Strandline

Strandline is an educational prototype that tells flood responders where a Himalayan flood reached, what it broke, and who it cut off, using only free satellite data. Built for the 26 August 2026 glacier lake outburst flood in Nepal’s Bhote Koshi–Trishuli valley, it takes any area and date. It detects flooding with Sentinel-1 radar, which sees through monsoon cloud, and refines the result with Copernicus elevation data, while marking terrain the radar cannot see as “not analysed” instead of guessing. Damaged buildings, cut roads and bridges come from a pre-event OpenStreetMap snapshot, and road-network routing with Monte Carlo uncertainty estimates which settlements are stranded. An AI copilot answers questions in English and Nepali without ever writing a number itself: code fills every figure from the evidence, and a verifier rejects anything untraceable. It scored 453/453 with zero invented numbers. A 3D dashboard and a one-page bilingual situation report deliver the results. We report every result honestly: 0.48% false flood on no-flood valleys, flood F1 of 0.53 on dev events, and 0.88 precision on a sealed, pre-registered test against Copernicus EMS. A Kuro Siwo deep-learning experiment (0.41) did not beat physics (0.45) on the same gorge floods, so physics remains the shipped default.