IsdaWatch: Safer seas for small fishers, sharper eyes on illegal fishing

Problem Statement

Fisherfolk remain one of the poorest sectors in the Philippines, with poverty incidence often reported at around a third of fishers. Small motorized bancas go out with little information on the weather. Every typhoon season, fishers go missing because gale warnings did not reach them in time or in a language they understand. At the same time, under the Fisheries Code (RA 8550 as amended by RA 10654), the 15-km municipal waters are reserved for small fishers, yet commercial vessels intrude, often at night with bright lights, and deplete stocks. BFAR and LGU Bantay Dagat teams have few patrol boats and cannot watch thousands of kilometers of coastline.

Proposed Solution

IsdaWatch has two sides. For fisherfolk: each morning, registered fishers receive an SMS or voice advisory in Cebuano, Waray, Tagalog, or another local language. It combines PAGASA gale warnings, wave and wind forecasts, and a “go / caution / stay ashore” safety rating with potential fishing zones estimated from sea surface temperature and chlorophyll satellite data. Fishers can text back “SOS” or their position. For BFAR/LGUs: an AI pipeline ingests VIIRS Day/Night Band boat detections (night lights from fishing vessels) and AIS/vessel tracks, overlays municipal water boundaries, and flags likely commercial vessels operating inside them. Each alert gets a confidence score, an estimated vessel size, and a recommended patrol route, so Bantay Dagat teams can act fast and build cases.

Target Market

• Primary users: Municipal fisherfolk registered in BFAR’s FishR (on the order of two million fishers) and fisherfolk associations.
• B2G buyers: BFAR, coastal LGUs and their Bantay Dagat task forces, the Philippine Coast Guard, and the NDRRMC/OCD for safety-at-sea.
• Other buyers: Marine conservation NGOs, protected-area management boards, and development partners (e.g., USAID Fish Right-type programs).
• Size: 800+ coastal municipalities and cities nationwide.

How AI Is Used

• Vessel detection classification: ML classifies VIIRS light detections (commercial light-boat vs. small banca vs. noise such as offshore platforms or moonlight cloud glare) using brightness, clustering, and persistence. Without it, analysts would face thousands of raw detections every night.
• Anomaly and behavior detection on AIS tracks (speed/turning patterns that indicate trawling or ring-netting, AIS gaps) flags suspicious behavior.
• Fishing-zone prediction: ML on SST, chlorophyll-a, and thermal fronts estimates productive zones, so fishers burn less fuel searching.
• Safety risk model turns forecast wind and wave data into simple risk ratings specific to small bancas.
• LLM and TTS write short advisories in local languages and voice notes for fishers who cannot read well.
• Patrol route optimization prioritizes intrusions by risk and distance from the Bantay Dagat port.

PLDT / Smart / ePLDT / Rokid Angle

• Smart SMS/broadcast and IVR are the delivery backbone in coastal barangays. They are a natural B2G messaging service and can be tied into emergency cell broadcast.
• Smart coastal coverage and satellite connectivity for patrol boats and fish ports. This creates connectivity demand in coastal GIDAs.
• ePLDT cloud hosts satellite ingestion, models, and the BFAR/LGU dashboard. Government data stays in the country.
• PLDT Enterprise B2G sells to BFAR and coastal LGUs as maritime domain awareness SaaS.
• Rokid AR glasses let Bantay Dagat officers on patrol see the flagged vessel’s bearing and the municipal water boundary overlaid at sea.

Demo Scenario (3 minutes for judges)

1. Story: Mang Elmer, a fisher from Guiuan, Eastern Samar, lost a friend at sea when a gale warning never reached their barangay.
2. A basic phone on stage buzzes: “IsdaWatch: Bukas, PELIGRO. Habagat, 2.5 m na balod. Ayaw paglawod” (Waray: “Tomorrow, DANGER. Southwest monsoon, 2.5 m waves. Do not go to sea.”). A voice note plays.
3. On a calm day, the SMS instead shows a “GO” rating and the nearest likely fishing zone, 8 km northeast.
4. Switch to the BFAR dashboard. A replay of last night’s VIIRS detections shows a bright cluster inside municipal waters, which the AI marks as a “likely commercial light-fishing vessel, 91% confidence.”
5. Click “Dispatch.” A patrol route is generated and the alert goes to the Bantay Dagat team leader by SMS.
6. Close: “Safer bancas, fuller nets, emptier seas for illegal fishers.”

30-Hour Build Plan

Suggested stack: Python, GeoPandas, xarray, scikit-learn/LightGBM, OR-Tools, FastAPI, React + Leaflet/Mapbox, Gemini or an open LLM, open TTS, SMS gateway mock. Public data / APIs: NOAA/EOG VIIRS Boat Detection (VBD), Global Fishing Watch API (AIS-based fishing activity, research/non-commercial terms), Copernicus Marine Service (SST, chlorophyll, waves), NOAA ERDDAP, Open-Meteo Marine API, PAGASA gale warnings and forecasts, NAMRIA/BFAR municipal water boundaries where available, OpenStreetMap coastlines, PSA fisheries statistics.

Risks & Privacy Notes

• False positives: VIIRS cannot identify vessels, and lights may be legal or come from non-fishing sources. Present outputs as leads for verification by patrols, never as proof, and combine them with AIS and persistence.
• Boundary accuracy: official municipal water delineations are incomplete for some LGUs. Mark boundary confidence and use approximate 15-km buffers only for the demo.
• Safety liability: advisories always defer to official PAGASA warnings and say so explicitly.
• Data Privacy Act of 2012: fisher phone numbers and FishR data are collected only with consent and used only for advisories. SOS locations are shared only with responders. Vessel data is public AIS and satellite data.
• Clause 24 compliance: no FishR or personal data goes into third-party AI tools. Data terms (Global Fishing Watch, Copernicus) are respected and credited, and AI use is disclosed.

AI Use Disclosure (draft)

We use scikit-learn/LightGBM for vessel detection classification and fishing-zone prediction, and Gemini/an open LLM with TTS for dialect advisories. Global Fishing Watch’s published AI-derived fishing activity data is used as an input and credited. ChatGPT and GitHub Copilot helped with code scaffolding and the pitch draft. All alerts, translations, and code were reviewed by team members.

Rubric Scoring