Where to look for a missing boat or person over the next three days.

Eight fishers from Konaseema district in Andhra Pradesh sailed from Odalarevu jetty at 10 a.m. on 3 August 2026. Phone contact ended an hour later, and the boat was due back on 12 August. By 18 August Coast Guard vessels were searching, and fishing boats from Kakinada to Paradip had been alerted (PTI, 18 August 2026).

At sea the area to search grows every hour. The current carries everything, and the wind then pushes each kind of object differently. In a 10 m/s wind (about 20 knots) the wind alone moves a person in the water about 4.5 nautical miles a day relative to the water, a small open boat about 15 and a life raft without ballast about 17, and each may veer to either side of the wind (Allen 2005). The data needed to work this out are public, but putting them together has needed specialist software and a specialist to run it.

• Takes what a first report gives: the last known position, typed as decimal degrees or as degrees and minutes from a GPS set, the time in IST, how uncertain both are, and what is adrift, from 13 object classes.
• Fetches surface currents (circulation, tides and wave drift) and 10 m winds for the area from Open-Meteo, straight from the browser. The request names a 0.25 degree grid around the area, never the position.
• Drifts 1,000 possible histories of the object for 72 hours, each with its own drift coefficients, its own side of the wind and its own forecast error, and stops those that reach the coast.
• Shows the cloud, a density map and the boxes that hold 50% and 90% of the particles, hour by hour, with corner positions in degrees and minutes and the share that reached the shore.
• Writes a two-part text message in the plain SMS alphabet and a full briefing, and names the likely rescue centre and the nearest sub-centres with numbers checked against the Coast Guard's own pages.
• Close case wipes the position, the results, the map and any copied text.

1. Read the reportLast known position and time, the uncertainty in each, and the object: a person in the water, a skiff, a fishing vessel, a life raft, a swamped boat or debris.
2. Fetch the forcingHourly currents from the Copernicus SMOC model at 1/12 degree and 10 m winds, for 74 hours, in two Open-Meteo requests.
3. DriftA Monte Carlo leeway model (Breivik and Allen 2008): 10 minute steps, coefficients from Allen (2005), 4% of particles switch sides each hour.
4. SummariseBoxes, ellipses, density and shoreline at every hour; the text message and briefing in IST, positions in degrees and minutes.
5. CheckThe same model runs in Python and in the browser. The test suite compares them particle by particle on both demo cases.

Equations, coefficients, parameters, verification and limits: methods page and methods note.

• Coast Guard rescue centres and sub-centres: a quick, independent estimate of where to look when a report comes in, to set beside their own planning.
• State fisheries departments and district control rooms: what to tell the rescue centre, and which stretch of coast to alert when particles reach the shore.
• Fishing cooperatives and boat owners: what a useful report contains (time, position, what is adrift) and a text with the boxes that can be passed on.
• Trainers: exercises with simulated cases on the day's modelled currents and winds.

• Positions stay on the device. The tool stores nothing, puts nothing in the address bar and sends nothing to Trinetra; its audit log records only that a case ran, with the object and settings.
• Close case clears the form, results, map, text and the clipboard.
• Advisory only. Every screen and every text names the Coast Guard rescue centre as the coordinator.
• The areas describe where people may need help. They are never used to enforce a maritime boundary.
• Demo cases are marked Simulated and do not represent the missing Konaseema boat.
• The method is published: equations, coefficients, parameters and known failure modes, with tests that anyone can run.

• Coastal currents are poorly resolved. The current model runs at about 9 km, so river plumes, estuaries, harbour mouths and the surf zone are missing.
• No leeway trials exist for Indian fishing craft. Mechanised boats and fibreglass boats with outboards use the nearest published classes, so their boxes are wider than the data alone would give.
• A boat is assumed adrift with no engine, anchor or sail after its last known time.
• The horizon is 72 hours. A boat reported days late, as in Konaseema, needs a new run from each sighting or debris find.
• The coastline is Natural Earth 1:10m; small islands, reefs and harbours may be missing.
• The free Open-Meteo API is for non-commercial use. An operational deployment needs a subscription or a self-hosted Open-Meteo on Indian soil.

We are looking for a Coast Guard district, a state fisheries department or a fishing cooperative to test Anvesha against drifter tracks and past search records, and to help measure drift for the boats Indian fishers use. contact@trinetra.in