Counterfactual Evaluation of a Satellite-Connected Multi-Sensor Early-Warning System During the 26 August 2026 Bhote Koshi–Trishuli Disaster

Basanta Paudyal

The 26 August 2026 Bhote Koshi–Trishuli disaster raised an important question: could a warning system continue to reach people if local infrastructure failed? This study tests a proposed system using satellite communication, a reviewed river network, 15 radar-gauge stations, 2 candidate geophone stations, a camera site and 32 mapped target references. The analysis consists of two segments. The first compares three possible routes from source references to Friendship Bridge at Gyirong port, with lengths of 21.764510–22.653727 km. The second uses a separately reviewed 80.614468 km river line from the Friendship-side reference to Galchhi bridge. Gyirong Port is a separate border-area site. The actual disaster source, upstream channel and arrival times remain unverified. The warning calculations therefore start when an operating gauge is assumed to detect the hazard, rather than when the disaster starts at origin. Of the 32 target points, 27 lie inside UNOSAT’s preliminary mapped mudflow/rockflow area. The other five points, Timure, Syabrubesi, Trishuli HPS powerhouse/tailrace (HP09), Battar Bazaar and Devighat, lie inside the area UNOSAT studied but outside its mapped affected area. Being inside or outside the mapped affected area does not confirm whether an individual site was damaged. We tested assumed propagation speeds of 2, 5 and 10 m/s, alert delays of 2, 10 and 30 min, and action times of 5, 15 and 30 min. In the downstream analysis, 696 of 864 tested combinations left enough calculated time for action, assuming working satellite communication and a channel that could reach recipients. This count shows how many tested scenarios gave enough time to act. It does not tell us the chance of success or whether warnings worked during the actual event. At Friendship Bridge and Rasuwagadhi, an earlier assumed trigger from RG-01 left enough calculated time in 14 of 27 combinations. RG-02 alone left no positive action margin in any of the 27 combinations. These results show how detector location, alert delay and delivery to recipients affect the model. Hydrographs, calibrated triggers, outage logs, verified recipient coverage and site-specific damage records are needed to assess whether the system could have worked during the event.