What free satellite images can and cannot show before Himalayan rock–ice collapses: four test sites and a screen of twelve Nepal corridors

Rock–ice collapses at Chamoli (2021), Langtang Lirung (2026) and Budhi Gandaki (2026) have raised the question of whether satellite images can show which Himalayan slope will fail next. I test what free Sentinel-2 images can and cannot show, in two parts. First, I measure the three sources in the years and months before failure with image pairs whose spacing suits each surface. The Chamoli block moved at 2–13 m yr⁻¹, fastest in 2017–2018, with no final speed-up. The Budhi Gandaki glacier flowed at 80–164 m yr⁻¹ and was slower in its last year than before. The Langtang outlet rose 2.4× from spring to June 2026, the largest of five years but not clearly abnormal (ordinary years 1.1–2.0×). None of the free optical records shows a distinctive precursor. Second, I build a blind multi-year creep map from autumn image pairs and test it at four collapse sources. It flags the Chamoli source, cannot measure the fast-ice source at Budhi Gandaki, and finds the Langtang failed slope and the slopes around the 2024 Thame outburst lakes quiet. I then run it over twelve corridors in Nepal, from Tamor to West Seti. The share of steep slopes it can measure falls from 99 % at 4,000 m to about 35 % at 5,500–5,750 m and 10–17 % above 6,000 m; the four test sources span about 4,300–6,600 m, across this decline. It flags 94 clusters inside Nepal, mostly steady creep of north-facing steep ice. A sun-angle test, and a repeat with sunlit spring images that confirmed 17 of 20 tested candidates, show that most flags are motion, not shadow. The Chamoli source looked like the many slopes that have not failed. A free-data screen therefore delivers a coverage map and a list of candidates, not a warning. Any null result should state its detection limit and its coverage at source height. Code and data are open, and one script reproduces the main numbers.