Calibrated detection limits in routine Sentinel-1 products 50.6 hours before the Langtang Lirung rock–ice avalanche, Nepal
Abstract
On 26 August 2026 a rock–ice avalanche detached from Langtang Lirung, Nepal, and released a debris flood that crossed the Nepal–China border. Deformation preceding the failure has since been reported from a multi-month Sentinel-1 stack, and pre-event indications have been identified in commercial optical imagery. Both analyses began from a known failure location. This paper asks a different question: what routine, freely available Sentinel-1 products contained at the last acquisition before failure, and how large a signal would have had to be before they contained something. Three quantities are tested on a 500 m disc over the detachment — backscatter amplitude, interferometric coherence, and line-of-sight displacement from unwrapped phase — each against 120 control discs in the same scene matched on local incidence angle, aspect and slope, and against the source's own history over the preceding two months. The last usable descending acquisition precedes failure by 50.6 hours. No quantity distinguishes the source from comparable slopes or from its own quiet behaviour: amplitude at the 67.5th percentile of controls, coherence fourth of five quiet intervals, cumulative displacement at the 31st percentile. The same statistics identify the collapse decisively once it falls inside the observation interval, at z = +3.4 in coherence and below all 120 controls in amplitude, so these are calibrated limits rather than uninformative nulls. No disturbance decorrelating more than about seventeen per cent of the usable source disc occurred in the twelve days ending 50.6 hours before failure, and no line-of-sight displacement above approximately 13 mm per month occurred over the preceding seven weeks. That bound lies above the rate reported from the multi-month stack, which is the substantive result: the deformation was measurable from orbit using free archived data, but not by the twelve-day products a monitoring service would have been screening.
Related
- No detectable precursor in Sentinel-1 backscatter before the 26 August 2026 Langtang Lirung rock–ice avalanche, Nepal
- Very-high-resolution optical imagery reveals key clues to the Langtang Lirung ice–rock avalanche
- The 26 August 2026 Langtang Lirung ice–rock avalanche and its transboundary cascade: a multi-source observational reconstruction in the Bhote Koshi basin
- Bounded at both ends: the detection window of routine 12-day C-band InSAR at rock–ice avalanche sources
- Observation operator and detection limits for MODIS and VIIRS Fire Radiative Power products
BibTeX
@misc{sah2026calibrated,
title = {{Calibrated detection limits in routine Sentinel-1 products 50.6 hours before the Langtang Lirung rock–ice avalanche, Nepal}},
author = {Ajay Kumar Sah},
howpublished = {EarthArXiv},
year = {2026},
month = sep,
doi = {10.31223/x5kz3g},
url = {https://eartharxiv.org/repository/view/14902/},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Calibrated detection limits in routine Sentinel-1 products 50.6 hours before the Langtang Lirung rock–ice avalanche, Nepal
AU - Ajay Kumar Sah
PY - 2026
DA - 2026/09/09
PB - EarthArXiv
DO - 10.31223/x5kz3g
UR - https://eartharxiv.org/repository/view/14902/
AB - On 26 August 2026 a rock–ice avalanche detached from Langtang Lirung, Nepal, and released a debris flood that crossed the Nepal–China border. Deformation preceding the failure has since been reported from a multi-month Sentinel-1 stack, and pre-event indications have been identified in commercial optical imagery. Both analyses began from a known failure location. This paper asks a different question: what routine, freely available Sentinel-1 products contained at the last acquisition before failure, and how large a signal would have had to be before they contained something. Three quantities are tested on a 500 m disc over the detachment — backscatter amplitude, interferometric coherence, and line-of-sight displacement from unwrapped phase — each against 120 control discs in the same scene matched on local incidence angle, aspect and slope, and against the source's own history over the preceding two months. The last usable descending acquisition precedes failure by 50.6 hours. No quantity distinguishes the source from comparable slopes or from its own quiet behaviour: amplitude at the 67.5th percentile of controls, coherence fourth of five quiet intervals, cumulative displacement at the 31st percentile. The same statistics identify the collapse decisively once it falls inside the observation interval, at z = +3.4 in coherence and below all 120 controls in amplitude, so these are calibrated limits rather than uninformative nulls. No disturbance decorrelating more than about seventeen per cent of the usable source disc occurred in the twelve days ending 50.6 hours before failure, and no line-of-sight displacement above approximately 13 mm per month occurred over the preceding seven weeks. That bound lies above the rate reported from the multi-month stack, which is the substantive result: the deformation was measurable from orbit using free archived data, but not by the twelve-day products a monitoring service would have been screening.
ER -