rarely geolocated, and almost never survive as defensible evidence. AquaGuard Sentinel closes that loop: intake from anyone with a phone, cryptographic sealing at the moment of capture, spatial and AI triage, and an auditable EPA workflow from report to remediation.
LCDS 2030 alignment
- Valuing forests & blue carbon: every incident is priced in tCO₂e at risk so mangrove loss enters national carbon accounts.
- Climate resilience: geofenced sanctuaries and drift projection protect the mangrove sea defence that shields coastal settlements.
- Clean energy transition safeguards: vessel eco-scoring and spill triage give the EPA oversight over expanding offshore activity.
- Inclusive digital state: SMS/USSD intake and an offline queue keep low-connectivity communities inside the reporting system.
Methodology
Ecological Vulnerability Score (EVS), 0–100
Baseline 50, +25 for chemical or oil categories, +20 for mangrove clearing, +20 for critical severity, +10 for high severity, capped at 100. Deterministic, so the same report always scores identically and the score is auditable.
Blue-carbon value at risk (USD)
incidents × 18.4 tCO₂e at risk per incident × US$645.38 per tCO₂e unit value. The per-incident tonnage is a conservative mangrove-stand estimate; the unit value mirrors Guyana's forest-carbon reference pricing. Illustrative, not an audited valuation.
Drift projection
Slick displacement is projected from a prevailing North Equatorial Current component and wind stress over the selected horizon, producing the 12-hour cone and impacted mangrove hectares in the simulator.
Evidence integrity
A SHA-256 digest of reporter, category, severity, narrative and timestamp is computed client-side before transmission. Database triggers reject any later change to that hash or to the reporter identity, so the seal proves the record has not been edited after capture.
What is genuinely live
- Postgres persistence with row-level security and role-based access control (admin / EPA officer / citizen)
- Realtime streaming of incidents, alerts and audit rows over a websocket channel with exponential-backoff reconnect
- SHA-256 evidence sealing computed in the browser and stored immutably; database triggers block tampering with reporter identity or hash
- Append-only audit trail written by database triggers on every status transition
- Offline intake queue in local storage with de-duplicated auto-sync on reconnect
- AI mangrove diagnostics, EcoAssist chat and evidence analysis calling a real Gemini model server-side, with heuristic fallbacks
- Alert rule engine evaluated in the database, plus CSV/JSON audit exports and KML drone tasking
- Real OpenStreetMap basemap with geofence polygons, incident heat/points and vessel markers
What is simulated (and why)
- Vessel AIS telemetry (positions, speed, fuel burn) is simulated; the ingest layer is shaped for a real AIS/VMS feed
- SAR radar, drift-cone and acoustic overlays are visual proxies for Copernicus Sentinel-1 and hydrophone feeds
- SMS/USSD intake is exercised through the in-app gateway console rather than a carrier short code
- Blue-carbon and EVS figures are illustrative, computed with the stated formulas below
Each simulated feed sits behind the same interface a production feed would use, so swapping in AIS, Sentinel-1 or a carrier gateway is a connector change, not a rewrite.